1 2018-02-06 Jakub Jelinek <jakub@redhat.com>
4 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
5 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
6 and skip it regardless of bb boundaries. Use CALL_P macro,
7 don't test INSN_P (insn) together with CALL_P or JUMP_P check
8 unnecessarily, formatting fix.
10 2018-02-06 Michael Collison <michael.collison@arm.com>
12 * config/arm/thumb2.md:
13 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
14 (*thumb_mov_notscc): Ditto.
16 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
19 * config/rs6000/rs6000.md (su code attribute): Use "u" for
20 unsigned_fix, not "s".
22 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
24 * configure.ac (gcc_fn_eh_frame_ro): New function.
25 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
26 correct .eh_frame permissions.
27 * configure: Regenerate.
29 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
31 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
34 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
36 * config/rs6000/rs6000.c (rs6000_option_override_internal):
37 Display warning message for -mno-speculate-indirect-jumps.
39 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
41 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
43 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
45 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
47 PR tree-optimization/84225
48 * tree-eh.c (find_trapping_overflow): Only call
49 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
51 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
54 * config/i386/i386.c: Reimplement the check of possible options
55 -mibt/-mshstk conbination. Change error messages.
56 * doc/invoke.texi: Fix a typo: remove extra '='.
58 2018-02-06 Marek Polacek <polacek@redhat.com>
60 PR tree-optimization/84228
61 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
63 2018-02-06 Tamar Christina <tamar.christina@arm.com>
66 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
67 emitted arch directives.
68 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
69 __ARM_FEATURE_COPROC before changing architectures.
71 2018-02-06 Richard Biener <rguenther@suse.de>
73 * config/i386/i386.c (print_reg): Fix typo.
74 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
76 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
78 * configure: Regenerate.
80 2018-02-05 Martin Sebor <msebor@redhat.com>
82 PR tree-optimization/83369
83 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
86 2018-02-05 Martin Liska <mliska@suse.cz>
88 * doc/invoke.texi: Cherry-pick upstream r323995.
90 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
92 * ira.c (ira_init_register_move_cost): Adjust comment.
94 2018-02-05 Martin Liska <mliska@suse.cz>
97 * doc/gcov.texi: Fix typo in documentation.
99 2018-02-05 Martin Liska <mliska@suse.cz>
101 PR gcov-profile/83879
102 * doc/gcov.texi: Document necessity of --dynamic-list-data when
103 using dlopen functionality.
105 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
107 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
108 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
109 _mm_maskz_range_ss, _mm_mask_range_round_ss,
110 _mm_maskz_range_round_ss): New intrinsics.
111 (__builtin_ia32_rangesd128_round)
112 (__builtin_ia32_rangess128_round): Remove.
113 (__builtin_ia32_rangesd128_mask_round,
114 __builtin_ia32_rangess128_mask_round): New builtins.
115 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
116 __builtin_ia32_rangess128_round): Remove.
117 (__builtin_ia32_rangesd128_mask_round,
118 __builtin_ia32_rangess128_mask_round): New builtins.
119 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
120 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
121 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
122 "<round_saeonly_constraint>")): Changed to ...
123 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
124 "<round_saeonly_scalar_constraint>")): ... this.
125 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
126 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
127 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
128 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
129 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
132 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
134 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
136 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
137 Remove all values except native, 8540 and 8548.
139 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
141 * config/i386/i386.c (ix86_output_function_return): Pass
142 INVALID_REGNUM, instead of -1, as invalid register number to
143 indirect_thunk_name and output_indirect_thunk.
145 2018-02-02 Julia Koval <julia.koval@intel.com>
147 * config.gcc: Add -march=icelake.
148 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
149 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
150 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
151 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
152 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
153 (processor_target_table): Add icelake.
154 (ix86_option_override_internal): Handle new PTAs.
155 (get_builtin_code_for_version): Handle icelake.
156 (M_INTEL_COREI7_ICELAKE): New.
157 (fold_builtin_cpu): Handle icelake.
158 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
159 * doc/invoke.texi: Add -march=icelake.
161 2018-02-02 Julia Koval <julia.koval@intel.com>
163 * config/i386/i386.c (ix86_option_override_internal): Change flags type
165 * wide-int-bitmask.h: New.
167 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
170 * config/i386/i386.md: Replace Pmode with word_mode in
171 builtin_setjmp_setup and builtin_longjmp to support x32.
173 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
177 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
179 (rs6000_supported_cpu_names): New static variable.
180 (linux_cpu_translation_table): Likewise.
181 (elf_platform) <cpu>: Define new static variable and use it.
182 Translate kernel AT_PLATFORM name to canonical name if needed.
183 Error if platform name is unknown.
185 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
188 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
190 2018-02-01 Jeff Law <law@redhat.com>
193 * config/i386/i386.c (release_scratch_register_on_entry): Add new
194 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
195 the scratch if RELEASE_VIA_POP is false.
196 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
197 If we have to save a temporary register, decrement SIZE appropriately.
198 Pass new arguments to release_scratch_register_on_entry.
199 (ix86_adjust_stack_and_probe): Likewise.
200 (ix86_emit_probe_stack_range): Pass new arguments to
201 release_scratch_register_on_entry.
203 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
205 PR rtl-optimization/84157
206 * combine.c (change_zero_ext): Use REG_P predicate in
207 front of HARD_REGISTER_P predicate.
209 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
211 * config/avr/avr.c (avr_option_override): Move disabling of
212 -fdelete-null-pointer-checks to...
213 * common/config/avr/avr-common.c (avr_option_optimization_table):
216 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
218 PR tree-optimization/81635
219 * tree-data-ref.c (split_constant_offset_1): For types that
220 wrap on overflow, try to use range info to prove that wrapping
223 2018-02-01 Renlin Li <renlin.li@arm.com>
226 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
228 (aarch64_register_move_cost): Likewise.
229 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
231 (REG_CLASS_NAMES): Likewise.
232 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
233 TAILCALL_ADDR_REGS. Remove IP registers.
234 * config/aarch64/aarch64.md (Ucs): Update register constraint.
236 2018-02-01 Richard Biener <rguenther@suse.de>
238 * domwalk.h (dom_walker::dom_walker): Add additional constructor
239 for specifying RPO order and allow NULL for that.
240 * domwalk.c (dom_walker::dom_walker): Likewise.
241 (dom_walker::walk): Handle NULL RPO order.
242 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
244 (rewrite_update_dom_walker): Likewise.
245 (mark_def_dom_walker): Likewise.
247 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
249 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
250 (aarch64_maybe_expand_sve_subreg_move): Declare.
251 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
252 * config/aarch64/predicates.md (aarch64_any_register_operand): New
254 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
255 that are semantically a reverse operation.
256 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
257 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
258 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
260 (aarch64_can_change_mode_class): For big-endian, forbid changes
261 between two SVE modes if they have different element sizes.
263 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
265 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
266 the TImode handling for big-endian targets.
268 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
270 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
271 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
273 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
274 Remove BSWAP handing for big-endian targets and use the form of
275 LD1RQ appropariate for the mode.
277 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
279 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
280 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
283 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
286 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
287 check for operands that need to go through aarch64_sve_reload_be.
289 2018-02-01 Jakub Jelinek <jakub@redhat.com>
291 PR tree-optimization/81661
292 PR tree-optimization/84117
293 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
294 * tree-eh.c: Include gimplify.h.
295 (find_trapping_overflow, replace_trapping_overflow,
296 rewrite_to_non_trapping_overflow): New functions.
297 * tree-vect-loop.c: Include tree-eh.h.
298 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
299 * tree-data-ref.c: Include tree-eh.h.
300 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
302 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
304 PR rtl-optimization/84123
305 * combine.c (change_zero_ext): Check if hard register satisfies
306 can_change_dest_mode before calling gen_lowpart_SUBREG.
308 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
311 * ira.c (ira_init_register_move_cost): Remove assert.
313 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
315 PR rtl-optimization/84071
316 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
317 * doc/tm.texi: Regenerate.
319 2018-01-31 Richard Biener <rguenther@suse.de>
321 PR tree-optimization/84132
322 * tree-data-ref.c (analyze_miv_subscript): Properly
323 check whether evolution_function_is_affine_multivariate_p
324 before calling gcd_of_steps_may_divide_p.
326 2018-01-31 Julia Koval <julia.koval@intel.com>
329 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
330 * config/i386/i386.md (rdpid_rex64) New.
331 (rdpid): Make 32bit only.
333 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
336 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
337 an IDENTIFIER_NODE for FUNCTION_TYPE's.
339 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
342 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
344 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
346 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
348 PR rtl-optimization/84071
349 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
350 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
352 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
354 * config/arc/arc.c (arc_handle_aux_attribute): New function.
355 (arc_attribute_table): Add 'aux' attribute.
356 (arc_in_small_data_p): Consider aux like variables.
357 (arc_is_aux_reg_p): New function.
358 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
359 (arc_get_aux_arg): New function.
360 (prepare_move_operands): Handle aux-register access.
361 (arc_handle_aux_attribute): New function.
362 * doc/extend.texi (ARC Variable attributes): Add subsection.
364 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
366 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
367 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
368 (arc_attribute_table): Add 'uncached' attribute.
369 (arc_print_operand): Print '.di' flag for uncached memory
371 (arc_in_small_data_p): Do not consider for small data the uncached
373 (arc_is_uncached_mem_p): New function.
374 * config/arc/predicates.md (compact_store_memory_operand): Check
375 for uncached memory accesses.
376 (nonvol_nonimm_operand): Likewise.
377 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
379 2018-01-31 Jakub Jelinek <jakub@redhat.com>
382 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
383 falign-loops=): Add Optimization flag.
385 2018-01-30 Jeff Law <law@redhat.com>
388 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
389 INT_REGISTERS_SAVED. Check it prior to calling
390 get_scratch_register_on_entry.
391 (ix86_adjust_stack_and_probe): Similarly.
392 (ix86_emit_probe_stack_range): Similarly.
393 (ix86_expand_prologue): Corresponding changes.
395 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
398 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
399 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
401 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
404 * lra-constraints.c (curr_insn_transform): Process AND in the
407 2018-01-30 Jakub Jelinek <jakub@redhat.com>
409 PR rtl-optimization/83986
410 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
411 dependence against last_pending_memory_flush in addition to
414 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
416 PR tree-optimization/81611
417 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
420 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
422 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
425 2018-01-30 Richard Biener <rguenther@suse.de>
426 Jakub Jelinek <jakub@redhat.com>
428 PR tree-optimization/84111
429 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
430 inner loops added during recursion, as they don't have up-to-date
433 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
436 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
437 (can_inline_edge_by_limits_p): ... here.
438 (can_early_inline_edge_p, check_callers,
439 update_caller_keys, update_callee_keys, recursive_inlining,
440 add_new_edges_to_heap, speculation_useful_p,
441 inline_small_functions,
442 inline_small_functions, flatten_function,
443 inline_to_all_callers_1): Update.
445 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
447 * profile-count.c (profile_count::combine_with_ipa_count): Handle
450 2018-01-30 Richard Biener <rguenther@suse.de>
452 PR tree-optimization/83008
453 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
454 invariant and constant vector uses in stmts when they need
457 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
460 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
461 * configure: Regenerate.
463 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
465 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
467 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
468 Use gen_rtx_REG rather than gen_lowpart.
470 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
472 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
473 rather than 0 when creating partial subregs.
475 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
477 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
480 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
483 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
484 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
485 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
486 flags. This restores the settings used before the 2017-07-24.
487 Turning off pre increment/decrement/modify allows IVOPTS to
488 optimize DF/SF loops where the index is an int.
490 2018-01-29 Richard Biener <rguenther@suse.de>
491 Kelvin Nilsen <kelvin@gcc.gnu.org>
494 * tree-vect-stmts.c (vectorizable_call): Don't call
495 targetm.vectorize_builtin_md_vectorized_function if callee is
498 2018-01-22 Carl Love <cel@us.ibm.com>
500 * doc/extend.tex: Fix typo in second arg in
501 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
503 2018-01-29 Richard Biener <rguenther@suse.de>
505 PR tree-optimization/84086
506 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
507 (flush_ssaname_freelist): When SSA names were released reset
510 2018-01-29 Richard Biener <rguenther@suse.de>
512 PR tree-optimization/84057
513 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
514 removed paths when removing edges.
516 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
518 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
519 -mfunction-return=@var{choice}.
521 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
524 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
526 (layout::print_source_line): Likewise.
527 (test_get_line_width_without_trailing_whitespace): Add test cases.
529 2018-01-27 Jakub Jelinek <jakub@redhat.com>
532 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
535 2018-01-26 Jim Wilson <jimw@sifive.com>
537 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
539 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
542 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
544 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
545 and CMP + SUB-immediate -> SUBS.
547 2018-01-26 Martin Sebor <msebor@redhat.com>
549 PR tree-optimization/83896
550 * tree-ssa-strlen.c (get_string_len): Rename...
551 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
552 Avoid assuming length is constant.
553 (handle_char_store): Use HOST_WIDE_INT for string length.
555 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
558 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
559 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
561 2018-01-26 Richard Biener <rguenther@suse.de>
563 PR rtl-optimization/84003
564 * dse.c (record_store): Only record redundant stores when
565 the earlier store aliases at least all accesses the later one does.
567 2018-01-26 Jakub Jelinek <jakub@redhat.com>
569 PR rtl-optimization/83985
570 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
571 REG_CFA_RESTORE insns.
572 (delete_unmarked_insns): Don't ignore separate shrink wrapping
573 REG_CFA_RESTORE insns here.
576 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
577 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
579 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
581 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
582 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
583 (arc_init): Likewise.
584 (arc_override_options): Likewise.
585 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
587 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
589 * config/arc/arc.h (TARGET_DBNZ): Define.
590 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
591 properly set the tune attribute.
592 (dbnz): Use TARGET_DBNZ guard.
593 * config/arc/arc.opt (mtune): Add core3 option.
595 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
597 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
598 recognize new pic like addresses.
599 (arc_delegitimize_address): Clean up.
601 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
603 * config/arc/arc-arches.def: Option mrf16 valid for all
605 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
606 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
607 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
608 * config/arc/arc-tables.opt: Regenerate.
609 * config/arc/arc.c (arc_conditional_register_usage): Handle
610 reduced register file case.
611 (arc_file_start): Set must have build attributes.
612 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
614 * config/arc/arc.opt (mrf16): Add new option.
615 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
616 * config/arc/genmultilib.awk: Handle new mrf16 option.
617 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
618 * config/arc/t-multilib: Regenerate.
619 * doc/invoke.texi (ARC Options): Document mrf16 option.
621 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
623 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
624 * config/arc/arc.c (arc_handle_secure_attribute): New function.
625 (arc_attribute_table): Add 'secure_call' attribute.
626 (arc_print_operand): Print secure call operand.
627 (arc_function_ok_for_sibcall): Don't optimize tail calls when
629 (arc_is_secure_call_p): New function. * config/arc/arc.md
630 (call_i): Add support for sjli instruction.
631 (call_value_i): Likewise.
632 * config/arc/constraints.md (Csc): New constraint.
634 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
635 John Eric Martin <John.Martin@emmicro-us.com>
637 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
638 * config/arc/arc.c (_arc_jli_section): New struct.
639 (arc_jli_section): New type.
640 (rc_jli_sections): New static variable.
641 (arc_handle_jli_attribute): New function.
642 (arc_attribute_table): Add jli_always and jli_fixed attribute.
643 (arc_file_end): New function.
644 (TARGET_ASM_FILE_END): Define.
645 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
646 (arc_add_jli_section): New function.
647 (jli_call_scan): Likewise.
648 (arc_reorg): Call jli_call_scan.
649 (arc_output_addsi): Remove 'S' from printing asm operand.
650 (arc_is_jli_call_p): New function.
651 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
653 (movhi_insn): Likewise.
654 (movsi_insn): Likewise.
655 (movsi_set_cc_insn): Likewise.
656 (loadqi_update): Likewise.
657 (load_zeroextendqisi_update): Likewise.
658 (load_signextendqisi_update): Likewise.
659 (loadhi_update): Likewise.
660 (load_zeroextendhisi_update): Likewise.
661 (load_signextendhisi_update): Likewise.
662 (loadsi_update): Likewise.
663 (loadsf_update): Likewise.
664 (movsicc_insn): Likewise.
665 (bset_insn): Likewise.
666 (bxor_insn): Likewise.
667 (bclr_insn): Likewise.
668 (bmsk_insn): Likewise.
669 (bicsi3_insn): Likewise.
670 (cmpsi_cc_c_insn): Likewise.
671 (movsi_ne): Likewise.
672 (movsi_cond_exec): Likewise.
673 (clrsbsi2): Likewise.
684 (call_i): Remove 'S' asm letter, add jli instruction.
685 (call_value_i): Likewise.
686 * config/arc/arc.op (mjli-always): New option.
687 * config/arc/constraints.md (Cji): New constraint.
688 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
690 (subsf3_fpx): Likewise.
691 (mulsf3_fpx): Likewise.
692 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
694 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
696 * doc/invoke.texi (ARC): Document mjli-always option.
698 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
700 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
701 and use incw and decw where possible
702 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
704 2018-01-26 Richard Biener <rguenther@suse.de>
706 PR tree-optimization/81082
707 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
708 association if it requires casting to unsigned.
709 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
710 from fold_plusminus_mult_expr to catch important cases late when
711 range info is available.
713 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
715 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
716 * configure.ac (hidden_linkonce): New test.
717 * configure: Regenerate.
718 * config.in: Regenerate.
720 2018-01-26 Julia Koval <julia.koval@intel.com>
722 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
723 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
724 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
725 _mm_mask_bitshuffle_epi64_mask): Fix type.
726 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
727 USI_FTYPE_V4DI_V4DI_USI): Remove.
728 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
729 __builtin_ia32_vpshufbitqmb256_mask,
730 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
731 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
732 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
734 2018-01-26 Alan Modra <amodra@gmail.com>
737 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
738 UNSPEC_VBPERMQ. Sort other unspecs.
740 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
742 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
744 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
747 * predict.c (drop_profile): Do not push/pop cfun; update also
749 (handle_missing_profiles): Fix logic looking for zero profiles.
751 2018-01-25 Jakub Jelinek <jakub@redhat.com>
754 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
755 on functions with #pragma omp declare simd or functions with simd
757 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
758 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
759 Remove trailing \n from warning_at calls.
761 2018-01-25 Tom de Vries <tom@codesourcery.com>
764 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
765 for neutered workers.
767 2018-01-24 Joseph Myers <joseph@codesourcery.com>
770 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
771 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
773 2018-01-24 Jeff Law <law@redhat.com>
776 * i386.c (get_probe_interval): Move to earlier point.
777 (ix86_compute_frame_layout): If -fstack-clash-protection and
778 the frame is larger than the probe interval, then use pushes
779 to save registers rather than reg->mem moves.
780 (ix86_expand_prologue): Remove conditional for int_registers_saved
783 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
786 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
787 min/max for never referenced object.
789 2018-01-24 Jakub Jelinek <jakub@redhat.com>
792 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
794 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
795 attributes from DECL_ATTRIBUTES (decl) without affecting
796 DECL_ATTRIBUTES (current_function_decl).
797 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
798 functions with non-NULL DECL_ABSTRACT_ORIGIN.
