1 2018-02-06 Richard Biener <rguenther@suse.de>
3 * config/i386/i386.c (print_reg): Fix typo.
4 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
6 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
8 * configure: Regenerate.
10 2018-02-05 Martin Sebor <msebor@redhat.com>
12 PR tree-optimization/83369
13 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
16 2018-02-05 Martin Liska <mliska@suse.cz>
18 * doc/invoke.texi: Cherry-pick upstream r323995.
20 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
22 * ira.c (ira_init_register_move_cost): Adjust comment.
24 2018-02-05 Martin Liska <mliska@suse.cz>
27 * doc/gcov.texi: Fix typo in documentation.
29 2018-02-05 Martin Liska <mliska@suse.cz>
32 * doc/gcov.texi: Document necessity of --dynamic-list-data when
33 using dlopen functionality.
35 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
37 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
38 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
39 _mm_maskz_range_ss, _mm_mask_range_round_ss,
40 _mm_maskz_range_round_ss): New intrinsics.
41 (__builtin_ia32_rangesd128_round)
42 (__builtin_ia32_rangess128_round): Remove.
43 (__builtin_ia32_rangesd128_mask_round,
44 __builtin_ia32_rangess128_mask_round): New builtins.
45 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
46 __builtin_ia32_rangess128_round): Remove.
47 (__builtin_ia32_rangesd128_mask_round,
48 __builtin_ia32_rangess128_mask_round): New builtins.
49 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
50 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
51 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
52 "<round_saeonly_constraint>")): Changed to ...
53 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
54 "<round_saeonly_scalar_constraint>")): ... this.
55 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
56 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
57 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
58 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
59 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
61 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
63 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
65 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
66 Remove all values except native, 8540 and 8548.
68 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
70 * config/i386/i386.c (ix86_output_function_return): Pass
71 INVALID_REGNUM, instead of -1, as invalid register number to
72 indirect_thunk_name and output_indirect_thunk.
74 2018-02-02 Julia Koval <julia.koval@intel.com>
76 * config.gcc: Add -march=icelake.
77 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
78 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
79 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
80 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
81 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
82 (processor_target_table): Add icelake.
83 (ix86_option_override_internal): Handle new PTAs.
84 (get_builtin_code_for_version): Handle icelake.
85 (M_INTEL_COREI7_ICELAKE): New.
86 (fold_builtin_cpu): Handle icelake.
87 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
88 * doc/invoke.texi: Add -march=icelake.
90 2018-02-02 Julia Koval <julia.koval@intel.com>
92 * config/i386/i386.c (ix86_option_override_internal): Change flags type
94 * wide-int-bitmask.h: New.
96 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
99 * config/i386/i386.md: Replace Pmode with word_mode in
100 builtin_setjmp_setup and builtin_longjmp to support x32.
102 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
106 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
108 (rs6000_supported_cpu_names): New static variable.
109 (linux_cpu_translation_table): Likewise.
110 (elf_platform) <cpu>: Define new static variable and use it.
111 Translate kernel AT_PLATFORM name to canonical name if needed.
112 Error if platform name is unknown.
114 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
117 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
119 2018-02-01 Jeff Law <law@redhat.com>
122 * config/i386/i386.c (release_scratch_register_on_entry): Add new
123 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
124 the scratch if RELEASE_VIA_POP is false.
125 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
126 If we have to save a temporary register, decrement SIZE appropriately.
127 Pass new arguments to release_scratch_register_on_entry.
128 (ix86_adjust_stack_and_probe): Likewise.
129 (ix86_emit_probe_stack_range): Pass new arguments to
130 release_scratch_register_on_entry.
132 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
134 PR rtl-optimization/84157
135 * combine.c (change_zero_ext): Use REG_P predicate in
136 front of HARD_REGISTER_P predicate.
138 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
140 * config/avr/avr.c (avr_option_override): Move disabling of
141 -fdelete-null-pointer-checks to...
142 * common/config/avr/avr-common.c (avr_option_optimization_table):
145 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
147 PR tree-optimization/81635
148 * tree-data-ref.c (split_constant_offset_1): For types that
149 wrap on overflow, try to use range info to prove that wrapping
152 2018-02-01 Renlin Li <renlin.li@arm.com>
155 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
157 (aarch64_register_move_cost): Likewise.
158 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
160 (REG_CLASS_NAMES): Likewise.
161 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
162 TAILCALL_ADDR_REGS. Remove IP registers.
163 * config/aarch64/aarch64.md (Ucs): Update register constraint.
165 2018-02-01 Richard Biener <rguenther@suse.de>
167 * domwalk.h (dom_walker::dom_walker): Add additional constructor
168 for specifying RPO order and allow NULL for that.
169 * domwalk.c (dom_walker::dom_walker): Likewise.
170 (dom_walker::walk): Handle NULL RPO order.
171 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
173 (rewrite_update_dom_walker): Likewise.
174 (mark_def_dom_walker): Likewise.
176 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
178 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
179 (aarch64_maybe_expand_sve_subreg_move): Declare.
180 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
181 * config/aarch64/predicates.md (aarch64_any_register_operand): New
183 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
184 that are semantically a reverse operation.
185 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
186 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
187 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
189 (aarch64_can_change_mode_class): For big-endian, forbid changes
190 between two SVE modes if they have different element sizes.
192 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
194 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
195 the TImode handling for big-endian targets.
197 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
199 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
200 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
202 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
203 Remove BSWAP handing for big-endian targets and use the form of
204 LD1RQ appropariate for the mode.
206 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
208 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
209 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
212 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
215 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
216 check for operands that need to go through aarch64_sve_reload_be.
218 2018-02-01 Jakub Jelinek <jakub@redhat.com>
220 PR tree-optimization/81661
221 PR tree-optimization/84117
222 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
223 * tree-eh.c: Include gimplify.h.
224 (find_trapping_overflow, replace_trapping_overflow,
225 rewrite_to_non_trapping_overflow): New functions.
226 * tree-vect-loop.c: Include tree-eh.h.
227 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
228 * tree-data-ref.c: Include tree-eh.h.
229 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
231 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
233 PR rtl-optimization/84123
234 * combine.c (change_zero_ext): Check if hard register satisfies
235 can_change_dest_mode before calling gen_lowpart_SUBREG.
237 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
240 * ira.c (ira_init_register_move_cost): Remove assert.
242 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
244 PR rtl-optimization/84071
245 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
246 * doc/tm.texi: Regenerate.
248 2018-01-31 Richard Biener <rguenther@suse.de>
250 PR tree-optimization/84132
251 * tree-data-ref.c (analyze_miv_subscript): Properly
252 check whether evolution_function_is_affine_multivariate_p
253 before calling gcd_of_steps_may_divide_p.
255 2018-01-31 Julia Koval <julia.koval@intel.com>
258 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
259 * config/i386/i386.md (rdpid_rex64) New.
260 (rdpid): Make 32bit only.
262 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
265 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
266 an IDENTIFIER_NODE for FUNCTION_TYPE's.
268 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
271 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
273 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
275 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
277 PR rtl-optimization/84071
278 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
279 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
281 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
283 * config/arc/arc.c (arc_handle_aux_attribute): New function.
284 (arc_attribute_table): Add 'aux' attribute.
285 (arc_in_small_data_p): Consider aux like variables.
286 (arc_is_aux_reg_p): New function.
287 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
288 (arc_get_aux_arg): New function.
289 (prepare_move_operands): Handle aux-register access.
290 (arc_handle_aux_attribute): New function.
291 * doc/extend.texi (ARC Variable attributes): Add subsection.
293 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
295 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
296 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
297 (arc_attribute_table): Add 'uncached' attribute.
298 (arc_print_operand): Print '.di' flag for uncached memory
300 (arc_in_small_data_p): Do not consider for small data the uncached
302 (arc_is_uncached_mem_p): New function.
303 * config/arc/predicates.md (compact_store_memory_operand): Check
304 for uncached memory accesses.
305 (nonvol_nonimm_operand): Likewise.
306 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
308 2018-01-31 Jakub Jelinek <jakub@redhat.com>
311 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
312 falign-loops=): Add Optimization flag.
314 2018-01-30 Jeff Law <law@redhat.com>
317 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
318 INT_REGISTERS_SAVED. Check it prior to calling
319 get_scratch_register_on_entry.
320 (ix86_adjust_stack_and_probe): Similarly.
321 (ix86_emit_probe_stack_range): Similarly.
322 (ix86_expand_prologue): Corresponding changes.
324 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
327 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
328 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
330 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
333 * lra-constraints.c (curr_insn_transform): Process AND in the
336 2018-01-30 Jakub Jelinek <jakub@redhat.com>
338 PR rtl-optimization/83986
339 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
340 dependence against last_pending_memory_flush in addition to
343 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
345 PR tree-optimization/81611
346 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
349 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
351 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
354 2018-01-30 Richard Biener <rguenther@suse.de>
355 Jakub Jelinek <jakub@redhat.com>
357 PR tree-optimization/84111
358 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
359 inner loops added during recursion, as they don't have up-to-date
362 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
365 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
366 (can_inline_edge_by_limits_p): ... here.
367 (can_early_inline_edge_p, check_callers,
368 update_caller_keys, update_callee_keys, recursive_inlining,
369 add_new_edges_to_heap, speculation_useful_p,
370 inline_small_functions,
371 inline_small_functions, flatten_function,
372 inline_to_all_callers_1): Update.
374 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
376 * profile-count.c (profile_count::combine_with_ipa_count): Handle
379 2018-01-30 Richard Biener <rguenther@suse.de>
381 PR tree-optimization/83008
382 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
383 invariant and constant vector uses in stmts when they need
386 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
389 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
390 * configure: Regenerate.
392 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
394 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
396 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
397 Use gen_rtx_REG rather than gen_lowpart.
399 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
401 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
402 rather than 0 when creating partial subregs.
404 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
406 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
409 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
412 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
413 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
414 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
415 flags. This restores the settings used before the 2017-07-24.
416 Turning off pre increment/decrement/modify allows IVOPTS to
417 optimize DF/SF loops where the index is an int.
419 2018-01-29 Richard Biener <rguenther@suse.de>
420 Kelvin Nilsen <kelvin@gcc.gnu.org>
423 * tree-vect-stmts.c (vectorizable_call): Don't call
424 targetm.vectorize_builtin_md_vectorized_function if callee is
427 2018-01-22 Carl Love <cel@us.ibm.com>
429 * doc/extend.tex: Fix typo in second arg in
430 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
432 2018-01-29 Richard Biener <rguenther@suse.de>
434 PR tree-optimization/84086
435 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
436 (flush_ssaname_freelist): When SSA names were released reset
439 2018-01-29 Richard Biener <rguenther@suse.de>
441 PR tree-optimization/84057
442 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
443 removed paths when removing edges.
445 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
447 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
448 -mfunction-return=@var{choice}.
450 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
453 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
455 (layout::print_source_line): Likewise.
456 (test_get_line_width_without_trailing_whitespace): Add test cases.
458 2018-01-27 Jakub Jelinek <jakub@redhat.com>
461 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
464 2018-01-26 Jim Wilson <jimw@sifive.com>
466 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
468 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
471 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
473 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
474 and CMP + SUB-immediate -> SUBS.
476 2018-01-26 Martin Sebor <msebor@redhat.com>
478 PR tree-optimization/83896
479 * tree-ssa-strlen.c (get_string_len): Rename...
480 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
481 Avoid assuming length is constant.
482 (handle_char_store): Use HOST_WIDE_INT for string length.
484 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
487 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
488 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
490 2018-01-26 Richard Biener <rguenther@suse.de>
492 PR rtl-optimization/84003
493 * dse.c (record_store): Only record redundant stores when
494 the earlier store aliases at least all accesses the later one does.
496 2018-01-26 Jakub Jelinek <jakub@redhat.com>
498 PR rtl-optimization/83985
499 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
500 REG_CFA_RESTORE insns.
501 (delete_unmarked_insns): Don't ignore separate shrink wrapping
502 REG_CFA_RESTORE insns here.
505 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
506 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
508 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
510 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
511 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
512 (arc_init): Likewise.
513 (arc_override_options): Likewise.
514 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
516 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
518 * config/arc/arc.h (TARGET_DBNZ): Define.
519 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
520 properly set the tune attribute.
521 (dbnz): Use TARGET_DBNZ guard.
522 * config/arc/arc.opt (mtune): Add core3 option.
524 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
526 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
527 recognize new pic like addresses.
528 (arc_delegitimize_address): Clean up.
530 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
532 * config/arc/arc-arches.def: Option mrf16 valid for all
534 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
535 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
536 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
537 * config/arc/arc-tables.opt: Regenerate.
538 * config/arc/arc.c (arc_conditional_register_usage): Handle
539 reduced register file case.
540 (arc_file_start): Set must have build attributes.
541 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
543 * config/arc/arc.opt (mrf16): Add new option.
544 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
545 * config/arc/genmultilib.awk: Handle new mrf16 option.
546 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
547 * config/arc/t-multilib: Regenerate.
548 * doc/invoke.texi (ARC Options): Document mrf16 option.
550 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
552 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
553 * config/arc/arc.c (arc_handle_secure_attribute): New function.
554 (arc_attribute_table): Add 'secure_call' attribute.
555 (arc_print_operand): Print secure call operand.
556 (arc_function_ok_for_sibcall): Don't optimize tail calls when
558 (arc_is_secure_call_p): New function. * config/arc/arc.md
559 (call_i): Add support for sjli instruction.
560 (call_value_i): Likewise.
561 * config/arc/constraints.md (Csc): New constraint.
563 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
564 John Eric Martin <John.Martin@emmicro-us.com>
566 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
567 * config/arc/arc.c (_arc_jli_section): New struct.
568 (arc_jli_section): New type.
569 (rc_jli_sections): New static variable.
570 (arc_handle_jli_attribute): New function.
571 (arc_attribute_table): Add jli_always and jli_fixed attribute.
572 (arc_file_end): New function.
573 (TARGET_ASM_FILE_END): Define.
574 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
575 (arc_add_jli_section): New function.
576 (jli_call_scan): Likewise.
577 (arc_reorg): Call jli_call_scan.
578 (arc_output_addsi): Remove 'S' from printing asm operand.
579 (arc_is_jli_call_p): New function.
580 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
582 (movhi_insn): Likewise.
583 (movsi_insn): Likewise.
584 (movsi_set_cc_insn): Likewise.
585 (loadqi_update): Likewise.
586 (load_zeroextendqisi_update): Likewise.
587 (load_signextendqisi_update): Likewise.
588 (loadhi_update): Likewise.
589 (load_zeroextendhisi_update): Likewise.
590 (load_signextendhisi_update): Likewise.
591 (loadsi_update): Likewise.
592 (loadsf_update): Likewise.
593 (movsicc_insn): Likewise.
594 (bset_insn): Likewise.
595 (bxor_insn): Likewise.
596 (bclr_insn): Likewise.
597 (bmsk_insn): Likewise.
598 (bicsi3_insn): Likewise.
599 (cmpsi_cc_c_insn): Likewise.
600 (movsi_ne): Likewise.
601 (movsi_cond_exec): Likewise.
602 (clrsbsi2): Likewise.
613 (call_i): Remove 'S' asm letter, add jli instruction.
614 (call_value_i): Likewise.
615 * config/arc/arc.op (mjli-always): New option.
616 * config/arc/constraints.md (Cji): New constraint.
617 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
619 (subsf3_fpx): Likewise.
620 (mulsf3_fpx): Likewise.
621 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
623 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
625 * doc/invoke.texi (ARC): Document mjli-always option.
627 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
629 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
630 and use incw and decw where possible
631 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
633 2018-01-26 Richard Biener <rguenther@suse.de>
635 PR tree-optimization/81082
636 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
637 association if it requires casting to unsigned.
638 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
639 from fold_plusminus_mult_expr to catch important cases late when
640 range info is available.
642 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
644 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
645 * configure.ac (hidden_linkonce): New test.
646 * configure: Regenerate.
647 * config.in: Regenerate.
649 2018-01-26 Julia Koval <julia.koval@intel.com>
651 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
652 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
653 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
654 _mm_mask_bitshuffle_epi64_mask): Fix type.
655 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
656 USI_FTYPE_V4DI_V4DI_USI): Remove.
657 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
658 __builtin_ia32_vpshufbitqmb256_mask,
659 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
660 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
661 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
663 2018-01-26 Alan Modra <amodra@gmail.com>
666 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
667 UNSPEC_VBPERMQ. Sort other unspecs.
669 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
671 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
673 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
676 * predict.c (drop_profile): Do not push/pop cfun; update also
678 (handle_missing_profiles): Fix logic looking for zero profiles.
680 2018-01-25 Jakub Jelinek <jakub@redhat.com>
683 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
684 on functions with #pragma omp declare simd or functions with simd
686 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
687 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
688 Remove trailing \n from warning_at calls.
690 2018-01-25 Tom de Vries <tom@codesourcery.com>
693 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
694 for neutered workers.
696 2018-01-24 Joseph Myers <joseph@codesourcery.com>
699 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
700 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
702 2018-01-24 Jeff Law <law@redhat.com>
705 * i386.c (get_probe_interval): Move to earlier point.
706 (ix86_compute_frame_layout): If -fstack-clash-protection and
707 the frame is larger than the probe interval, then use pushes
708 to save registers rather than reg->mem moves.
709 (ix86_expand_prologue): Remove conditional for int_registers_saved
712 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
715 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
716 min/max for never referenced object.
718 2018-01-24 Jakub Jelinek <jakub@redhat.com>
721 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
723 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
724 attributes from DECL_ATTRIBUTES (decl) without affecting
725 DECL_ATTRIBUTES (current_function_decl).
726 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
727 functions with non-NULL DECL_ABSTRACT_ORIGIN.
729 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
731 PR tree-optimization/83979
732 * fold-const.c (fold_comparison): Use constant_boolean_node
733 instead of boolean_{true,false}_node.
735 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
737 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
740 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
742 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
743 Simplify the clause that sets the length attribute.
744 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
745 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
746 clause that sets the length attribute.
747 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
749 2018-01-24 Tom de Vries <tom@codesourcery.com>
752 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
753 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
754 Add strict parameter.
755 (prevent_branch_around_nothing): Insert dummy insn between branch to
756 label and label with no ptx insn inbetween.
757 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
759 2018-01-24 Tom de Vries <tom@codesourcery.com>
762 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
763 for neutered threads in warp.
764 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
766 2018-01-24 Richard Biener <rguenther@suse.de>
768 PR tree-optimization/83176
769 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
772 2018-01-24 Richard Biener <rguenther@suse.de>
774 PR tree-optimization/82819
775 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
776 code generating pluses that are no-ops in the target precision.
778 2018-01-24 Richard Biener <rguenther@suse.de>
781 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
783 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
785 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
786 to merge probabilities.
787 * predict.c (probably_never_executed): Also mark as cold functions
788 with global 0 profile and guessed local profile.
789 * profile-count.c (profile_probability::combine_with_count): New
791 * profile-count.h (profile_probability::operator*,
792 profile_probability::operator*=, profile_probability::operator/,
793 profile_probability::operator/=): Reduce precision to adjusted
794 and set value to guessed on contradictory divisions.
795 (profile_probability::combine_with_freq): Remove.
796 (profile_probability::combine_wiht_count): Declare.
797 (profile_count::force_nonzero):: Set to adjusted.
798 (profile_count::probability_in):: Set quality to adjusted.
799 * tree-ssa-tail-merge.c (replace_block_by): Use
802 2018-01-23 Andrew Waterman <andrew@sifive.com>
803 Jim Wilson <jimw@sifive.com>
805 * config/riscv/riscv.c (riscv_stack_boundary): New.
806 (riscv_option_override): Set riscv_stack_boundary. Handle
807 riscv_preferred_stack_boundary_arg.
808 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
809 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
810 (STACK_BOUNDARY): Set to riscv_stack_boundary.
811 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
812 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
813 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
815 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
818 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
819 of struct ix86_frame.
820 (ix86_expand_epilogue): Likewise. Add a local variable for
821 the reg_save_offset field in struct ix86_frame.
823 2018-01-23 Bin Cheng <bin.cheng@arm.com>
825 PR tree-optimization/82604
826 * tree-loop-distribution.c (enum partition_kind): New enum item
827 PKIND_PARTIAL_MEMSET.
828 (partition_builtin_p): Support above new enum item.
829 (generate_code_for_partition): Ditto.
830 (compute_access_range): Differentiate cases that equality can be
831 proven at all loops, the innermost loops or no loops.
832 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
833 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
834 (finalize_partitions, distribute_loop): Don't fuse partition of
835 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
836 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
839 2018-01-23 Martin Liska <mliska@suse.cz>
841 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
842 order to ignore the predictor.
843 (PRED_POLYMORPHIC_CALL): Likewise.
844 (PRED_RECURSIVE_CALL): Likewise.
846 2018-01-23 Martin Liska <mliska@suse.cz>
848 * tree-profile.c (tree_profiling): Print function header to
849 aware reader which function we are working on.
850 * value-prof.c (gimple_find_values_to_profile): Do not print
851 not interesting value histograms.
853 2018-01-23 Martin Liska <mliska@suse.cz>
855 * profile-count.h (enum profile_quality): Add
856 profile_uninitialized as the first value. Do not number values
857 as they are zero based.
858 (profile_count::verify): Update sanity check.
859 (profile_probability::verify): Likewise.
861 2018-01-23 Nathan Sidwell <nathan@acm.org>
863 * doc/invoke.texi (ffor-scope): Deprecate.
865 2018-01-23 David Malcolm <dmalcolm@redhat.com>
867 PR tree-optimization/83510
868 * domwalk.c (set_all_edges_as_executable): New function.