800 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
802 PR tree-optimization/83979
803 * fold-const.c (fold_comparison): Use constant_boolean_node
804 instead of boolean_{true,false}_node.
806 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
808 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
811 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
813 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
814 Simplify the clause that sets the length attribute.
815 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
816 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
817 clause that sets the length attribute.
818 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
820 2018-01-24 Tom de Vries <tom@codesourcery.com>
823 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
824 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
825 Add strict parameter.
826 (prevent_branch_around_nothing): Insert dummy insn between branch to
827 label and label with no ptx insn inbetween.
828 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
830 2018-01-24 Tom de Vries <tom@codesourcery.com>
833 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
834 for neutered threads in warp.
835 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
837 2018-01-24 Richard Biener <rguenther@suse.de>
839 PR tree-optimization/83176
840 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
843 2018-01-24 Richard Biener <rguenther@suse.de>
845 PR tree-optimization/82819
846 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
847 code generating pluses that are no-ops in the target precision.
849 2018-01-24 Richard Biener <rguenther@suse.de>
852 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
854 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
856 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
857 to merge probabilities.
858 * predict.c (probably_never_executed): Also mark as cold functions
859 with global 0 profile and guessed local profile.
860 * profile-count.c (profile_probability::combine_with_count): New
862 * profile-count.h (profile_probability::operator*,
863 profile_probability::operator*=, profile_probability::operator/,
864 profile_probability::operator/=): Reduce precision to adjusted
865 and set value to guessed on contradictory divisions.
866 (profile_probability::combine_with_freq): Remove.
867 (profile_probability::combine_wiht_count): Declare.
868 (profile_count::force_nonzero):: Set to adjusted.
869 (profile_count::probability_in):: Set quality to adjusted.
870 * tree-ssa-tail-merge.c (replace_block_by): Use
873 2018-01-23 Andrew Waterman <andrew@sifive.com>
874 Jim Wilson <jimw@sifive.com>
876 * config/riscv/riscv.c (riscv_stack_boundary): New.
877 (riscv_option_override): Set riscv_stack_boundary. Handle
878 riscv_preferred_stack_boundary_arg.
879 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
880 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
881 (STACK_BOUNDARY): Set to riscv_stack_boundary.
882 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
883 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
884 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
886 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
889 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
890 of struct ix86_frame.
891 (ix86_expand_epilogue): Likewise. Add a local variable for
892 the reg_save_offset field in struct ix86_frame.
894 2018-01-23 Bin Cheng <bin.cheng@arm.com>
896 PR tree-optimization/82604
897 * tree-loop-distribution.c (enum partition_kind): New enum item
898 PKIND_PARTIAL_MEMSET.
899 (partition_builtin_p): Support above new enum item.
900 (generate_code_for_partition): Ditto.
901 (compute_access_range): Differentiate cases that equality can be
902 proven at all loops, the innermost loops or no loops.
903 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
904 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
905 (finalize_partitions, distribute_loop): Don't fuse partition of
906 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
907 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
910 2018-01-23 Martin Liska <mliska@suse.cz>
912 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
913 order to ignore the predictor.
914 (PRED_POLYMORPHIC_CALL): Likewise.
915 (PRED_RECURSIVE_CALL): Likewise.
917 2018-01-23 Martin Liska <mliska@suse.cz>
919 * tree-profile.c (tree_profiling): Print function header to
920 aware reader which function we are working on.
921 * value-prof.c (gimple_find_values_to_profile): Do not print
922 not interesting value histograms.
924 2018-01-23 Martin Liska <mliska@suse.cz>
926 * profile-count.h (enum profile_quality): Add
927 profile_uninitialized as the first value. Do not number values
928 as they are zero based.
929 (profile_count::verify): Update sanity check.
930 (profile_probability::verify): Likewise.
932 2018-01-23 Nathan Sidwell <nathan@acm.org>
934 * doc/invoke.texi (ffor-scope): Deprecate.
936 2018-01-23 David Malcolm <dmalcolm@redhat.com>
938 PR tree-optimization/83510
939 * domwalk.c (set_all_edges_as_executable): New function.
940 (dom_walker::dom_walker): Convert bool param
941 "skip_unreachable_blocks" to enum reachability. Move setup of
942 edge flags to set_all_edges_as_executable and only do it when
943 reachability is REACHABLE_BLOCKS.
944 * domwalk.h (enum dom_walker::reachability): New enum.
945 (dom_walker::dom_walker): Convert bool param
946 "skip_unreachable_blocks" to enum reachability.
947 (set_all_edges_as_executable): New decl.
948 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
949 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
951 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
952 but converting true to REACHABLE_BLOCKS.
953 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
955 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
956 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
957 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
959 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
960 if check_all_array_refs will be called.
962 2018-01-23 David Malcolm <dmalcolm@redhat.com>
964 * tree.c (selftest::test_location_wrappers): Add more test
967 2018-01-23 David Malcolm <dmalcolm@redhat.com>
969 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
970 (selftest::test_bit_in_range): Likewise.
972 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
975 * doc/sourcebuild.texi (vect_float): Say that the selector
976 only describes the situation when -funsafe-math-optimizations is on.
977 (vect_float_strict): Document.
979 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
981 PR tree-optimization/83965
982 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
983 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
984 instead of checking only for a reduction.
985 (vect_recog_widen_sum_pattern): Likewise.
987 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
989 * predict.c (probably_never_executed): Only use precise profile info.
990 (compute_function_frequency): Skip after inlining hack since we now
991 have quality checking.
993 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
995 * profile-count.h (profile_probability::very_unlikely,
996 profile_probability::unlikely, profile_probability::even): Set
997 precision to guessed.
999 2018-01-23 Richard Biener <rguenther@suse.de>
1001 PR tree-optimization/83963
1002 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1003 Properly terminate dominator walk when crossing the exit edge not
1004 when visiting its source block.
1006 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1009 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1010 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1012 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1014 PR tree-optimization/83957
1015 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1016 semicolon after for body surrounded by braces.
1018 PR tree-optimization/83081
1019 * profile-count.h (profile_probability::split): New method.
1020 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1021 Use profile_probability::split.
1022 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1023 when splitting a single conditional jump into 2.
1025 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1027 PR tree-optimization/69452
1028 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1031 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1033 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1034 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1035 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1037 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1039 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1040 * config/rl78/rl78.md: New define_expand "movdi"
1041 * config/rl78/rl78.c: New function definition rl78_split_movdi
1043 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1046 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1048 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1049 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1050 128-bit to produce an UNSPEC move to get the double word with the
1051 signbit and then a shift directly to do signbit.
1052 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1053 implementation with a new version that just does either a direct
1054 move or a regular move. Move memory interface to separate insns.
1055 Move insns so they are next to the expander.
1056 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1057 with signbit move. Split big and little endian case.
1058 (signbit<mode>2_dm_mem_le): Likewise.
1059 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1060 (signbit<mode>2_dm2): Likewise.
1062 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1064 * config/rl78/rl78.md: New define_expand "anddi3".
1066 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1068 * config/rl78/rl78.md: New define_expand "umindi3".
1070 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1072 * config/rl78/rl78.md: New define_expand "smindi3".
1074 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1076 * config/rl78/rl78.md: New define_expand "smaxdi3".
1078 2018-01-22 Carl Love <cel@us.ibm.com>
1080 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1081 LVX_V1TI): Add macro expansion.
1082 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1083 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1084 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1085 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1086 Change check to determine if the instruction is a byte reversing
1087 entry. Fix typo in comment.
1088 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1089 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1090 Add def_builtin calls for new builtins.
1091 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1092 Add define_insn expansion.
1094 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1096 * config/rl78/rl78.md: New define_expand "umaxdi3".
1098 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1100 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1101 for non-QImode registers
1103 2018-01-22 Richard Biener <rguenther@suse.de>
1105 PR tree-optimization/83963
1106 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1107 including the loop exit block.
1108 (scop_detection::merge_sese): Likewise.
1109 (scop_detection::add_scop): Do it here instead.
1111 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1113 * doc/sourcebuild.texi (arm_softfloat): Document.
1115 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1118 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1119 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1120 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1122 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1123 David Edelsohn <dje.gcc@gmail.com>
1126 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1127 Change "crset eq" to "crset 2".
1128 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1129 (*call_indirect_aix<mode>_nospec): Likewise.
1130 (*call_value_indirect_aix<mode>_nospec): Likewise.
1131 (*call_indirect_elfv2<mode>_nospec): Likewise.
1132 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1133 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1134 change assembly output from . to $.
1135 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1136 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1137 (*tablejump<mode>_internal1_nospec): Likewise.
1139 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1142 * config/sh/sh_optimize_sett_clrt.cc:
1143 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1145 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1147 PR tree-optimization/83940
1148 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1149 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1150 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1151 use it to pass back the definition type.
1152 (vect_check_store_rhs): Likewise rhs_dt_out.
1153 (vect_build_gather_load_calls): Add a mask_dt argument and use
1154 it instead of a call to vect_is_simple_use.
1155 (vectorizable_store): Update calls to vect_check_load_store_mask
1156 and vect_check_store_rhs. Use the dt returned by the latter instead
1157 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1158 instead of calls to vect_is_simple_use. Pass the scalar rather
1159 than the vector operand to vect_is_simple_use when handling
1160 second and subsequent copies of an rhs value.
1161 (vectorizable_load): Update calls to vect_check_load_store_mask
1162 and vect_build_gather_load_calls. Use the cached mask_dt and
1163 gs_info.offset_dt instead of calls to vect_is_simple_use.
1165 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1168 * tree-emutls.c: Include gimplify.h.
1169 (lower_emutls_2): New function.
1170 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1171 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1172 it before further processing.
1175 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1176 UINTVAL (trueop1) instead of INTVAL (op1).
1178 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1182 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1183 INCOMING_FRAME_SP_OFFSET if not defined.
1184 (scan_trace): Add ENTRY argument. If true and
1185 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1186 emit a note to adjust the CFA offset.
1187 (create_cfi_notes): Adjust scan_trace callers.
1188 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1189 INCOMING_FRAME_SP_OFFSET in the CIE.
1190 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1191 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1193 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1194 * doc/tm.texi: Regenerated.
1196 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1198 PR rtl-optimization/83147
1199 * lra-constraints.c (remove_inheritance_pseudos): Use
1200 lra_substitute_pseudo_within_insn.
1202 2018-01-19 Tom de Vries <tom@codesourcery.com>
1203 Cesar Philippidis <cesar@codesourcery.com>
1206 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1208 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1211 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1212 spaces for function labels.
1214 2018-01-19 Martin Liska <mliska@suse.cz>
1216 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1217 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1218 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1219 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1220 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1221 (PRED_CONST_RETURN): Change from 69 to 65.
1222 (PRED_NULL_RETURN): Change from 91 to 71.
1223 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1224 (PRED_LOOP_GUARD): Change from 66 to 73.
1226 2018-01-19 Martin Liska <mliska@suse.cz>
1228 * predict.c (predict_insn_def): Add new assert.
1229 (struct branch_predictor): Change type to signed integer.
1230 (test_prediction_value_range): Amend test to cover
1232 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1233 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1234 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1235 (PRED_LOOP_IV_COMPARE): Likewise.
1236 * predict.h (PROB_UNINITIALIZED): Define new constant.
1238 2018-01-19 Martin Liska <mliska@suse.cz>
1240 * predict.c (dump_prediction): Add new format for
1241 analyze_brprob.py script which is enabled with -details
1243 * profile-count.h (precise_p): New function.
1245 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1247 PR tree-optimization/83922
1248 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1249 there are no statements that need masking.
1250 (vect_active_double_reduction_p): New function.
1251 (vect_analyze_loop_operations): Use it when handling phis that
1252 are not in the loop header.
1254 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1256 PR tree-optimization/83914
1257 * tree-vect-loop.c (vectorizable_induction): Don't convert
1258 init_expr or apply the peeling adjustment for inductions
1259 that are nested within the vectorized loop.
1261 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1263 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1266 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1270 * function.h (gimplify_parameters): Add gimple_seq * argument.
1271 * function.c: Include gimple.h and options.h.
1272 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1273 for the added local temporaries if needed.
1274 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1275 if there are any parameter cleanups, wrap whole body into a
1276 try/finally with the cleanups.
1278 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1281 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1282 Use GET_MODE_CLASS for scalar floating point.
1284 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1288 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1289 Fix call of call_cgraph_insertion_hooks.
1291 2018-01-18 Martin Sebor <msebor@redhat.com>
1293 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1295 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1298 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1301 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1304 * common.opt: (-ffile-prefix-map): New option.
1305 * opts.c (common_handle_option): Defer it.
1306 * opts-global.c (handle_common_deferred_options): Handle it.
1307 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1308 * file-prefix-map.h: New file.
1309 (remap_debug_filename, add_debug_prefix_map): ...here.
1310 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1311 * final.c (debug_prefix_map, add_debug_prefix_map
1312 remap_debug_filename): Move to...
1313 * file-prefix-map.c: New file.
1314 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1315 generalize, get rid of alloca(), use strrchr() instead of strchr().
1316 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1317 Implement in terms of add_prefix_map().
1318 (remap_macro_filename, remap_debug_filename): Implement in term of
1320 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1321 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1322 * dbxout.c: Include file-prefix-map.h.
1323 * varasm.c: Likewise.
1324 * vmsdbgout.c: Likewise.
1325 * xcoffout.c: Likewise.
1326 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1327 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1328 * doc/invoke.texi (-ffile-prefix-map): Document.
1329 (-fdebug-prefix-map): Update description.
1331 2018-01-18 Martin Liska <mliska@suse.cz>
1333 * config/i386/i386.c (indirect_thunk_name): Document that also
1335 (output_indirect_thunk): Document why both instructions
1336 (pause and lfence) are generated.
1338 2018-01-18 Richard Biener <rguenther@suse.de>
1340 PR tree-optimization/83887
1341 * graphite-scop-detection.c
1342 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1343 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1344 (scop_detection::merge_sese): Re-implement with a flood-fill
1345 algorithm that properly finds a SESE region if it exists.
1347 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1350 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1351 pointer_diff optimizations use view_convert instead of convert.
1353 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1355 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1356 Generate different code for -mno-speculate-indirect-jumps.
1357 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1358 (*call_indirect_aix<mode>): Disable for
1359 -mno-speculate-indirect-jumps.
1360 (*call_indirect_aix<mode>_nospec): New define_insn.
1361 (*call_value_indirect_aix<mode>): Disable for
1362 -mno-speculate-indirect-jumps.
1363 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1364 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1365 -mno-speculate-indirect-jumps.
1366 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1368 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1370 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1371 long double type, set the flags for noting the default long double
1372 type, even if we don't pass or return a long double type.
1374 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1377 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1380 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1382 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1383 support for merge[hl].
1384 (fold_mergehl_helper): New helper function.
1385 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1386 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1387 (altivec_vmrglw_direct): Add xxmrglw insn.
1389 2018-01-17 Andrew Waterman <andrew@sifive.com>
1391 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1392 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1394 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1397 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1398 call the lto_location_cache before reading the
1399 DECL_SOURCE_LOCATION of the types.
1401 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1402 Richard Sandiford <richard.sandiford@linaro.org>
1404 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1405 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1406 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1407 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1408 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1410 * config/aarch64/constraints.md (aarch64_movti_operand):
1412 * config/aarch64/predicates.md (Uti): Add new constraint.
1414 2018-01-17 Carl Love <cel@us.ibm.com>
1415 * config/rs6000/vsx.md (define_expand xl_len_r,
1416 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1417 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1419 (define_expand, define_insn): Move the shift left from the
1420 define_insn to the define_expand for lxvl and stxvl instructions.
1421 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1422 and XL_LEN_R definitions to PURE.
1424 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1426 * config/i386/i386.c (indirect_thunk_name): Declare regno
1427 as unsigned int. Compare regno with INVALID_REGNUM.
1428 (output_indirect_thunk): Ditto.
1429 (output_indirect_thunk_function): Ditto.
1430 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1431 in the call to output_indirect_thunk_function.
1433 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1436 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1437 rather than the size of inner_type to determine the stack slot size
1438 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1440 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1443 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1446 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1448 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1449 endian Linux systems to optionally enable multilibs for selecting
1450 the long double type if the user configured an explicit type.
1451 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1452 have no long double multilibs if not defined.
1453 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1454 warn if the user used -mabi={ieee,ibm}longdouble and we built
1455 multilibs for long double.
1456 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1457 appropriate multilib option.
1458 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1460 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1461 for building long double multilibs.
1462 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1464 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1466 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1469 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1471 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1474 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1477 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1480 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1482 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1483 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1485 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1487 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1488 different rtl trees depending on TARGET_64BIT.
1489 (rs6000_gen_lvx): Likewise.
1491 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1493 * config/visium/visium.md (nop): Tweak comment.
1494 (hazard_nop): Likewise.
1496 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1498 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1499 -mspeculate-indirect-jumps.
1500 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1501 for -mno-speculate-indirect-jumps.
1502 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1503 (*call_value_indirect_elfv2<mode>): Disable for
1504 -mno-speculate-indirect-jumps.
1505 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1506 (indirect_jump): Emit different RTL for
1507 -mno-speculate-indirect-jumps.
1508 (*indirect_jump<mode>): Disable for
1509 -mno-speculate-indirect-jumps.
1510 (*indirect_jump<mode>_nospec): New define_insn.
1511 (tablejump): Emit different RTL for
1512 -mno-speculate-indirect-jumps.
1513 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1514 (tablejumpsi_nospec): New define_expand.
1515 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1516 (tablejumpdi_nospec): New define_expand.
1517 (*tablejump<mode>_internal1): Disable for
1518 -mno-speculate-indirect-jumps.
1519 (*tablejump<mode>_internal1_nospec): New define_insn.
1520 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1523 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1525 * caller-save.c (insert_save): Drop unnecessary parameter. All
1528 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1529 Richard Biener <rguenth@suse.de>
1532 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1533 return early, inline manually is_gimple_sizepos. Make sure if we
1534 call gimplify_expr we don't end up with a gimple constant.
1535 * tree.c (variably_modified_type_p): Don't return true for
1536 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1537 * gimplify.h (is_gimple_sizepos): Remove.
1539 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1541 PR tree-optimization/83857
1542 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1543 vectorizable_live_operation for pure SLP statements.
1544 (vectorizable_live_operation): Handle PHIs.
1546 2018-01-16 Richard Biener <rguenther@suse.de>
1548 PR tree-optimization/83867
1549 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1550 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1552 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1555 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1556 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1557 If off is not INTEGER_CST, issue a may not be aligned warning
1558 rather than isn't aligned. Use isn%'t rather than isn't.
1559 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1561 <case MULT_EXPR>: Improve the case when bottom and one of the
1562 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1563 operand, in that case check if the other operand is multiple of
1564 bottom divided by the INTEGER_CST operand.
1566 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1569 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1570 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1571 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1572 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1573 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1574 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1575 (pa_function_arg_size): New function.
1577 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1579 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1580 in a separate statement.
1582 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1584 PR tree-optimization/83847
1585 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1586 group gathers and scatters.
1588 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1590 PR rtl-optimization/86620
1591 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1593 PR rtl-optimization/83213
1594 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1595 to last if both are JUMP_INSNs.
1597 PR tree-optimization/83843
1598 * gimple-ssa-store-merging.c
1599 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1600 store_immediate_info for bswap/nop orig_stores.
1602 2018-01-15 Andrew Waterman <andrew@sifive.com>
1604 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1606 <UDIV>: Increase cost if !TARGET_DIV.
1608 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1610 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1611 (define_attr "cr_logical_3op"): New.
1612 (cceq_ior_compare): Adjust.
1613 (cceq_ior_compare_complement): Adjust.
1614 (*cceq_rev_compare): Adjust.
1615 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1616 (is_cracked_insn): Adjust.
1617 (insn_must_be_first_in_group): Adjust.
1618 * config/rs6000/40x.md: Adjust.
1619 * config/rs6000/440.md: Adjust.
1620 * config/rs6000/476.md: Adjust.
1621 * config/rs6000/601.md: Adjust.
1622 * config/rs6000/603.md: Adjust.
1623 * config/rs6000/6xx.md: Adjust.
1624 * config/rs6000/7450.md: Adjust.
1625 * config/rs6000/7xx.md: Adjust.
1626 * config/rs6000/8540.md: Adjust.
1627 * config/rs6000/cell.md: Adjust.
1628 * config/rs6000/e300c2c3.md: Adjust.
1629 * config/rs6000/e500mc.md: Adjust.
1630 * config/rs6000/e500mc64.md: Adjust.
1631 * config/rs6000/e5500.md: Adjust.
1632 * config/rs6000/e6500.md: Adjust.
1633 * config/rs6000/mpc.md: Adjust.
1634 * config/rs6000/power4.md: Adjust.
1635 * config/rs6000/power5.md: Adjust.
1636 * config/rs6000/power6.md: Adjust.
1637 * config/rs6000/power7.md: Adjust.
1638 * config/rs6000/power8.md: Adjust.
1639 * config/rs6000/power9.md: Adjust.
1640 * config/rs6000/rs64.md: Adjust.
1641 * config/rs6000/titan.md: Adjust.
1643 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1645 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1646 ix86_indirect_branch_register logic.
1648 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1650 * config/i386/constraints.md (Bs): Update
1651 ix86_indirect_branch_register check. Don't check
1652 ix86_indirect_branch_register with GOT_memory_operand.
1654 * config/i386/predicates.md (GOT_memory_operand): Don't check
1655 ix86_indirect_branch_register here.
1656 (GOT32_symbol_operand): Likewise.
1658 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1660 * config/i386/predicates.md (constant_call_address_operand):
1661 Rewrite ix86_indirect_branch_register logic.
1662 (sibcall_insn_operand): Likewise.
1664 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1666 * config/i386/constraints.md (Bs): Replace
1667 ix86_indirect_branch_thunk_register with
1668 ix86_indirect_branch_register.
1670 * config/i386/i386.md (indirect_jump): Likewise.
1671 (tablejump): Likewise.
1672 (*sibcall_memory): Likewise.
1673 (*sibcall_value_memory): Likewise.
1674 Peepholes of indirect call and jump via memory: Likewise.
1675 * config/i386/i386.opt: Likewise.
1676 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1677 (GOT_memory_operand): Likewise.
1678 (call_insn_operand): Likewise.
1679 (sibcall_insn_operand): Likewise.
1680 (GOT32_symbol_operand): Likewise.
1682 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1685 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1686 type rather than type addr's type points to.
1687 (expand_omp_atomic_mutex): Likewise.
1688 (expand_omp_atomic): Likewise.
1690 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1693 * config/i386/i386.c (output_indirect_thunk_function): Use
1694 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1695 for __x86_return_thunk.
1697 2018-01-15 Richard Biener <rguenther@suse.de>
1700 * expmed.c (extract_bit_field_1): Fix typo.
1702 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1705 * config/arm/iterators.md (VF): New mode iterator.
1706 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1707 Remove integer-related logic from pattern.
1708 (neon_vabd<mode>_3): Likewise.
1710 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1713 * common.opt (fstrict-overflow): No longer an alias.
1714 (fwrapv-pointer): New option.
1715 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1716 also for pointer types based on flag_wrapv_pointer.
1717 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1718 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1719 opts->x_flag_wrapv got set.
1720 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1721 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1722 POINTER_TYPE_OVERFLOW_UNDEFINED.
1723 * match.pd: Likewise in address comparison pattern.
1724 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1726 2018-01-15 Richard Biener <rguenther@suse.de>
1729 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1730 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1731 Reset type names to their identifier if their TYPE_DECL doesn't
1732 have linkage (and thus is used for ODR and devirt).
1733 (save_debug_info_for_decl): Remove.
1734 (save_debug_info_for_type): Likewise.
1735 (add_tree_to_fld_list): Adjust.
1736 * tree-pretty-print.c (dump_generic_node): Make dumping of
1737 type names more robust.
1739 2018-01-15 Richard Biener <rguenther@suse.de>
1741 * BASE-VER: Bump to 8.0.1.
1743 2018-01-14 Martin Sebor <msebor@redhat.com>
1746 * builtins.c (check_access): Avoid warning when the no-warning bit
1749 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1751 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1752 * ira-color (allocno_hard_regs_compare): Likewise.
1754 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1757 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1758 Use .pushsection/.popsection.
1760 2018-01-14 Martin Sebor <msebor@redhat.com>
1763 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1765 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1767 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1768 entry from extra_headers.
1769 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1770 extra_headers, make the list bitwise identical to the i?86-*-* one.
1772 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1774 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1775 -mcmodel=large with -mindirect-branch=thunk,
1776 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1777 -mfunction-return=thunk-extern.
1778 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1779 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1780 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1782 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1784 * config/i386/i386.c (print_reg): Print the name of the full
1785 integer register without '%'.
1786 (ix86_print_operand): Handle 'V'.
1787 * doc/extend.texi: Document 'V' modifier.
1789 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1791 * config/i386/constraints.md (Bs): Disallow memory operand for
1792 -mindirect-branch-register.
1794 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1795 (GOT_memory_operand): Likewise.
1796 (call_insn_operand): Likewise.
1797 (sibcall_insn_operand): Likewise.
1798 (GOT32_symbol_operand): Likewise.
1799 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1800 for -mindirect-branch-register.
1801 (tablejump): Likewise.
1802 (*sibcall_memory): Likewise.
1803 (*sibcall_value_memory): Likewise.
1804 Disallow peepholes of indirect call and jump via memory for
1805 -mindirect-branch-register.
1806 (*call_pop): Replace m with Bw.
1807 (*call_value_pop): Likewise.
1808 (*sibcall_pop_memory): Replace m with Bs.
1809 * config/i386/i386.opt (mindirect-branch-register): New option.
1810 * doc/invoke.texi: Document -mindirect-branch-register option.
1812 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1814 * config/i386/i386-protos.h (ix86_output_function_return): New.
1815 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1816 set function_return_type.
1817 (indirect_thunk_name): Add ret_p to indicate thunk for function
1819 (output_indirect_thunk_function): Pass false to
1820 indirect_thunk_name.
1821 (ix86_output_indirect_branch_via_reg): Likewise.
1822 (ix86_output_indirect_branch_via_push): Likewise.
1823 (output_indirect_thunk_function): Create alias for function
1824 return thunk if regno < 0.
1825 (ix86_output_function_return): New function.
1826 (ix86_handle_fndecl_attribute): Handle function_return.
1827 (ix86_attribute_table): Add function_return.
1828 * config/i386/i386.h (machine_function): Add
1829 function_return_type.
1830 * config/i386/i386.md (simple_return_internal): Use
1831 ix86_output_function_return.
1832 (simple_return_internal_long): Likewise.
1833 * config/i386/i386.opt (mfunction-return=): New option.
1834 (indirect_branch): Mention -mfunction-return=.
1835 * doc/extend.texi: Document function_return function attribute.
1836 * doc/invoke.texi: Document -mfunction-return= option.
1838 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1840 * config/i386/i386-opts.h (indirect_branch): New.
1841 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1842 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1843 with local indirect jump when converting indirect call and jump.
1844 (ix86_set_indirect_branch_type): New.
1845 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1846 (indirectlabelno): New.
1847 (indirect_thunk_needed): Likewise.
1848 (indirect_thunk_bnd_needed): Likewise.
1849 (indirect_thunks_used): Likewise.
1850 (indirect_thunks_bnd_used): Likewise.
1851 (INDIRECT_LABEL): Likewise.
1852 (indirect_thunk_name): Likewise.
1853 (output_indirect_thunk): Likewise.
1854 (output_indirect_thunk_function): Likewise.
1855 (ix86_output_indirect_branch_via_reg): Likewise.
1856 (ix86_output_indirect_branch_via_push): Likewise.
1857 (ix86_output_indirect_branch): Likewise.
1858 (ix86_output_indirect_jmp): Likewise.
1859 (ix86_code_end): Call output_indirect_thunk_function if needed.
1860 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1862 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1863 (ix86_attribute_table): Add indirect_branch.
1864 * config/i386/i386.h (machine_function): Add indirect_branch_type
1865 and has_local_indirect_jump.
1866 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1868 (tablejump): Likewise.
1869 (*indirect_jump): Use ix86_output_indirect_jmp.
1870 (*tablejump_1): Likewise.
1871 (simple_return_indirect_internal): Likewise.
1872 * config/i386/i386.opt (mindirect-branch=): New option.
1873 (indirect_branch): New.
1876 (thunk-inline): Likewise.
1877 (thunk-extern): Likewise.
1878 * doc/extend.texi: Document indirect_branch function attribute.
1879 * doc/invoke.texi: Document -mindirect-branch= option.
1881 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1884 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1886 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1888 * ipa-inline.c (want_inline_small_function_p): Return false if
1889 inlining has already failed with CIF_FINAL_ERROR.
1890 (update_caller_keys): Call want_inline_small_function_p before
1892 (update_callee_keys): Likewise.
1894 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1896 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1898 (rs6000_quadword_masked_address_p): Likewise.
1899 (quad_aligned_load_p): Likewise.
1900 (quad_aligned_store_p): Likewise.
1901 (const_load_sequence_p): Add comment to describe the outer-most loop.
1902 (mimic_memory_attributes_and_flags): New function.
1903 (rs6000_gen_stvx): Likewise.
1904 (replace_swapped_aligned_store): Likewise.
1905 (rs6000_gen_lvx): Likewise.
1906 (replace_swapped_aligned_load): Likewise.
1907 (replace_swapped_load_constant): Capitalize argument name in
1908 comment describing this function.
1909 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1910 and stores that access quad-word aligned addresses and replace
1911 with stvx or lvx instructions when appropriate.
1912 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1913 New function prototype.
1914 (rs6000_quadword_masked_address_p): Likewise.
1915 (rs6000_gen_lvx): Likewise.
1916 (rs6000_gen_stvx): Likewise.
1917 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1918 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1919 when memory address is aligned.
1920 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1921 this split to select lvx instruction when memory address is aligned.
1922 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1923 instruction when memory address is aligned.
1924 (*vsx_le_perm_load_v16qi): Likewise.
1925 (four unnamed splitters): Modify to select the stvx instruction
1926 when memory is aligned.
1928 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1930 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1931 which appears in the queue multiple times.
1933 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1934 Alan Hayward <alan.hayward@arm.com>
1935 David Sherwood <david.sherwood@arm.com>
1937 * tree-vectorizer.h (vec_lower_bound): New structure.
1938 (_loop_vec_info): Add check_nonzero and lower_bounds.
1939 (LOOP_VINFO_CHECK_NONZERO): New macro.
1940 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1941 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1942 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1943 fields. Make seg_len the distance travelled, not including the
1945 (dr_direction_indicator): Declare.
1946 (dr_zero_step_indicator): Likewise.
1947 (dr_known_forward_stride_p): Likewise.
1948 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1950 (runtime_alias_check_p): Allow runtime alias checks with
1952 (operator ==): Compare access_size and align.
1953 (prune_runtime_alias_test_list): Rework for new distinction between
1954 the access_size and seg_len.
1955 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1957 (get_segment_min_max): New function.
1958 (create_intersect_range_checks): Use it.
1959 (dr_step_indicator): New function.
1960 (dr_direction_indicator): Likewise.
1961 (dr_zero_step_indicator): Likewise.
1962 (dr_known_forward_stride_p): Likewise.
1963 * tree-loop-distribution.c (data_ref_segment_size): Return
1964 DR_STEP * (niters - 1).
1965 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1967 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1968 (vect_preserves_scalar_order_p): New function, split out from...
1969 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1970 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1971 (vect_vfa_access_size): New function.
1972 (vect_vfa_align): Likewise.
1973 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1974 (dump_lower_bound): New function.
1975 (vect_check_lower_bound): Likewise.
1976 (vect_small_gap_p): Likewise.
1977 (vectorizable_with_step_bound_p): Likewise.
1978 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1979 depencies if the vectorization factor is 1. Convert the checks
1980 for nonzero steps into checks on the bounds of DR_STEP. Try using
1981 a bunds check for variable steps if the minimum required step is
1982 relatively small. Update calls to the dr_with_seg_len
1983 constructor and to vect_compile_time_alias.
1984 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1986 (vect_loop_versioning): Call it.
1987 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1989 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1991 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1992 Alan Hayward <alan.hayward@arm.com>
1993 David Sherwood <david.sherwood@arm.com>
1995 * doc/sourcebuild.texi (vect_scatter_store): Document.
1996 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1998 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2000 * genopinit.c (main): Add supports_vec_scatter_store and
2001 supports_vec_scatter_store_cached to target_optabs.
2002 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2003 IFN_MASK_SCATTER_STORE.
2004 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2006 * internal-fn.h (internal_store_fn_p): Declare.
2007 (internal_fn_stored_value_index): Likewise.
2008 * internal-fn.c (scatter_store_direct): New macro.
2009 (expand_scatter_store_optab_fn): New function.
2010 (direct_scatter_store_optab_supported_p): New macro.
2011 (internal_store_fn_p): New function.
2012 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2013 IFN_MASK_SCATTER_STORE.
2014 (internal_fn_mask_index): Likewise.
2015 (internal_fn_stored_value_index): New function.
2016 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2018 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2019 * optabs-query.c (supports_vec_scatter_store_p): New function.
2020 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2021 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2022 true for scatter stores.
2023 (vect_gather_scatter_fn_p): Handle scatter stores too.
2024 (vect_check_gather_scatter): Consider using scatter stores if
2025 supports_vec_scatter_store_p.
2026 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2028 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2029 internal_fn_stored_value_index.
2030 (check_load_store_masking): Handle scatter stores too.
2031 (vect_get_store_rhs): Make public.
2032 (vectorizable_call): Use internal_store_fn_p.
2033 (vectorizable_store): Handle scatter store internal functions.
2034 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2035 when deciding whether the end of the group has been reached.
2036 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2037 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2038 (mask_scatter_store<mode>): New insns.
2040 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2041 Alan Hayward <alan.hayward@arm.com>
2042 David Sherwood <david.sherwood@arm.com>
2044 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2045 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2046 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2048 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2049 Use vect_truncate_gather_scatter_offset if we can't treat the
2050 operation as a normal gather load or scatter store.
2051 (get_group_load_store_type): Take the gather_scatter_info
2052 as argument. Try using a gather load or scatter store for
2053 single-element groups.
2054 (get_load_store_type): Update calls to get_group_load_store_type
2055 and vect_use_strided_gather_scatters_p.
2057 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2058 Alan Hayward <alan.hayward@arm.com>
2059 David Sherwood <david.sherwood@arm.com>
2061 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2062 optional tree argument.
2063 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2065 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2066 but continue to use the current value as a fallback.
2067 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2068 to compare the updates.
2069 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2070 (get_load_store_type): Use it when handling a strided access.
2071 (vect_get_strided_load_store_ops): New function.
2072 (vect_get_data_ptr_increment): Likewise.
2073 (vectorizable_load): Handle strided gather loads. Always pass
2074 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2076 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2077 Alan Hayward <alan.hayward@arm.com>
2078 David Sherwood <david.sherwood@arm.com>
2080 * doc/md.texi (gather_load@var{m}): Document.
2081 (mask_gather_load@var{m}): Likewise.
2082 * genopinit.c (main): Add supports_vec_gather_load and
2083 supports_vec_gather_load_cached to target_optabs.
2084 * optabs-tree.c (init_tree_optimization_optabs): Use
2085 ggc_cleared_alloc to allocate target_optabs.
2086 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2087 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2089 * internal-fn.h (internal_load_fn_p): Declare.
2090 (internal_gather_scatter_fn_p): Likewise.
2091 (internal_fn_mask_index): Likewise.
2092 (internal_gather_scatter_fn_supported_p): Likewise.
2093 * internal-fn.c (gather_load_direct): New macro.
2094 (expand_gather_load_optab_fn): New function.
2095 (direct_gather_load_optab_supported_p): New macro.
2096 (direct_internal_fn_optab): New function.
2097 (internal_load_fn_p): Likewise.
2098 (internal_gather_scatter_fn_p): Likewise.
2099 (internal_fn_mask_index): Likewise.
2100 (internal_gather_scatter_fn_supported_p): Likewise.
2101 * optabs-query.c (supports_at_least_one_mode_p): New function.
2102 (supports_vec_gather_load_p): Likewise.
2103 * optabs-query.h (supports_vec_gather_load_p): Declare.
2104 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2105 and memory_type field.
2106 (NUM_PATTERNS): Bump to 15.
2107 * tree-vect-data-refs.c: Include internal-fn.h.
2108 (vect_gather_scatter_fn_p): New function.
2109 (vect_describe_gather_scatter_call): Likewise.
2110 (vect_check_gather_scatter): Try using internal functions for
2111 gather loads. Recognize existing calls to a gather load function.
2112 (vect_analyze_data_refs): Consider using gather loads if
2113 supports_vec_gather_load_p.
2114 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2115 (vect_get_gather_scatter_offset_type): Likewise.
2116 (vect_convert_mask_for_vectype): Likewise.
2117 (vect_add_conversion_to_patterm): Likewise.
2118 (vect_try_gather_scatter_pattern): Likewise.
2119 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2120 (vect_vect_recog_func_ptrs): Add it.
2121 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2122 internal_fn_mask_index and internal_gather_scatter_fn_p.
2123 (check_load_store_masking): Take the gather_scatter_info as an
2124 argument and handle gather loads.
2125 (vect_get_gather_scatter_ops): New function.
2126 (vectorizable_call): Check internal_load_fn_p.
2127 (vectorizable_load): Likewise. Handle gather load internal
2129 (vectorizable_store): Update call to check_load_store_masking.
2130 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2131 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2132 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2133 (aarch64_gather_scale_operand_d): New predicates.
2134 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2135 (mask_gather_load<mode>): New insns.
2137 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2138 Alan Hayward <alan.hayward@arm.com>
2139 David Sherwood <david.sherwood@arm.com>
2141 * optabs.def (fold_left_plus_optab): New optab.
2142 * doc/md.texi (fold_left_plus_@var{m}): Document.
2143 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2144 * internal-fn.c (fold_left_direct): Define.
2145 (expand_fold_left_optab_fn): Likewise.