869 (dom_walker::dom_walker): Convert bool param
870 "skip_unreachable_blocks" to enum reachability. Move setup of
871 edge flags to set_all_edges_as_executable and only do it when
872 reachability is REACHABLE_BLOCKS.
873 * domwalk.h (enum dom_walker::reachability): New enum.
874 (dom_walker::dom_walker): Convert bool param
875 "skip_unreachable_blocks" to enum reachability.
876 (set_all_edges_as_executable): New decl.
877 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
878 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
880 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
881 but converting true to REACHABLE_BLOCKS.
882 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
884 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
885 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
886 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
888 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
889 if check_all_array_refs will be called.
891 2018-01-23 David Malcolm <dmalcolm@redhat.com>
893 * tree.c (selftest::test_location_wrappers): Add more test
896 2018-01-23 David Malcolm <dmalcolm@redhat.com>
898 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
899 (selftest::test_bit_in_range): Likewise.
901 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
904 * doc/sourcebuild.texi (vect_float): Say that the selector
905 only describes the situation when -funsafe-math-optimizations is on.
906 (vect_float_strict): Document.
908 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
910 PR tree-optimization/83965
911 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
912 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
913 instead of checking only for a reduction.
914 (vect_recog_widen_sum_pattern): Likewise.
916 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
918 * predict.c (probably_never_executed): Only use precise profile info.
919 (compute_function_frequency): Skip after inlining hack since we now
920 have quality checking.
922 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
924 * profile-count.h (profile_probability::very_unlikely,
925 profile_probability::unlikely, profile_probability::even): Set
926 precision to guessed.
928 2018-01-23 Richard Biener <rguenther@suse.de>
930 PR tree-optimization/83963
931 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
932 Properly terminate dominator walk when crossing the exit edge not
933 when visiting its source block.
935 2018-01-23 Jakub Jelinek <jakub@redhat.com>
938 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
939 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
941 2018-01-22 Jakub Jelinek <jakub@redhat.com>
943 PR tree-optimization/83957
944 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
945 semicolon after for body surrounded by braces.
947 PR tree-optimization/83081
948 * profile-count.h (profile_probability::split): New method.
949 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
950 Use profile_probability::split.
951 (do_compare_rtx_and_jump): Fix adjustment of probabilities
952 when splitting a single conditional jump into 2.
954 2018-01-22 David Malcolm <dmalcolm@redhat.com>
956 PR tree-optimization/69452
957 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
960 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
962 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
963 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
964 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
966 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
968 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
969 * config/rl78/rl78.md: New define_expand "movdi"
970 * config/rl78/rl78.c: New function definition rl78_split_movdi
972 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
975 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
977 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
978 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
979 128-bit to produce an UNSPEC move to get the double word with the
980 signbit and then a shift directly to do signbit.
981 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
982 implementation with a new version that just does either a direct
983 move or a regular move. Move memory interface to separate insns.
984 Move insns so they are next to the expander.
985 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
986 with signbit move. Split big and little endian case.
987 (signbit<mode>2_dm_mem_le): Likewise.
988 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
989 (signbit<mode>2_dm2): Likewise.
991 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
993 * config/rl78/rl78.md: New define_expand "anddi3".
995 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
997 * config/rl78/rl78.md: New define_expand "umindi3".
999 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1001 * config/rl78/rl78.md: New define_expand "smindi3".
1003 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1005 * config/rl78/rl78.md: New define_expand "smaxdi3".
1007 2018-01-22 Carl Love <cel@us.ibm.com>
1009 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1010 LVX_V1TI): Add macro expansion.
1011 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1012 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1013 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1014 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1015 Change check to determine if the instruction is a byte reversing
1016 entry. Fix typo in comment.
1017 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1018 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1019 Add def_builtin calls for new builtins.
1020 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1021 Add define_insn expansion.
1023 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1025 * config/rl78/rl78.md: New define_expand "umaxdi3".
1027 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1029 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1030 for non-QImode registers
1032 2018-01-22 Richard Biener <rguenther@suse.de>
1034 PR tree-optimization/83963
1035 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1036 including the loop exit block.
1037 (scop_detection::merge_sese): Likewise.
1038 (scop_detection::add_scop): Do it here instead.
1040 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1042 * doc/sourcebuild.texi (arm_softfloat): Document.
1044 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1047 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1048 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1049 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1051 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1052 David Edelsohn <dje.gcc@gmail.com>
1055 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1056 Change "crset eq" to "crset 2".
1057 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1058 (*call_indirect_aix<mode>_nospec): Likewise.
1059 (*call_value_indirect_aix<mode>_nospec): Likewise.
1060 (*call_indirect_elfv2<mode>_nospec): Likewise.
1061 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1062 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1063 change assembly output from . to $.
1064 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1065 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1066 (*tablejump<mode>_internal1_nospec): Likewise.
1068 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1071 * config/sh/sh_optimize_sett_clrt.cc:
1072 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1074 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1076 PR tree-optimization/83940
1077 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1078 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1079 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1080 use it to pass back the definition type.
1081 (vect_check_store_rhs): Likewise rhs_dt_out.
1082 (vect_build_gather_load_calls): Add a mask_dt argument and use
1083 it instead of a call to vect_is_simple_use.
1084 (vectorizable_store): Update calls to vect_check_load_store_mask
1085 and vect_check_store_rhs. Use the dt returned by the latter instead
1086 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1087 instead of calls to vect_is_simple_use. Pass the scalar rather
1088 than the vector operand to vect_is_simple_use when handling
1089 second and subsequent copies of an rhs value.
1090 (vectorizable_load): Update calls to vect_check_load_store_mask
1091 and vect_build_gather_load_calls. Use the cached mask_dt and
1092 gs_info.offset_dt instead of calls to vect_is_simple_use.
1094 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1097 * tree-emutls.c: Include gimplify.h.
1098 (lower_emutls_2): New function.
1099 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1100 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1101 it before further processing.
1104 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1105 UINTVAL (trueop1) instead of INTVAL (op1).
1107 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1111 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1112 INCOMING_FRAME_SP_OFFSET if not defined.
1113 (scan_trace): Add ENTRY argument. If true and
1114 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1115 emit a note to adjust the CFA offset.
1116 (create_cfi_notes): Adjust scan_trace callers.
1117 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1118 INCOMING_FRAME_SP_OFFSET in the CIE.
1119 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1120 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1122 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1123 * doc/tm.texi: Regenerated.
1125 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1127 PR rtl-optimization/83147
1128 * lra-constraints.c (remove_inheritance_pseudos): Use
1129 lra_substitute_pseudo_within_insn.
1131 2018-01-19 Tom de Vries <tom@codesourcery.com>
1132 Cesar Philippidis <cesar@codesourcery.com>
1135 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1137 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1140 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1141 spaces for function labels.
1143 2018-01-19 Martin Liska <mliska@suse.cz>
1145 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1146 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1147 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1148 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1149 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1150 (PRED_CONST_RETURN): Change from 69 to 65.
1151 (PRED_NULL_RETURN): Change from 91 to 71.
1152 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1153 (PRED_LOOP_GUARD): Change from 66 to 73.
1155 2018-01-19 Martin Liska <mliska@suse.cz>
1157 * predict.c (predict_insn_def): Add new assert.
1158 (struct branch_predictor): Change type to signed integer.
1159 (test_prediction_value_range): Amend test to cover
1161 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1162 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1163 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1164 (PRED_LOOP_IV_COMPARE): Likewise.
1165 * predict.h (PROB_UNINITIALIZED): Define new constant.
1167 2018-01-19 Martin Liska <mliska@suse.cz>
1169 * predict.c (dump_prediction): Add new format for
1170 analyze_brprob.py script which is enabled with -details
1172 * profile-count.h (precise_p): New function.
1174 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1176 PR tree-optimization/83922
1177 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1178 there are no statements that need masking.
1179 (vect_active_double_reduction_p): New function.
1180 (vect_analyze_loop_operations): Use it when handling phis that
1181 are not in the loop header.
1183 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1185 PR tree-optimization/83914
1186 * tree-vect-loop.c (vectorizable_induction): Don't convert
1187 init_expr or apply the peeling adjustment for inductions
1188 that are nested within the vectorized loop.
1190 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1192 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1195 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1199 * function.h (gimplify_parameters): Add gimple_seq * argument.
1200 * function.c: Include gimple.h and options.h.
1201 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1202 for the added local temporaries if needed.
1203 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1204 if there are any parameter cleanups, wrap whole body into a
1205 try/finally with the cleanups.
1207 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1210 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1211 Use GET_MODE_CLASS for scalar floating point.
1213 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1217 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1218 Fix call of call_cgraph_insertion_hooks.
1220 2018-01-18 Martin Sebor <msebor@redhat.com>
1222 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1224 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1227 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1230 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1233 * common.opt: (-ffile-prefix-map): New option.
1234 * opts.c (common_handle_option): Defer it.
1235 * opts-global.c (handle_common_deferred_options): Handle it.
1236 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1237 * file-prefix-map.h: New file.
1238 (remap_debug_filename, add_debug_prefix_map): ...here.
1239 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1240 * final.c (debug_prefix_map, add_debug_prefix_map
1241 remap_debug_filename): Move to...
1242 * file-prefix-map.c: New file.
1243 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1244 generalize, get rid of alloca(), use strrchr() instead of strchr().
1245 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1246 Implement in terms of add_prefix_map().
1247 (remap_macro_filename, remap_debug_filename): Implement in term of
1249 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1250 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1251 * dbxout.c: Include file-prefix-map.h.
1252 * varasm.c: Likewise.
1253 * vmsdbgout.c: Likewise.
1254 * xcoffout.c: Likewise.
1255 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1256 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1257 * doc/invoke.texi (-ffile-prefix-map): Document.
1258 (-fdebug-prefix-map): Update description.
1260 2018-01-18 Martin Liska <mliska@suse.cz>
1262 * config/i386/i386.c (indirect_thunk_name): Document that also
1264 (output_indirect_thunk): Document why both instructions
1265 (pause and lfence) are generated.
1267 2018-01-18 Richard Biener <rguenther@suse.de>
1269 PR tree-optimization/83887
1270 * graphite-scop-detection.c
1271 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1272 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1273 (scop_detection::merge_sese): Re-implement with a flood-fill
1274 algorithm that properly finds a SESE region if it exists.
1276 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1279 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1280 pointer_diff optimizations use view_convert instead of convert.
1282 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1284 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1285 Generate different code for -mno-speculate-indirect-jumps.
1286 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1287 (*call_indirect_aix<mode>): Disable for
1288 -mno-speculate-indirect-jumps.
1289 (*call_indirect_aix<mode>_nospec): New define_insn.
1290 (*call_value_indirect_aix<mode>): Disable for
1291 -mno-speculate-indirect-jumps.
1292 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1293 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1294 -mno-speculate-indirect-jumps.
1295 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1297 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1299 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1300 long double type, set the flags for noting the default long double
1301 type, even if we don't pass or return a long double type.
1303 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1306 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1309 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1311 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1312 support for merge[hl].
1313 (fold_mergehl_helper): New helper function.
1314 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1315 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1316 (altivec_vmrglw_direct): Add xxmrglw insn.
1318 2018-01-17 Andrew Waterman <andrew@sifive.com>
1320 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1321 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1323 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1326 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1327 call the lto_location_cache before reading the
1328 DECL_SOURCE_LOCATION of the types.
1330 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1331 Richard Sandiford <richard.sandiford@linaro.org>
1333 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1334 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1335 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1336 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1337 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1339 * config/aarch64/constraints.md (aarch64_movti_operand):
1341 * config/aarch64/predicates.md (Uti): Add new constraint.
1343 2018-01-17 Carl Love <cel@us.ibm.com>
1344 * config/rs6000/vsx.md (define_expand xl_len_r,
1345 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1346 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1348 (define_expand, define_insn): Move the shift left from the
1349 define_insn to the define_expand for lxvl and stxvl instructions.
1350 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1351 and XL_LEN_R definitions to PURE.
1353 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1355 * config/i386/i386.c (indirect_thunk_name): Declare regno
1356 as unsigned int. Compare regno with INVALID_REGNUM.
1357 (output_indirect_thunk): Ditto.
1358 (output_indirect_thunk_function): Ditto.
1359 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1360 in the call to output_indirect_thunk_function.
1362 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1365 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1366 rather than the size of inner_type to determine the stack slot size
1367 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1369 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1372 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1375 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1377 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1378 endian Linux systems to optionally enable multilibs for selecting
1379 the long double type if the user configured an explicit type.
1380 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1381 have no long double multilibs if not defined.
1382 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1383 warn if the user used -mabi={ieee,ibm}longdouble and we built
1384 multilibs for long double.
1385 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1386 appropriate multilib option.
1387 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1389 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1390 for building long double multilibs.
1391 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1393 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1395 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1398 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1400 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1403 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1406 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1409 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1411 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1412 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1414 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1416 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1417 different rtl trees depending on TARGET_64BIT.
1418 (rs6000_gen_lvx): Likewise.
1420 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1422 * config/visium/visium.md (nop): Tweak comment.
1423 (hazard_nop): Likewise.
1425 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1427 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1428 -mspeculate-indirect-jumps.
1429 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1430 for -mno-speculate-indirect-jumps.
1431 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1432 (*call_value_indirect_elfv2<mode>): Disable for
1433 -mno-speculate-indirect-jumps.
1434 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1435 (indirect_jump): Emit different RTL for
1436 -mno-speculate-indirect-jumps.
1437 (*indirect_jump<mode>): Disable for
1438 -mno-speculate-indirect-jumps.
1439 (*indirect_jump<mode>_nospec): New define_insn.
1440 (tablejump): Emit different RTL for
1441 -mno-speculate-indirect-jumps.
1442 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1443 (tablejumpsi_nospec): New define_expand.
1444 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1445 (tablejumpdi_nospec): New define_expand.
1446 (*tablejump<mode>_internal1): Disable for
1447 -mno-speculate-indirect-jumps.
1448 (*tablejump<mode>_internal1_nospec): New define_insn.
1449 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1452 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1454 * caller-save.c (insert_save): Drop unnecessary parameter. All
1457 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1458 Richard Biener <rguenth@suse.de>
1461 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1462 return early, inline manually is_gimple_sizepos. Make sure if we
1463 call gimplify_expr we don't end up with a gimple constant.
1464 * tree.c (variably_modified_type_p): Don't return true for
1465 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1466 * gimplify.h (is_gimple_sizepos): Remove.
1468 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1470 PR tree-optimization/83857
1471 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1472 vectorizable_live_operation for pure SLP statements.
1473 (vectorizable_live_operation): Handle PHIs.
1475 2018-01-16 Richard Biener <rguenther@suse.de>
1477 PR tree-optimization/83867
1478 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1479 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1481 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1484 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1485 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1486 If off is not INTEGER_CST, issue a may not be aligned warning
1487 rather than isn't aligned. Use isn%'t rather than isn't.
1488 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1490 <case MULT_EXPR>: Improve the case when bottom and one of the
1491 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1492 operand, in that case check if the other operand is multiple of
1493 bottom divided by the INTEGER_CST operand.
1495 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1498 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1499 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1500 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1501 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1502 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1503 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1504 (pa_function_arg_size): New function.
1506 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1508 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1509 in a separate statement.
1511 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1513 PR tree-optimization/83847
1514 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1515 group gathers and scatters.
1517 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1519 PR rtl-optimization/86620
1520 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1522 PR rtl-optimization/83213
1523 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1524 to last if both are JUMP_INSNs.
1526 PR tree-optimization/83843
1527 * gimple-ssa-store-merging.c
1528 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1529 store_immediate_info for bswap/nop orig_stores.
1531 2018-01-15 Andrew Waterman <andrew@sifive.com>
1533 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1535 <UDIV>: Increase cost if !TARGET_DIV.
1537 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1539 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1540 (define_attr "cr_logical_3op"): New.
1541 (cceq_ior_compare): Adjust.
1542 (cceq_ior_compare_complement): Adjust.
1543 (*cceq_rev_compare): Adjust.
1544 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1545 (is_cracked_insn): Adjust.
1546 (insn_must_be_first_in_group): Adjust.
1547 * config/rs6000/40x.md: Adjust.
1548 * config/rs6000/440.md: Adjust.
1549 * config/rs6000/476.md: Adjust.
1550 * config/rs6000/601.md: Adjust.
1551 * config/rs6000/603.md: Adjust.
1552 * config/rs6000/6xx.md: Adjust.
1553 * config/rs6000/7450.md: Adjust.
1554 * config/rs6000/7xx.md: Adjust.
1555 * config/rs6000/8540.md: Adjust.
1556 * config/rs6000/cell.md: Adjust.
1557 * config/rs6000/e300c2c3.md: Adjust.
1558 * config/rs6000/e500mc.md: Adjust.
1559 * config/rs6000/e500mc64.md: Adjust.
1560 * config/rs6000/e5500.md: Adjust.
1561 * config/rs6000/e6500.md: Adjust.
1562 * config/rs6000/mpc.md: Adjust.
1563 * config/rs6000/power4.md: Adjust.
1564 * config/rs6000/power5.md: Adjust.
1565 * config/rs6000/power6.md: Adjust.
1566 * config/rs6000/power7.md: Adjust.
1567 * config/rs6000/power8.md: Adjust.
1568 * config/rs6000/power9.md: Adjust.
1569 * config/rs6000/rs64.md: Adjust.
1570 * config/rs6000/titan.md: Adjust.
1572 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1574 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1575 ix86_indirect_branch_register logic.
1577 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1579 * config/i386/constraints.md (Bs): Update
1580 ix86_indirect_branch_register check. Don't check
1581 ix86_indirect_branch_register with GOT_memory_operand.
1583 * config/i386/predicates.md (GOT_memory_operand): Don't check
1584 ix86_indirect_branch_register here.
1585 (GOT32_symbol_operand): Likewise.
1587 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1589 * config/i386/predicates.md (constant_call_address_operand):
1590 Rewrite ix86_indirect_branch_register logic.
1591 (sibcall_insn_operand): Likewise.
1593 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1595 * config/i386/constraints.md (Bs): Replace
1596 ix86_indirect_branch_thunk_register with
1597 ix86_indirect_branch_register.
1599 * config/i386/i386.md (indirect_jump): Likewise.
1600 (tablejump): Likewise.
1601 (*sibcall_memory): Likewise.
1602 (*sibcall_value_memory): Likewise.
1603 Peepholes of indirect call and jump via memory: Likewise.
1604 * config/i386/i386.opt: Likewise.
1605 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1606 (GOT_memory_operand): Likewise.
1607 (call_insn_operand): Likewise.
1608 (sibcall_insn_operand): Likewise.
1609 (GOT32_symbol_operand): Likewise.
1611 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1614 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1615 type rather than type addr's type points to.
1616 (expand_omp_atomic_mutex): Likewise.
1617 (expand_omp_atomic): Likewise.
1619 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1622 * config/i386/i386.c (output_indirect_thunk_function): Use
1623 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1624 for __x86_return_thunk.
1626 2018-01-15 Richard Biener <rguenther@suse.de>
1629 * expmed.c (extract_bit_field_1): Fix typo.
1631 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1634 * config/arm/iterators.md (VF): New mode iterator.
1635 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1636 Remove integer-related logic from pattern.
1637 (neon_vabd<mode>_3): Likewise.
1639 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1642 * common.opt (fstrict-overflow): No longer an alias.
1643 (fwrapv-pointer): New option.
1644 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1645 also for pointer types based on flag_wrapv_pointer.
1646 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1647 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1648 opts->x_flag_wrapv got set.
1649 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1650 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1651 POINTER_TYPE_OVERFLOW_UNDEFINED.
1652 * match.pd: Likewise in address comparison pattern.
1653 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1655 2018-01-15 Richard Biener <rguenther@suse.de>
1658 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1659 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1660 Reset type names to their identifier if their TYPE_DECL doesn't
1661 have linkage (and thus is used for ODR and devirt).
1662 (save_debug_info_for_decl): Remove.
1663 (save_debug_info_for_type): Likewise.
1664 (add_tree_to_fld_list): Adjust.
1665 * tree-pretty-print.c (dump_generic_node): Make dumping of
1666 type names more robust.
1668 2018-01-15 Richard Biener <rguenther@suse.de>
1670 * BASE-VER: Bump to 8.0.1.
1672 2018-01-14 Martin Sebor <msebor@redhat.com>
1675 * builtins.c (check_access): Avoid warning when the no-warning bit
1678 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1680 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1681 * ira-color (allocno_hard_regs_compare): Likewise.
1683 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1686 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1687 Use .pushsection/.popsection.
1689 2018-01-14 Martin Sebor <msebor@redhat.com>
1692 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1694 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1696 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1697 entry from extra_headers.
1698 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1699 extra_headers, make the list bitwise identical to the i?86-*-* one.
1701 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1703 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1704 -mcmodel=large with -mindirect-branch=thunk,
1705 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1706 -mfunction-return=thunk-extern.
1707 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1708 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1709 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1711 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1713 * config/i386/i386.c (print_reg): Print the name of the full
1714 integer register without '%'.
1715 (ix86_print_operand): Handle 'V'.
1716 * doc/extend.texi: Document 'V' modifier.
1718 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1720 * config/i386/constraints.md (Bs): Disallow memory operand for
1721 -mindirect-branch-register.
1723 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1724 (GOT_memory_operand): Likewise.
1725 (call_insn_operand): Likewise.
1726 (sibcall_insn_operand): Likewise.
1727 (GOT32_symbol_operand): Likewise.
1728 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1729 for -mindirect-branch-register.
1730 (tablejump): Likewise.
1731 (*sibcall_memory): Likewise.
1732 (*sibcall_value_memory): Likewise.
1733 Disallow peepholes of indirect call and jump via memory for
1734 -mindirect-branch-register.
1735 (*call_pop): Replace m with Bw.
1736 (*call_value_pop): Likewise.
1737 (*sibcall_pop_memory): Replace m with Bs.
1738 * config/i386/i386.opt (mindirect-branch-register): New option.
1739 * doc/invoke.texi: Document -mindirect-branch-register option.
1741 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1743 * config/i386/i386-protos.h (ix86_output_function_return): New.
1744 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1745 set function_return_type.
1746 (indirect_thunk_name): Add ret_p to indicate thunk for function
1748 (output_indirect_thunk_function): Pass false to
1749 indirect_thunk_name.
1750 (ix86_output_indirect_branch_via_reg): Likewise.
1751 (ix86_output_indirect_branch_via_push): Likewise.
1752 (output_indirect_thunk_function): Create alias for function
1753 return thunk if regno < 0.
1754 (ix86_output_function_return): New function.
1755 (ix86_handle_fndecl_attribute): Handle function_return.
1756 (ix86_attribute_table): Add function_return.
1757 * config/i386/i386.h (machine_function): Add
1758 function_return_type.
1759 * config/i386/i386.md (simple_return_internal): Use
1760 ix86_output_function_return.
1761 (simple_return_internal_long): Likewise.
1762 * config/i386/i386.opt (mfunction-return=): New option.
1763 (indirect_branch): Mention -mfunction-return=.
1764 * doc/extend.texi: Document function_return function attribute.
1765 * doc/invoke.texi: Document -mfunction-return= option.
1767 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1769 * config/i386/i386-opts.h (indirect_branch): New.
1770 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1771 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1772 with local indirect jump when converting indirect call and jump.
1773 (ix86_set_indirect_branch_type): New.
1774 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1775 (indirectlabelno): New.
1776 (indirect_thunk_needed): Likewise.
1777 (indirect_thunk_bnd_needed): Likewise.
1778 (indirect_thunks_used): Likewise.
1779 (indirect_thunks_bnd_used): Likewise.
1780 (INDIRECT_LABEL): Likewise.
1781 (indirect_thunk_name): Likewise.
1782 (output_indirect_thunk): Likewise.
1783 (output_indirect_thunk_function): Likewise.
1784 (ix86_output_indirect_branch_via_reg): Likewise.
1785 (ix86_output_indirect_branch_via_push): Likewise.
1786 (ix86_output_indirect_branch): Likewise.
1787 (ix86_output_indirect_jmp): Likewise.
1788 (ix86_code_end): Call output_indirect_thunk_function if needed.
1789 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1791 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1792 (ix86_attribute_table): Add indirect_branch.
1793 * config/i386/i386.h (machine_function): Add indirect_branch_type
1794 and has_local_indirect_jump.
1795 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1797 (tablejump): Likewise.
1798 (*indirect_jump): Use ix86_output_indirect_jmp.
1799 (*tablejump_1): Likewise.
1800 (simple_return_indirect_internal): Likewise.
1801 * config/i386/i386.opt (mindirect-branch=): New option.
1802 (indirect_branch): New.
1805 (thunk-inline): Likewise.
1806 (thunk-extern): Likewise.
1807 * doc/extend.texi: Document indirect_branch function attribute.
1808 * doc/invoke.texi: Document -mindirect-branch= option.
1810 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1813 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1815 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1817 * ipa-inline.c (want_inline_small_function_p): Return false if
1818 inlining has already failed with CIF_FINAL_ERROR.
1819 (update_caller_keys): Call want_inline_small_function_p before
1821 (update_callee_keys): Likewise.
1823 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1825 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1827 (rs6000_quadword_masked_address_p): Likewise.
1828 (quad_aligned_load_p): Likewise.
1829 (quad_aligned_store_p): Likewise.
1830 (const_load_sequence_p): Add comment to describe the outer-most loop.
1831 (mimic_memory_attributes_and_flags): New function.
1832 (rs6000_gen_stvx): Likewise.
1833 (replace_swapped_aligned_store): Likewise.
1834 (rs6000_gen_lvx): Likewise.
1835 (replace_swapped_aligned_load): Likewise.
1836 (replace_swapped_load_constant): Capitalize argument name in
1837 comment describing this function.
1838 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1839 and stores that access quad-word aligned addresses and replace
1840 with stvx or lvx instructions when appropriate.
1841 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1842 New function prototype.
1843 (rs6000_quadword_masked_address_p): Likewise.
1844 (rs6000_gen_lvx): Likewise.
1845 (rs6000_gen_stvx): Likewise.
1846 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1847 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1848 when memory address is aligned.
1849 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1850 this split to select lvx instruction when memory address is aligned.
1851 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1852 instruction when memory address is aligned.
1853 (*vsx_le_perm_load_v16qi): Likewise.
1854 (four unnamed splitters): Modify to select the stvx instruction
1855 when memory is aligned.
1857 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1859 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1860 which appears in the queue multiple times.
1862 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1863 Alan Hayward <alan.hayward@arm.com>
1864 David Sherwood <david.sherwood@arm.com>
1866 * tree-vectorizer.h (vec_lower_bound): New structure.
1867 (_loop_vec_info): Add check_nonzero and lower_bounds.
1868 (LOOP_VINFO_CHECK_NONZERO): New macro.
1869 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1870 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1871 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1872 fields. Make seg_len the distance travelled, not including the
1874 (dr_direction_indicator): Declare.
1875 (dr_zero_step_indicator): Likewise.
1876 (dr_known_forward_stride_p): Likewise.
1877 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1879 (runtime_alias_check_p): Allow runtime alias checks with
1881 (operator ==): Compare access_size and align.
1882 (prune_runtime_alias_test_list): Rework for new distinction between
1883 the access_size and seg_len.
1884 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1886 (get_segment_min_max): New function.
1887 (create_intersect_range_checks): Use it.
1888 (dr_step_indicator): New function.
1889 (dr_direction_indicator): Likewise.
1890 (dr_zero_step_indicator): Likewise.
1891 (dr_known_forward_stride_p): Likewise.
1892 * tree-loop-distribution.c (data_ref_segment_size): Return
1893 DR_STEP * (niters - 1).
1894 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1896 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1897 (vect_preserves_scalar_order_p): New function, split out from...
1898 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1899 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1900 (vect_vfa_access_size): New function.
1901 (vect_vfa_align): Likewise.
1902 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1903 (dump_lower_bound): New function.
1904 (vect_check_lower_bound): Likewise.
1905 (vect_small_gap_p): Likewise.
1906 (vectorizable_with_step_bound_p): Likewise.
1907 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1908 depencies if the vectorization factor is 1. Convert the checks
1909 for nonzero steps into checks on the bounds of DR_STEP. Try using
1910 a bunds check for variable steps if the minimum required step is
1911 relatively small. Update calls to the dr_with_seg_len
1912 constructor and to vect_compile_time_alias.
1913 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1915 (vect_loop_versioning): Call it.
1916 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1918 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1920 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1921 Alan Hayward <alan.hayward@arm.com>
1922 David Sherwood <david.sherwood@arm.com>
1924 * doc/sourcebuild.texi (vect_scatter_store): Document.
1925 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1927 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1929 * genopinit.c (main): Add supports_vec_scatter_store and
1930 supports_vec_scatter_store_cached to target_optabs.
1931 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1932 IFN_MASK_SCATTER_STORE.
1933 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1935 * internal-fn.h (internal_store_fn_p): Declare.
1936 (internal_fn_stored_value_index): Likewise.
1937 * internal-fn.c (scatter_store_direct): New macro.
1938 (expand_scatter_store_optab_fn): New function.
1939 (direct_scatter_store_optab_supported_p): New macro.
1940 (internal_store_fn_p): New function.
1941 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1942 IFN_MASK_SCATTER_STORE.
1943 (internal_fn_mask_index): Likewise.
1944 (internal_fn_stored_value_index): New function.
1945 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1947 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1948 * optabs-query.c (supports_vec_scatter_store_p): New function.
1949 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1950 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1951 true for scatter stores.
1952 (vect_gather_scatter_fn_p): Handle scatter stores too.
1953 (vect_check_gather_scatter): Consider using scatter stores if
1954 supports_vec_scatter_store_p.
1955 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1957 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1958 internal_fn_stored_value_index.
1959 (check_load_store_masking): Handle scatter stores too.
1960 (vect_get_store_rhs): Make public.
1961 (vectorizable_call): Use internal_store_fn_p.
1962 (vectorizable_store): Handle scatter store internal functions.
1963 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1964 when deciding whether the end of the group has been reached.
1965 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1966 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1967 (mask_scatter_store<mode>): New insns.
1969 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1970 Alan Hayward <alan.hayward@arm.com>
1971 David Sherwood <david.sherwood@arm.com>
1973 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1974 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1975 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1977 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1978 Use vect_truncate_gather_scatter_offset if we can't treat the
1979 operation as a normal gather load or scatter store.
1980 (get_group_load_store_type): Take the gather_scatter_info
1981 as argument. Try using a gather load or scatter store for
1982 single-element groups.
1983 (get_load_store_type): Update calls to get_group_load_store_type
1984 and vect_use_strided_gather_scatters_p.
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 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1991 optional tree argument.
1992 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1994 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1995 but continue to use the current value as a fallback.
1996 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1997 to compare the updates.
1998 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1999 (get_load_store_type): Use it when handling a strided access.
2000 (vect_get_strided_load_store_ops): New function.
2001 (vect_get_data_ptr_increment): Likewise.
2002 (vectorizable_load): Handle strided gather loads. Always pass
2003 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2005 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2006 Alan Hayward <alan.hayward@arm.com>
2007 David Sherwood <david.sherwood@arm.com>
2009 * doc/md.texi (gather_load@var{m}): Document.
2010 (mask_gather_load@var{m}): Likewise.
2011 * genopinit.c (main): Add supports_vec_gather_load and
2012 supports_vec_gather_load_cached to target_optabs.
2013 * optabs-tree.c (init_tree_optimization_optabs): Use
2014 ggc_cleared_alloc to allocate target_optabs.
2015 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2016 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2018 * internal-fn.h (internal_load_fn_p): Declare.
2019 (internal_gather_scatter_fn_p): Likewise.
2020 (internal_fn_mask_index): Likewise.
2021 (internal_gather_scatter_fn_supported_p): Likewise.
2022 * internal-fn.c (gather_load_direct): New macro.
2023 (expand_gather_load_optab_fn): New function.
2024 (direct_gather_load_optab_supported_p): New macro.
2025 (direct_internal_fn_optab): New function.
2026 (internal_load_fn_p): Likewise.
2027 (internal_gather_scatter_fn_p): Likewise.
2028 (internal_fn_mask_index): Likewise.
2029 (internal_gather_scatter_fn_supported_p): Likewise.
2030 * optabs-query.c (supports_at_least_one_mode_p): New function.
2031 (supports_vec_gather_load_p): Likewise.
2032 * optabs-query.h (supports_vec_gather_load_p): Declare.
2033 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2034 and memory_type field.
2035 (NUM_PATTERNS): Bump to 15.
2036 * tree-vect-data-refs.c: Include internal-fn.h.
2037 (vect_gather_scatter_fn_p): New function.
2038 (vect_describe_gather_scatter_call): Likewise.
2039 (vect_check_gather_scatter): Try using internal functions for
2040 gather loads. Recognize existing calls to a gather load function.
2041 (vect_analyze_data_refs): Consider using gather loads if
2042 supports_vec_gather_load_p.
2043 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2044 (vect_get_gather_scatter_offset_type): Likewise.
2045 (vect_convert_mask_for_vectype): Likewise.
2046 (vect_add_conversion_to_patterm): Likewise.
2047 (vect_try_gather_scatter_pattern): Likewise.
2048 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2049 (vect_vect_recog_func_ptrs): Add it.
2050 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2051 internal_fn_mask_index and internal_gather_scatter_fn_p.
2052 (check_load_store_masking): Take the gather_scatter_info as an
2053 argument and handle gather loads.
2054 (vect_get_gather_scatter_ops): New function.
2055 (vectorizable_call): Check internal_load_fn_p.
2056 (vectorizable_load): Likewise. Handle gather load internal
2058 (vectorizable_store): Update call to check_load_store_masking.
2059 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2060 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2061 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2062 (aarch64_gather_scale_operand_d): New predicates.
2063 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2064 (mask_gather_load<mode>): New insns.
2066 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2067 Alan Hayward <alan.hayward@arm.com>
2068 David Sherwood <david.sherwood@arm.com>
2070 * optabs.def (fold_left_plus_optab): New optab.
2071 * doc/md.texi (fold_left_plus_@var{m}): Document.
2072 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2073 * internal-fn.c (fold_left_direct): Define.
2074 (expand_fold_left_optab_fn): Likewise.
2075 (direct_fold_left_optab_supported_p): Likewise.
2076 * fold-const-call.c (fold_const_fold_left): New function.
2077 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2078 * tree-parloops.c (valid_reduction_p): New function.
2079 (gather_scalar_reductions): Use it.
2080 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2081 (vect_finish_replace_stmt): Declare.
2082 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2083 (needs_fold_left_reduction_p): New function, split out from...
2084 (vect_is_simple_reduction): ...here. Accept reductions that
2085 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2086 (vect_force_simple_reduction): Also store the reduction type in
2087 the assignment's STMT_VINFO_REDUC_TYPE.
2088 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2089 (merge_with_identity): New function.
2090 (vect_expand_fold_left): Likewise.
2091 (vectorize_fold_left_reduction): Likewise.
2092 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2093 scalar phi in place for it. Check for target support and reject
2094 cases that would reassociate the operation. Defer the transform
2095 phase to vectorize_fold_left_reduction.
2096 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2097 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2098 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2100 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2102 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2103 call to ifc_temp_var.
2105 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2106 Alan Hayward <alan.hayward@arm.com>
2107 David Sherwood <david.sherwood@arm.com>
2109 * target.def (legitimize_address_displacement): Take the original
2110 offset as a poly_int.
2111 * targhooks.h (default_legitimize_address_displacement): Update
2113 * targhooks.c (default_legitimize_address_displacement): Likewise.
2114 * doc/tm.texi: Regenerate.
2115 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2116 as an argument, moving assert of ad->disp == ad->disp_term to...
2117 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2118 Try calling targetm.legitimize_address_displacement before expanding
2119 the address rather than afterwards, and adjust for the new interface.
2120 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2121 Match the new hook interface. Handle SVE addresses.
2122 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2125 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2127 * Makefile.in (OBJS): Add early-remat.o.
2128 * target.def (select_early_remat_modes): New hook.
2129 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2130 * doc/tm.texi: Regenerate.
2131 * targhooks.h (default_select_early_remat_modes): Declare.
2132 * targhooks.c (default_select_early_remat_modes): New function.
2133 * timevar.def (TV_EARLY_REMAT): New timevar.
2134 * passes.def (pass_early_remat): New pass.
2135 * tree-pass.h (make_pass_early_remat): Declare.
2136 * early-remat.c: New file.
2137 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2139 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2141 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2142 Alan Hayward <alan.hayward@arm.com>
2143 David Sherwood <david.sherwood@arm.com>
2145 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2146 vfm1 with a bound_epilog parameter.
2147 (vect_do_peeling): Update calls accordingly, and move the prologue
2148 call earlier in the function. Treat the base bound_epilog as 0 for
2149 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2150 this base when peeling for gaps.
2151 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2152 with fully-masked loops.
2153 (vect_estimate_min_profitable_iters): Handle the single peeled
2154 iteration in that case.
2156 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2157 Alan Hayward <alan.hayward@arm.com>
2158 David Sherwood <david.sherwood@arm.com>
2160 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2161 single-element interleaving even if the size is not a power of 2.
2162 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2163 accesses for single-element interleaving if the group size is
2166 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2167 Alan Hayward <alan.hayward@arm.com>
2168 David Sherwood <david.sherwood@arm.com>
2170 * doc/md.texi (fold_extract_last_@var{m}): Document.
2171 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2172 * optabs.def (fold_extract_last_optab): New optab.
2173 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2174 * internal-fn.c (fold_extract_direct): New macro.
2175 (expand_fold_extract_optab_fn): Likewise.
2176 (direct_fold_extract_optab_supported_p): Likewise.
2177 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2178 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2179 EXTRACT_LAST_REDUCTION.
2180 (get_initial_def_for_reduction): Do not create an initial vector
2181 for EXTRACT_LAST_REDUCTION reductions.
2182 (vectorizable_reduction): Leave the scalar phi in place for
2183 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2184 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2185 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2186 transform phase to vectorizable_condition.
2187 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2189 (vect_finish_stmt_generation): ...here.
2190 (vect_finish_replace_stmt): New function.
2191 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2192 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2194 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2196 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2197 Alan Hayward <alan.hayward@arm.com>
2198 David Sherwood <david.sherwood@arm.com>
2200 * doc/md.texi (extract_last_@var{m}): Document.
2201 * optabs.def (extract_last_optab): New optab.
2202 * internal-fn.def (EXTRACT_LAST): New internal function.
2203 * internal-fn.c (cond_unary_direct): New macro.
2204 (expand_cond_unary_optab_fn): Likewise.
2205 (direct_cond_unary_optab_supported_p): Likewise.
2206 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2207 loops using EXTRACT_LAST.
2208 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2209 (extract_last_<mode>): ...this optab.
2210 (vec_extract<mode><Vel>): Update accordingly.
2212 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2213 Alan Hayward <alan.hayward@arm.com>
2214 David Sherwood <david.sherwood@arm.com>
2216 * target.def (empty_mask_is_expensive): New hook.
2217 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2218 * doc/tm.texi: Regenerate.
2219 * targhooks.h (default_empty_mask_is_expensive): Declare.
2220 * targhooks.c (default_empty_mask_is_expensive): New function.
2221 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2222 if the target says that empty masks are expensive.
2223 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2225 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2227 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2228 Alan Hayward <alan.hayward@arm.com>
2229 David Sherwood <david.sherwood@arm.com>
2231 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2232 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2233 (vect_use_loop_mask_for_alignment_p): New function.
2234 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2235 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2236 niters_skip argument. Make sure that the first niters_skip elements
2237 of the first iteration are inactive.
2238 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2239 Update call to vect_set_loop_masks_directly.
2240 (get_misalign_in_elems): New function, split out from...
2241 (vect_gen_prolog_loop_niters): ...here.
2242 (vect_update_init_of_dr): Take a code argument that specifies whether
2243 the adjustment should be added or subtracted.
2244 (vect_update_init_of_drs): Likewise.
2245 (vect_prepare_for_masked_peels): New function.
2246 (vect_do_peeling): Skip prologue peeling if we're using a mask
2247 instead. Update call to vect_update_inits_of_drs.
2248 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2250 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2251 alignment. Do not include the number of peeled iterations in
2252 the minimum threshold in that case.
2253 (vectorizable_induction): Adjust the start value down by
2254 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2255 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2256 Take the number of skipped iterations into account when calculating
2258 * tree-vect-stmts.c (vect_gen_while_not): New function.
2260 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2261 Alan Hayward <alan.hayward@arm.com>
2262 David Sherwood <david.sherwood@arm.com>
2264 * doc/sourcebuild.texi (vect_fully_masked): Document.
2265 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2267 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2269 (vect_analyze_loop_2): ...here. Don't check the vectorization
2270 factor against the number of loop iterations if the loop is
2273 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2274 Alan Hayward <alan.hayward@arm.com>
2275 David Sherwood <david.sherwood@arm.com>
2277 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2278 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2279 (dump_groups): Update accordingly.
2280 (iv_use::mem_type): New member variable.
2281 (address_p): New function.
2282 (record_use): Add a mem_type argument and initialize the new
2284 (record_group_use): Add a mem_type argument. Use address_p.
2285 Remove obsolete null checks of base_object. Update call to record_use.
2286 (find_interesting_uses_op): Update call to record_group_use.
2287 (find_interesting_uses_cond): Likewise.
2288 (find_interesting_uses_address): Likewise.
2289 (get_mem_type_for_internal_fn): New function.
2290 (find_address_like_use): Likewise.
2291 (find_interesting_uses_stmt): Try find_address_like_use before
2292 calling find_interesting_uses_op.
2293 (addr_offset_valid_p): Use the iv mem_type field as the type
2294 of the addressed memory.
2295 (add_autoinc_candidates): Likewise.
2296 (get_address_cost): Likewise.
2297 (split_small_address_groups_p): Use address_p.
2298 (split_address_groups): Likewise.
2299 (add_iv_candidate_for_use): Likewise.
2300 (autoinc_possible_for_pair): Likewise.
2301 (rewrite_groups): Likewise.
2302 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2303 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2304 (get_alias_ptr_type_for_ptr_address): New function.
2305 (rewrite_use_address): Rewrite address uses in calls that were
2306 identified by find_address_like_use.
2308 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2309 Alan Hayward <alan.hayward@arm.com>
2310 David Sherwood <david.sherwood@arm.com>
2312 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2314 * gimple-expr.h (is_gimple_addressable: Likewise.
2315 * gimple-expr.c (is_gimple_address): Likewise.
2316 * internal-fn.c (expand_call_mem_ref): New function.
2317 (expand_mask_load_optab_fn): Use it.
2318 (expand_mask_store_optab_fn): Likewise.
2320 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2321 Alan Hayward <alan.hayward@arm.com>
2322 David Sherwood <david.sherwood@arm.com>
2324 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2325 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2326 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2327 (cond_umax@var{mode}): Document.
2328 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2329 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2330 (cond_umin_optab, cond_umax_optab): New optabs.