2146 (direct_fold_left_optab_supported_p): Likewise.
2147 * fold-const-call.c (fold_const_fold_left): New function.
2148 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2149 * tree-parloops.c (valid_reduction_p): New function.
2150 (gather_scalar_reductions): Use it.
2151 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2152 (vect_finish_replace_stmt): Declare.
2153 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2154 (needs_fold_left_reduction_p): New function, split out from...
2155 (vect_is_simple_reduction): ...here. Accept reductions that
2156 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2157 (vect_force_simple_reduction): Also store the reduction type in
2158 the assignment's STMT_VINFO_REDUC_TYPE.
2159 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2160 (merge_with_identity): New function.
2161 (vect_expand_fold_left): Likewise.
2162 (vectorize_fold_left_reduction): Likewise.
2163 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2164 scalar phi in place for it. Check for target support and reject
2165 cases that would reassociate the operation. Defer the transform
2166 phase to vectorize_fold_left_reduction.
2167 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2168 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2169 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2171 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2173 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2174 call to ifc_temp_var.
2176 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2177 Alan Hayward <alan.hayward@arm.com>
2178 David Sherwood <david.sherwood@arm.com>
2180 * target.def (legitimize_address_displacement): Take the original
2181 offset as a poly_int.
2182 * targhooks.h (default_legitimize_address_displacement): Update
2184 * targhooks.c (default_legitimize_address_displacement): Likewise.
2185 * doc/tm.texi: Regenerate.
2186 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2187 as an argument, moving assert of ad->disp == ad->disp_term to...
2188 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2189 Try calling targetm.legitimize_address_displacement before expanding
2190 the address rather than afterwards, and adjust for the new interface.
2191 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2192 Match the new hook interface. Handle SVE addresses.
2193 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2196 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2198 * Makefile.in (OBJS): Add early-remat.o.
2199 * target.def (select_early_remat_modes): New hook.
2200 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2201 * doc/tm.texi: Regenerate.
2202 * targhooks.h (default_select_early_remat_modes): Declare.
2203 * targhooks.c (default_select_early_remat_modes): New function.
2204 * timevar.def (TV_EARLY_REMAT): New timevar.
2205 * passes.def (pass_early_remat): New pass.
2206 * tree-pass.h (make_pass_early_remat): Declare.
2207 * early-remat.c: New file.
2208 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2210 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
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 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2217 vfm1 with a bound_epilog parameter.
2218 (vect_do_peeling): Update calls accordingly, and move the prologue
2219 call earlier in the function. Treat the base bound_epilog as 0 for
2220 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2221 this base when peeling for gaps.
2222 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2223 with fully-masked loops.
2224 (vect_estimate_min_profitable_iters): Handle the single peeled
2225 iteration in that case.
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-vect-data-refs.c (vect_analyze_group_access_1): Allow
2232 single-element interleaving even if the size is not a power of 2.
2233 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2234 accesses for single-element interleaving if the group size is
2237 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2238 Alan Hayward <alan.hayward@arm.com>
2239 David Sherwood <david.sherwood@arm.com>
2241 * doc/md.texi (fold_extract_last_@var{m}): Document.
2242 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2243 * optabs.def (fold_extract_last_optab): New optab.
2244 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2245 * internal-fn.c (fold_extract_direct): New macro.
2246 (expand_fold_extract_optab_fn): Likewise.
2247 (direct_fold_extract_optab_supported_p): Likewise.
2248 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2249 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2250 EXTRACT_LAST_REDUCTION.
2251 (get_initial_def_for_reduction): Do not create an initial vector
2252 for EXTRACT_LAST_REDUCTION reductions.
2253 (vectorizable_reduction): Leave the scalar phi in place for
2254 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2255 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2256 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2257 transform phase to vectorizable_condition.
2258 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2260 (vect_finish_stmt_generation): ...here.
2261 (vect_finish_replace_stmt): New function.
2262 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2263 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2265 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2267 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2268 Alan Hayward <alan.hayward@arm.com>
2269 David Sherwood <david.sherwood@arm.com>
2271 * doc/md.texi (extract_last_@var{m}): Document.
2272 * optabs.def (extract_last_optab): New optab.
2273 * internal-fn.def (EXTRACT_LAST): New internal function.
2274 * internal-fn.c (cond_unary_direct): New macro.
2275 (expand_cond_unary_optab_fn): Likewise.
2276 (direct_cond_unary_optab_supported_p): Likewise.
2277 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2278 loops using EXTRACT_LAST.
2279 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2280 (extract_last_<mode>): ...this optab.
2281 (vec_extract<mode><Vel>): Update accordingly.
2283 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2284 Alan Hayward <alan.hayward@arm.com>
2285 David Sherwood <david.sherwood@arm.com>
2287 * target.def (empty_mask_is_expensive): New hook.
2288 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2289 * doc/tm.texi: Regenerate.
2290 * targhooks.h (default_empty_mask_is_expensive): Declare.
2291 * targhooks.c (default_empty_mask_is_expensive): New function.
2292 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2293 if the target says that empty masks are expensive.
2294 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2296 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2298 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2299 Alan Hayward <alan.hayward@arm.com>
2300 David Sherwood <david.sherwood@arm.com>
2302 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2303 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2304 (vect_use_loop_mask_for_alignment_p): New function.
2305 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2306 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2307 niters_skip argument. Make sure that the first niters_skip elements
2308 of the first iteration are inactive.
2309 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2310 Update call to vect_set_loop_masks_directly.
2311 (get_misalign_in_elems): New function, split out from...
2312 (vect_gen_prolog_loop_niters): ...here.
2313 (vect_update_init_of_dr): Take a code argument that specifies whether
2314 the adjustment should be added or subtracted.
2315 (vect_update_init_of_drs): Likewise.
2316 (vect_prepare_for_masked_peels): New function.
2317 (vect_do_peeling): Skip prologue peeling if we're using a mask
2318 instead. Update call to vect_update_inits_of_drs.
2319 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2321 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2322 alignment. Do not include the number of peeled iterations in
2323 the minimum threshold in that case.
2324 (vectorizable_induction): Adjust the start value down by
2325 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2326 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2327 Take the number of skipped iterations into account when calculating
2329 * tree-vect-stmts.c (vect_gen_while_not): New function.
2331 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2332 Alan Hayward <alan.hayward@arm.com>
2333 David Sherwood <david.sherwood@arm.com>
2335 * doc/sourcebuild.texi (vect_fully_masked): Document.
2336 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2338 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2340 (vect_analyze_loop_2): ...here. Don't check the vectorization
2341 factor against the number of loop iterations if the loop is
2344 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2345 Alan Hayward <alan.hayward@arm.com>
2346 David Sherwood <david.sherwood@arm.com>
2348 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2349 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2350 (dump_groups): Update accordingly.
2351 (iv_use::mem_type): New member variable.
2352 (address_p): New function.
2353 (record_use): Add a mem_type argument and initialize the new
2355 (record_group_use): Add a mem_type argument. Use address_p.
2356 Remove obsolete null checks of base_object. Update call to record_use.
2357 (find_interesting_uses_op): Update call to record_group_use.
2358 (find_interesting_uses_cond): Likewise.
2359 (find_interesting_uses_address): Likewise.
2360 (get_mem_type_for_internal_fn): New function.
2361 (find_address_like_use): Likewise.
2362 (find_interesting_uses_stmt): Try find_address_like_use before
2363 calling find_interesting_uses_op.
2364 (addr_offset_valid_p): Use the iv mem_type field as the type
2365 of the addressed memory.
2366 (add_autoinc_candidates): Likewise.
2367 (get_address_cost): Likewise.
2368 (split_small_address_groups_p): Use address_p.
2369 (split_address_groups): Likewise.
2370 (add_iv_candidate_for_use): Likewise.
2371 (autoinc_possible_for_pair): Likewise.
2372 (rewrite_groups): Likewise.
2373 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2374 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2375 (get_alias_ptr_type_for_ptr_address): New function.
2376 (rewrite_use_address): Rewrite address uses in calls that were
2377 identified by find_address_like_use.
2379 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2380 Alan Hayward <alan.hayward@arm.com>
2381 David Sherwood <david.sherwood@arm.com>
2383 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2385 * gimple-expr.h (is_gimple_addressable: Likewise.
2386 * gimple-expr.c (is_gimple_address): Likewise.
2387 * internal-fn.c (expand_call_mem_ref): New function.
2388 (expand_mask_load_optab_fn): Use it.
2389 (expand_mask_store_optab_fn): Likewise.
2391 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2392 Alan Hayward <alan.hayward@arm.com>
2393 David Sherwood <david.sherwood@arm.com>
2395 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2396 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2397 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2398 (cond_umax@var{mode}): Document.
2399 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2400 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2401 (cond_umin_optab, cond_umax_optab): New optabs.
2402 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2403 (COND_IOR, COND_XOR): New internal functions.
2404 * internal-fn.h (get_conditional_internal_fn): Declare.
2405 * internal-fn.c (cond_binary_direct): New macro.
2406 (expand_cond_binary_optab_fn): Likewise.
2407 (direct_cond_binary_optab_supported_p): Likewise.
2408 (get_conditional_internal_fn): New function.
2409 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2410 Cope with reduction statements that are vectorized as calls rather
2412 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2413 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2414 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2415 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2416 (UNSPEC_COND_EOR): New unspecs.
2417 (optab): Add mappings for them.
2418 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2419 (sve_int_op, sve_fp_op): New int attributes.
2421 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2422 Alan Hayward <alan.hayward@arm.com>
2423 David Sherwood <david.sherwood@arm.com>
2425 * optabs.def (while_ult_optab): New optab.
2426 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2427 * internal-fn.def (WHILE_ULT): New internal function.
2428 * internal-fn.h (direct_internal_fn_supported_p): New override
2429 that takes two types as argument.
2430 * internal-fn.c (while_direct): New macro.
2431 (expand_while_optab_fn): New function.
2432 (convert_optab_supported_p): Likewise.
2433 (direct_while_optab_supported_p): New macro.
2434 * wide-int.h (wi::udiv_ceil): New function.
2435 * tree-vectorizer.h (rgroup_masks): New structure.
2436 (vec_loop_masks): New typedef.
2437 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2439 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2440 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2441 (vect_max_vf): New function.
2442 (slpeel_make_loop_iterate_ntimes): Delete.
2443 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2444 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2445 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2446 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2447 internal-fn.h, stor-layout.h and optabs-query.h.
2448 (vect_set_loop_mask): New function.
2449 (add_preheader_seq): Likewise.
2450 (add_header_seq): Likewise.
2451 (interleave_supported_p): Likewise.
2452 (vect_maybe_permute_loop_masks): Likewise.
2453 (vect_set_loop_masks_directly): Likewise.
2454 (vect_set_loop_condition_masked): Likewise.
2455 (vect_set_loop_condition_unmasked): New function, split out from
2456 slpeel_make_loop_iterate_ntimes.
2457 (slpeel_make_loop_iterate_ntimes): Rename to..
2458 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2459 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2460 (vect_do_peeling): Update call accordingly.
2461 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2463 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2464 mask_compare_type, can_fully_mask_p and fully_masked_p.
2465 (release_vec_loop_masks): New function.
2466 (_loop_vec_info): Use it to free the loop masks.
2467 (can_produce_all_loop_masks_p): New function.
2468 (vect_get_max_nscalars_per_iter): Likewise.
2469 (vect_verify_full_masking): Likewise.
2470 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2471 retries, and free the mask rgroups before retrying. Check loop-wide
2472 reasons for disallowing fully-masked loops. Make the final decision
2473 about whether use a fully-masked loop or not.
2474 (vect_estimate_min_profitable_iters): Do not assume that peeling
2475 for the number of iterations will be needed for fully-masked loops.
2476 (vectorizable_reduction): Disable fully-masked loops.
2477 (vectorizable_live_operation): Likewise.
2478 (vect_halve_mask_nunits): New function.
2479 (vect_double_mask_nunits): Likewise.
2480 (vect_record_loop_mask): Likewise.
2481 (vect_get_loop_mask): Likewise.
2482 (vect_transform_loop): Handle the case in which the final loop
2483 iteration might handle a partial vector. Call vect_set_loop_condition
2484 instead of slpeel_make_loop_iterate_ntimes.
2485 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2486 (check_load_store_masking): New function.
2487 (prepare_load_store_mask): Likewise.
2488 (vectorizable_store): Handle fully-masked loops.
2489 (vectorizable_load): Likewise.
2490 (supportable_widening_operation): Use vect_halve_mask_nunits for
2492 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2493 (vect_gen_while): New function.
2494 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2495 (aarch64_uqdec<mode>): New insn.
2497 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2498 Alan Hayward <alan.hayward@arm.com>
2499 David Sherwood <david.sherwood@arm.com>
2501 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2502 (reduc_xor_scal_optab): New optabs.
2503 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2504 (reduc_xor_scal_@var{m}): Document.
2505 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2506 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2508 * fold-const-call.c (fold_const_call): Handle them.
2509 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2510 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2511 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2512 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2513 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2514 (UNSPEC_XORV): New unspecs.
2515 (optab): Add entries for them.
2516 (BITWISEV): New int iterator.
2517 (bit_reduc_op): New int attributes.
2519 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2520 Alan Hayward <alan.hayward@arm.com>
2521 David Sherwood <david.sherwood@arm.com>
2523 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2524 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2525 * optabs.def (vec_shl_insert_optab): New optab.
2526 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2527 (duplicate_and_interleave): Likewise.
2528 * tree-vect-loop.c: Include internal-fn.h.
2529 (neutral_op_for_slp_reduction): New function, split out from
2530 get_initial_defs_for_reduction.
2531 (get_initial_def_for_reduction): Handle option 2 for variable-length
2532 vectors by loading the neutral value into a vector and then shifting
2533 the initial value into element 0.
2534 (get_initial_defs_for_reduction): Replace the code argument with
2535 the neutral value calculated by neutral_op_for_slp_reduction.
2536 Use gimple_build_vector for constant-length vectors.
2537 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2538 but the first group_size elements have a neutral value.
2539 Use duplicate_and_interleave otherwise.
2540 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2541 Update call to get_initial_defs_for_reduction. Handle SLP
2542 reductions for variable-length vectors by creating one vector
2543 result for each scalar result, with the elements associated
2544 with other scalar results stubbed out with the neutral value.
2545 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2546 Require IFN_VEC_SHL_INSERT for double reductions on
2547 variable-length vectors, or SLP reductions that have
2548 a neutral value. Require can_duplicate_and_interleave_p
2549 support for variable-length unchained SLP reductions if there
2550 is no neutral value, such as for MIN/MAX reductions. Also require
2551 the number of vector elements to be a multiple of the number of
2552 SLP statements when doing variable-length unchained SLP reductions.
2553 Update call to vect_create_epilog_for_reduction.
2554 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2555 and remove initial values.
2556 (duplicate_and_interleave): Make public.
2557 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2558 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2560 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2561 Alan Hayward <alan.hayward@arm.com>
2562 David Sherwood <david.sherwood@arm.com>
2564 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2565 (can_duplicate_and_interleave_p): New function.
2566 (vect_get_and_check_slp_defs): Take the vector of statements
2567 rather than just the current one. Remove excess parentheses.
2568 Restriction rejectinon of vect_constant_def and vect_external_def
2569 for variable-length vectors to boolean types, or types for which
2570 can_duplicate_and_interleave_p is false.
2571 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2572 (duplicate_and_interleave): New function.
2573 (vect_get_constant_vectors): Use gimple_build_vector for
2574 constant-length vectors and suitable variable-length constant
2575 vectors. Use duplicate_and_interleave for other variable-length
2576 vectors. Don't defer the update when inserting new statements.
2578 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2579 Alan Hayward <alan.hayward@arm.com>
2580 David Sherwood <david.sherwood@arm.com>
2582 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2583 min_profitable_iters doesn't go negative.
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 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2590 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2591 * optabs.def (vec_mask_load_lanes_optab): New optab.
2592 (vec_mask_store_lanes_optab): Likewise.
2593 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2594 (MASK_STORE_LANES): Likewise.
2595 * internal-fn.c (mask_load_lanes_direct): New macro.
2596 (mask_store_lanes_direct): Likewise.
2597 (expand_mask_load_optab_fn): Handle masked operations.
2598 (expand_mask_load_lanes_optab_fn): New macro.
2599 (expand_mask_store_optab_fn): Handle masked operations.
2600 (expand_mask_store_lanes_optab_fn): New macro.
2601 (direct_mask_load_lanes_optab_supported_p): Likewise.
2602 (direct_mask_store_lanes_optab_supported_p): Likewise.
2603 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2605 (vect_load_lanes_supported): Likewise.
2606 * tree-vect-data-refs.c (strip_conversion): New function.
2607 (can_group_stmts_p): Likewise.
2608 (vect_analyze_data_ref_accesses): Use it instead of checking
2609 for a pair of assignments.
2610 (vect_store_lanes_supported): Take a masked_p parameter.
2611 (vect_load_lanes_supported): Likewise.
2612 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2613 vect_store_lanes_supported and vect_load_lanes_supported.
2614 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2615 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2616 parameter. Don't allow gaps for masked accesses.
2617 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2618 and vect_load_lanes_supported.
2619 (get_load_store_type): Take a masked_p parameter and update
2620 call to get_group_load_store_type.
2621 (vectorizable_store): Update call to get_load_store_type.
2622 Handle IFN_MASK_STORE_LANES.
2623 (vectorizable_load): Update call to get_load_store_type.
2624 Handle IFN_MASK_LOAD_LANES.
2626 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2627 Alan Hayward <alan.hayward@arm.com>
2628 David Sherwood <david.sherwood@arm.com>
2630 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2632 * config/aarch64/aarch64-protos.h
2633 (aarch64_sve_struct_memory_operand_p): Declare.
2634 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2635 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2636 (VPRED, vpred): Handle SVE structure modes.
2637 * config/aarch64/constraints.md (Utx): New constraint.
2638 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2639 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2640 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2641 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2642 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2643 structure modes. Split into pieces after RA.
2644 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2645 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2647 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2648 SVE structure modes.
2649 (aarch64_classify_address): Likewise.
2650 (sizetochar): Move earlier in file.
2651 (aarch64_print_operand): Handle SVE register lists.
2652 (aarch64_array_mode): New function.
2653 (aarch64_sve_struct_memory_operand_p): Likewise.
2654 (TARGET_ARRAY_MODE): Redefine.
2656 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2657 Alan Hayward <alan.hayward@arm.com>
2658 David Sherwood <david.sherwood@arm.com>
2660 * target.def (array_mode): New target hook.
2661 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2662 * doc/tm.texi: Regenerate.
2663 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2664 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2665 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2667 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2670 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2671 Alan Hayward <alan.hayward@arm.com>
2672 David Sherwood <david.sherwood@arm.com>
2674 * fold-const.c (fold_binary_loc): Check the argument types
2675 rather than the result type when testing for a vector operation.
2677 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2679 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2680 * doc/tm.texi: Regenerate.
2682 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2683 Alan Hayward <alan.hayward@arm.com>
2684 David Sherwood <david.sherwood@arm.com>
2686 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2687 (sve): Document new AArch64 extension.
2688 * doc/md.texi (w): Extend the description of the AArch64
2689 constraint to include SVE vectors.