2331 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2332 (COND_IOR, COND_XOR): New internal functions.
2333 * internal-fn.h (get_conditional_internal_fn): Declare.
2334 * internal-fn.c (cond_binary_direct): New macro.
2335 (expand_cond_binary_optab_fn): Likewise.
2336 (direct_cond_binary_optab_supported_p): Likewise.
2337 (get_conditional_internal_fn): New function.
2338 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2339 Cope with reduction statements that are vectorized as calls rather
2341 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2342 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2343 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2344 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2345 (UNSPEC_COND_EOR): New unspecs.
2346 (optab): Add mappings for them.
2347 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2348 (sve_int_op, sve_fp_op): New int attributes.
2350 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2351 Alan Hayward <alan.hayward@arm.com>
2352 David Sherwood <david.sherwood@arm.com>
2354 * optabs.def (while_ult_optab): New optab.
2355 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2356 * internal-fn.def (WHILE_ULT): New internal function.
2357 * internal-fn.h (direct_internal_fn_supported_p): New override
2358 that takes two types as argument.
2359 * internal-fn.c (while_direct): New macro.
2360 (expand_while_optab_fn): New function.
2361 (convert_optab_supported_p): Likewise.
2362 (direct_while_optab_supported_p): New macro.
2363 * wide-int.h (wi::udiv_ceil): New function.
2364 * tree-vectorizer.h (rgroup_masks): New structure.
2365 (vec_loop_masks): New typedef.
2366 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2368 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2369 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2370 (vect_max_vf): New function.
2371 (slpeel_make_loop_iterate_ntimes): Delete.
2372 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2373 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2374 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2375 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2376 internal-fn.h, stor-layout.h and optabs-query.h.
2377 (vect_set_loop_mask): New function.
2378 (add_preheader_seq): Likewise.
2379 (add_header_seq): Likewise.
2380 (interleave_supported_p): Likewise.
2381 (vect_maybe_permute_loop_masks): Likewise.
2382 (vect_set_loop_masks_directly): Likewise.
2383 (vect_set_loop_condition_masked): Likewise.
2384 (vect_set_loop_condition_unmasked): New function, split out from
2385 slpeel_make_loop_iterate_ntimes.
2386 (slpeel_make_loop_iterate_ntimes): Rename to..
2387 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2388 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2389 (vect_do_peeling): Update call accordingly.
2390 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2392 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2393 mask_compare_type, can_fully_mask_p and fully_masked_p.
2394 (release_vec_loop_masks): New function.
2395 (_loop_vec_info): Use it to free the loop masks.
2396 (can_produce_all_loop_masks_p): New function.
2397 (vect_get_max_nscalars_per_iter): Likewise.
2398 (vect_verify_full_masking): Likewise.
2399 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2400 retries, and free the mask rgroups before retrying. Check loop-wide
2401 reasons for disallowing fully-masked loops. Make the final decision
2402 about whether use a fully-masked loop or not.
2403 (vect_estimate_min_profitable_iters): Do not assume that peeling
2404 for the number of iterations will be needed for fully-masked loops.
2405 (vectorizable_reduction): Disable fully-masked loops.
2406 (vectorizable_live_operation): Likewise.
2407 (vect_halve_mask_nunits): New function.
2408 (vect_double_mask_nunits): Likewise.
2409 (vect_record_loop_mask): Likewise.
2410 (vect_get_loop_mask): Likewise.
2411 (vect_transform_loop): Handle the case in which the final loop
2412 iteration might handle a partial vector. Call vect_set_loop_condition
2413 instead of slpeel_make_loop_iterate_ntimes.
2414 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2415 (check_load_store_masking): New function.
2416 (prepare_load_store_mask): Likewise.
2417 (vectorizable_store): Handle fully-masked loops.
2418 (vectorizable_load): Likewise.
2419 (supportable_widening_operation): Use vect_halve_mask_nunits for
2421 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2422 (vect_gen_while): New function.
2423 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2424 (aarch64_uqdec<mode>): New insn.
2426 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2427 Alan Hayward <alan.hayward@arm.com>
2428 David Sherwood <david.sherwood@arm.com>
2430 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2431 (reduc_xor_scal_optab): New optabs.
2432 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2433 (reduc_xor_scal_@var{m}): Document.
2434 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2435 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2437 * fold-const-call.c (fold_const_call): Handle them.
2438 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2439 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2440 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2441 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2442 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2443 (UNSPEC_XORV): New unspecs.
2444 (optab): Add entries for them.
2445 (BITWISEV): New int iterator.
2446 (bit_reduc_op): New int attributes.
2448 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2449 Alan Hayward <alan.hayward@arm.com>
2450 David Sherwood <david.sherwood@arm.com>
2452 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2453 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2454 * optabs.def (vec_shl_insert_optab): New optab.
2455 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2456 (duplicate_and_interleave): Likewise.
2457 * tree-vect-loop.c: Include internal-fn.h.
2458 (neutral_op_for_slp_reduction): New function, split out from
2459 get_initial_defs_for_reduction.
2460 (get_initial_def_for_reduction): Handle option 2 for variable-length
2461 vectors by loading the neutral value into a vector and then shifting
2462 the initial value into element 0.
2463 (get_initial_defs_for_reduction): Replace the code argument with
2464 the neutral value calculated by neutral_op_for_slp_reduction.
2465 Use gimple_build_vector for constant-length vectors.
2466 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2467 but the first group_size elements have a neutral value.
2468 Use duplicate_and_interleave otherwise.
2469 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2470 Update call to get_initial_defs_for_reduction. Handle SLP
2471 reductions for variable-length vectors by creating one vector
2472 result for each scalar result, with the elements associated
2473 with other scalar results stubbed out with the neutral value.
2474 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2475 Require IFN_VEC_SHL_INSERT for double reductions on
2476 variable-length vectors, or SLP reductions that have
2477 a neutral value. Require can_duplicate_and_interleave_p
2478 support for variable-length unchained SLP reductions if there
2479 is no neutral value, such as for MIN/MAX reductions. Also require
2480 the number of vector elements to be a multiple of the number of
2481 SLP statements when doing variable-length unchained SLP reductions.
2482 Update call to vect_create_epilog_for_reduction.
2483 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2484 and remove initial values.
2485 (duplicate_and_interleave): Make public.
2486 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2487 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2489 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2490 Alan Hayward <alan.hayward@arm.com>
2491 David Sherwood <david.sherwood@arm.com>
2493 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2494 (can_duplicate_and_interleave_p): New function.
2495 (vect_get_and_check_slp_defs): Take the vector of statements
2496 rather than just the current one. Remove excess parentheses.
2497 Restriction rejectinon of vect_constant_def and vect_external_def
2498 for variable-length vectors to boolean types, or types for which
2499 can_duplicate_and_interleave_p is false.
2500 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2501 (duplicate_and_interleave): New function.
2502 (vect_get_constant_vectors): Use gimple_build_vector for
2503 constant-length vectors and suitable variable-length constant
2504 vectors. Use duplicate_and_interleave for other variable-length
2505 vectors. Don't defer the update when inserting new statements.
2507 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2508 Alan Hayward <alan.hayward@arm.com>
2509 David Sherwood <david.sherwood@arm.com>
2511 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2512 min_profitable_iters doesn't go negative.
2514 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2515 Alan Hayward <alan.hayward@arm.com>
2516 David Sherwood <david.sherwood@arm.com>
2518 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2519 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2520 * optabs.def (vec_mask_load_lanes_optab): New optab.
2521 (vec_mask_store_lanes_optab): Likewise.
2522 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2523 (MASK_STORE_LANES): Likewise.
2524 * internal-fn.c (mask_load_lanes_direct): New macro.
2525 (mask_store_lanes_direct): Likewise.
2526 (expand_mask_load_optab_fn): Handle masked operations.
2527 (expand_mask_load_lanes_optab_fn): New macro.
2528 (expand_mask_store_optab_fn): Handle masked operations.
2529 (expand_mask_store_lanes_optab_fn): New macro.
2530 (direct_mask_load_lanes_optab_supported_p): Likewise.
2531 (direct_mask_store_lanes_optab_supported_p): Likewise.
2532 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2534 (vect_load_lanes_supported): Likewise.
2535 * tree-vect-data-refs.c (strip_conversion): New function.
2536 (can_group_stmts_p): Likewise.
2537 (vect_analyze_data_ref_accesses): Use it instead of checking
2538 for a pair of assignments.
2539 (vect_store_lanes_supported): Take a masked_p parameter.
2540 (vect_load_lanes_supported): Likewise.
2541 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2542 vect_store_lanes_supported and vect_load_lanes_supported.
2543 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2544 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2545 parameter. Don't allow gaps for masked accesses.
2546 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2547 and vect_load_lanes_supported.
2548 (get_load_store_type): Take a masked_p parameter and update
2549 call to get_group_load_store_type.
2550 (vectorizable_store): Update call to get_load_store_type.
2551 Handle IFN_MASK_STORE_LANES.
2552 (vectorizable_load): Update call to get_load_store_type.
2553 Handle IFN_MASK_LOAD_LANES.
2555 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2556 Alan Hayward <alan.hayward@arm.com>
2557 David Sherwood <david.sherwood@arm.com>
2559 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2561 * config/aarch64/aarch64-protos.h
2562 (aarch64_sve_struct_memory_operand_p): Declare.
2563 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2564 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2565 (VPRED, vpred): Handle SVE structure modes.
2566 * config/aarch64/constraints.md (Utx): New constraint.
2567 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2568 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2569 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2570 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2571 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2572 structure modes. Split into pieces after RA.
2573 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2574 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2576 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2577 SVE structure modes.
2578 (aarch64_classify_address): Likewise.
2579 (sizetochar): Move earlier in file.
2580 (aarch64_print_operand): Handle SVE register lists.
2581 (aarch64_array_mode): New function.
2582 (aarch64_sve_struct_memory_operand_p): Likewise.
2583 (TARGET_ARRAY_MODE): Redefine.
2585 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2586 Alan Hayward <alan.hayward@arm.com>
2587 David Sherwood <david.sherwood@arm.com>
2589 * target.def (array_mode): New target hook.
2590 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2591 * doc/tm.texi: Regenerate.
2592 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2593 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2594 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2596 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2599 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2600 Alan Hayward <alan.hayward@arm.com>
2601 David Sherwood <david.sherwood@arm.com>
2603 * fold-const.c (fold_binary_loc): Check the argument types
2604 rather than the result type when testing for a vector operation.
2606 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2608 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2609 * doc/tm.texi: Regenerate.
2611 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2612 Alan Hayward <alan.hayward@arm.com>
2613 David Sherwood <david.sherwood@arm.com>
2615 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2616 (sve): Document new AArch64 extension.
2617 * doc/md.texi (w): Extend the description of the AArch64
2618 constraint to include SVE vectors.
2619 (Upl, Upa): Document new AArch64 predicate constraints.
2620 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2622 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2623 (msve-vector-bits=): New option.
2624 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2625 SVE when these are disabled.
2626 (sve): New extension.
2627 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2628 modes. Adjust their number of units based on aarch64_sve_vg.
2629 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2630 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2631 aarch64_addr_query_type.
2632 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2633 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2634 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2635 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2636 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2637 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2638 (aarch64_simd_imm_zero_p): Delete.
2639 (aarch64_check_zero_based_sve_index_immediate): Declare.
2640 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2641 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2642 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2643 (aarch64_sve_float_mul_immediate_p): Likewise.
2644 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2646 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2647 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2648 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2649 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2650 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2651 (aarch64_regmode_natural_size): Likewise.
2652 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2653 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2655 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2656 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2657 for VG and the SVE predicate registers.
2658 (V_ALIASES): Add a "z"-prefixed alias.
2659 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2660 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2661 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2662 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2663 (REG_CLASS_NAMES): Add entries for them.
2664 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2665 and the predicate registers.
2666 (aarch64_sve_vg): Declare.
2667 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2668 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2669 (REGMODE_NATURAL_SIZE): Define.
2670 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2672 * config/aarch64/aarch64.c: Include cfgrtl.h.
2673 (simd_immediate_info): Add a constructor for series vectors,
2674 and an associated step field.
2675 (aarch64_sve_vg): New variable.
2676 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2677 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2678 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2679 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2680 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2681 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2682 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2683 (aarch64_get_mask_mode): New functions.
2684 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2685 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2686 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2687 predicate modes and predicate registers. Explicitly restrict
2688 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2689 to store a vector mode if it is recognized by
2690 aarch64_classify_vector_mode.
2691 (aarch64_regmode_natural_size): New function.
2692 (aarch64_hard_regno_caller_save_mode): Return the original mode
2694 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2695 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2696 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2697 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2699 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2700 does not overlap dest if the function is frame-related. Handle
2702 (aarch64_split_add_offset): New function.
2703 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2704 them aarch64_add_offset.
2705 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2706 and update call to aarch64_sub_sp.
2707 (aarch64_add_cfa_expression): New function.
2708 (aarch64_expand_prologue): Pass extra temporary registers to the
2709 functions above. Handle the case in which we need to emit new
2710 DW_CFA_expressions for registers that were originally saved
2711 relative to the stack pointer, but now have to be expressed
2712 relative to the frame pointer.
2713 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2715 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2716 IP0 and IP1 values for SVE frames.
2717 (aarch64_expand_vec_series): New function.
2718 (aarch64_expand_sve_widened_duplicate): Likewise.
2719 (aarch64_expand_sve_const_vector): Likewise.
2720 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2721 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2722 into the register, rather than emitting a SET directly.
2723 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2724 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2725 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2726 (offset_9bit_signed_scaled_p): New functions.
2727 (aarch64_replicate_bitmask_imm): New function.
2728 (aarch64_bitmask_imm): Use it.
2729 (aarch64_cannot_force_const_mem): Reject expressions involving
2730 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2731 (aarch64_classify_index): Handle SVE indices, by requiring
2732 a plain register index with a scale that matches the element size.
2733 (aarch64_classify_address): Handle SVE addresses. Assert that
2734 the mode of the address is VOIDmode or an integer mode.
2735 Update call to aarch64_classify_symbol.
2736 (aarch64_classify_symbolic_expression): Update call to
2737 aarch64_classify_symbol.
2738 (aarch64_const_vec_all_in_range_p): New function.
2739 (aarch64_print_vector_float_operand): Likewise.
2740 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2741 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2742 and the FP immediates 1.0 and 0.5.
2743 (aarch64_print_address_internal): Handle SVE addresses.
2744 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2745 (aarch64_regno_regclass): Handle predicate registers.
2746 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2748 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2749 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2750 (aarch64_convert_sve_vector_bits): New function.
2751 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2752 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2754 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2755 Handle SVE vector and predicate modes. Accept VL-based constants
2756 that need only one temporary register, and VL offsets that require
2757 no temporary registers.
2758 (aarch64_conditional_register_usage): Mark the predicate registers
2759 as fixed if SVE isn't available.
2760 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2761 Return true for SVE vector and predicate modes.
2762 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2763 rather than an unsigned int. Handle SVE modes.
2764 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2766 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2768 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2769 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2770 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2771 (aarch64_sve_float_mul_immediate_p): New functions.
2772 (aarch64_sve_valid_immediate): New function.
2773 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2774 Explicitly reject structure modes. Check for INDEX constants.
2775 Handle PTRUE and PFALSE constants.
2776 (aarch64_check_zero_based_sve_index_immediate): New function.
2777 (aarch64_simd_imm_zero_p): Delete.
2778 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2779 vector modes. Accept constants in the range of CNT[BHWD].
2780 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2781 ask for an Advanced SIMD mode.
2782 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2783 (aarch64_simd_vector_alignment): Handle SVE predicates.
2784 (aarch64_vectorize_preferred_vector_alignment): New function.
2785 (aarch64_simd_vector_alignment_reachable): Use it instead of
2787 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2788 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2790 (MAX_VECT_LEN): Delete.
2791 (expand_vec_perm_d): Add a vec_flags field.
2792 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2793 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2794 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2796 (aarch64_evpc_rev): Rename to...
2797 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2798 (aarch64_evpc_rev_global): New function.
2799 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2800 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2802 (aarch64_evpc_sve_tbl): New function.
2803 (aarch64_expand_vec_perm_const_1): Update after rename of
2804 aarch64_evpc_rev. Handle SVE permutes too, trying
2805 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2806 than aarch64_evpc_tbl.
2807 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2808 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2809 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2810 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2811 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2812 (aarch64_expand_sve_vcond): New functions.
2813 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2814 of aarch64_vector_mode_p.
2815 (aarch64_dwarf_poly_indeterminate_value): New function.
2816 (aarch64_compute_pressure_classes): Likewise.
2817 (aarch64_can_change_mode_class): Likewise.
2818 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2819 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2820 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2821 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2822 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2823 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2824 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2825 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2827 (Dn, Dl, Dr): Accept const as well as const_vector.
2828 (Dz): Likewise. Compare against CONST0_RTX.
2829 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2830 of "vector" where appropriate.
2831 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2832 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2833 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2834 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2835 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2836 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2837 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2838 (v_int_equiv): Extend to SVE modes.
2839 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2841 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2842 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2843 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2844 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2845 (SVE_COND_FP_CMP): New int iterators.
2846 (perm_hilo): Handle the new unpack unspecs.
2847 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2849 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2850 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2851 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2852 (aarch64_equality_operator, aarch64_constant_vector_operand)
2853 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2854 (aarch64_sve_nonimmediate_operand): Likewise.
2855 (aarch64_sve_general_operand): Likewise.
2856 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2857 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2858 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2859 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2860 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2861 (aarch64_sve_float_arith_immediate): Likewise.
2862 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2863 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2864 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2865 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2866 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2867 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2868 (aarch64_sve_float_arith_operand): Likewise.
2869 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2870 (aarch64_sve_float_mul_operand): Likewise.
2871 (aarch64_sve_vec_perm_operand): Likewise.
2872 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2873 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2874 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2875 as well as const_vector.
2876 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2877 in file. Use CONST0_RTX and CONSTM1_RTX.
2878 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2879 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2880 Use aarch64_simd_imm_zero.
2881 * config/aarch64/aarch64-sve.md: New file.
2882 * config/aarch64/aarch64.md: Include it.
2883 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2884 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2885 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2886 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2887 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2888 (sve): New attribute.
2889 (enabled): Disable instructions with the sve attribute unless
2891 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2892 aarch64_expand_mov_immediate.
2893 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2894 CNT[BHSD] immediates.
2895 (movti): Split CONST_POLY_INT moves into two halves.
2896 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2897 Split additions that need a temporary here if the destination
2898 is the stack pointer.
2899 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2900 (*add<mode>3_poly_1): New instruction.
2901 (set_clobber_cc): New expander.
2903 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2905 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2906 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2907 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2908 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2909 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2910 Change innermode from fixed_mode_size to machine_mode.
2911 (simplify_subreg): Update call accordingly. Handle a constant-sized
2912 subreg of a variable-length CONST_VECTOR.
2914 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2915 Alan Hayward <alan.hayward@arm.com>
2916 David Sherwood <david.sherwood@arm.com>
2918 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2919 (add_offset_to_base): New function, split out from...
2920 (create_mem_ref): ...here. When handling a scale other than 1,
2921 check first whether the address is valid without the offset.
2922 Add it into the base if so, leaving the index and scale as-is.
2924 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2927 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2928 fold_for_warn before checking if arg2 is INTEGER_CST.
2930 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2932 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2933 (store_multiple_operation): Delete.
2934 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2935 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2936 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2937 guarded by TARGET_STRING.
2938 (rs6000_output_load_multiple): Delete.
2939 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2940 OPTION_MASK_STRING / TARGET_STRING handling.
2941 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2942 (const rs6000_opt_masks) <"string">: Change mask to 0.
2943 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2944 (MASK_STRING): Delete.
2945 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2947 (load_multiple): Delete.
2954 (store_multiple): Delete.
2961 (movmemsi_8reg): Delete.
2962 (corresponding unnamed define_insn): Delete.
2963 (movmemsi_6reg): Delete.
2964 (corresponding unnamed define_insn): Delete.
2965 (movmemsi_4reg): Delete.
2966 (corresponding unnamed define_insn): Delete.
2967 (movmemsi_2reg): Delete.
2968 (corresponding unnamed define_insn): Delete.
2969 (movmemsi_1reg): Delete.
2970 (corresponding unnamed define_insn): Delete.
2971 * config/rs6000/rs6000.opt (mno-string): New.
2972 (mstring): Replace by deprecation warning stub.
2973 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2975 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2977 * regrename.c (regrename_do_replace): If replacing the same
2978 reg multiple times, try to reuse last created gen_raw_REG.
2981 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2982 main to workaround a bug in GDB.
2984 2018-01-12 Tom de Vries <tom@codesourcery.com>
2987 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2989 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2991 PR rtl-optimization/80481
2992 * ira-color.c (get_cap_member): New function.
2993 (allocnos_conflict_by_live_ranges_p): Use it.
2994 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2995 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2997 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3000 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3001 (*saddl_se_1): Ditto.
3003 (*ssubl_se_1): Ditto.
3005 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3007 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3008 rather than wi::to_widest for DR_INITs.
3009 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3010 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3011 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3013 (vect_analyze_group_access_1): Note that here.
3015 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3017 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3018 polynomial type sizes.
3020 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3022 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3023 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3024 (gimple_add_tmp_var): Likewise.
3026 2018-01-12 Martin Liska <mliska@suse.cz>
3028 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3029 (gimple_alloc_sizes): Likewise.
3030 (dump_gimple_statistics): Use PRIu64 in printf format.
3031 * gimple.h: Change uint64_t to int.
3033 2018-01-12 Martin Liska <mliska@suse.cz>
3035 * tree-core.h: Use uint64_t instead of int.