2690 (Upl, Upa): Document new AArch64 predicate constraints.
2691 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2693 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2694 (msve-vector-bits=): New option.
2695 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2696 SVE when these are disabled.
2697 (sve): New extension.
2698 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2699 modes. Adjust their number of units based on aarch64_sve_vg.
2700 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2701 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2702 aarch64_addr_query_type.
2703 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2704 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2705 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2706 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2707 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2708 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2709 (aarch64_simd_imm_zero_p): Delete.
2710 (aarch64_check_zero_based_sve_index_immediate): Declare.
2711 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2712 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2713 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2714 (aarch64_sve_float_mul_immediate_p): Likewise.
2715 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2717 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2718 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2719 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2720 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2721 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2722 (aarch64_regmode_natural_size): Likewise.
2723 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2724 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2726 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2727 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2728 for VG and the SVE predicate registers.
2729 (V_ALIASES): Add a "z"-prefixed alias.
2730 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2731 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2732 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2733 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2734 (REG_CLASS_NAMES): Add entries for them.
2735 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2736 and the predicate registers.
2737 (aarch64_sve_vg): Declare.
2738 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2739 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2740 (REGMODE_NATURAL_SIZE): Define.
2741 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2743 * config/aarch64/aarch64.c: Include cfgrtl.h.
2744 (simd_immediate_info): Add a constructor for series vectors,
2745 and an associated step field.
2746 (aarch64_sve_vg): New variable.
2747 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2748 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2749 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2750 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2751 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2752 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2753 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2754 (aarch64_get_mask_mode): New functions.
2755 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2756 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2757 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2758 predicate modes and predicate registers. Explicitly restrict
2759 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2760 to store a vector mode if it is recognized by
2761 aarch64_classify_vector_mode.
2762 (aarch64_regmode_natural_size): New function.
2763 (aarch64_hard_regno_caller_save_mode): Return the original mode
2765 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2766 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2767 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2768 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2770 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2771 does not overlap dest if the function is frame-related. Handle
2773 (aarch64_split_add_offset): New function.
2774 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2775 them aarch64_add_offset.
2776 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2777 and update call to aarch64_sub_sp.
2778 (aarch64_add_cfa_expression): New function.
2779 (aarch64_expand_prologue): Pass extra temporary registers to the
2780 functions above. Handle the case in which we need to emit new
2781 DW_CFA_expressions for registers that were originally saved
2782 relative to the stack pointer, but now have to be expressed
2783 relative to the frame pointer.
2784 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2786 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2787 IP0 and IP1 values for SVE frames.
2788 (aarch64_expand_vec_series): New function.
2789 (aarch64_expand_sve_widened_duplicate): Likewise.
2790 (aarch64_expand_sve_const_vector): Likewise.
2791 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2792 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2793 into the register, rather than emitting a SET directly.
2794 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2795 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2796 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2797 (offset_9bit_signed_scaled_p): New functions.
2798 (aarch64_replicate_bitmask_imm): New function.
2799 (aarch64_bitmask_imm): Use it.
2800 (aarch64_cannot_force_const_mem): Reject expressions involving
2801 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2802 (aarch64_classify_index): Handle SVE indices, by requiring
2803 a plain register index with a scale that matches the element size.
2804 (aarch64_classify_address): Handle SVE addresses. Assert that
2805 the mode of the address is VOIDmode or an integer mode.
2806 Update call to aarch64_classify_symbol.
2807 (aarch64_classify_symbolic_expression): Update call to
2808 aarch64_classify_symbol.
2809 (aarch64_const_vec_all_in_range_p): New function.
2810 (aarch64_print_vector_float_operand): Likewise.
2811 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2812 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2813 and the FP immediates 1.0 and 0.5.
2814 (aarch64_print_address_internal): Handle SVE addresses.
2815 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2816 (aarch64_regno_regclass): Handle predicate registers.
2817 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2819 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2820 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2821 (aarch64_convert_sve_vector_bits): New function.
2822 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2823 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2825 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2826 Handle SVE vector and predicate modes. Accept VL-based constants
2827 that need only one temporary register, and VL offsets that require
2828 no temporary registers.
2829 (aarch64_conditional_register_usage): Mark the predicate registers
2830 as fixed if SVE isn't available.
2831 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2832 Return true for SVE vector and predicate modes.
2833 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2834 rather than an unsigned int. Handle SVE modes.
2835 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2837 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2839 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2840 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2841 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2842 (aarch64_sve_float_mul_immediate_p): New functions.
2843 (aarch64_sve_valid_immediate): New function.
2844 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2845 Explicitly reject structure modes. Check for INDEX constants.
2846 Handle PTRUE and PFALSE constants.
2847 (aarch64_check_zero_based_sve_index_immediate): New function.
2848 (aarch64_simd_imm_zero_p): Delete.
2849 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2850 vector modes. Accept constants in the range of CNT[BHWD].
2851 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2852 ask for an Advanced SIMD mode.
2853 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2854 (aarch64_simd_vector_alignment): Handle SVE predicates.
2855 (aarch64_vectorize_preferred_vector_alignment): New function.
2856 (aarch64_simd_vector_alignment_reachable): Use it instead of
2858 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2859 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2861 (MAX_VECT_LEN): Delete.
2862 (expand_vec_perm_d): Add a vec_flags field.
2863 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2864 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2865 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2867 (aarch64_evpc_rev): Rename to...
2868 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2869 (aarch64_evpc_rev_global): New function.
2870 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2871 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2873 (aarch64_evpc_sve_tbl): New function.
2874 (aarch64_expand_vec_perm_const_1): Update after rename of
2875 aarch64_evpc_rev. Handle SVE permutes too, trying
2876 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2877 than aarch64_evpc_tbl.
2878 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2879 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2880 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2881 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2882 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2883 (aarch64_expand_sve_vcond): New functions.
2884 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2885 of aarch64_vector_mode_p.
2886 (aarch64_dwarf_poly_indeterminate_value): New function.
2887 (aarch64_compute_pressure_classes): Likewise.
2888 (aarch64_can_change_mode_class): Likewise.
2889 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2890 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2891 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2892 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2893 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2894 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2895 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2896 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2898 (Dn, Dl, Dr): Accept const as well as const_vector.
2899 (Dz): Likewise. Compare against CONST0_RTX.
2900 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2901 of "vector" where appropriate.
2902 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2903 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2904 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2905 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2906 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2907 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2908 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2909 (v_int_equiv): Extend to SVE modes.
2910 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2912 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2913 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2914 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2915 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2916 (SVE_COND_FP_CMP): New int iterators.
2917 (perm_hilo): Handle the new unpack unspecs.
2918 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2920 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2921 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2922 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2923 (aarch64_equality_operator, aarch64_constant_vector_operand)
2924 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2925 (aarch64_sve_nonimmediate_operand): Likewise.
2926 (aarch64_sve_general_operand): Likewise.
2927 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2928 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2929 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2930 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2931 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2932 (aarch64_sve_float_arith_immediate): Likewise.
2933 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2934 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2935 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2936 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2937 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2938 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2939 (aarch64_sve_float_arith_operand): Likewise.
2940 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2941 (aarch64_sve_float_mul_operand): Likewise.
2942 (aarch64_sve_vec_perm_operand): Likewise.
2943 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2944 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2945 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2946 as well as const_vector.
2947 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2948 in file. Use CONST0_RTX and CONSTM1_RTX.
2949 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2950 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2951 Use aarch64_simd_imm_zero.
2952 * config/aarch64/aarch64-sve.md: New file.
2953 * config/aarch64/aarch64.md: Include it.
2954 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2955 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2956 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2957 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2958 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2959 (sve): New attribute.
2960 (enabled): Disable instructions with the sve attribute unless
2962 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2963 aarch64_expand_mov_immediate.
2964 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2965 CNT[BHSD] immediates.
2966 (movti): Split CONST_POLY_INT moves into two halves.
2967 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2968 Split additions that need a temporary here if the destination
2969 is the stack pointer.
2970 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2971 (*add<mode>3_poly_1): New instruction.
2972 (set_clobber_cc): New expander.
2974 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2976 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2977 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2978 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2979 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2980 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2981 Change innermode from fixed_mode_size to machine_mode.
2982 (simplify_subreg): Update call accordingly. Handle a constant-sized
2983 subreg of a variable-length CONST_VECTOR.
2985 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2986 Alan Hayward <alan.hayward@arm.com>
2987 David Sherwood <david.sherwood@arm.com>
2989 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2990 (add_offset_to_base): New function, split out from...
2991 (create_mem_ref): ...here. When handling a scale other than 1,
2992 check first whether the address is valid without the offset.
2993 Add it into the base if so, leaving the index and scale as-is.
2995 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2998 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2999 fold_for_warn before checking if arg2 is INTEGER_CST.
3001 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3003 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3004 (store_multiple_operation): Delete.
3005 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3006 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3007 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3008 guarded by TARGET_STRING.
3009 (rs6000_output_load_multiple): Delete.
3010 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3011 OPTION_MASK_STRING / TARGET_STRING handling.
3012 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3013 (const rs6000_opt_masks) <"string">: Change mask to 0.
3014 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3015 (MASK_STRING): Delete.
3016 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3018 (load_multiple): Delete.
3025 (store_multiple): Delete.
3032 (movmemsi_8reg): Delete.
3033 (corresponding unnamed define_insn): Delete.
3034 (movmemsi_6reg): Delete.
3035 (corresponding unnamed define_insn): Delete.
3036 (movmemsi_4reg): Delete.
3037 (corresponding unnamed define_insn): Delete.
3038 (movmemsi_2reg): Delete.
3039 (corresponding unnamed define_insn): Delete.
3040 (movmemsi_1reg): Delete.
3041 (corresponding unnamed define_insn): Delete.
3042 * config/rs6000/rs6000.opt (mno-string): New.
3043 (mstring): Replace by deprecation warning stub.
3044 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3046 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3048 * regrename.c (regrename_do_replace): If replacing the same
3049 reg multiple times, try to reuse last created gen_raw_REG.
3052 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3053 main to workaround a bug in GDB.
3055 2018-01-12 Tom de Vries <tom@codesourcery.com>
3058 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3060 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3062 PR rtl-optimization/80481
3063 * ira-color.c (get_cap_member): New function.
3064 (allocnos_conflict_by_live_ranges_p): Use it.
3065 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3066 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3068 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3071 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3072 (*saddl_se_1): Ditto.
3074 (*ssubl_se_1): Ditto.
3076 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3078 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3079 rather than wi::to_widest for DR_INITs.
3080 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3081 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3082 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3084 (vect_analyze_group_access_1): Note that here.
3086 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3088 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3089 polynomial type sizes.
3091 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3093 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3094 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3095 (gimple_add_tmp_var): Likewise.
3097 2018-01-12 Martin Liska <mliska@suse.cz>
3099 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3100 (gimple_alloc_sizes): Likewise.
3101 (dump_gimple_statistics): Use PRIu64 in printf format.
3102 * gimple.h: Change uint64_t to int.
3104 2018-01-12 Martin Liska <mliska@suse.cz>
3106 * tree-core.h: Use uint64_t instead of int.
3107 * tree.c (tree_node_counts): Likewise.
3108 (tree_node_sizes): Likewise.
3109 (dump_tree_statistics): Use PRIu64 in printf format.
3111 2018-01-12 Martin Liska <mliska@suse.cz>
3113 * Makefile.in: As qsort_chk is implemented in vec.c, add
3114 vec.o to linkage of gencfn-macros.
3115 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3116 passing the info to record_node_allocation_statistics.
3117 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3119 * ggc-common.c (struct ggc_usage): Add operator== and use
3120 it in operator< and compare function.
3121 * mem-stats.h (struct mem_usage): Likewise.
3122 * vec.c (struct vec_usage): Remove operator< and compare
3123 function. Can be simply inherited.
3125 2018-01-12 Martin Jambor <mjambor@suse.cz>
3128 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3129 * tree-ssa-math-opts.c: Include domwalk.h.
3130 (convert_mult_to_fma_1): New function.
3131 (fma_transformation_info): New type.
3132 (fma_deferring_state): Likewise.
3133 (cancel_fma_deferring): New function.
3134 (result_of_phi): Likewise.
3135 (last_fma_candidate_feeds_initial_phi): Likewise.
3136 (convert_mult_to_fma): Added deferring logic, split actual
3137 transformation to convert_mult_to_fma_1.
3138 (math_opts_dom_walker): New type.
3139 (math_opts_dom_walker::after_dom_children): New method, body moved
3140 here from pass_optimize_widening_mul::execute, added deferring logic
3142 (pass_optimize_widening_mul::execute): Moved most of code to
3143 math_opts_dom_walker::after_dom_children.
3144 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3145 * config/i386/i386.c (ix86_option_override_internal): Added
3146 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3148 2018-01-12 Richard Biener <rguenther@suse.de>
3151 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3152 inline instance vars.
3154 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3157 * config/rx/rx.c (rx_is_restricted_memory_address):
3160 2018-01-12 Richard Biener <rguenther@suse.de>
3162 PR tree-optimization/80846
3163 * target.def (split_reduction): New target hook.
3164 * targhooks.c (default_split_reduction): New function.
3165 * targhooks.h (default_split_reduction): Declare.
3166 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3167 target requests first reduce vectors by combining low and high
3169 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3170 (get_vectype_for_scalar_type_and_size): Export.
3171 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3172 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3173 * doc/tm.texi: Regenerate.
3174 * config/i386/i386.c (ix86_split_reduction): Implement
3175 TARGET_VECTORIZE_SPLIT_REDUCTION.
3177 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3180 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3181 in PIC mode except for TARGET_VXWORKS_RTP.
3182 * config/sparc/sparc.c: Include cfgrtl.h.
3183 (TARGET_INIT_PIC_REG): Define.
3184 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3185 (sparc_pic_register_p): New predicate.
3186 (sparc_legitimate_address_p): Use it.
3187 (sparc_legitimize_pic_address): Likewise.
3188 (sparc_delegitimize_address): Likewise.
3189 (sparc_mode_dependent_address_p): Likewise.
3190 (gen_load_pcrel_sym): Remove 4th parameter.
3191 (load_got_register): Adjust call to above. Remove obsolete stuff.
3192 (sparc_expand_prologue): Do not call load_got_register here.
3193 (sparc_flat_expand_prologue): Likewise.
3194 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3195 (sparc_use_pseudo_pic_reg): New function.
3196 (sparc_init_pic_reg): Likewise.
3197 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3198 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3200 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3202 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3203 Add item for branch_cost.
3205 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3207 PR rtl-optimization/83565
3208 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3209 not extend the result to a larger mode for rotate operations.
3210 (num_sign_bit_copies1): Likewise.
3212 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3215 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3217 Use values-Xc.o for -pedantic.
3218 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3220 2018-01-12 Martin Liska <mliska@suse.cz>
3223 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3225 (possible_polymorphic_call_targets): Use it.
3226 (ipa_devirt): Likewise.
3228 2018-01-12 Martin Liska <mliska@suse.cz>
3230 * profile-count.h (enum profile_quality): Use 0 as invalid
3231 enum value of profile_quality.
3233 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3235 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3236 -mext-string options.
3238 2018-01-12 Richard Biener <rguenther@suse.de>
3240 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3241 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3242 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3244 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3246 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3248 * configure.ac (--with-long-double-format): Add support for the
3249 configuration option to change the default long double format on
3251 * config.gcc (powerpc*-linux*-*): Likewise.
3252 * configure: Regenerate.
3253 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3254 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3255 used without modification.
3257 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3259 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3260 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3261 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3262 MISC_BUILTIN_SPEC_BARRIER.
3263 (rs6000_init_builtins): Likewise.
3264 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3266 (speculation_barrier): New define_insn.
3267 * doc/extend.texi: Document __builtin_speculation_barrier.
3269 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3272 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3273 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3274 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3276 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3277 integral modes instead of "ss" and "sd".
3278 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3279 vectors with 32-bit and 64-bit elements.
3280 (vecdupssescalarmodesuffix): New mode attribute.
3281 (vec_dup<mode>): Use it.
3283 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3286 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3287 frame if argument is passed on stack.
3289 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3292 * ree.c (combine_reaching_defs): Optimize also
3293 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3294 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3296 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3299 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3301 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3304 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3305 after they are computed.
3307 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3309 PR tree-optimization/83695
3310 * gimple-loop-linterchange.cc
3311 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3312 reset cached scev information after interchange.
3313 (pass_linterchange::execute): Remove call to scev_reset_htab.
3315 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3317 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3318 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3319 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3320 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3321 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3322 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3323 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3324 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3325 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3326 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3327 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3328 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3329 (V_lane_reg): Likewise.
3330 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3332 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3333 (vfmal_lane_low<mode>_intrinsic,
3334 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3335 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3336 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3337 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3338 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3339 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3341 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3343 * config/arm/arm-cpus.in (fp16fml): New feature.
3344 (ALL_SIMD): Add fp16fml.
3345 (armv8.2-a): Add fp16fml as an option.
3346 (armv8.3-a): Likewise.
3347 (armv8.4-a): Add fp16fml as part of fp16.
3348 * config/arm/arm.h (TARGET_FP16FML): Define.
3349 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3351 * config/arm/arm-modes.def (V2HF): Define.
3352 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3353 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3354 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3355 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3356 vfmsl_low, vfmsl_high): New set of builtins.
3357 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3358 (vfml_op): New code attribute.
3359 (VFMLHALVES): New int iterator.
3360 (VFML, VFMLSEL): New mode attributes.
3361 (V_reg): Define mapping for V2HF.
3362 (V_hi, V_lo): New mode attributes.
3363 (VF_constraint): Likewise.
3364 (vfml_half, vfml_half_selector): New int attributes.
3365 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3367 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3368 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3370 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3371 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3372 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3373 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3375 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3376 Document new effective target and option set.
3378 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3380 * config/arm/arm-cpus.in (armv8_4): New feature.
3381 (ARMv8_4a): New fgroup.
3382 (armv8.4-a): New arch.
3383 * config/arm/arm-tables.opt: Regenerate.
3384 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3385 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3386 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3387 Add matching rules for -march=armv8.4-a and extensions.
3388 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3390 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3393 * config/rx/rx.md (BW): New mode attribute.
3394 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3396 2018-01-11 Richard Biener <rguenther@suse.de>
3398 PR tree-optimization/83435
3399 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3400 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3401 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3403 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3404 Alan Hayward <alan.hayward@arm.com>
3405 David Sherwood <david.sherwood@arm.com>
3407 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3409 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3410 (aarch64_print_address_internal): Use it to check for a zero offset.
3412 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3413 Alan Hayward <alan.hayward@arm.com>
3414 David Sherwood <david.sherwood@arm.com>
3416 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3417 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3418 Return a poly_int64 rather than a HOST_WIDE_INT.
3419 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3420 rather than a HOST_WIDE_INT.
3421 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3422 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3423 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3424 final_offset from HOST_WIDE_INT to poly_int64.
3425 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3426 to_constant when getting the number of units in an Advanced SIMD
3428 (aarch64_builtin_vectorized_function): Check for a constant number
3430 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3432 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3433 attribute instead of GET_MODE_NUNITS.
3434 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3435 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3436 GET_MODE_SIZE for fixed-size registers.