3036 * tree.c (tree_node_counts): Likewise.
3037 (tree_node_sizes): Likewise.
3038 (dump_tree_statistics): Use PRIu64 in printf format.
3040 2018-01-12 Martin Liska <mliska@suse.cz>
3042 * Makefile.in: As qsort_chk is implemented in vec.c, add
3043 vec.o to linkage of gencfn-macros.
3044 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3045 passing the info to record_node_allocation_statistics.
3046 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3048 * ggc-common.c (struct ggc_usage): Add operator== and use
3049 it in operator< and compare function.
3050 * mem-stats.h (struct mem_usage): Likewise.
3051 * vec.c (struct vec_usage): Remove operator< and compare
3052 function. Can be simply inherited.
3054 2018-01-12 Martin Jambor <mjambor@suse.cz>
3057 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3058 * tree-ssa-math-opts.c: Include domwalk.h.
3059 (convert_mult_to_fma_1): New function.
3060 (fma_transformation_info): New type.
3061 (fma_deferring_state): Likewise.
3062 (cancel_fma_deferring): New function.
3063 (result_of_phi): Likewise.
3064 (last_fma_candidate_feeds_initial_phi): Likewise.
3065 (convert_mult_to_fma): Added deferring logic, split actual
3066 transformation to convert_mult_to_fma_1.
3067 (math_opts_dom_walker): New type.
3068 (math_opts_dom_walker::after_dom_children): New method, body moved
3069 here from pass_optimize_widening_mul::execute, added deferring logic
3071 (pass_optimize_widening_mul::execute): Moved most of code to
3072 math_opts_dom_walker::after_dom_children.
3073 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3074 * config/i386/i386.c (ix86_option_override_internal): Added
3075 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3077 2018-01-12 Richard Biener <rguenther@suse.de>
3080 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3081 inline instance vars.
3083 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3086 * config/rx/rx.c (rx_is_restricted_memory_address):
3089 2018-01-12 Richard Biener <rguenther@suse.de>
3091 PR tree-optimization/80846
3092 * target.def (split_reduction): New target hook.
3093 * targhooks.c (default_split_reduction): New function.
3094 * targhooks.h (default_split_reduction): Declare.
3095 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3096 target requests first reduce vectors by combining low and high
3098 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3099 (get_vectype_for_scalar_type_and_size): Export.
3100 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3101 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3102 * doc/tm.texi: Regenerate.
3103 * config/i386/i386.c (ix86_split_reduction): Implement
3104 TARGET_VECTORIZE_SPLIT_REDUCTION.
3106 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3109 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3110 in PIC mode except for TARGET_VXWORKS_RTP.
3111 * config/sparc/sparc.c: Include cfgrtl.h.
3112 (TARGET_INIT_PIC_REG): Define.
3113 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3114 (sparc_pic_register_p): New predicate.
3115 (sparc_legitimate_address_p): Use it.
3116 (sparc_legitimize_pic_address): Likewise.
3117 (sparc_delegitimize_address): Likewise.
3118 (sparc_mode_dependent_address_p): Likewise.
3119 (gen_load_pcrel_sym): Remove 4th parameter.
3120 (load_got_register): Adjust call to above. Remove obsolete stuff.
3121 (sparc_expand_prologue): Do not call load_got_register here.
3122 (sparc_flat_expand_prologue): Likewise.
3123 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3124 (sparc_use_pseudo_pic_reg): New function.
3125 (sparc_init_pic_reg): Likewise.
3126 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3127 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3129 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3131 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3132 Add item for branch_cost.
3134 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3136 PR rtl-optimization/83565
3137 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3138 not extend the result to a larger mode for rotate operations.
3139 (num_sign_bit_copies1): Likewise.
3141 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3144 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3146 Use values-Xc.o for -pedantic.
3147 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3149 2018-01-12 Martin Liska <mliska@suse.cz>
3152 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3154 (possible_polymorphic_call_targets): Use it.
3155 (ipa_devirt): Likewise.
3157 2018-01-12 Martin Liska <mliska@suse.cz>
3159 * profile-count.h (enum profile_quality): Use 0 as invalid
3160 enum value of profile_quality.
3162 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3164 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3165 -mext-string options.
3167 2018-01-12 Richard Biener <rguenther@suse.de>
3169 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3170 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3171 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3173 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3175 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3177 * configure.ac (--with-long-double-format): Add support for the
3178 configuration option to change the default long double format on
3180 * config.gcc (powerpc*-linux*-*): Likewise.
3181 * configure: Regenerate.
3182 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3183 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3184 used without modification.
3186 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3188 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3189 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3190 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3191 MISC_BUILTIN_SPEC_BARRIER.
3192 (rs6000_init_builtins): Likewise.
3193 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3195 (speculation_barrier): New define_insn.
3196 * doc/extend.texi: Document __builtin_speculation_barrier.
3198 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3201 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3202 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3203 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3205 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3206 integral modes instead of "ss" and "sd".
3207 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3208 vectors with 32-bit and 64-bit elements.
3209 (vecdupssescalarmodesuffix): New mode attribute.
3210 (vec_dup<mode>): Use it.
3212 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3215 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3216 frame if argument is passed on stack.
3218 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3221 * ree.c (combine_reaching_defs): Optimize also
3222 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3223 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3225 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3228 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3230 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3233 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3234 after they are computed.
3236 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3238 PR tree-optimization/83695
3239 * gimple-loop-linterchange.cc
3240 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3241 reset cached scev information after interchange.
3242 (pass_linterchange::execute): Remove call to scev_reset_htab.
3244 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3246 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3247 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3248 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3249 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3250 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3251 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3252 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3253 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3254 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3255 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3256 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3257 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3258 (V_lane_reg): Likewise.
3259 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3261 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3262 (vfmal_lane_low<mode>_intrinsic,
3263 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3264 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3265 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3266 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3267 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3268 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3270 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3272 * config/arm/arm-cpus.in (fp16fml): New feature.
3273 (ALL_SIMD): Add fp16fml.
3274 (armv8.2-a): Add fp16fml as an option.
3275 (armv8.3-a): Likewise.
3276 (armv8.4-a): Add fp16fml as part of fp16.
3277 * config/arm/arm.h (TARGET_FP16FML): Define.
3278 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3280 * config/arm/arm-modes.def (V2HF): Define.
3281 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3282 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3283 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3284 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3285 vfmsl_low, vfmsl_high): New set of builtins.
3286 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3287 (vfml_op): New code attribute.
3288 (VFMLHALVES): New int iterator.
3289 (VFML, VFMLSEL): New mode attributes.
3290 (V_reg): Define mapping for V2HF.
3291 (V_hi, V_lo): New mode attributes.
3292 (VF_constraint): Likewise.
3293 (vfml_half, vfml_half_selector): New int attributes.
3294 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3296 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3297 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3299 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3300 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3301 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3302 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3304 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3305 Document new effective target and option set.
3307 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3309 * config/arm/arm-cpus.in (armv8_4): New feature.
3310 (ARMv8_4a): New fgroup.
3311 (armv8.4-a): New arch.
3312 * config/arm/arm-tables.opt: Regenerate.
3313 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3314 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3315 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3316 Add matching rules for -march=armv8.4-a and extensions.
3317 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3319 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3322 * config/rx/rx.md (BW): New mode attribute.
3323 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3325 2018-01-11 Richard Biener <rguenther@suse.de>
3327 PR tree-optimization/83435
3328 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3329 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3330 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3332 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3333 Alan Hayward <alan.hayward@arm.com>
3334 David Sherwood <david.sherwood@arm.com>
3336 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3338 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3339 (aarch64_print_address_internal): Use it to check for a zero offset.
3341 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3342 Alan Hayward <alan.hayward@arm.com>
3343 David Sherwood <david.sherwood@arm.com>
3345 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3346 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3347 Return a poly_int64 rather than a HOST_WIDE_INT.
3348 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3349 rather than a HOST_WIDE_INT.
3350 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3351 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3352 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3353 final_offset from HOST_WIDE_INT to poly_int64.
3354 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3355 to_constant when getting the number of units in an Advanced SIMD
3357 (aarch64_builtin_vectorized_function): Check for a constant number
3359 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3361 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3362 attribute instead of GET_MODE_NUNITS.
3363 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3364 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3365 GET_MODE_SIZE for fixed-size registers.
3366 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3367 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3368 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3369 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3370 (aarch64_print_operand, aarch64_print_address_internal)
3371 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3372 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3373 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3374 Handle polynomial GET_MODE_SIZE.
3375 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3376 wider than SImode without modification.
3377 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3378 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3379 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3380 passing and returning SVE modes.
3381 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3382 rather than GEN_INT.
3383 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3384 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3385 (aarch64_allocate_and_probe_stack_space): Likewise.
3386 (aarch64_layout_frame): Cope with polynomial offsets.
3387 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3388 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3390 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3391 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3392 poly_int64 rather than a HOST_WIDE_INT.
3393 (aarch64_get_separate_components, aarch64_process_components)
3394 (aarch64_expand_prologue, aarch64_expand_epilogue)
3395 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3396 (aarch64_anchor_offset): New function, split out from...
3397 (aarch64_legitimize_address): ...here.
3398 (aarch64_builtin_vectorization_cost): Handle polynomial
3399 TYPE_VECTOR_SUBPARTS.
3400 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3402 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3403 number of elements from the PARALLEL rather than the mode.
3404 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3405 rather than GET_MODE_BITSIZE.
3406 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3407 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3408 (aarch64_expand_vec_perm_const_1): Handle polynomial
3409 d->perm.length () and d->perm elements.
3410 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3411 Apply to_constant to d->perm elements.
3412 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3413 polynomial CONST_VECTOR_NUNITS.
3414 (aarch64_move_pointer): Take amount as a poly_int64 rather
3416 (aarch64_progress_pointer): Avoid temporary variable.
3417 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3418 the mode attribute instead of GET_MODE.
3420 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3421 Alan Hayward <alan.hayward@arm.com>
3422 David Sherwood <david.sherwood@arm.com>
3424 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3425 x exists before using it.
3426 (aarch64_add_constant_internal): Rename to...
3427 (aarch64_add_offset_1): ...this. Replace regnum with separate
3428 src and dest rtxes. Handle the case in which they're different,
3429 including when the offset is zero. Replace scratchreg with an rtx.
3430 Use 2 additions if there is no spare register into which we can
3431 move a 16-bit constant.
3432 (aarch64_add_constant): Delete.
3433 (aarch64_add_offset): Replace reg with separate src and dest
3434 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3435 Use aarch64_add_offset_1.
3436 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3437 an rtx rather than an int. Take the delta as a poly_int64
3438 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3439 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3440 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3441 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3442 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3444 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3445 aarch64_add_constant.
3447 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3449 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3450 Use scalar_float_mode.
3452 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3454 * config/aarch64/aarch64-simd.md
3455 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3456 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3457 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3458 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3459 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3460 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3461 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3462 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3463 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3464 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3466 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3469 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3470 targ_options->x_arm_arch_string is non NULL.
3472 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3474 * config/aarch64/aarch64.h
3475 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3477 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3480 * expmed.c (emit_store_flag_force): Swap if const op0
3481 and change VOIDmode to mode of op0.
3483 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3485 PR rtl-optimization/83761
3486 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3487 than bytes to mode_for_size.
3489 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3492 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3493 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3496 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3499 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3500 when in layout mode.
3501 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3502 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3505 2018-01-10 Michael Collison <michael.collison@arm.com>
3507 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3508 * config/aarch64/aarch64-option-extension.def: Add
3509 AARCH64_OPT_EXTENSION of 'fp16fml'.
3510 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3511 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3512 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3513 * config/aarch64/constraints.md (Ui7): New constraint.
3514 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3515 (VFMLA_SEL_W): Ditto.
3518 (VFMLA16_LOW): New int iterator.
3519 (VFMLA16_HIGH): Ditto.
3520 (UNSPEC_FMLAL): New unspec.
3521 (UNSPEC_FMLSL): Ditto.
3522 (UNSPEC_FMLAL2): Ditto.
3523 (UNSPEC_FMLSL2): Ditto.
3524 (f16mac): New code attribute.
3525 * config/aarch64/aarch64-simd-builtins.def
3526 (aarch64_fmlal_lowv2sf): Ditto.
3527 (aarch64_fmlsl_lowv2sf): Ditto.
3528 (aarch64_fmlalq_lowv4sf): Ditto.
3529 (aarch64_fmlslq_lowv4sf): Ditto.
3530 (aarch64_fmlal_highv2sf): Ditto.
3531 (aarch64_fmlsl_highv2sf): Ditto.
3532 (aarch64_fmlalq_highv4sf): Ditto.
3533 (aarch64_fmlslq_highv4sf): Ditto.
3534 (aarch64_fmlal_lane_lowv2sf): Ditto.
3535 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3536 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3537 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3538 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3539 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3540 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3541 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3542 (aarch64_fmlal_lane_highv2sf): Ditto.
3543 (aarch64_fmlsl_lane_highv2sf): Ditto.
3544 (aarch64_fmlal_laneq_highv2sf): Ditto.
3545 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3546 (aarch64_fmlalq_lane_highv4sf): Ditto.
3547 (aarch64_fmlsl_lane_highv4sf): Ditto.
3548 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3549 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3550 * config/aarch64/aarch64-simd.md:
3551 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3552 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3553 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3554 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3555 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3556 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3557 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3558 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3559 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3560 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3561 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3562 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3563 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3564 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3565 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3566 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3567 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3568 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3569 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3570 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3571 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3572 (vfmlsl_low_u32): Ditto.
3573 (vfmlalq_low_u32): Ditto.
3574 (vfmlslq_low_u32): Ditto.
3575 (vfmlal_high_u32): Ditto.
3576 (vfmlsl_high_u32): Ditto.
3577 (vfmlalq_high_u32): Ditto.
3578 (vfmlslq_high_u32): Ditto.
3579 (vfmlal_lane_low_u32): Ditto.
3580 (vfmlsl_lane_low_u32): Ditto.
3581 (vfmlal_laneq_low_u32): Ditto.
3582 (vfmlsl_laneq_low_u32): Ditto.
3583 (vfmlalq_lane_low_u32): Ditto.
3584 (vfmlslq_lane_low_u32): Ditto.
3585 (vfmlalq_laneq_low_u32): Ditto.
3586 (vfmlslq_laneq_low_u32): Ditto.
3587 (vfmlal_lane_high_u32): Ditto.
3588 (vfmlsl_lane_high_u32): Ditto.
3589 (vfmlal_laneq_high_u32): Ditto.
3590 (vfmlsl_laneq_high_u32): Ditto.
3591 (vfmlalq_lane_high_u32): Ditto.
3592 (vfmlslq_lane_high_u32): Ditto.
3593 (vfmlalq_laneq_high_u32): Ditto.
3594 (vfmlslq_laneq_high_u32): Ditto.
3595 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3596 (AARCH64_FL_FOR_ARCH8_4): New.
3597 (AARCH64_ISA_F16FML): New ISA flag.
3598 (TARGET_F16FML): New feature flag for fp16fml.
3599 (doc/invoke.texi): Document new fp16fml option.
3601 2018-01-10 Michael Collison <michael.collison@arm.com>
3603 * config/aarch64/aarch64-builtins.c:
3604 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3605 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3606 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3607 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3608 (AARCH64_ISA_SHA3): New ISA flag.
3609 (TARGET_SHA3): New feature flag for sha3.
3610 * config/aarch64/iterators.md (sha512_op): New int attribute.
3611 (CRYPTO_SHA512): New int iterator.
3612 (UNSPEC_SHA512H): New unspec.
3613 (UNSPEC_SHA512H2): Ditto.
3614 (UNSPEC_SHA512SU0): Ditto.
3615 (UNSPEC_SHA512SU1): Ditto.
3616 * config/aarch64/aarch64-simd-builtins.def
3617 (aarch64_crypto_sha512hqv2di): New builtin.
3618 (aarch64_crypto_sha512h2qv2di): Ditto.
3619 (aarch64_crypto_sha512su0qv2di): Ditto.
3620 (aarch64_crypto_sha512su1qv2di): Ditto.
3621 (aarch64_eor3qv8hi): Ditto.
3622 (aarch64_rax1qv2di): Ditto.
3623 (aarch64_xarqv2di): Ditto.
3624 (aarch64_bcaxqv8hi): Ditto.
3625 * config/aarch64/aarch64-simd.md:
3626 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3627 (aarch64_crypto_sha512su0qv2di): Ditto.
3628 (aarch64_crypto_sha512su1qv2di): Ditto.
3629 (aarch64_eor3qv8hi): Ditto.
3630 (aarch64_rax1qv2di): Ditto.
3631 (aarch64_xarqv2di): Ditto.
3632 (aarch64_bcaxqv8hi): Ditto.
3633 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3634 (vsha512h2q_u64): Ditto.
3635 (vsha512su0q_u64): Ditto.
3636 (vsha512su1q_u64): Ditto.
3637 (veor3q_u16): Ditto.
3638 (vrax1q_u64): Ditto.
3640 (vbcaxq_u16): Ditto.
3641 * config/arm/types.md (crypto_sha512): New type attribute.
3642 (crypto_sha3): Ditto.
3643 (doc/invoke.texi): Document new sha3 option.
3645 2018-01-10 Michael Collison <michael.collison@arm.com>
3647 * config/aarch64/aarch64-builtins.c:
3648 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3649 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3650 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3651 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3652 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3653 (AARCH64_ISA_SM4): New ISA flag.
3654 (TARGET_SM4): New feature flag for sm4.
3655 * config/aarch64/aarch64-simd-builtins.def
3656 (aarch64_sm3ss1qv4si): Ditto.
3657 (aarch64_sm3tt1aq4si): Ditto.
3658 (aarch64_sm3tt1bq4si): Ditto.
3659 (aarch64_sm3tt2aq4si): Ditto.
3660 (aarch64_sm3tt2bq4si): Ditto.
3661 (aarch64_sm3partw1qv4si): Ditto.
3662 (aarch64_sm3partw2qv4si): Ditto.
3663 (aarch64_sm4eqv4si): Ditto.
3664 (aarch64_sm4ekeyqv4si): Ditto.
3665 * config/aarch64/aarch64-simd.md:
3666 (aarch64_sm3ss1qv4si): Ditto.
3667 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3668 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3669 (aarch64_sm4eqv4si): Ditto.
3670 (aarch64_sm4ekeyqv4si): Ditto.
3671 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3672 (sm3part_op): Ditto.
3673 (CRYPTO_SM3TT): Ditto.
3674 (CRYPTO_SM3PART): Ditto.
3675 (UNSPEC_SM3SS1): New unspec.
3676 (UNSPEC_SM3TT1A): Ditto.
3677 (UNSPEC_SM3TT1B): Ditto.
3678 (UNSPEC_SM3TT2A): Ditto.
3679 (UNSPEC_SM3TT2B): Ditto.
3680 (UNSPEC_SM3PARTW1): Ditto.
3681 (UNSPEC_SM3PARTW2): Ditto.
3682 (UNSPEC_SM4E): Ditto.
3683 (UNSPEC_SM4EKEY): Ditto.
3684 * config/aarch64/constraints.md (Ui2): New constraint.
3685 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3686 * config/arm/types.md (crypto_sm3): New type attribute.
3687 (crypto_sm4): Ditto.
3688 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3689 (vsm3tt1aq_u32): Ditto.
3690 (vsm3tt1bq_u32): Ditto.
3691 (vsm3tt2aq_u32): Ditto.
3692 (vsm3tt2bq_u32): Ditto.
3693 (vsm3partw1q_u32): Ditto.
3694 (vsm3partw2q_u32): Ditto.
3695 (vsm4eq_u32): Ditto.
3696 (vsm4ekeyq_u32): Ditto.
3697 (doc/invoke.texi): Document new sm4 option.
3699 2018-01-10 Michael Collison <michael.collison@arm.com>
3701 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3702 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3703 (AARCH64_FL_FOR_ARCH8_4): New.
3704 (AARCH64_FL_V8_4): New flag.
3705 (doc/invoke.texi): Document new armv8.4-a option.
3707 2018-01-10 Michael Collison <michael.collison@arm.com>
3709 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3710 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3711 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3712 * config/aarch64/aarch64-option-extension.def: Add
3713 AARCH64_OPT_EXTENSION of 'sha2'.
3714 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3715 (crypto): Disable sha2 and aes if crypto disabled.
3716 (crypto): Enable aes and sha2 if enabled.
3717 (simd): Disable sha2 and aes if simd disabled.
3718 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3720 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3721 (TARGET_SHA2): New feature flag for sha2.
3722 (TARGET_AES): New feature flag for aes.
3723 * config/aarch64/aarch64-simd.md:
3724 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3725 conditional on TARGET_AES.
3726 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3727 (aarch64_crypto_sha1hsi): Make pattern conditional
3729 (aarch64_crypto_sha1hv4si): Ditto.
3730 (aarch64_be_crypto_sha1hv4si): Ditto.
3731 (aarch64_crypto_sha1su1v4si): Ditto.
3732 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3733 (aarch64_crypto_sha1su0v4si): Ditto.
3734 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3735 (aarch64_crypto_sha256su0v4si): Ditto.
3736 (aarch64_crypto_sha256su1v4si): Ditto.
3737 (doc/invoke.texi): Document new aes and sha2 options.
3739 2018-01-10 Martin Sebor <msebor@redhat.com>
3741 PR tree-optimization/83781
3742 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3745 2018-01-11 Martin Sebor <msebor@gmail.com>
3746 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3748 PR tree-optimization/83501
3749 PR tree-optimization/81703
3751 * tree-ssa-strlen.c (get_string_cst): Rename...