3437 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3438 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3439 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3440 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3441 (aarch64_print_operand, aarch64_print_address_internal)
3442 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3443 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3444 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3445 Handle polynomial GET_MODE_SIZE.
3446 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3447 wider than SImode without modification.
3448 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3449 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3450 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3451 passing and returning SVE modes.
3452 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3453 rather than GEN_INT.
3454 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3455 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3456 (aarch64_allocate_and_probe_stack_space): Likewise.
3457 (aarch64_layout_frame): Cope with polynomial offsets.
3458 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3459 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3461 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3462 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3463 poly_int64 rather than a HOST_WIDE_INT.
3464 (aarch64_get_separate_components, aarch64_process_components)
3465 (aarch64_expand_prologue, aarch64_expand_epilogue)
3466 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3467 (aarch64_anchor_offset): New function, split out from...
3468 (aarch64_legitimize_address): ...here.
3469 (aarch64_builtin_vectorization_cost): Handle polynomial
3470 TYPE_VECTOR_SUBPARTS.
3471 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3473 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3474 number of elements from the PARALLEL rather than the mode.
3475 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3476 rather than GET_MODE_BITSIZE.
3477 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3478 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3479 (aarch64_expand_vec_perm_const_1): Handle polynomial
3480 d->perm.length () and d->perm elements.
3481 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3482 Apply to_constant to d->perm elements.
3483 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3484 polynomial CONST_VECTOR_NUNITS.
3485 (aarch64_move_pointer): Take amount as a poly_int64 rather
3487 (aarch64_progress_pointer): Avoid temporary variable.
3488 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3489 the mode attribute instead of GET_MODE.
3491 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3492 Alan Hayward <alan.hayward@arm.com>
3493 David Sherwood <david.sherwood@arm.com>
3495 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3496 x exists before using it.
3497 (aarch64_add_constant_internal): Rename to...
3498 (aarch64_add_offset_1): ...this. Replace regnum with separate
3499 src and dest rtxes. Handle the case in which they're different,
3500 including when the offset is zero. Replace scratchreg with an rtx.
3501 Use 2 additions if there is no spare register into which we can
3502 move a 16-bit constant.
3503 (aarch64_add_constant): Delete.
3504 (aarch64_add_offset): Replace reg with separate src and dest
3505 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3506 Use aarch64_add_offset_1.
3507 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3508 an rtx rather than an int. Take the delta as a poly_int64
3509 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3510 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3511 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3512 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3513 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3515 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3516 aarch64_add_constant.
3518 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3520 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3521 Use scalar_float_mode.
3523 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3525 * config/aarch64/aarch64-simd.md
3526 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3527 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3528 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3529 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3530 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3531 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3532 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3533 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3534 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3535 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3537 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3540 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3541 targ_options->x_arm_arch_string is non NULL.
3543 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3545 * config/aarch64/aarch64.h
3546 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3548 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3551 * expmed.c (emit_store_flag_force): Swap if const op0
3552 and change VOIDmode to mode of op0.
3554 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3556 PR rtl-optimization/83761
3557 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3558 than bytes to mode_for_size.
3560 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3563 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3564 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3567 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3570 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3571 when in layout mode.
3572 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3573 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3576 2018-01-10 Michael Collison <michael.collison@arm.com>
3578 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3579 * config/aarch64/aarch64-option-extension.def: Add
3580 AARCH64_OPT_EXTENSION of 'fp16fml'.
3581 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3582 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3583 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3584 * config/aarch64/constraints.md (Ui7): New constraint.
3585 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3586 (VFMLA_SEL_W): Ditto.
3589 (VFMLA16_LOW): New int iterator.
3590 (VFMLA16_HIGH): Ditto.
3591 (UNSPEC_FMLAL): New unspec.
3592 (UNSPEC_FMLSL): Ditto.
3593 (UNSPEC_FMLAL2): Ditto.
3594 (UNSPEC_FMLSL2): Ditto.
3595 (f16mac): New code attribute.
3596 * config/aarch64/aarch64-simd-builtins.def
3597 (aarch64_fmlal_lowv2sf): Ditto.
3598 (aarch64_fmlsl_lowv2sf): Ditto.
3599 (aarch64_fmlalq_lowv4sf): Ditto.
3600 (aarch64_fmlslq_lowv4sf): Ditto.
3601 (aarch64_fmlal_highv2sf): Ditto.
3602 (aarch64_fmlsl_highv2sf): Ditto.
3603 (aarch64_fmlalq_highv4sf): Ditto.
3604 (aarch64_fmlslq_highv4sf): Ditto.
3605 (aarch64_fmlal_lane_lowv2sf): Ditto.
3606 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3607 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3608 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3609 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3610 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3611 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3612 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3613 (aarch64_fmlal_lane_highv2sf): Ditto.
3614 (aarch64_fmlsl_lane_highv2sf): Ditto.
3615 (aarch64_fmlal_laneq_highv2sf): Ditto.
3616 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3617 (aarch64_fmlalq_lane_highv4sf): Ditto.
3618 (aarch64_fmlsl_lane_highv4sf): Ditto.
3619 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3620 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3621 * config/aarch64/aarch64-simd.md:
3622 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3623 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3624 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3625 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3626 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3627 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3628 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3629 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3630 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3631 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3632 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3633 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3634 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3635 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3636 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3637 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3638 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3639 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3640 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3641 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3642 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3643 (vfmlsl_low_u32): Ditto.
3644 (vfmlalq_low_u32): Ditto.
3645 (vfmlslq_low_u32): Ditto.
3646 (vfmlal_high_u32): Ditto.
3647 (vfmlsl_high_u32): Ditto.
3648 (vfmlalq_high_u32): Ditto.
3649 (vfmlslq_high_u32): Ditto.
3650 (vfmlal_lane_low_u32): Ditto.
3651 (vfmlsl_lane_low_u32): Ditto.
3652 (vfmlal_laneq_low_u32): Ditto.
3653 (vfmlsl_laneq_low_u32): Ditto.
3654 (vfmlalq_lane_low_u32): Ditto.
3655 (vfmlslq_lane_low_u32): Ditto.
3656 (vfmlalq_laneq_low_u32): Ditto.
3657 (vfmlslq_laneq_low_u32): Ditto.
3658 (vfmlal_lane_high_u32): Ditto.
3659 (vfmlsl_lane_high_u32): Ditto.
3660 (vfmlal_laneq_high_u32): Ditto.
3661 (vfmlsl_laneq_high_u32): Ditto.
3662 (vfmlalq_lane_high_u32): Ditto.
3663 (vfmlslq_lane_high_u32): Ditto.
3664 (vfmlalq_laneq_high_u32): Ditto.
3665 (vfmlslq_laneq_high_u32): Ditto.
3666 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3667 (AARCH64_FL_FOR_ARCH8_4): New.
3668 (AARCH64_ISA_F16FML): New ISA flag.
3669 (TARGET_F16FML): New feature flag for fp16fml.
3670 (doc/invoke.texi): Document new fp16fml option.
3672 2018-01-10 Michael Collison <michael.collison@arm.com>
3674 * config/aarch64/aarch64-builtins.c:
3675 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3676 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3677 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3678 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3679 (AARCH64_ISA_SHA3): New ISA flag.
3680 (TARGET_SHA3): New feature flag for sha3.
3681 * config/aarch64/iterators.md (sha512_op): New int attribute.
3682 (CRYPTO_SHA512): New int iterator.
3683 (UNSPEC_SHA512H): New unspec.
3684 (UNSPEC_SHA512H2): Ditto.
3685 (UNSPEC_SHA512SU0): Ditto.
3686 (UNSPEC_SHA512SU1): Ditto.
3687 * config/aarch64/aarch64-simd-builtins.def
3688 (aarch64_crypto_sha512hqv2di): New builtin.
3689 (aarch64_crypto_sha512h2qv2di): Ditto.
3690 (aarch64_crypto_sha512su0qv2di): Ditto.
3691 (aarch64_crypto_sha512su1qv2di): Ditto.
3692 (aarch64_eor3qv8hi): Ditto.
3693 (aarch64_rax1qv2di): Ditto.
3694 (aarch64_xarqv2di): Ditto.
3695 (aarch64_bcaxqv8hi): Ditto.
3696 * config/aarch64/aarch64-simd.md:
3697 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3698 (aarch64_crypto_sha512su0qv2di): Ditto.
3699 (aarch64_crypto_sha512su1qv2di): Ditto.
3700 (aarch64_eor3qv8hi): Ditto.
3701 (aarch64_rax1qv2di): Ditto.
3702 (aarch64_xarqv2di): Ditto.
3703 (aarch64_bcaxqv8hi): Ditto.
3704 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3705 (vsha512h2q_u64): Ditto.
3706 (vsha512su0q_u64): Ditto.
3707 (vsha512su1q_u64): Ditto.
3708 (veor3q_u16): Ditto.
3709 (vrax1q_u64): Ditto.
3711 (vbcaxq_u16): Ditto.
3712 * config/arm/types.md (crypto_sha512): New type attribute.
3713 (crypto_sha3): Ditto.
3714 (doc/invoke.texi): Document new sha3 option.
3716 2018-01-10 Michael Collison <michael.collison@arm.com>
3718 * config/aarch64/aarch64-builtins.c:
3719 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3720 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3721 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3722 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3723 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3724 (AARCH64_ISA_SM4): New ISA flag.
3725 (TARGET_SM4): New feature flag for sm4.
3726 * config/aarch64/aarch64-simd-builtins.def
3727 (aarch64_sm3ss1qv4si): Ditto.
3728 (aarch64_sm3tt1aq4si): Ditto.
3729 (aarch64_sm3tt1bq4si): Ditto.
3730 (aarch64_sm3tt2aq4si): Ditto.
3731 (aarch64_sm3tt2bq4si): Ditto.
3732 (aarch64_sm3partw1qv4si): Ditto.
3733 (aarch64_sm3partw2qv4si): Ditto.
3734 (aarch64_sm4eqv4si): Ditto.
3735 (aarch64_sm4ekeyqv4si): Ditto.
3736 * config/aarch64/aarch64-simd.md:
3737 (aarch64_sm3ss1qv4si): Ditto.
3738 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3739 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3740 (aarch64_sm4eqv4si): Ditto.
3741 (aarch64_sm4ekeyqv4si): Ditto.
3742 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3743 (sm3part_op): Ditto.
3744 (CRYPTO_SM3TT): Ditto.
3745 (CRYPTO_SM3PART): Ditto.
3746 (UNSPEC_SM3SS1): New unspec.
3747 (UNSPEC_SM3TT1A): Ditto.
3748 (UNSPEC_SM3TT1B): Ditto.
3749 (UNSPEC_SM3TT2A): Ditto.
3750 (UNSPEC_SM3TT2B): Ditto.
3751 (UNSPEC_SM3PARTW1): Ditto.
3752 (UNSPEC_SM3PARTW2): Ditto.
3753 (UNSPEC_SM4E): Ditto.
3754 (UNSPEC_SM4EKEY): Ditto.
3755 * config/aarch64/constraints.md (Ui2): New constraint.
3756 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3757 * config/arm/types.md (crypto_sm3): New type attribute.
3758 (crypto_sm4): Ditto.
3759 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3760 (vsm3tt1aq_u32): Ditto.
3761 (vsm3tt1bq_u32): Ditto.
3762 (vsm3tt2aq_u32): Ditto.
3763 (vsm3tt2bq_u32): Ditto.
3764 (vsm3partw1q_u32): Ditto.
3765 (vsm3partw2q_u32): Ditto.
3766 (vsm4eq_u32): Ditto.
3767 (vsm4ekeyq_u32): Ditto.
3768 (doc/invoke.texi): Document new sm4 option.
3770 2018-01-10 Michael Collison <michael.collison@arm.com>
3772 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3773 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3774 (AARCH64_FL_FOR_ARCH8_4): New.
3775 (AARCH64_FL_V8_4): New flag.
3776 (doc/invoke.texi): Document new armv8.4-a option.
3778 2018-01-10 Michael Collison <michael.collison@arm.com>
3780 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3781 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3782 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3783 * config/aarch64/aarch64-option-extension.def: Add
3784 AARCH64_OPT_EXTENSION of 'sha2'.
3785 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3786 (crypto): Disable sha2 and aes if crypto disabled.
3787 (crypto): Enable aes and sha2 if enabled.
3788 (simd): Disable sha2 and aes if simd disabled.
3789 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3791 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3792 (TARGET_SHA2): New feature flag for sha2.
3793 (TARGET_AES): New feature flag for aes.
3794 * config/aarch64/aarch64-simd.md:
3795 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3796 conditional on TARGET_AES.
3797 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3798 (aarch64_crypto_sha1hsi): Make pattern conditional
3800 (aarch64_crypto_sha1hv4si): Ditto.
3801 (aarch64_be_crypto_sha1hv4si): Ditto.
3802 (aarch64_crypto_sha1su1v4si): Ditto.
3803 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3804 (aarch64_crypto_sha1su0v4si): Ditto.
3805 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3806 (aarch64_crypto_sha256su0v4si): Ditto.
3807 (aarch64_crypto_sha256su1v4si): Ditto.
3808 (doc/invoke.texi): Document new aes and sha2 options.
3810 2018-01-10 Martin Sebor <msebor@redhat.com>
3812 PR tree-optimization/83781
3813 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3816 2018-01-11 Martin Sebor <msebor@gmail.com>
3817 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3819 PR tree-optimization/83501
3820 PR tree-optimization/81703
3822 * tree-ssa-strlen.c (get_string_cst): Rename...
3823 (get_string_len): ...to this. Handle global constants.
3824 (handle_char_store): Adjust.
3826 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3827 Jim Wilson <jimw@sifive.com>
3829 * config/riscv/riscv-protos.h (riscv_output_return): New.
3830 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3831 (riscv_attribute_table, riscv_output_return),
3832 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3833 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3834 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3836 (riscv_expand_prologue): Add early return for naked function.
3837 (riscv_expand_epilogue): Likewise.
3838 (riscv_function_ok_for_sibcall): Return false for naked function.
3839 (riscv_set_current_function): New.
3840 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3841 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3842 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3843 * doc/extend.texi (RISC-V Function Attributes): New.
3845 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3847 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3848 check for 128-bit long double before checking TCmode.
3849 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3850 128-bit long doubles before checking TFmode or TCmode.
3851 (FLOAT128_IBM_P): Likewise.
3853 2018-01-10 Martin Sebor <msebor@redhat.com>
3855 PR tree-optimization/83671
3856 * builtins.c (c_strlen): Unconditionally return zero for the empty
3858 Use -Warray-bounds for warnings.
3859 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3860 for non-constant array indices with COMPONENT_REF, arrays of
3861 arrays, and pointers to arrays.
3862 (gimple_fold_builtin_strlen): Determine and set length range for
3863 non-constant character arrays.
3865 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3868 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3871 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3873 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3875 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3878 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3879 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3880 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3881 indexed_or_indirect_operand predicate.
3882 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3883 (*vsx_le_perm_load_v8hi): Likewise.
3884 (*vsx_le_perm_load_v16qi): Likewise.
3885 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3886 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3887 (*vsx_le_perm_store_v8hi): Likewise.
3888 (*vsx_le_perm_store_v16qi): Likewise.
3889 (eight unnamed splitters): Likewise.
3891 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3893 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3894 * config/rs6000/emmintrin.h: Likewise.
3895 * config/rs6000/mmintrin.h: Likewise.
3896 * config/rs6000/xmmintrin.h: Likewise.
3898 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3901 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3903 * tree.c (tree_nop_conversion): Return true for location wrapper
3905 (maybe_wrap_with_location): New function.
3906 (selftest::check_strip_nops): New function.
3907 (selftest::test_location_wrappers): New function.
3908 (selftest::tree_c_tests): Call it.
3909 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3910 (maybe_wrap_with_location): New decl.
3911 (EXPR_LOCATION_WRAPPER_P): New macro.
3912 (location_wrapper_p): New inline function.
3913 (tree_strip_any_location_wrapper): New inline function.
3915 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3918 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3919 stack_realign_offset for the largest alignment of stack slot
3921 (ix86_find_max_used_stack_alignment): New function.
3922 (ix86_finalize_stack_frame_flags): Use it. Set
3923 max_used_stack_alignment if we don't realign stack.
3924 * config/i386/i386.h (machine_function): Add
3925 max_used_stack_alignment.
3927 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3929 * config/arm/arm.opt (-mbranch-cost): New option.
3930 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3933 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3936 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3937 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3939 2018-01-10 Richard Biener <rguenther@suse.de>
3942 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3943 early out so it also covers the case where we have a non-NULL
3946 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3948 PR tree-optimization/83753
3949 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3950 for non-strided grouped accesses if the number of elements is 1.
3952 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3955 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3956 * i386.h (TARGET_USE_GATHER): Define.
3957 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3959 2018-01-10 Martin Liska <mliska@suse.cz>
3962 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3963 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3965 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3966 CLEANUP_NO_PARTITIONING is not set.
3968 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3970 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3971 for vectors, as a partial revert of r254296.
3972 * rtl.h (const_vec_p): Delete.
3973 (const_vec_duplicate_p): Don't test for vector CONSTs.
3974 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3975 * expmed.c (make_tree): Likewise.
3978 * common.md (E, F): Use CONSTANT_P instead of checking for
3980 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3981 checking for CONST_VECTOR.
3983 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3986 * predict.c (force_edge_cold): Handle in more sane way edges
3989 2018-01-09 Carl Love <cel@us.ibm.com>
3991 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3993 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3994 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3995 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3996 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3997 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3998 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3999 * config/rs6000/rs6000-protos.h: Add extern defition for
4000 rs6000_generate_float2_double_code.
4001 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4003 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4004 (float2_v2df): Add define_expand.
4006 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4009 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4010 op_mode in the force_to_mode call.
4012 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4014 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4015 instead of checking each element individually.
4016 (aarch64_evpc_uzp): Likewise.
4017 (aarch64_evpc_zip): Likewise.
4018 (aarch64_evpc_ext): Likewise.
4019 (aarch64_evpc_rev): Likewise.
4020 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4021 instead of checking each element individually. Return true without
4023 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4024 whether all selected elements come from the same input, instead of
4025 checking each element individually. Remove calls to gen_rtx_REG,
4026 start_sequence and end_sequence and instead assert that no rtl is
4029 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4031 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4032 order of HIGH and CONST checks.
4034 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4036 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4037 if the destination isn't an SSA_NAME.
4039 2018-01-09 Richard Biener <rguenther@suse.de>
4041 PR tree-optimization/83668
4042 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4044 (canonicalize_loop_form): ... here, renamed from ...
4045 (canonicalize_loop_closed_ssa_form): ... this and amended to
4046 swap successor edges for loop exit blocks to make us use
4047 the RPO order we need for initial schedule generation.
4049 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4051 PR tree-optimization/64811
4052 * match.pd: When optimizing comparisons with Inf, avoid
4053 introducing or losing exceptions from comparisons with NaN.
4055 2018-01-09 Martin Liska <mliska@suse.cz>
4058 * asan.c (shadow_mem_size): Add gcc_assert.
4060 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4062 Don't save registers in main().