3752 (get_string_len): ...to this. Handle global constants.
3753 (handle_char_store): Adjust.
3755 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3756 Jim Wilson <jimw@sifive.com>
3758 * config/riscv/riscv-protos.h (riscv_output_return): New.
3759 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3760 (riscv_attribute_table, riscv_output_return),
3761 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3762 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3763 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3765 (riscv_expand_prologue): Add early return for naked function.
3766 (riscv_expand_epilogue): Likewise.
3767 (riscv_function_ok_for_sibcall): Return false for naked function.
3768 (riscv_set_current_function): New.
3769 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3770 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3771 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3772 * doc/extend.texi (RISC-V Function Attributes): New.
3774 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3776 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3777 check for 128-bit long double before checking TCmode.
3778 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3779 128-bit long doubles before checking TFmode or TCmode.
3780 (FLOAT128_IBM_P): Likewise.
3782 2018-01-10 Martin Sebor <msebor@redhat.com>
3784 PR tree-optimization/83671
3785 * builtins.c (c_strlen): Unconditionally return zero for the empty
3787 Use -Warray-bounds for warnings.
3788 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3789 for non-constant array indices with COMPONENT_REF, arrays of
3790 arrays, and pointers to arrays.
3791 (gimple_fold_builtin_strlen): Determine and set length range for
3792 non-constant character arrays.
3794 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3797 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3800 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3802 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3804 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3807 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3808 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3809 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3810 indexed_or_indirect_operand predicate.
3811 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3812 (*vsx_le_perm_load_v8hi): Likewise.
3813 (*vsx_le_perm_load_v16qi): Likewise.
3814 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3815 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3816 (*vsx_le_perm_store_v8hi): Likewise.
3817 (*vsx_le_perm_store_v16qi): Likewise.
3818 (eight unnamed splitters): Likewise.
3820 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3822 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3823 * config/rs6000/emmintrin.h: Likewise.
3824 * config/rs6000/mmintrin.h: Likewise.
3825 * config/rs6000/xmmintrin.h: Likewise.
3827 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3830 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3832 * tree.c (tree_nop_conversion): Return true for location wrapper
3834 (maybe_wrap_with_location): New function.
3835 (selftest::check_strip_nops): New function.
3836 (selftest::test_location_wrappers): New function.
3837 (selftest::tree_c_tests): Call it.
3838 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3839 (maybe_wrap_with_location): New decl.
3840 (EXPR_LOCATION_WRAPPER_P): New macro.
3841 (location_wrapper_p): New inline function.
3842 (tree_strip_any_location_wrapper): New inline function.
3844 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3847 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3848 stack_realign_offset for the largest alignment of stack slot
3850 (ix86_find_max_used_stack_alignment): New function.
3851 (ix86_finalize_stack_frame_flags): Use it. Set
3852 max_used_stack_alignment if we don't realign stack.
3853 * config/i386/i386.h (machine_function): Add
3854 max_used_stack_alignment.
3856 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3858 * config/arm/arm.opt (-mbranch-cost): New option.
3859 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3862 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3865 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3866 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3868 2018-01-10 Richard Biener <rguenther@suse.de>
3871 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3872 early out so it also covers the case where we have a non-NULL
3875 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3877 PR tree-optimization/83753
3878 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3879 for non-strided grouped accesses if the number of elements is 1.
3881 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3884 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3885 * i386.h (TARGET_USE_GATHER): Define.
3886 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3888 2018-01-10 Martin Liska <mliska@suse.cz>
3891 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3892 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3894 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3895 CLEANUP_NO_PARTITIONING is not set.
3897 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3899 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3900 for vectors, as a partial revert of r254296.
3901 * rtl.h (const_vec_p): Delete.
3902 (const_vec_duplicate_p): Don't test for vector CONSTs.
3903 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3904 * expmed.c (make_tree): Likewise.
3907 * common.md (E, F): Use CONSTANT_P instead of checking for
3909 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3910 checking for CONST_VECTOR.
3912 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3915 * predict.c (force_edge_cold): Handle in more sane way edges
3918 2018-01-09 Carl Love <cel@us.ibm.com>
3920 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3922 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3923 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3924 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3925 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3926 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3927 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3928 * config/rs6000/rs6000-protos.h: Add extern defition for
3929 rs6000_generate_float2_double_code.
3930 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3932 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3933 (float2_v2df): Add define_expand.
3935 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3938 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3939 op_mode in the force_to_mode call.
3941 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3943 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3944 instead of checking each element individually.
3945 (aarch64_evpc_uzp): Likewise.
3946 (aarch64_evpc_zip): Likewise.
3947 (aarch64_evpc_ext): Likewise.
3948 (aarch64_evpc_rev): Likewise.
3949 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3950 instead of checking each element individually. Return true without
3952 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3953 whether all selected elements come from the same input, instead of
3954 checking each element individually. Remove calls to gen_rtx_REG,
3955 start_sequence and end_sequence and instead assert that no rtl is
3958 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3960 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3961 order of HIGH and CONST checks.
3963 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3965 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3966 if the destination isn't an SSA_NAME.
3968 2018-01-09 Richard Biener <rguenther@suse.de>
3970 PR tree-optimization/83668
3971 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3973 (canonicalize_loop_form): ... here, renamed from ...
3974 (canonicalize_loop_closed_ssa_form): ... this and amended to
3975 swap successor edges for loop exit blocks to make us use
3976 the RPO order we need for initial schedule generation.
3978 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3980 PR tree-optimization/64811
3981 * match.pd: When optimizing comparisons with Inf, avoid
3982 introducing or losing exceptions from comparisons with NaN.
3984 2018-01-09 Martin Liska <mliska@suse.cz>
3987 * asan.c (shadow_mem_size): Add gcc_assert.
3989 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3991 Don't save registers in main().
3994 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3995 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3996 * config/avr/avr.c (avr_set_current_function): Don't error if
3997 naked, OS_task or OS_main are specified at the same time.
3998 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4000 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4002 * common/config/avr/avr-common.c (avr_option_optimization_table):
4003 Switch on -mmain-is-OS_task for optimizing compilations.
4005 2018-01-09 Richard Biener <rguenther@suse.de>
4007 PR tree-optimization/83572
4008 * graphite.c: Include cfganal.h.
4009 (graphite_transform_loops): Connect infinite loops to exit
4010 and remove fake edges at the end.
4012 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4014 * ipa-inline.c (edge_badness): Revert accidental checkin.
4016 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4019 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4020 symbols; not inline clones.
4022 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4025 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4026 hard registers. Formatting fixes.
4028 PR preprocessor/83722
4029 * gcc.c (try_generate_repro): Pass
4030 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4031 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4034 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4035 Kito Cheng <kito.cheng@gmail.com>
4037 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4038 (riscv_leaf_function_p): Delete.
4039 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4041 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4043 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4045 (do_ifelse): New function.
4046 (do_isel): New function.
4047 (do_sub3): New function.
4048 (do_add3): New function.
4049 (do_load_mask_compare): New function.
4050 (do_overlap_load_compare): New function.
4051 (expand_compare_loop): New function.
4052 (expand_block_compare): Call expand_compare_loop() when appropriate.
4053 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4055 (-mblock-compare-inline-loop-limit): New option.
4057 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4060 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4061 Reverse order of second and third operands in first alternative.
4062 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4063 of first and second elements in UNSPEC_VPERMR vector.
4064 (altivec_expand_vec_perm_le): Likewise.
4066 2018-01-08 Jeff Law <law@redhat.com>
4068 PR rtl-optimizatin/81308
4069 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4070 (process_switch): If group_case_labels makes a change, then set
4072 (pass_convert_switch::execute): If a switch is converted, then
4073 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4075 PR rtl-optimization/81308
4076 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4079 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4081 PR target/83663 - Revert r255946
4082 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4083 generation for cases where splatting a value is not useful.
4084 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4085 across a vec_duplicate and a paradoxical subreg forming a vector
4086 mode to a vec_concat.
4088 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4090 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4091 -march=armv8.3-a variants.
4092 * config/arm/t-multilib: Likewise.
4093 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4095 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4097 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4099 (cceq_ior_compare_complement): Give it a name so I can use it, and
4100 change boolean_or_operator predicate to boolean_operator so it can
4101 be used to generate a crand.
4102 (eqne): New code iterator.
4103 (bd/bd_neg): New code_attrs.
4104 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4105 a single define_insn.
4106 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4107 decrement (bdnzt/bdnzf/bdzt/bdzf).
4108 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4109 with the new names of the branch decrement patterns, and added the
4110 names of the branch decrement conditional patterns.
4112 2018-01-08 Richard Biener <rguenther@suse.de>
4114 PR tree-optimization/83563
4115 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4118 2018-01-08 Richard Biener <rguenther@suse.de>
4121 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4123 2018-01-08 Richard Biener <rguenther@suse.de>
4125 PR tree-optimization/83685
4126 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4127 references to abnormals.
4129 2018-01-08 Richard Biener <rguenther@suse.de>
4132 * dwarf2out.c (output_indirect_strings): Handle empty
4133 skeleton_debug_str_hash.
4134 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4136 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4138 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4139 (emit_store_direct): Likewise.
4140 (arc_trampoline_adjust_address): Likewise.
4141 (arc_asm_trampoline_template): New function.
4142 (arc_initialize_trampoline): Use asm_trampoline_template.
4143 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4144 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4145 * config/arc/arc.md (flush_icache): Delete pattern.
4147 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4149 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4150 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4153 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4156 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4157 by not USED_FOR_TARGET.
4158 (make_pass_resolve_sw_modes): Likewise.
4160 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4162 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4165 2018-01-08 Richard Biener <rguenther@suse.de>
4168 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4170 2018-01-08 Richard Biener <rguenther@suse.de>
4173 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4175 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4178 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4179 basic blocks with a small number of successors.
4180 (convert_control_dep_chain_into_preds): Improve handling of
4182 (dump_predicates): Split apart into...
4183 (dump_pred_chain): ...here...
4184 (dump_pred_info): ...and here.
4185 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4186 (can_chain_union_be_invalidated_p): Improve check for invalidation
4188 (uninit_uses_cannot_happen): Avoid unnecessary if
4189 convert_control_dep_chain_into_preds yielded nothing.
4191 2018-01-06 Martin Sebor <msebor@redhat.com>
4193 PR tree-optimization/83640
4194 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4195 subtracting negative offset from size.
4196 (builtin_access::overlap): Adjust offset bounds of the access to fall
4197 within the size of the object if possible.
4199 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4201 PR rtl-optimization/83699
4202 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4203 extract_bit_field_as_subreg to cases in which the extracted
4204 value is also a vector.
4206 * lra-constraints.c (process_alt_operands): Test for the equivalence
4207 substitutions when detecting a possible reload cycle.
4209 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4212 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4213 by default if flag_selective_schedling{,2}. Formatting fixes.
4215 PR rtl-optimization/83682
4216 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4217 if it has non-VECTOR_MODE element mode.
4218 (vec_duplicate_p): Likewise.
4221 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4222 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4224 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4227 * config/i386/i386-builtin.def
4228 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4229 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4230 Require also OPTION_MASK_ISA_AVX512F in addition to
4231 OPTION_MASK_ISA_GFNI.
4232 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4233 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4234 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4235 to OPTION_MASK_ISA_GFNI.
4236 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4237 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4238 OPTION_MASK_ISA_AVX512BW.
4239 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4240 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4241 addition to OPTION_MASK_ISA_GFNI.
4242 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4243 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4244 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4245 to OPTION_MASK_ISA_GFNI.
4246 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4247 a requirement for all ISAs rather than any of them with a few
4249 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4251 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4252 bitmasks to be enabled with 3 exceptions, instead of requiring any
4253 enabled ISA with lots of exceptions.
4254 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4255 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4256 Change avx512bw in isa attribute to avx512f.
4257 * config/i386/sgxintrin.h: Add license boilerplate.
4258 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4259 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4260 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4261 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4263 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4264 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4265 temporarily sse2 rather than sse if not enabled already.
4268 * config/i386/sse.md (VI248_VLBW): Rename to ...
4269 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4270 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4271 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4272 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4273 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4274 mode iterator instead of VI248_VLBW.
4276 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4278 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4279 (record_modified): Skip clobbers; add debug output.
4280 (param_change_prob): Use sreal frequencies.
4282 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4284 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4285 punt for user-aligned variables.
4287 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4289 * tree-chrec.c (chrec_contains_symbols): Return true for
4292 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4295 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4296 of (x|y) == x for BICS pattern.
4298 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4300 PR tree-optimization/83605
4301 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4302 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4305 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4307 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4308 * config/epiphany/rtems.h: New file.
4310 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4311 Uros Bizjak <ubizjak@gmail.com>
4314 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4315 QIreg_operand instead of register_operand predicate.
4316 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4317 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4318 comments instead of -fmitigate[-_]rop.
4320 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4323 * cgraphunit.c (symbol_table::compile): Switch to text_section
4324 before calling assembly_start debug hook.
4325 * run-rtl-passes.c (run_rtl_passes): Likewise.
4328 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4330 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4331 range_int_cst_p rather than !symbolic_range_p before calling
4332 extract_range_from_multiplicative_op_1.
4334 2018-01-04 Jeff Law <law@redhat.com>
4336 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4337 redundant test in assertion.
4339 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4341 * doc/rtl.texi: Document machine_mode wrapper classes.
4343 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4345 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4348 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4350 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4351 the VEC_PERM_EXPR fold to fail.
4353 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4356 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4357 to switched_sections.
4359 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4362 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4365 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4368 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4369 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4371 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4374 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4375 is BLKmode and bitpos not zero or mode change is needed.
4377 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4380 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4383 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4386 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4387 instead of MULT rtx. Update all corresponding splitters.
4389 (*ssub<modesuffix>): Ditto.
4391 (*cmp_sadd_di): Update split patterns.
4392 (*cmp_sadd_si): Ditto.
4393 (*cmp_sadd_sidi): Ditto.
4394 (*cmp_ssub_di): Ditto.
4395 (*cmp_ssub_si): Ditto.
4396 (*cmp_ssub_sidi): Ditto.
4397 * config/alpha/predicates.md (const23_operand): New predicate.
4398 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4399 Look for ASHIFT, not MULT inner operand.
4400 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4402 2018-01-04 Martin Liska <mliska@suse.cz>
4404 PR gcov-profile/83669
4405 * gcov.c (output_intermediate_file): Add version to intermediate
4407 * doc/gcov.texi: Document new field 'version' in intermediate
4408 file format. Fix location of '-k' option of gcov command.
4410 2018-01-04 Martin Liska <mliska@suse.cz>
4413 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4415 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4417 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4419 2018-01-03 Martin Sebor <msebor@redhat.com>
4421 PR tree-optimization/83655
4422 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4423 checking calls with invalid arguments.
4425 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4427 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4428 (vectorizable_mask_load_store): Delete.
4429 (vectorizable_call): Return false for masked loads and stores.
4430 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4431 instead of gimple_assign_rhs1.
4432 (vectorizable_load): Handle IFN_MASK_LOAD.
4433 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4435 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4437 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4439 (vectorizable_mask_load_store): ...here.
4440 (vectorizable_load): ...and here.
4442 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4444 * tree-vect-stmts.c (vect_build_all_ones_mask)
4445 (vect_build_zero_merge_argument): New functions, split out from...
4446 (vectorizable_load): ...here.
4448 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4450 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4452 (vectorizable_mask_load_store): ...here.
4453 (vectorizable_store): ...and here.
4455 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4457 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4459 (vectorizable_mask_load_store): ...here.
4461 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4463 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4464 (vect_model_store_cost): Take a vec_load_store_type instead of a
4466 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4467 (vect_model_store_cost): Take a vec_load_store_type instead of a
4469 (vectorizable_mask_load_store): Update accordingly.
4470 (vectorizable_store): Likewise.
4471 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4473 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4475 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4476 IFN_MASK_LOAD calls here rather than...
4477 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4479 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4480 Alan Hayward <alan.hayward@arm.com>
4481 David Sherwood <david.sherwood@arm.com>
4483 * expmed.c (extract_bit_field_1): For vector extracts,
4484 fall back to extract_bit_field_as_subreg if vec_extract
4487 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4488 Alan Hayward <alan.hayward@arm.com>
4489 David Sherwood <david.sherwood@arm.com>
4491 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4492 they are variable or constant sized.
4493 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4494 slots for constant-sized data.
4496 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4497 Alan Hayward <alan.hayward@arm.com>
4498 David Sherwood <david.sherwood@arm.com>
4500 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4501 handling COND_EXPRs with boolean comparisons, try to find a better
4502 basis for the mask type than the boolean itself.
4504 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4506 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4507 is calculated and how it can be overridden.
4508 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4509 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4511 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4514 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4515 Alan Hayward <alan.hayward@arm.com>
4516 David Sherwood <david.sherwood@arm.com>
4518 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4519 Remove the mode argument.
4520 (aarch64_simd_valid_immediate): Remove the mode and inverse
4522 * config/aarch64/iterators.md (bitsize): New iterator.
4523 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4524 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4525 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4526 aarch64_simd_valid_immediate.
4527 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4528 (aarch64_reg_or_bic_imm): Likewise.
4529 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4530 with an insn_type enum and msl with a modifier_type enum.
4531 Replace element_width with a scalar_mode. Change the shift
4532 to unsigned int. Add constructors for scalar_float_mode and
4533 scalar_int_mode elements.
4534 (aarch64_vect_float_const_representable_p): Delete.
4535 (aarch64_can_const_movi_rtx_p)
4536 (aarch64_simd_scalar_immediate_valid_for_move)
4537 (aarch64_simd_make_constant): Update call to
4538 aarch64_simd_valid_immediate.
4539 (aarch64_advsimd_valid_immediate_hs): New function.
4540 (aarch64_advsimd_valid_immediate): Likewise.
4541 (aarch64_simd_valid_immediate): Remove mode and inverse
4542 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4543 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4544 and aarch64_float_const_representable_p on the result.
4545 (aarch64_output_simd_mov_immediate): Remove mode argument.
4546 Update call to aarch64_simd_valid_immediate and use of
4547 simd_immediate_info.
4548 (aarch64_output_scalar_simd_mov_immediate): Update call
4551 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4552 Alan Hayward <alan.hayward@arm.com>
4553 David Sherwood <david.sherwood@arm.com>
4555 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4556 (mode_nunits): Likewise CONST_MODE_NUNITS.
4557 * machmode.def (ADJUST_NUNITS): Document.
4558 * genmodes.c (mode_data::need_nunits_adj): New field.
4559 (blank_mode): Update accordingly.
4560 (adj_nunits): New variable.
4561 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4563 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4564 listed in adj_nunits.
4565 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4566 listed in adj_nunits. Don't emit case statements for such modes.
4567 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4568 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4569 nothing if adj_nunits is nonnull.
4570 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4571 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4572 (emit_mode_fbit): Update use of print_maybe_const_decl.
4573 (emit_move_size): Likewise. Treat the array as non-const
4575 (emit_mode_adjustments): Handle adj_nunits.
4577 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4579 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4580 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4581 (VECTOR_MODES): Use it.
4582 (make_vector_modes): Take the prefix as an argument.
4584 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4585 Alan Hayward <alan.hayward@arm.com>
4586 David Sherwood <david.sherwood@arm.com>
4588 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4589 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4590 for MODE_VECTOR_BOOL.
4591 * machmode.def (VECTOR_BOOL_MODE): Document.
4592 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4593 (make_vector_bool_mode): New function.
4594 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4596 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4597 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4599 * stor-layout.c (int_mode_for_mode): Likewise.
4600 * tree.c (build_vector_type_for_mode): Likewise.
4601 * varasm.c (output_constant_pool_2): Likewise.
4602 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4603 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4604 for MODE_VECTOR_BOOL.
4605 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4606 of mode class checks.
4607 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4608 instead of a list of mode class checks.
4609 (expand_vector_scalar_condition): Likewise.
4610 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4612 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4613 Alan Hayward <alan.hayward@arm.com>
4614 David Sherwood <david.sherwood@arm.com>
4616 * machmode.h (mode_size): Change from unsigned short to
4618 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4619 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4620 or if measurement_type is not polynomial.
4621 (fixed_size_mode::includes_p): Check for constant-sized modes.
4622 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4623 return a poly_uint16 rather than an unsigned short.
4624 (emit_mode_size): Change the type of mode_size from unsigned short
4625 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4626 (emit_mode_adjustments): Cope with polynomial vector sizes.
4627 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4629 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4631 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4632 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4633 * caller-save.c (setup_save_areas): Likewise.
4634 (replace_reg_with_saved_mem): Likewise.
4635 * calls.c (emit_library_call_value_1): Likewise.
4636 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4637 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4638 (gen_lowpart_for_combine): Likewise.
4639 * convert.c (convert_to_integer_1): Likewise.
4640 * cse.c (equiv_constant, cse_insn): Likewise.
4641 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4642 (cselib_subst_to_values): Likewise.
4643 * dce.c (word_dce_process_block): Likewise.
4644 * df-problems.c (df_word_lr_mark_ref): Likewise.
4645 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4646 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4647 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4648 (rtl_for_decl_location): Likewise.
4649 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4650 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4651 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4652 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4653 (expand_expr_real_1): Likewise.
4654 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4655 (pad_below): Likewise.
4656 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4657 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4658 * ira.c (get_subreg_tracking_sizes): Likewise.
4659 * ira-build.c (ira_create_allocno_objects): Likewise.
4660 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4661 (ira_sort_regnos_for_alter_reg): Likewise.
4662 * ira-costs.c (record_operand_costs): Likewise.
4663 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4664 (resolve_simple_move): Likewise.
4665 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4666 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4667 (lra_constraints): Likewise.
4668 (CONST_POOL_OK_P): Reject variable-sized modes.
4669 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4670 (add_pseudo_to_slot, lra_spill): Likewise.
4671 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4672 * optabs-query.c (get_best_extraction_insn): Likewise.
4673 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4674 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4675 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4676 * recog.c (offsettable_address_addr_space_p): Likewise.
4677 * regcprop.c (maybe_mode_change): Likewise.
4678 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4679 * regrename.c (build_def_use): Likewise.
4680 * regstat.c (dump_reg_info): Likewise.
4681 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4682 (find_reloads, find_reloads_subreg_address): Likewise.
4683 * reload1.c (eliminate_regs_1): Likewise.
4684 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4685 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4686 (simplify_binary_operation_1, simplify_subreg): Likewise.
4687 * targhooks.c (default_function_arg_padding): Likewise.
4688 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4689 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4690 (verify_gimple_assign_ternary): Likewise.
4691 * tree-inline.c (estimate_move_cost): Likewise.
4692 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4693 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4694 (get_address_cost_ainc): Likewise.
4695 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4696 (vect_supportable_dr_alignment): Likewise.
4697 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4698 (vectorizable_reduction): Likewise.
4699 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4700 (vectorizable_operation, vectorizable_load): Likewise.
4701 * tree.c (build_same_sized_truth_vector_type): Likewise.
4702 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4703 * var-tracking.c (emit_note_insn_var_location): Likewise.
4704 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4705 (ADDR_VEC_ALIGN): Likewise.
4707 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4708 Alan Hayward <alan.hayward@arm.com>
4709 David Sherwood <david.sherwood@arm.com>
4711 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4713 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4714 or if measurement_type is polynomial.
4715 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4716 * combine.c (make_extraction): Likewise.
4717 * dse.c (find_shift_sequence): Likewise.
4718 * dwarf2out.c (mem_loc_descriptor): Likewise.
4719 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4720 (extract_bit_field, extract_low_bits): Likewise.
4721 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4722 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4723 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4724 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4725 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4726 * reload.c (find_reloads): Likewise.
4727 * reload1.c (alter_reg): Likewise.
4728 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4729 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4730 * tree-if-conv.c (predicate_mem_writes): Likewise.
4731 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4732 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4733 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4734 * valtrack.c (dead_debug_insert_temp): Likewise.
4735 * varasm.c (mergeable_constant_section): Likewise.
4736 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4738 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4739 Alan Hayward <alan.hayward@arm.com>
4740 David Sherwood <david.sherwood@arm.com>
4742 * expr.c (expand_assignment): Cope with polynomial mode sizes
4743 when assigning to a CONCAT.
4745 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4746 Alan Hayward <alan.hayward@arm.com>
4747 David Sherwood <david.sherwood@arm.com>
4749 * machmode.h (mode_precision): Change from unsigned short to
4751 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4753 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4754 or if measurement_type is not polynomial.
4755 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4756 in which the mode is already known to be a scalar_int_mode.
4757 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4758 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4760 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4761 for GET_MODE_PRECISION.
4762 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4763 for GET_MODE_PRECISION.
4764 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4766 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4767 (expand_field_assignment, make_extraction): Likewise.
4768 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4769 (get_last_value): Likewise.
4770 * convert.c (convert_to_integer_1): Likewise.
4771 * cse.c (cse_insn): Likewise.
4772 * expr.c (expand_expr_real_1): Likewise.
4773 * lra-constraints.c (simplify_operand_subreg): Likewise.
4774 * optabs-query.c (can_atomic_load_p): Likewise.
4775 * optabs.c (expand_atomic_load): Likewise.
4776 (expand_atomic_store): Likewise.
4777 * ree.c (combine_reaching_defs): Likewise.
4778 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4779 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4780 * tree.h (type_has_mode_precision_p): Likewise.
4781 * ubsan.c (instrument_si_overflow): Likewise.
4783 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4784 Alan Hayward <alan.hayward@arm.com>
4785 David Sherwood <david.sherwood@arm.com>
4787 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4788 polynomial numbers of units.
4789 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4790 (valid_vector_subparts_p): New function.
4791 (build_vector_type): Remove temporary shim and take the number
4792 of units as a poly_uint64 rather than an int.
4793 (build_opaque_vector_type): Take the number of units as a
4794 poly_uint64 rather than an int.
4795 * tree.c (build_vector_from_ctor): Handle polynomial
4796 TYPE_VECTOR_SUBPARTS.
4797 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4798 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4799 (build_vector_from_val): If the number of units is variable,
4800 use build_vec_duplicate_cst for constant operands and
4801 VEC_DUPLICATE_EXPR otherwise.
4802 (make_vector_type): Remove temporary is_constant ().
4803 (build_vector_type, build_opaque_vector_type): Take the number of
4804 units as a poly_uint64 rather than an int.
4805 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4807 * cfgexpand.c (expand_debug_expr): Likewise.
4808 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4809 (store_constructor, expand_expr_real_1): Likewise.
4810 (const_scalar_mask_from_tree): Likewise.
4811 * fold-const-call.c (fold_const_reduction): Likewise.
4812 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4813 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4814 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4815 (fold_relational_const): Likewise.
4816 (native_interpret_vector): Likewise. Change the size from an
4817 int to an unsigned int.
4818 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4819 TYPE_VECTOR_SUBPARTS.
4820 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4821 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4822 duplicating a non-constant operand into a variable-length vector.
4823 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4824 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4825 * ipa-icf.c (sem_variable::equals): Likewise.
4826 * match.pd: Likewise.
4827 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4828 * print-tree.c (print_node): Likewise.
4829 * stor-layout.c (layout_type): Likewise.
4830 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4831 * tree-cfg.c (verify_gimple_comparison): Likewise.
4832 (verify_gimple_assign_binary): Likewise.
4833 (verify_gimple_assign_ternary): Likewise.
4834 (verify_gimple_assign_single): Likewise.
4835 * tree-pretty-print.c (dump_generic_node): Likewise.
4836 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4837 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4838 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4839 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4840 (vect_shift_permute_load_chain): Likewise.
4841 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4842 (expand_vector_condition, optimize_vector_constructor): Likewise.
4843 (lower_vec_perm, get_compute_type): Likewise.
4844 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4845 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4846 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4847 (vect_recog_mask_conversion_pattern): Likewise.
4848 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4849 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4850 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4851 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4852 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4853 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4854 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4855 (supportable_widening_operation): Likewise.
4856 (supportable_narrowing_operation): Likewise.
4857 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4859 * varasm.c (output_constant): Likewise.
4861 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4862 Alan Hayward <alan.hayward@arm.com>
4863 David Sherwood <david.sherwood@arm.com>
4865 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4866 so that both the length == 3 and length != 3 cases set up their
4867 own permute vectors. Add comments explaining why we know the
4868 number of elements is constant.
4869 (vect_permute_load_chain): Likewise.
4871 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4872 Alan Hayward <alan.hayward@arm.com>
4873 David Sherwood <david.sherwood@arm.com>
4875 * machmode.h (mode_nunits): Change from unsigned char to
4877 (ONLY_FIXED_SIZE_MODES): New macro.
4878 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4879 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4880 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4882 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4883 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4884 or if measurement_type is not polynomial.
4885 * genmodes.c (ZERO_COEFFS): New macro.
4886 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4888 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4889 Use ZERO_COEFFS when emitting initializers.
4890 * data-streamer.h (bp_pack_poly_value): New function.
4891 (bp_unpack_poly_value): Likewise.
4892 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4893 for GET_MODE_NUNITS.
4894 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4895 for GET_MODE_NUNITS.
4896 * tree.c (make_vector_type): Remove temporary shim and make
4897 the real function take the number of units as a poly_uint64
4899 (build_vector_type_for_mode): Handle polynomial nunits.
4900 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4901 * emit-rtl.c (const_vec_series_p_1): Likewise.
4902 (gen_rtx_CONST_VECTOR): Likewise.
4903 * fold-const.c (test_vec_duplicate_folding): Likewise.
4904 * genrecog.c (validate_pattern): Likewise.
4905 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4906 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4907 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4908 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4909 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4910 * rtlanal.c (subreg_get_info): Likewise.
4911 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4912 (vect_grouped_load_supported): Likewise.
4913 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4914 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4915 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4916 (simplify_const_unary_operation, simplify_binary_operation_1)
4917 (simplify_const_binary_operation, simplify_ternary_operation)
4918 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4919 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4920 instead of CONST_VECTOR_NUNITS.
4921 * varasm.c (output_constant_pool_2): Likewise.
4922 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4923 explicit-encoded elements in the XVEC for variable-length vectors.
4925 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4927 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4929 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4930 Alan Hayward <alan.hayward@arm.com>
4931 David Sherwood <david.sherwood@arm.com>
4933 * coretypes.h (fixed_size_mode): Declare.
4934 (fixed_size_mode_pod): New typedef.
4935 * builtins.h (target_builtins::x_apply_args_mode)
4936 (target_builtins::x_apply_result_mode): Change type to
4937 fixed_size_mode_pod.
4938 * builtins.c (apply_args_size, apply_result_size, result_vector)
4939 (expand_builtin_apply_args_1, expand_builtin_apply)
4940 (expand_builtin_return): Update accordingly.
4942 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4944 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4945 * cselib.c (cselib_hash_rtx): Likewise.
4946 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4947 CONST_VECTOR encoding.
4949 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4950 Jeff Law <law@redhat.com>
4953 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4954 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4955 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4956 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4959 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4960 explicitly probe *sp in a noreturn function if there were any callee
4961 register saves or frame pointer is needed.
4963 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4966 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4967 BLKmode for ternary, binary or unary expressions.
4970 * var-tracking.c (delete_vta_debug_insn): New inline function.
4971 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4972 insns from get_insns () to NULL instead of each bb separately.
4973 Use delete_vta_debug_insn. No longer static.
4974 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4975 delete_vta_debug_insns callers.
4976 * rtl.h (delete_vta_debug_insns): Declare.
4977 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4978 instead of variable_tracking_main.
4980 2018-01-03 Martin Sebor <msebor@redhat.com>
4982 PR tree-optimization/83603
4983 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4984 arguments past the endof the argument list in functions declared
4985 without a prototype.
4986 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4987 Avoid checking when arguments are null.
4989 2018-01-03 Martin Sebor <msebor@redhat.com>
4992 * doc/extend.texi (attribute const): Fix a typo.
4993 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4994 issuing -Wsuggest-attribute for void functions.
4996 2018-01-03 Martin Sebor <msebor@redhat.com>
4998 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4999 offset_int::from instead of wide_int::to_shwi.
5000 (maybe_diag_overlap): Remove assertion.
5001 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5002 * gimple-ssa-sprintf.c (format_directive): Same.
5003 (parse_directive): Same.
5004 (sprintf_dom_walker::compute_format_length): Same.
5005 (try_substitute_return_value): Same.
5007 2018-01-03 Jeff Law <law@redhat.com>
5010 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5011 non-constant residual for zero at runtime and avoid probing in
5012 that case. Reorganize code for trailing problem to mirror handling
5015 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5017 PR tree-optimization/83501
5018 * tree-ssa-strlen.c (get_string_cst): New.
5019 (handle_char_store): Call get_string_cst.
5021 2018-01-03 Martin Liska <mliska@suse.cz>
5023 PR tree-optimization/83593
5024 * tree-ssa-strlen.c: Include tree-cfg.h.
5025 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5026 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5027 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5029 (strlen_dom_walker::before_dom_children): Call
5030 gimple_purge_dead_eh_edges. Dump tranformation with details
5032 (strlen_dom_walker::before_dom_children): Update call by adding
5033 new argument cleanup_eh.
5034 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5036 2018-01-03 Martin Liska <mliska@suse.cz>
5039 * cif-code.def (VARIADIC_THUNK): New enum value.
5040 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5043 2018-01-03 Jan Beulich <jbeulich@suse.com>
5045 * sse.md (mov<mode>_internal): Tighten condition for when to use
5046 vmovdqu<ssescalarsize> for TI and OI modes.
5048 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5050 Update copyright years.
5052 2018-01-03 Martin Liska <mliska@suse.cz>
5055 * ipa-visibility.c (function_and_variable_visibility): Skip
5056 functions with noipa attribure.
5058 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5060 * gcc.c (process_command): Update copyright notice dates.
5061 * gcov-dump.c (print_version): Ditto.
5062 * gcov.c (print_version): Ditto.
5063 * gcov-tool.c (print_version): Ditto.
5064 * gengtype.c (create_file): Ditto.
5065 * doc/cpp.texi: Bump @copying's copyright year.
5066 * doc/cppinternals.texi: Ditto.
5067 * doc/gcc.texi: Ditto.
5068 * doc/gccint.texi: Ditto.
5069 * doc/gcov.texi: Ditto.
5070 * doc/install.texi: Ditto.
5071 * doc/invoke.texi: Ditto.
5073 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5075 * vector-builder.h (vector_builder::m_full_nelts): Change from
5076 unsigned int to poly_uint64.
5077 (vector_builder::full_nelts): Update prototype accordingly.
5078 (vector_builder::new_vector): Likewise.
5079 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5080 (vector_builder::operator ==): Likewise.
5081 (vector_builder::finalize): Likewise.
5082 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5083 Take the number of elements as a poly_uint64 rather than an
5085 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5086 from unsigned int to poly_uint64.
5087 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5088 (vec_perm_indices::new_vector): Likewise.
5089 (vec_perm_indices::length): Likewise.
5090 (vec_perm_indices::nelts_per_input): Likewise.
5091 (vec_perm_indices::input_nelts): Likewise.
5092 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5093 number of elements per input as a poly_uint64 rather than an
5094 unsigned int. Use the original encoding for variable-length
5095 vectors, rather than clamping each individual element.
5096 For the second and subsequent elements in each pattern,
5097 clamp the step and base before clamping their sum.
5098 (vec_perm_indices::series_p): Handle polynomial element counts.
5099 (vec_perm_indices::all_in_range_p): Likewise.
5100 (vec_perm_indices_to_tree): Likewise.
5101 (vec_perm_indices_to_rtx): Likewise.
5102 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5103 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5104 (tree_vector_builder::new_binary_operation): Handle polynomial
5105 element counts. Return false if we need to know the number
5106 of elements at compile time.
5107 * fold-const.c (fold_vec_perm): Punt if the number of elements
5108 isn't known at compile time.
5110 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5112 * vec-perm-indices.h (vec_perm_builder): Change element type
5113 from HOST_WIDE_INT to poly_int64.
5114 (vec_perm_indices::element_type): Update accordingly.
5115 (vec_perm_indices::clamp): Handle polynomial element_types.
5116 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5117 (vec_perm_indices::all_in_range_p): Likewise.
5118 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5120 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5121 polynomial vec_perm_indices element types.
5122 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5123 * fold-const.c (fold_vec_perm): Likewise.
5124 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5125 * tree-vect-generic.c (lower_vec_perm): Likewise.
5126 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5127 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5128 element type to HOST_WIDE_INT.
5130 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5131 Alan Hayward <alan.hayward@arm.com>
5132 David Sherwood <david.sherwood@arm.com>
5134 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5135 rather than an int. Use plus_constant.
5136 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5137 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5139 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5140 Alan Hayward <alan.hayward@arm.com>
5141 David Sherwood <david.sherwood@arm.com>
5143 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5144 a HOST_WIDE_INT to a poly_int64.
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 * calls.c (load_register_parameters): Cope with polynomial
5151 mode sizes. Require a constant size for BLKmode parameters
5152 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5153 forces a parameter to be padded at the lsb end in order to
5154 fill a complete number of words, require the parameter size
5155 to be ordered wrt UNITS_PER_WORD.
5157 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5158 Alan Hayward <alan.hayward@arm.com>
5159 David Sherwood <david.sherwood@arm.com>
5161 * reload1.c (spill_stack_slot_width): Change element type
5162 from unsigned int to poly_uint64_pod.
5163 (alter_reg): Treat mode sizes as polynomial.
5165 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5166 Alan Hayward <alan.hayward@arm.com>
5167 David Sherwood <david.sherwood@arm.com>
5169 * reload.c (complex_word_subreg_p): New function.
5170 (reload_inner_reg_of_subreg, push_reload): Use it.
5172 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5173 Alan Hayward <alan.hayward@arm.com>
5174 David Sherwood <david.sherwood@arm.com>
5176 * lra-constraints.c (process_alt_operands): Reject matched
5177 operands whose sizes aren't ordered.
5178 (match_reload): Refer to this check here.
5180 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5181 Alan Hayward <alan.hayward@arm.com>
5182 David Sherwood <david.sherwood@arm.com>
5184 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5185 that the mode size is in the set {1, 2, 4, 8, 16}.
5187 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5188 Alan Hayward <alan.hayward@arm.com>
5189 David Sherwood <david.sherwood@arm.com>
5191 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5192 Use plus_constant instead of gen_rtx_PLUS.
5194 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5195 Alan Hayward <alan.hayward@arm.com>
5196 David Sherwood <david.sherwood@arm.com>
5198 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5199 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5200 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5201 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5202 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5203 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5204 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5205 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5206 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5207 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5209 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5210 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5211 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5212 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5213 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5214 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5215 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5216 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5217 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5218 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5220 * expr.c (emit_move_resolve_push): Treat the input and result
5221 of PUSH_ROUNDING as a poly_int64.
5222 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5223 (emit_push_insn): Likewise.
5224 * lra-eliminations.c (mark_not_eliminable): Likewise.
5225 * recog.c (push_operand): Likewise.
5226 * reload1.c (elimination_effects): Likewise.
5227 * rtlanal.c (nonzero_bits1): Likewise.
5228 * calls.c (store_one_arg): Likewise. Require the padding to be
5229 known at compile time.
5231 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5232 Alan Hayward <alan.hayward@arm.com>
5233 David Sherwood <david.sherwood@arm.com>
5235 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5236 Use plus_constant instead of gen_rtx_PLUS.
5238 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5239 Alan Hayward <alan.hayward@arm.com>
5240 David Sherwood <david.sherwood@arm.com>
5242 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5245 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5246 Alan Hayward <alan.hayward@arm.com>
5247 David Sherwood <david.sherwood@arm.com>
5249 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5250 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5251 via stack temporaries. Treat the mode size as polynomial too.
5253 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5254 Alan Hayward <alan.hayward@arm.com>
5255 David Sherwood <david.sherwood@arm.com>
5257 * expr.c (expand_expr_real_2): When handling conversions involving
5258 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5259 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5260 as a poly_uint64 too.
5262 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5263 Alan Hayward <alan.hayward@arm.com>
5264 David Sherwood <david.sherwood@arm.com>
5266 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5268 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5269 Alan Hayward <alan.hayward@arm.com>
5270 David Sherwood <david.sherwood@arm.com>
5272 * combine.c (can_change_dest_mode): Handle polynomial
5273 REGMODE_NATURAL_SIZE.
5274 * expmed.c (store_bit_field_1): Likewise.
5275 * expr.c (store_constructor): Likewise.
5276 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5277 and polynomial REGMODE_NATURAL_SIZE.
5278 (gen_lowpart_common): Likewise.
5279 * reginfo.c (record_subregs_of_mode): Likewise.
5280 * rtlanal.c (read_modify_subreg_p): Likewise.
5282 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5283 Alan Hayward <alan.hayward@arm.com>
5284 David Sherwood <david.sherwood@arm.com>
5286 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5287 numbers of elements.
5289 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5290 Alan Hayward <alan.hayward@arm.com>
5291 David Sherwood <david.sherwood@arm.com>
5293 * match.pd: Cope with polynomial numbers of vector elements.
5295 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5296 Alan Hayward <alan.hayward@arm.com>
5297 David Sherwood <david.sherwood@arm.com>
5299 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5300 in a POINTER_PLUS_EXPR.
5302 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5303 Alan Hayward <alan.hayward@arm.com>
5304 David Sherwood <david.sherwood@arm.com>
5306 * omp-simd-clone.c (simd_clone_subparts): New function.
5307 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5308 (ipa_simd_modify_function_body): Likewise.
5310 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5311 Alan Hayward <alan.hayward@arm.com>
5312 David Sherwood <david.sherwood@arm.com>
5314 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5315 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5316 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5317 (expand_vector_condition, vector_element): Likewise.
5318 (subparts_gt): New function.
5319 (get_compute_type): Use subparts_gt.
5320 (count_type_subparts): Delete.
5321 (expand_vector_operations_1): Use subparts_gt instead of
5322 count_type_subparts.
5324 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5325 Alan Hayward <alan.hayward@arm.com>
5326 David Sherwood <david.sherwood@arm.com>
5328 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5329 (vect_compile_time_alias): ...this new function. Do the calculation
5330 on poly_ints rather than trees.
5331 (vect_prune_runtime_alias_test_list): Update call accordingly.
5333 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5334 Alan Hayward <alan.hayward@arm.com>
5335 David Sherwood <david.sherwood@arm.com>
5337 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5339 (vect_schedule_slp_instance): Likewise.
5341 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5342 Alan Hayward <alan.hayward@arm.com>
5343 David Sherwood <david.sherwood@arm.com>
5345 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5346 constant and extern definitions for variable-length vectors.
5347 (vect_get_constant_vectors): Note that the number of units
5348 is known to be constant.
5350 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5351 Alan Hayward <alan.hayward@arm.com>
5352 David Sherwood <david.sherwood@arm.com>
5354 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5355 of units as polynomial. Choose between WIDE and NARROW based
5358 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5359 Alan Hayward <alan.hayward@arm.com>
5360 David Sherwood <david.sherwood@arm.com>
5362 * tree-vect-stmts.c (simd_clone_subparts): New function.