4065 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4066 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4067 * config/avr/avr.c (avr_set_current_function): Don't error if
4068 naked, OS_task or OS_main are specified at the same time.
4069 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4071 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4073 * common/config/avr/avr-common.c (avr_option_optimization_table):
4074 Switch on -mmain-is-OS_task for optimizing compilations.
4076 2018-01-09 Richard Biener <rguenther@suse.de>
4078 PR tree-optimization/83572
4079 * graphite.c: Include cfganal.h.
4080 (graphite_transform_loops): Connect infinite loops to exit
4081 and remove fake edges at the end.
4083 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4085 * ipa-inline.c (edge_badness): Revert accidental checkin.
4087 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4090 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4091 symbols; not inline clones.
4093 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4096 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4097 hard registers. Formatting fixes.
4099 PR preprocessor/83722
4100 * gcc.c (try_generate_repro): Pass
4101 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4102 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4105 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4106 Kito Cheng <kito.cheng@gmail.com>
4108 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4109 (riscv_leaf_function_p): Delete.
4110 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4112 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4114 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4116 (do_ifelse): New function.
4117 (do_isel): New function.
4118 (do_sub3): New function.
4119 (do_add3): New function.
4120 (do_load_mask_compare): New function.
4121 (do_overlap_load_compare): New function.
4122 (expand_compare_loop): New function.
4123 (expand_block_compare): Call expand_compare_loop() when appropriate.
4124 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4126 (-mblock-compare-inline-loop-limit): New option.
4128 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4131 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4132 Reverse order of second and third operands in first alternative.
4133 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4134 of first and second elements in UNSPEC_VPERMR vector.
4135 (altivec_expand_vec_perm_le): Likewise.
4137 2018-01-08 Jeff Law <law@redhat.com>
4139 PR rtl-optimizatin/81308
4140 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4141 (process_switch): If group_case_labels makes a change, then set
4143 (pass_convert_switch::execute): If a switch is converted, then
4144 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4146 PR rtl-optimization/81308
4147 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4150 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4152 PR target/83663 - Revert r255946
4153 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4154 generation for cases where splatting a value is not useful.
4155 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4156 across a vec_duplicate and a paradoxical subreg forming a vector
4157 mode to a vec_concat.
4159 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4161 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4162 -march=armv8.3-a variants.
4163 * config/arm/t-multilib: Likewise.
4164 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4166 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4168 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4170 (cceq_ior_compare_complement): Give it a name so I can use it, and
4171 change boolean_or_operator predicate to boolean_operator so it can
4172 be used to generate a crand.
4173 (eqne): New code iterator.
4174 (bd/bd_neg): New code_attrs.
4175 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4176 a single define_insn.
4177 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4178 decrement (bdnzt/bdnzf/bdzt/bdzf).
4179 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4180 with the new names of the branch decrement patterns, and added the
4181 names of the branch decrement conditional patterns.
4183 2018-01-08 Richard Biener <rguenther@suse.de>
4185 PR tree-optimization/83563
4186 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4189 2018-01-08 Richard Biener <rguenther@suse.de>
4192 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4194 2018-01-08 Richard Biener <rguenther@suse.de>
4196 PR tree-optimization/83685
4197 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4198 references to abnormals.
4200 2018-01-08 Richard Biener <rguenther@suse.de>
4203 * dwarf2out.c (output_indirect_strings): Handle empty
4204 skeleton_debug_str_hash.
4205 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4207 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4209 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4210 (emit_store_direct): Likewise.
4211 (arc_trampoline_adjust_address): Likewise.
4212 (arc_asm_trampoline_template): New function.
4213 (arc_initialize_trampoline): Use asm_trampoline_template.
4214 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4215 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4216 * config/arc/arc.md (flush_icache): Delete pattern.
4218 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4220 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4221 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4224 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4227 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4228 by not USED_FOR_TARGET.
4229 (make_pass_resolve_sw_modes): Likewise.
4231 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4233 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4236 2018-01-08 Richard Biener <rguenther@suse.de>
4239 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4241 2018-01-08 Richard Biener <rguenther@suse.de>
4244 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4246 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4249 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4250 basic blocks with a small number of successors.
4251 (convert_control_dep_chain_into_preds): Improve handling of
4253 (dump_predicates): Split apart into...
4254 (dump_pred_chain): ...here...
4255 (dump_pred_info): ...and here.
4256 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4257 (can_chain_union_be_invalidated_p): Improve check for invalidation
4259 (uninit_uses_cannot_happen): Avoid unnecessary if
4260 convert_control_dep_chain_into_preds yielded nothing.
4262 2018-01-06 Martin Sebor <msebor@redhat.com>
4264 PR tree-optimization/83640
4265 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4266 subtracting negative offset from size.
4267 (builtin_access::overlap): Adjust offset bounds of the access to fall
4268 within the size of the object if possible.
4270 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4272 PR rtl-optimization/83699
4273 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4274 extract_bit_field_as_subreg to cases in which the extracted
4275 value is also a vector.
4277 * lra-constraints.c (process_alt_operands): Test for the equivalence
4278 substitutions when detecting a possible reload cycle.
4280 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4283 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4284 by default if flag_selective_schedling{,2}. Formatting fixes.
4286 PR rtl-optimization/83682
4287 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4288 if it has non-VECTOR_MODE element mode.
4289 (vec_duplicate_p): Likewise.
4292 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4293 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4295 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4298 * config/i386/i386-builtin.def
4299 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4300 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4301 Require also OPTION_MASK_ISA_AVX512F in addition to
4302 OPTION_MASK_ISA_GFNI.
4303 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4304 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4305 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4306 to OPTION_MASK_ISA_GFNI.
4307 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4308 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4309 OPTION_MASK_ISA_AVX512BW.
4310 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4311 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4312 addition to OPTION_MASK_ISA_GFNI.
4313 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4314 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4315 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4316 to OPTION_MASK_ISA_GFNI.
4317 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4318 a requirement for all ISAs rather than any of them with a few
4320 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4322 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4323 bitmasks to be enabled with 3 exceptions, instead of requiring any
4324 enabled ISA with lots of exceptions.
4325 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4326 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4327 Change avx512bw in isa attribute to avx512f.
4328 * config/i386/sgxintrin.h: Add license boilerplate.
4329 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4330 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4331 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4332 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4334 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4335 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4336 temporarily sse2 rather than sse if not enabled already.
4339 * config/i386/sse.md (VI248_VLBW): Rename to ...
4340 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4341 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4342 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4343 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4344 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4345 mode iterator instead of VI248_VLBW.
4347 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4349 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4350 (record_modified): Skip clobbers; add debug output.
4351 (param_change_prob): Use sreal frequencies.
4353 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4355 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4356 punt for user-aligned variables.
4358 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4360 * tree-chrec.c (chrec_contains_symbols): Return true for
4363 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4366 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4367 of (x|y) == x for BICS pattern.
4369 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4371 PR tree-optimization/83605
4372 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4373 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4376 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4378 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4379 * config/epiphany/rtems.h: New file.
4381 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4382 Uros Bizjak <ubizjak@gmail.com>
4385 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4386 QIreg_operand instead of register_operand predicate.
4387 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4388 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4389 comments instead of -fmitigate[-_]rop.
4391 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4394 * cgraphunit.c (symbol_table::compile): Switch to text_section
4395 before calling assembly_start debug hook.
4396 * run-rtl-passes.c (run_rtl_passes): Likewise.
4399 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4401 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4402 range_int_cst_p rather than !symbolic_range_p before calling
4403 extract_range_from_multiplicative_op_1.
4405 2018-01-04 Jeff Law <law@redhat.com>
4407 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4408 redundant test in assertion.
4410 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4412 * doc/rtl.texi: Document machine_mode wrapper classes.
4414 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4416 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4419 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4421 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4422 the VEC_PERM_EXPR fold to fail.
4424 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4427 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4428 to switched_sections.
4430 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4433 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4436 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4439 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4440 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4442 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4445 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4446 is BLKmode and bitpos not zero or mode change is needed.
4448 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4451 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4454 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4457 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4458 instead of MULT rtx. Update all corresponding splitters.
4460 (*ssub<modesuffix>): Ditto.
4462 (*cmp_sadd_di): Update split patterns.
4463 (*cmp_sadd_si): Ditto.
4464 (*cmp_sadd_sidi): Ditto.
4465 (*cmp_ssub_di): Ditto.
4466 (*cmp_ssub_si): Ditto.
4467 (*cmp_ssub_sidi): Ditto.
4468 * config/alpha/predicates.md (const23_operand): New predicate.
4469 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4470 Look for ASHIFT, not MULT inner operand.
4471 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4473 2018-01-04 Martin Liska <mliska@suse.cz>
4475 PR gcov-profile/83669
4476 * gcov.c (output_intermediate_file): Add version to intermediate
4478 * doc/gcov.texi: Document new field 'version' in intermediate
4479 file format. Fix location of '-k' option of gcov command.
4481 2018-01-04 Martin Liska <mliska@suse.cz>
4484 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4486 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4488 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4490 2018-01-03 Martin Sebor <msebor@redhat.com>
4492 PR tree-optimization/83655
4493 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4494 checking calls with invalid arguments.
4496 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4498 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4499 (vectorizable_mask_load_store): Delete.
4500 (vectorizable_call): Return false for masked loads and stores.
4501 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4502 instead of gimple_assign_rhs1.
4503 (vectorizable_load): Handle IFN_MASK_LOAD.
4504 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4506 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4508 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4510 (vectorizable_mask_load_store): ...here.
4511 (vectorizable_load): ...and here.
4513 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4515 * tree-vect-stmts.c (vect_build_all_ones_mask)
4516 (vect_build_zero_merge_argument): New functions, split out from...
4517 (vectorizable_load): ...here.
4519 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4521 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4523 (vectorizable_mask_load_store): ...here.
4524 (vectorizable_store): ...and here.
4526 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4528 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4530 (vectorizable_mask_load_store): ...here.
4532 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4534 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4535 (vect_model_store_cost): Take a vec_load_store_type instead of a
4537 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4538 (vect_model_store_cost): Take a vec_load_store_type instead of a
4540 (vectorizable_mask_load_store): Update accordingly.
4541 (vectorizable_store): Likewise.
4542 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4544 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4546 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4547 IFN_MASK_LOAD calls here rather than...
4548 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4550 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4551 Alan Hayward <alan.hayward@arm.com>
4552 David Sherwood <david.sherwood@arm.com>
4554 * expmed.c (extract_bit_field_1): For vector extracts,
4555 fall back to extract_bit_field_as_subreg if vec_extract
4558 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4559 Alan Hayward <alan.hayward@arm.com>
4560 David Sherwood <david.sherwood@arm.com>
4562 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4563 they are variable or constant sized.
4564 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4565 slots for constant-sized data.
4567 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4568 Alan Hayward <alan.hayward@arm.com>
4569 David Sherwood <david.sherwood@arm.com>
4571 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4572 handling COND_EXPRs with boolean comparisons, try to find a better
4573 basis for the mask type than the boolean itself.
4575 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4577 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4578 is calculated and how it can be overridden.
4579 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4580 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4582 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4585 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4586 Alan Hayward <alan.hayward@arm.com>
4587 David Sherwood <david.sherwood@arm.com>
4589 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4590 Remove the mode argument.
4591 (aarch64_simd_valid_immediate): Remove the mode and inverse
4593 * config/aarch64/iterators.md (bitsize): New iterator.
4594 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4595 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4596 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4597 aarch64_simd_valid_immediate.
4598 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4599 (aarch64_reg_or_bic_imm): Likewise.
4600 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4601 with an insn_type enum and msl with a modifier_type enum.
4602 Replace element_width with a scalar_mode. Change the shift
4603 to unsigned int. Add constructors for scalar_float_mode and
4604 scalar_int_mode elements.
4605 (aarch64_vect_float_const_representable_p): Delete.
4606 (aarch64_can_const_movi_rtx_p)
4607 (aarch64_simd_scalar_immediate_valid_for_move)
4608 (aarch64_simd_make_constant): Update call to
4609 aarch64_simd_valid_immediate.
4610 (aarch64_advsimd_valid_immediate_hs): New function.
4611 (aarch64_advsimd_valid_immediate): Likewise.
4612 (aarch64_simd_valid_immediate): Remove mode and inverse
4613 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4614 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4615 and aarch64_float_const_representable_p on the result.
4616 (aarch64_output_simd_mov_immediate): Remove mode argument.
4617 Update call to aarch64_simd_valid_immediate and use of
4618 simd_immediate_info.
4619 (aarch64_output_scalar_simd_mov_immediate): Update call
4622 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4623 Alan Hayward <alan.hayward@arm.com>
4624 David Sherwood <david.sherwood@arm.com>
4626 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4627 (mode_nunits): Likewise CONST_MODE_NUNITS.
4628 * machmode.def (ADJUST_NUNITS): Document.
4629 * genmodes.c (mode_data::need_nunits_adj): New field.
4630 (blank_mode): Update accordingly.
4631 (adj_nunits): New variable.
4632 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4634 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4635 listed in adj_nunits.
4636 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4637 listed in adj_nunits. Don't emit case statements for such modes.
4638 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4639 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4640 nothing if adj_nunits is nonnull.
4641 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4642 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4643 (emit_mode_fbit): Update use of print_maybe_const_decl.
4644 (emit_move_size): Likewise. Treat the array as non-const
4646 (emit_mode_adjustments): Handle adj_nunits.
4648 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4650 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4651 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4652 (VECTOR_MODES): Use it.
4653 (make_vector_modes): Take the prefix as an argument.
4655 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4656 Alan Hayward <alan.hayward@arm.com>
4657 David Sherwood <david.sherwood@arm.com>
4659 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4660 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4661 for MODE_VECTOR_BOOL.
4662 * machmode.def (VECTOR_BOOL_MODE): Document.
4663 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4664 (make_vector_bool_mode): New function.
4665 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4667 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4668 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4670 * stor-layout.c (int_mode_for_mode): Likewise.
4671 * tree.c (build_vector_type_for_mode): Likewise.
4672 * varasm.c (output_constant_pool_2): Likewise.
4673 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4674 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4675 for MODE_VECTOR_BOOL.
4676 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4677 of mode class checks.
4678 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4679 instead of a list of mode class checks.
4680 (expand_vector_scalar_condition): Likewise.
4681 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4683 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4684 Alan Hayward <alan.hayward@arm.com>
4685 David Sherwood <david.sherwood@arm.com>
4687 * machmode.h (mode_size): Change from unsigned short to
4689 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4690 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4691 or if measurement_type is not polynomial.
4692 (fixed_size_mode::includes_p): Check for constant-sized modes.
4693 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4694 return a poly_uint16 rather than an unsigned short.
4695 (emit_mode_size): Change the type of mode_size from unsigned short
4696 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4697 (emit_mode_adjustments): Cope with polynomial vector sizes.
4698 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4700 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4702 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4703 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4704 * caller-save.c (setup_save_areas): Likewise.
4705 (replace_reg_with_saved_mem): Likewise.
4706 * calls.c (emit_library_call_value_1): Likewise.
4707 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4708 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4709 (gen_lowpart_for_combine): Likewise.
4710 * convert.c (convert_to_integer_1): Likewise.
4711 * cse.c (equiv_constant, cse_insn): Likewise.
4712 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4713 (cselib_subst_to_values): Likewise.
4714 * dce.c (word_dce_process_block): Likewise.
4715 * df-problems.c (df_word_lr_mark_ref): Likewise.
4716 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4717 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4718 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4719 (rtl_for_decl_location): Likewise.
4720 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4721 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4722 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4723 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4724 (expand_expr_real_1): Likewise.
4725 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4726 (pad_below): Likewise.
4727 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4728 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4729 * ira.c (get_subreg_tracking_sizes): Likewise.
4730 * ira-build.c (ira_create_allocno_objects): Likewise.
4731 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4732 (ira_sort_regnos_for_alter_reg): Likewise.
4733 * ira-costs.c (record_operand_costs): Likewise.
4734 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4735 (resolve_simple_move): Likewise.
4736 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4737 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4738 (lra_constraints): Likewise.
4739 (CONST_POOL_OK_P): Reject variable-sized modes.
4740 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4741 (add_pseudo_to_slot, lra_spill): Likewise.
4742 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4743 * optabs-query.c (get_best_extraction_insn): Likewise.
4744 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4745 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4746 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4747 * recog.c (offsettable_address_addr_space_p): Likewise.
4748 * regcprop.c (maybe_mode_change): Likewise.
4749 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4750 * regrename.c (build_def_use): Likewise.
4751 * regstat.c (dump_reg_info): Likewise.
4752 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4753 (find_reloads, find_reloads_subreg_address): Likewise.
4754 * reload1.c (eliminate_regs_1): Likewise.
4755 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4756 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4757 (simplify_binary_operation_1, simplify_subreg): Likewise.
4758 * targhooks.c (default_function_arg_padding): Likewise.
4759 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4760 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4761 (verify_gimple_assign_ternary): Likewise.
4762 * tree-inline.c (estimate_move_cost): Likewise.
4763 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4764 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4765 (get_address_cost_ainc): Likewise.
4766 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4767 (vect_supportable_dr_alignment): Likewise.
4768 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4769 (vectorizable_reduction): Likewise.
4770 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4771 (vectorizable_operation, vectorizable_load): Likewise.
4772 * tree.c (build_same_sized_truth_vector_type): Likewise.
4773 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4774 * var-tracking.c (emit_note_insn_var_location): Likewise.
4775 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4776 (ADDR_VEC_ALIGN): Likewise.
4778 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4779 Alan Hayward <alan.hayward@arm.com>
4780 David Sherwood <david.sherwood@arm.com>
4782 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4784 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4785 or if measurement_type is polynomial.
4786 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4787 * combine.c (make_extraction): Likewise.
4788 * dse.c (find_shift_sequence): Likewise.
4789 * dwarf2out.c (mem_loc_descriptor): Likewise.
4790 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4791 (extract_bit_field, extract_low_bits): Likewise.
4792 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4793 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4794 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4795 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4796 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4797 * reload.c (find_reloads): Likewise.
4798 * reload1.c (alter_reg): Likewise.
4799 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4800 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4801 * tree-if-conv.c (predicate_mem_writes): Likewise.
4802 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4803 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4804 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4805 * valtrack.c (dead_debug_insert_temp): Likewise.
4806 * varasm.c (mergeable_constant_section): Likewise.
4807 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4809 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4810 Alan Hayward <alan.hayward@arm.com>
4811 David Sherwood <david.sherwood@arm.com>
4813 * expr.c (expand_assignment): Cope with polynomial mode sizes
4814 when assigning to a CONCAT.
4816 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4817 Alan Hayward <alan.hayward@arm.com>
4818 David Sherwood <david.sherwood@arm.com>
4820 * machmode.h (mode_precision): Change from unsigned short to
4822 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4824 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4825 or if measurement_type is not polynomial.
4826 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4827 in which the mode is already known to be a scalar_int_mode.
4828 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4829 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4831 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4832 for GET_MODE_PRECISION.
4833 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4834 for GET_MODE_PRECISION.
4835 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4837 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4838 (expand_field_assignment, make_extraction): Likewise.
4839 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4840 (get_last_value): Likewise.
4841 * convert.c (convert_to_integer_1): Likewise.