5363 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5365 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5366 Alan Hayward <alan.hayward@arm.com>
5367 David Sherwood <david.sherwood@arm.com>
5369 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5370 vectors as polynomial. Use build_index_vector for
5373 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5374 Alan Hayward <alan.hayward@arm.com>
5375 David Sherwood <david.sherwood@arm.com>
5377 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5378 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5379 for variable-length vectors.
5380 (vectorizable_mask_load_store): Treat the number of units as
5381 polynomial, asserting that it is constant if the condition has
5382 already been enforced.
5383 (vectorizable_store, vectorizable_load): Likewise.
5385 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5386 Alan Hayward <alan.hayward@arm.com>
5387 David Sherwood <david.sherwood@arm.com>
5389 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5390 of units as polynomial. Punt if we can't tell at compile time
5391 which vector contains the final result.
5393 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5394 Alan Hayward <alan.hayward@arm.com>
5395 David Sherwood <david.sherwood@arm.com>
5397 * tree-vect-loop.c (vectorizable_induction): Treat the number
5398 of units as polynomial. Punt on SLP inductions. Use an integer
5399 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5400 cast of such a series for variable-length floating-point
5403 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5404 Alan Hayward <alan.hayward@arm.com>
5405 David Sherwood <david.sherwood@arm.com>
5407 * tree.h (build_index_vector): Declare.
5408 * tree.c (build_index_vector): New function.
5409 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5410 of units as polynomial, forcibly converting it to a constant if
5411 vectorizable_reduction has already enforced the condition.
5412 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5413 to create a {1,2,3,...} vector.
5414 (vectorizable_reduction): Treat the number of units as polynomial.
5415 Choose vectype_in based on the largest scalar element size rather
5416 than the smallest number of units. Enforce the restrictions
5419 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5420 Alan Hayward <alan.hayward@arm.com>
5421 David Sherwood <david.sherwood@arm.com>
5423 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5424 number of units as polynomial.
5426 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5427 Alan Hayward <alan.hayward@arm.com>
5428 David Sherwood <david.sherwood@arm.com>
5430 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5431 * target.def (autovectorize_vector_sizes): Return the vector sizes
5432 by pointer, using vector_sizes rather than a bitmask.
5433 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5434 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5435 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5437 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5438 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5439 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5440 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5441 * omp-general.c (omp_max_vf): Likewise.
5442 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5443 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5444 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5445 * tree-vect-slp.c (vect_slp_bb): Likewise.
5446 * doc/tm.texi: Regenerate.
5447 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5449 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5450 the vector size as a poly_uint64 rather than an unsigned int.
5451 (current_vector_size): Change from an unsigned int to a poly_uint64.
5452 (get_vectype_for_scalar_type): Update accordingly.
5453 * tree.h (build_truth_vector_type): Take the size and number of
5454 units as a poly_uint64 rather than an unsigned int.
5455 (build_vector_type): Add a temporary overload that takes
5456 the number of units as a poly_uint64 rather than an unsigned int.
5457 * tree.c (make_vector_type): Likewise.
5458 (build_truth_vector_type): Take the number of units as a poly_uint64
5459 rather than an unsigned int.
5461 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5462 Alan Hayward <alan.hayward@arm.com>
5463 David Sherwood <david.sherwood@arm.com>
5465 * target.def (get_mask_mode): Take the number of units and length
5466 as poly_uint64s rather than unsigned ints.
5467 * targhooks.h (default_get_mask_mode): Update accordingly.
5468 * targhooks.c (default_get_mask_mode): Likewise.
5469 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5470 * doc/tm.texi: Regenerate.
5472 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5473 Alan Hayward <alan.hayward@arm.com>
5474 David Sherwood <david.sherwood@arm.com>
5476 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5477 * omp-general.c (omp_max_vf): Likewise.
5478 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5479 (expand_omp_simd): Handle polynomial safelen.
5480 * omp-low.c (omplow_simd_context): Add a default constructor.
5481 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5482 (lower_rec_simd_input_clauses): Update accordingly.
5483 (lower_rec_input_clauses): Likewise.
5485 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5486 Alan Hayward <alan.hayward@arm.com>
5487 David Sherwood <david.sherwood@arm.com>
5489 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5490 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5491 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5492 (vect_analyze_slp_cost): Likewise.
5493 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5494 (vect_model_load_cost): Likewise.
5496 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5497 Alan Hayward <alan.hayward@arm.com>
5498 David Sherwood <david.sherwood@arm.com>
5500 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5501 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5502 from an unsigned int * to a poly_uint64_pod *.
5503 (calculate_unrolling_factor): New function.
5504 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5506 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5507 Alan Hayward <alan.hayward@arm.com>
5508 David Sherwood <david.sherwood@arm.com>
5510 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5511 from an unsigned int to a poly_uint64.
5512 (_loop_vec_info::slp_unrolling_factor): Likewise.
5513 (_loop_vec_info::vectorization_factor): Change from an int
5515 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5516 (vect_get_num_vectors): New function.
5517 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5518 (vect_get_num_copies): Use vect_get_num_vectors.
5519 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5520 to an unsigned int *.
5521 (vect_analyze_data_refs): Change min_vf from an int * to a
5523 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5524 than an unsigned HOST_WIDE_INT.
5525 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5526 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5527 to an unsigned int *.
5528 (vect_analyze_data_ref_dependences): Likewise.
5529 (vect_compute_data_ref_alignment): Handle polynomial vf.
5530 (vect_enhance_data_refs_alignment): Likewise.
5531 (vect_prune_runtime_alias_test_list): Likewise.
5532 (vect_shift_permute_load_chain): Likewise.
5533 (vect_supportable_dr_alignment): Likewise.
5534 (dependence_distance_ge_vf): Take the vectorization factor as a
5535 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5536 (vect_analyze_data_refs): Change min_vf from an int * to a
5538 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5539 vfm1 as a poly_uint64 rather than an int. Make the same change
5540 for the returned bound_scalar.
5541 (vect_gen_vector_loop_niters): Handle polynomial vf.
5542 (vect_do_peeling): Likewise. Update call to
5543 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5544 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5546 * tree-vect-loop.c (vect_determine_vectorization_factor)
5547 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5548 (vect_get_known_peeling_cost): Likewise.
5549 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5550 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5551 (vect_transform_loop): Likewise. Use the lowest possible VF when
5552 updating the upper bounds of the loop.
5553 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5555 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5556 polynomial unroll factors.
5557 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5558 (vect_make_slp_decision): Likewise.
5559 (vect_supported_load_permutation_p): Likewise, and polynomial
5561 (vect_analyze_slp_cost): Handle polynomial vf.
5562 (vect_slp_analyze_node_operations): Likewise.
5563 (vect_slp_analyze_bb_1): Likewise.
5564 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5565 than an unsigned HOST_WIDE_INT.
5566 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5567 (vectorizable_load): Handle polynomial vf.
5568 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5570 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5572 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5573 Alan Hayward <alan.hayward@arm.com>
5574 David Sherwood <david.sherwood@arm.com>
5576 * match.pd: Handle bit operations involving three constants
5577 and try to fold one pair.
5579 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5581 * tree-vect-loop-manip.c: Include gimple-fold.h.
5582 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5583 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5584 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5585 Add a path that uses a step of VF instead of 1, but disable it
5587 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5588 and niters_no_overflow parameters. Update calls to
5589 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5590 Create a new SSA name if the latter choses to use a ste other
5591 than zero, and return it via niters_vector_mult_vf_var.
5592 * tree-vect-loop.c (vect_transform_loop): Update calls to
5593 vect_do_peeling, vect_gen_vector_loop_niters and
5594 slpeel_make_loop_iterate_ntimes.
5595 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5596 (vect_gen_vector_loop_niters): Update declarations after above changes.
5598 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5600 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5601 128-bit round to integer instructions.
5602 (ceil<mode>2): Likewise.
5603 (btrunc<mode>2): Likewise.
5604 (round<mode>2): Likewise.
5606 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5608 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5609 unaligned VSX load/store on P8/P9.
5610 (expand_block_clear): Allow the use of unaligned VSX
5611 load/store on P8/P9.
5613 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5615 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5617 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5618 swap associated with both a load and a store.
5620 2018-01-02 Andrew Waterman <andrew@sifive.com>
5622 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5623 * config/riscv/riscv.md (clear_cache): Use it.
5625 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5627 * web.c: Remove out-of-date comment.
5629 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5631 * expr.c (fixup_args_size_notes): Check that any existing
5632 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5633 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5634 (emit_single_push_insn): ...here.
5636 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5638 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5639 (const_vector_encoded_nelts): New function.
5640 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5641 (const_vector_int_elt, const_vector_elt): Declare.
5642 * emit-rtl.c (const_vector_int_elt_1): New function.
5643 (const_vector_elt): Likewise.
5644 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5645 of CONST_VECTOR_ELT.
5647 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5649 * expr.c: Include rtx-vector-builder.h.
5650 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5651 directly on the tree encoding.
5652 (const_vector_from_tree): Likewise.
5653 * optabs.c: Include rtx-vector-builder.h.
5654 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5655 sequence of "u" values.
5656 * vec-perm-indices.c: Include rtx-vector-builder.h.
5657 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5658 directly on the vec_perm_indices encoding.
5660 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5662 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5663 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5664 * rtx-vector-builder.h: New file.
5665 * rtx-vector-builder.c: Likewise.
5666 * rtl.h (rtx_def::u2): Add a const_vector field.
5667 (CONST_VECTOR_NPATTERNS): New macro.
5668 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5669 (CONST_VECTOR_DUPLICATE_P): Likewise.
5670 (CONST_VECTOR_STEPPED_P): Likewise.
5671 (CONST_VECTOR_ENCODED_ELT): Likewise.
5672 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5673 (unwrap_const_vec_duplicate): Likewise.
5674 (const_vec_series_p): Check for a non-duplicated vector encoding.
5675 Say that the function only returns true for integer vectors.
5676 * emit-rtl.c: Include rtx-vector-builder.h.
5677 (gen_const_vec_duplicate_1): Delete.
5678 (gen_const_vector): Call gen_const_vec_duplicate instead of
5679 gen_const_vec_duplicate_1.
5680 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5681 (gen_const_vec_duplicate): Use rtx_vector_builder.
5682 (gen_const_vec_series): Likewise.
5683 (gen_rtx_CONST_VECTOR): Likewise.
5684 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5685 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5686 Build a new vector rather than modifying a CONST_VECTOR in-place.
5687 (handle_special_swappables): Update call accordingly.
5688 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5689 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5690 Build a new vector rather than modifying a CONST_VECTOR in-place.
5691 (handle_special_swappables): Update call accordingly.
5693 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5695 * simplify-rtx.c (simplify_const_binary_operation): Use
5696 CONST_VECTOR_ELT instead of XVECEXP.
5698 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5700 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5701 the selector elements to be different from the data elements
5702 if the selector is a VECTOR_CST.
5703 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5704 ssizetype for the selector.
5706 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5708 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5709 before testing each element individually.
5710 * tree-vect-generic.c (lower_vec_perm): Likewise.
5712 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5714 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5715 * selftest-run-tests.c (selftest::run_tests): Call it.
5716 * vector-builder.h (vector_builder::operator ==): New function.
5717 (vector_builder::operator !=): Likewise.
5718 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5719 (vec_perm_indices::all_from_input_p): New function.
5720 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5721 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5722 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5723 instead of reading the VECTOR_CST directly. Detect whether both
5724 vector inputs are the same before constructing the vec_perm_indices,
5725 and update the number of inputs argument accordingly. Use the
5726 utility functions added above. Only construct sel2 if we need to.
5728 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5730 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5731 the broadcast of the low byte.
5732 (expand_mult_highpart): Use an explicit encoding for the permutes.
5733 * optabs-query.c (can_mult_highpart_p): Likewise.
5734 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5735 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5736 (vectorizable_bswap): Likewise.
5737 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5738 explicit encoding for the power-of-2 permutes.
5739 (vect_permute_store_chain): Likewise.
5740 (vect_grouped_load_supported): Likewise.
5741 (vect_permute_load_chain): Likewise.
5743 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5745 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5746 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5747 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5748 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5749 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5750 (vect_gen_perm_mask_any): Likewise.
5752 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5754 * int-vector-builder.h: New file.
5755 * vec-perm-indices.h: Include int-vector-builder.h.
5756 (vec_perm_indices): Redefine as an int_vector_builder.
5757 (auto_vec_perm_indices): Delete.
5758 (vec_perm_builder): Redefine as a stand-alone class.
5759 (vec_perm_indices::vec_perm_indices): New function.
5760 (vec_perm_indices::clamp): Likewise.
5761 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5762 (vec_perm_indices::new_vector): New function.
5763 (vec_perm_indices::new_expanded_vector): Update for new
5764 vec_perm_indices class.
5765 (vec_perm_indices::rotate_inputs): New function.
5766 (vec_perm_indices::all_in_range_p): Operate directly on the
5767 encoded form, without computing elided elements.
5768 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5769 encoding. Update for new vec_perm_indices class.
5770 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5771 the given vec_perm_builder.
5772 (expand_vec_perm_var): Update vec_perm_builder constructor.
5773 (expand_mult_highpart): Use vec_perm_builder instead of
5774 auto_vec_perm_indices.
5775 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5776 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5777 or double series encoding as appropriate.
5778 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5779 vec_perm_indices instead of auto_vec_perm_indices.
5780 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5781 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5782 (vect_permute_store_chain): Likewise.
5783 (vect_grouped_load_supported): Likewise.
5784 (vect_permute_load_chain): Likewise.
5785 (vect_shift_permute_load_chain): Likewise.
5786 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5787 (vect_transform_slp_perm_load): Likewise.
5788 (vect_schedule_slp_instance): Likewise.
5789 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5790 (vectorizable_mask_load_store): Likewise.
5791 (vectorizable_bswap): Likewise.
5792 (vectorizable_store): Likewise.
5793 (vectorizable_load): Likewise.
5794 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5795 vec_perm_indices instead of auto_vec_perm_indices. Use
5796 tree_to_vec_perm_builder to read the vector from a tree.
5797 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5798 vec_perm_builder instead of a vec_perm_indices.
5799 (have_whole_vector_shift): Use vec_perm_builder and
5800 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5801 truncation to calc_vec_perm_mask_for_shift.
5802 (vect_create_epilog_for_reduction): Likewise.
5803 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5804 from auto_vec_perm_indices to vec_perm_indices.
5805 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5806 instead of changing individual elements.
5807 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5808 the vector in d.perm.
5809 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5810 from auto_vec_perm_indices to vec_perm_indices.
5811 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5812 instead of changing individual elements.
5813 (arm_vectorize_vec_perm_const): Use new_vector to install
5814 the vector in d.perm.
5815 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5816 Update vec_perm_builder constructor.
5817 (rs6000_expand_interleave): Likewise.
5818 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5819 (rs6000_expand_interleave): Likewise.
5821 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5823 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5824 to qimode could truncate the indices.
5825 * optabs.c (expand_vec_perm_var): Likewise.
5827 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5829 * Makefile.in (OBJS): Add vec-perm-indices.o.
5830 * vec-perm-indices.h: New file.
5831 * vec-perm-indices.c: Likewise.
5832 * target.h (vec_perm_indices): Replace with a forward class
5834 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5835 * optabs.h: Include vec-perm-indices.h.
5836 (expand_vec_perm): Delete.
5837 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5838 (expand_vec_perm_const): Declare.
5839 * target.def (vec_perm_const_ok): Replace with...
5840 (vec_perm_const): ...this new hook.
5841 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5842 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5843 * doc/tm.texi: Regenerate.
5844 * optabs.def (vec_perm_const): Delete.
5845 * doc/md.texi (vec_perm_const): Likewise.
5846 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5847 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5848 expand_vec_perm for constant permutation vectors. Assert that
5849 the mode of variable permutation vectors is the integer equivalent
5850 of the mode that is being permuted.
5851 * optabs-query.h (selector_fits_mode_p): Declare.
5852 * optabs-query.c: Include vec-perm-indices.h.
5853 (selector_fits_mode_p): New function.
5854 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5855 is defined, instead of checking whether the vec_perm_const_optab
5856 exists. Use targetm.vectorize.vec_perm_const instead of
5857 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5858 fit in the vector mode before using a variable permute.
5859 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5860 vec_perm_indices instead of an rtx.
5861 (expand_vec_perm): Replace with...
5862 (expand_vec_perm_const): ...this new function. Take the selector
5863 as a vec_perm_indices rather than an rtx. Also take the mode of
5864 the selector. Update call to shift_amt_for_vec_perm_mask.
5865 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5866 Use vec_perm_indices::new_expanded_vector to expand the original
5867 selector into bytes. Check whether the indices fit in the vector
5868 mode before using a variable permute.
5869 (expand_vec_perm_var): Make global.
5870 (expand_mult_highpart): Use expand_vec_perm_const.
5871 * fold-const.c: Includes vec-perm-indices.h.
5872 * tree-ssa-forwprop.c: Likewise.
5873 * tree-vect-data-refs.c: Likewise.
5874 * tree-vect-generic.c: Likewise.
5875 * tree-vect-loop.c: Likewise.
5876 * tree-vect-slp.c: Likewise.
5877 * tree-vect-stmts.c: Likewise.
5878 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5880 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5881 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5882 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5883 (aarch64_vectorize_vec_perm_const): ...this new function.
5884 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5885 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5886 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5887 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5888 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5889 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5890 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5892 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5893 check for NEON modes.
5894 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5895 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5896 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5897 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5899 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5900 the old VEC_PERM_CONST conditions.
5901 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5902 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5903 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5904 (ia64_vectorize_vec_perm_const_ok): Merge into...
5905 (ia64_vectorize_vec_perm_const): ...this new function.
5906 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5907 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5908 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5909 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5910 * config/mips/mips.c (mips_expand_vec_perm_const)
5911 (mips_vectorize_vec_perm_const_ok): Merge into...
5912 (mips_vectorize_vec_perm_const): ...this new function.
5913 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5914 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5915 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5916 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5917 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5918 (rs6000_expand_vec_perm_const): Delete.
5919 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5921 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5922 (altivec_expand_vec_perm_const_le): Take each operand individually.
5923 Operate on constant selectors rather than rtxes.
5924 (altivec_expand_vec_perm_const): Likewise. Update call to
5925 altivec_expand_vec_perm_const_le.
5926 (rs6000_expand_vec_perm_const): Delete.
5927 (rs6000_vectorize_vec_perm_const_ok): Delete.
5928 (rs6000_vectorize_vec_perm_const): New function.
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/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5934 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5935 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5936 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5937 (rs6000_expand_vec_perm_const): Delete.
5938 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5939 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5940 (altivec_expand_vec_perm_const_le): Take each operand individually.
5941 Operate on constant selectors rather than rtxes.
5942 (altivec_expand_vec_perm_const): Likewise. Update call to
5943 altivec_expand_vec_perm_const_le.
5944 (rs6000_expand_vec_perm_const): Delete.
5945 (rs6000_vectorize_vec_perm_const_ok): Delete.
5946 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5947 reference to the SPE evmerge intructions.
5948 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5949 an element count and rtx array.
5950 (rs6000_expand_extract_even): Update call accordingly.
5951 (rs6000_expand_interleave): Likewise.
5952 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5953 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5955 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5957 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5959 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5960 vector mode and that that mode matches the mode of the data
5962 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5963 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5964 directly using expand_vec_perm_1 when forcing selectors into
5966 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5968 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5970 * optabs-query.h (can_vec_perm_p): Delete.
5971 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5972 * optabs-query.c (can_vec_perm_p): Split into...
5973 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5974 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5975 particular selector is valid.
5976 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5977 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5978 (vect_grouped_load_supported): Likewise.
5979 (vect_shift_permute_load_chain): Likewise.
5980 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5981 (vect_transform_slp_perm_load): Likewise.
5982 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5983 (vectorizable_bswap): Likewise.
5984 (vect_gen_perm_mask_checked): Likewise.
5985 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5986 implementations of variable permutation vectors into account
5987 when deciding which selector to use.
5988 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5989 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5990 with a false third argument.
5991 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5992 to test whether the constant selector is valid and can_vec_perm_var_p
5993 to test whether a variable selector is valid.
5995 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5997 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5998 * optabs-query.c (can_vec_perm_p): Likewise.
5999 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6000 instead of vec_perm_indices.
6001 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6002 (vect_gen_perm_mask_checked): Likewise,
6003 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6004 (vect_gen_perm_mask_checked): Likewise,
6006 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6008 * optabs-query.h (qimode_for_vec_perm): Declare.
6009 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6010 (qimode_for_vec_perm): ...this new function.
6011 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6013 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6015 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6016 does not have a conditional at the top.
6018 2018-01-02 Richard Biener <rguenther@suse.de>
6020 * ipa-inline.c (big_speedup_p): Fix expression.
6022 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6025 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6028 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6032 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6033 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6034 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6035 cond_taken_branch_cost 3->4.
6037 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6039 PR tree-optimization/83581
6040 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6041 TODO_cleanup_cfg if any changes have been made.
6044 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6045 convert_modes if target mode has the right side, but different mode
6049 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6050 last argument when extracting from CONCAT. If either from_real or
6051 from_imag is NULL, use expansion through memory. If result is not
6052 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6053 the parts directly to inner mode, if even that fails, use expansion
6057 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6058 check for bswap in mode rather than HImode and use that in expand_unop
6061 Copyright (C) 2018 Free Software Foundation, Inc.
6063 Copying and distribution of this file, with or without modification,
6064 are permitted in any medium without royalty provided the copyright
6065 notice and this notice are preserved.