4842 * cse.c (cse_insn): Likewise.
4843 * expr.c (expand_expr_real_1): Likewise.
4844 * lra-constraints.c (simplify_operand_subreg): Likewise.
4845 * optabs-query.c (can_atomic_load_p): Likewise.
4846 * optabs.c (expand_atomic_load): Likewise.
4847 (expand_atomic_store): Likewise.
4848 * ree.c (combine_reaching_defs): Likewise.
4849 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4850 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4851 * tree.h (type_has_mode_precision_p): Likewise.
4852 * ubsan.c (instrument_si_overflow): Likewise.
4854 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4855 Alan Hayward <alan.hayward@arm.com>
4856 David Sherwood <david.sherwood@arm.com>
4858 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4859 polynomial numbers of units.
4860 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4861 (valid_vector_subparts_p): New function.
4862 (build_vector_type): Remove temporary shim and take the number
4863 of units as a poly_uint64 rather than an int.
4864 (build_opaque_vector_type): Take the number of units as a
4865 poly_uint64 rather than an int.
4866 * tree.c (build_vector_from_ctor): Handle polynomial
4867 TYPE_VECTOR_SUBPARTS.
4868 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4869 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4870 (build_vector_from_val): If the number of units is variable,
4871 use build_vec_duplicate_cst for constant operands and
4872 VEC_DUPLICATE_EXPR otherwise.
4873 (make_vector_type): Remove temporary is_constant ().
4874 (build_vector_type, build_opaque_vector_type): Take the number of
4875 units as a poly_uint64 rather than an int.
4876 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4878 * cfgexpand.c (expand_debug_expr): Likewise.
4879 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4880 (store_constructor, expand_expr_real_1): Likewise.
4881 (const_scalar_mask_from_tree): Likewise.
4882 * fold-const-call.c (fold_const_reduction): Likewise.
4883 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4884 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4885 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4886 (fold_relational_const): Likewise.
4887 (native_interpret_vector): Likewise. Change the size from an
4888 int to an unsigned int.
4889 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4890 TYPE_VECTOR_SUBPARTS.
4891 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4892 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4893 duplicating a non-constant operand into a variable-length vector.
4894 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4895 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4896 * ipa-icf.c (sem_variable::equals): Likewise.
4897 * match.pd: Likewise.
4898 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4899 * print-tree.c (print_node): Likewise.
4900 * stor-layout.c (layout_type): Likewise.
4901 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4902 * tree-cfg.c (verify_gimple_comparison): Likewise.
4903 (verify_gimple_assign_binary): Likewise.
4904 (verify_gimple_assign_ternary): Likewise.
4905 (verify_gimple_assign_single): Likewise.
4906 * tree-pretty-print.c (dump_generic_node): Likewise.
4907 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4908 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4909 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4910 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4911 (vect_shift_permute_load_chain): Likewise.
4912 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4913 (expand_vector_condition, optimize_vector_constructor): Likewise.
4914 (lower_vec_perm, get_compute_type): Likewise.
4915 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4916 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4917 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4918 (vect_recog_mask_conversion_pattern): Likewise.
4919 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4920 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4921 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4922 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4923 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4924 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4925 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4926 (supportable_widening_operation): Likewise.
4927 (supportable_narrowing_operation): Likewise.
4928 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4930 * varasm.c (output_constant): Likewise.
4932 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4933 Alan Hayward <alan.hayward@arm.com>
4934 David Sherwood <david.sherwood@arm.com>
4936 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4937 so that both the length == 3 and length != 3 cases set up their
4938 own permute vectors. Add comments explaining why we know the
4939 number of elements is constant.
4940 (vect_permute_load_chain): Likewise.
4942 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4943 Alan Hayward <alan.hayward@arm.com>
4944 David Sherwood <david.sherwood@arm.com>
4946 * machmode.h (mode_nunits): Change from unsigned char to
4948 (ONLY_FIXED_SIZE_MODES): New macro.
4949 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4950 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4951 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4953 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4954 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4955 or if measurement_type is not polynomial.
4956 * genmodes.c (ZERO_COEFFS): New macro.
4957 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4959 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4960 Use ZERO_COEFFS when emitting initializers.
4961 * data-streamer.h (bp_pack_poly_value): New function.
4962 (bp_unpack_poly_value): Likewise.
4963 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4964 for GET_MODE_NUNITS.
4965 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4966 for GET_MODE_NUNITS.
4967 * tree.c (make_vector_type): Remove temporary shim and make
4968 the real function take the number of units as a poly_uint64
4970 (build_vector_type_for_mode): Handle polynomial nunits.
4971 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4972 * emit-rtl.c (const_vec_series_p_1): Likewise.
4973 (gen_rtx_CONST_VECTOR): Likewise.
4974 * fold-const.c (test_vec_duplicate_folding): Likewise.
4975 * genrecog.c (validate_pattern): Likewise.
4976 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4977 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4978 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4979 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4980 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4981 * rtlanal.c (subreg_get_info): Likewise.
4982 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4983 (vect_grouped_load_supported): Likewise.
4984 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4985 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4986 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4987 (simplify_const_unary_operation, simplify_binary_operation_1)
4988 (simplify_const_binary_operation, simplify_ternary_operation)
4989 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4990 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4991 instead of CONST_VECTOR_NUNITS.
4992 * varasm.c (output_constant_pool_2): Likewise.
4993 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4994 explicit-encoded elements in the XVEC for variable-length vectors.
4996 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4998 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5000 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5001 Alan Hayward <alan.hayward@arm.com>
5002 David Sherwood <david.sherwood@arm.com>
5004 * coretypes.h (fixed_size_mode): Declare.
5005 (fixed_size_mode_pod): New typedef.
5006 * builtins.h (target_builtins::x_apply_args_mode)
5007 (target_builtins::x_apply_result_mode): Change type to
5008 fixed_size_mode_pod.
5009 * builtins.c (apply_args_size, apply_result_size, result_vector)
5010 (expand_builtin_apply_args_1, expand_builtin_apply)
5011 (expand_builtin_return): Update accordingly.
5013 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5015 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5016 * cselib.c (cselib_hash_rtx): Likewise.
5017 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5018 CONST_VECTOR encoding.
5020 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5021 Jeff Law <law@redhat.com>
5024 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5025 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5026 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5027 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5030 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5031 explicitly probe *sp in a noreturn function if there were any callee
5032 register saves or frame pointer is needed.
5034 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5037 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5038 BLKmode for ternary, binary or unary expressions.
5041 * var-tracking.c (delete_vta_debug_insn): New inline function.
5042 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5043 insns from get_insns () to NULL instead of each bb separately.
5044 Use delete_vta_debug_insn. No longer static.
5045 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5046 delete_vta_debug_insns callers.
5047 * rtl.h (delete_vta_debug_insns): Declare.
5048 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5049 instead of variable_tracking_main.
5051 2018-01-03 Martin Sebor <msebor@redhat.com>
5053 PR tree-optimization/83603
5054 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5055 arguments past the endof the argument list in functions declared
5056 without a prototype.
5057 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5058 Avoid checking when arguments are null.
5060 2018-01-03 Martin Sebor <msebor@redhat.com>
5063 * doc/extend.texi (attribute const): Fix a typo.
5064 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5065 issuing -Wsuggest-attribute for void functions.
5067 2018-01-03 Martin Sebor <msebor@redhat.com>
5069 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5070 offset_int::from instead of wide_int::to_shwi.
5071 (maybe_diag_overlap): Remove assertion.
5072 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5073 * gimple-ssa-sprintf.c (format_directive): Same.
5074 (parse_directive): Same.
5075 (sprintf_dom_walker::compute_format_length): Same.
5076 (try_substitute_return_value): Same.
5078 2018-01-03 Jeff Law <law@redhat.com>
5081 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5082 non-constant residual for zero at runtime and avoid probing in
5083 that case. Reorganize code for trailing problem to mirror handling
5086 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5088 PR tree-optimization/83501
5089 * tree-ssa-strlen.c (get_string_cst): New.
5090 (handle_char_store): Call get_string_cst.
5092 2018-01-03 Martin Liska <mliska@suse.cz>
5094 PR tree-optimization/83593
5095 * tree-ssa-strlen.c: Include tree-cfg.h.
5096 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5097 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5098 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5100 (strlen_dom_walker::before_dom_children): Call
5101 gimple_purge_dead_eh_edges. Dump tranformation with details
5103 (strlen_dom_walker::before_dom_children): Update call by adding
5104 new argument cleanup_eh.
5105 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5107 2018-01-03 Martin Liska <mliska@suse.cz>
5110 * cif-code.def (VARIADIC_THUNK): New enum value.
5111 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5114 2018-01-03 Jan Beulich <jbeulich@suse.com>
5116 * sse.md (mov<mode>_internal): Tighten condition for when to use
5117 vmovdqu<ssescalarsize> for TI and OI modes.
5119 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5121 Update copyright years.
5123 2018-01-03 Martin Liska <mliska@suse.cz>
5126 * ipa-visibility.c (function_and_variable_visibility): Skip
5127 functions with noipa attribure.
5129 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5131 * gcc.c (process_command): Update copyright notice dates.
5132 * gcov-dump.c (print_version): Ditto.
5133 * gcov.c (print_version): Ditto.
5134 * gcov-tool.c (print_version): Ditto.
5135 * gengtype.c (create_file): Ditto.
5136 * doc/cpp.texi: Bump @copying's copyright year.
5137 * doc/cppinternals.texi: Ditto.
5138 * doc/gcc.texi: Ditto.
5139 * doc/gccint.texi: Ditto.
5140 * doc/gcov.texi: Ditto.
5141 * doc/install.texi: Ditto.
5142 * doc/invoke.texi: Ditto.
5144 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5146 * vector-builder.h (vector_builder::m_full_nelts): Change from
5147 unsigned int to poly_uint64.
5148 (vector_builder::full_nelts): Update prototype accordingly.
5149 (vector_builder::new_vector): Likewise.
5150 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5151 (vector_builder::operator ==): Likewise.
5152 (vector_builder::finalize): Likewise.
5153 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5154 Take the number of elements as a poly_uint64 rather than an
5156 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5157 from unsigned int to poly_uint64.
5158 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5159 (vec_perm_indices::new_vector): Likewise.
5160 (vec_perm_indices::length): Likewise.
5161 (vec_perm_indices::nelts_per_input): Likewise.
5162 (vec_perm_indices::input_nelts): Likewise.
5163 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5164 number of elements per input as a poly_uint64 rather than an
5165 unsigned int. Use the original encoding for variable-length
5166 vectors, rather than clamping each individual element.
5167 For the second and subsequent elements in each pattern,
5168 clamp the step and base before clamping their sum.
5169 (vec_perm_indices::series_p): Handle polynomial element counts.
5170 (vec_perm_indices::all_in_range_p): Likewise.
5171 (vec_perm_indices_to_tree): Likewise.
5172 (vec_perm_indices_to_rtx): Likewise.
5173 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5174 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5175 (tree_vector_builder::new_binary_operation): Handle polynomial
5176 element counts. Return false if we need to know the number
5177 of elements at compile time.
5178 * fold-const.c (fold_vec_perm): Punt if the number of elements
5179 isn't known at compile time.
5181 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5183 * vec-perm-indices.h (vec_perm_builder): Change element type
5184 from HOST_WIDE_INT to poly_int64.
5185 (vec_perm_indices::element_type): Update accordingly.
5186 (vec_perm_indices::clamp): Handle polynomial element_types.
5187 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5188 (vec_perm_indices::all_in_range_p): Likewise.
5189 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5191 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5192 polynomial vec_perm_indices element types.
5193 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5194 * fold-const.c (fold_vec_perm): Likewise.
5195 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5196 * tree-vect-generic.c (lower_vec_perm): Likewise.
5197 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5198 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5199 element type to HOST_WIDE_INT.
5201 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5202 Alan Hayward <alan.hayward@arm.com>
5203 David Sherwood <david.sherwood@arm.com>
5205 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5206 rather than an int. Use plus_constant.
5207 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5208 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5210 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5211 Alan Hayward <alan.hayward@arm.com>
5212 David Sherwood <david.sherwood@arm.com>
5214 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5215 a HOST_WIDE_INT to a poly_int64.
5217 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5218 Alan Hayward <alan.hayward@arm.com>
5219 David Sherwood <david.sherwood@arm.com>
5221 * calls.c (load_register_parameters): Cope with polynomial
5222 mode sizes. Require a constant size for BLKmode parameters
5223 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5224 forces a parameter to be padded at the lsb end in order to
5225 fill a complete number of words, require the parameter size
5226 to be ordered wrt UNITS_PER_WORD.
5228 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5229 Alan Hayward <alan.hayward@arm.com>
5230 David Sherwood <david.sherwood@arm.com>
5232 * reload1.c (spill_stack_slot_width): Change element type
5233 from unsigned int to poly_uint64_pod.
5234 (alter_reg): Treat mode sizes as polynomial.
5236 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5237 Alan Hayward <alan.hayward@arm.com>
5238 David Sherwood <david.sherwood@arm.com>
5240 * reload.c (complex_word_subreg_p): New function.
5241 (reload_inner_reg_of_subreg, push_reload): Use it.
5243 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5244 Alan Hayward <alan.hayward@arm.com>
5245 David Sherwood <david.sherwood@arm.com>
5247 * lra-constraints.c (process_alt_operands): Reject matched
5248 operands whose sizes aren't ordered.
5249 (match_reload): Refer to this check here.
5251 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5252 Alan Hayward <alan.hayward@arm.com>
5253 David Sherwood <david.sherwood@arm.com>
5255 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5256 that the mode size is in the set {1, 2, 4, 8, 16}.
5258 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5259 Alan Hayward <alan.hayward@arm.com>
5260 David Sherwood <david.sherwood@arm.com>
5262 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5263 Use plus_constant instead of gen_rtx_PLUS.
5265 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5266 Alan Hayward <alan.hayward@arm.com>
5267 David Sherwood <david.sherwood@arm.com>
5269 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5270 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5271 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5272 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5273 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5274 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5275 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5276 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5277 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5278 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5280 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5281 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5282 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5283 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5284 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5285 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5286 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5287 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5288 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5289 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5291 * expr.c (emit_move_resolve_push): Treat the input and result
5292 of PUSH_ROUNDING as a poly_int64.
5293 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5294 (emit_push_insn): Likewise.
5295 * lra-eliminations.c (mark_not_eliminable): Likewise.
5296 * recog.c (push_operand): Likewise.
5297 * reload1.c (elimination_effects): Likewise.
5298 * rtlanal.c (nonzero_bits1): Likewise.
5299 * calls.c (store_one_arg): Likewise. Require the padding to be
5300 known at compile time.
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 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5307 Use plus_constant instead of gen_rtx_PLUS.
5309 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5310 Alan Hayward <alan.hayward@arm.com>
5311 David Sherwood <david.sherwood@arm.com>
5313 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5316 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5317 Alan Hayward <alan.hayward@arm.com>
5318 David Sherwood <david.sherwood@arm.com>
5320 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5321 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5322 via stack temporaries. Treat the mode size as polynomial too.
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 * expr.c (expand_expr_real_2): When handling conversions involving
5329 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5330 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5331 as a poly_uint64 too.
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 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5339 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5340 Alan Hayward <alan.hayward@arm.com>
5341 David Sherwood <david.sherwood@arm.com>
5343 * combine.c (can_change_dest_mode): Handle polynomial
5344 REGMODE_NATURAL_SIZE.
5345 * expmed.c (store_bit_field_1): Likewise.
5346 * expr.c (store_constructor): Likewise.
5347 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5348 and polynomial REGMODE_NATURAL_SIZE.
5349 (gen_lowpart_common): Likewise.
5350 * reginfo.c (record_subregs_of_mode): Likewise.
5351 * rtlanal.c (read_modify_subreg_p): Likewise.
5353 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5354 Alan Hayward <alan.hayward@arm.com>
5355 David Sherwood <david.sherwood@arm.com>
5357 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5358 numbers of elements.
5360 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5361 Alan Hayward <alan.hayward@arm.com>
5362 David Sherwood <david.sherwood@arm.com>
5364 * match.pd: Cope with polynomial numbers of vector elements.
5366 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5367 Alan Hayward <alan.hayward@arm.com>
5368 David Sherwood <david.sherwood@arm.com>
5370 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5371 in a POINTER_PLUS_EXPR.
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 * omp-simd-clone.c (simd_clone_subparts): New function.
5378 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5379 (ipa_simd_modify_function_body): Likewise.
5381 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5382 Alan Hayward <alan.hayward@arm.com>
5383 David Sherwood <david.sherwood@arm.com>
5385 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5386 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5387 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5388 (expand_vector_condition, vector_element): Likewise.
5389 (subparts_gt): New function.
5390 (get_compute_type): Use subparts_gt.
5391 (count_type_subparts): Delete.
5392 (expand_vector_operations_1): Use subparts_gt instead of
5393 count_type_subparts.
5395 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5396 Alan Hayward <alan.hayward@arm.com>
5397 David Sherwood <david.sherwood@arm.com>
5399 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5400 (vect_compile_time_alias): ...this new function. Do the calculation
5401 on poly_ints rather than trees.
5402 (vect_prune_runtime_alias_test_list): Update call accordingly.
5404 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5405 Alan Hayward <alan.hayward@arm.com>
5406 David Sherwood <david.sherwood@arm.com>
5408 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5410 (vect_schedule_slp_instance): Likewise.
5412 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5413 Alan Hayward <alan.hayward@arm.com>
5414 David Sherwood <david.sherwood@arm.com>
5416 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5417 constant and extern definitions for variable-length vectors.
5418 (vect_get_constant_vectors): Note that the number of units
5419 is known to be constant.
5421 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5422 Alan Hayward <alan.hayward@arm.com>
5423 David Sherwood <david.sherwood@arm.com>
5425 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5426 of units as polynomial. Choose between WIDE and NARROW based
5429 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5430 Alan Hayward <alan.hayward@arm.com>
5431 David Sherwood <david.sherwood@arm.com>
5433 * tree-vect-stmts.c (simd_clone_subparts): New function.
5434 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5436 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5437 Alan Hayward <alan.hayward@arm.com>
5438 David Sherwood <david.sherwood@arm.com>
5440 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5441 vectors as polynomial. Use build_index_vector for
5444 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5445 Alan Hayward <alan.hayward@arm.com>
5446 David Sherwood <david.sherwood@arm.com>
5448 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5449 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5450 for variable-length vectors.
5451 (vectorizable_mask_load_store): Treat the number of units as
5452 polynomial, asserting that it is constant if the condition has
5453 already been enforced.
5454 (vectorizable_store, vectorizable_load): Likewise.
5456 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5457 Alan Hayward <alan.hayward@arm.com>
5458 David Sherwood <david.sherwood@arm.com>
5460 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5461 of units as polynomial. Punt if we can't tell at compile time
5462 which vector contains the final result.
5464 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5465 Alan Hayward <alan.hayward@arm.com>
5466 David Sherwood <david.sherwood@arm.com>
5468 * tree-vect-loop.c (vectorizable_induction): Treat the number
5469 of units as polynomial. Punt on SLP inductions. Use an integer
5470 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5471 cast of such a series for variable-length floating-point
5474 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5475 Alan Hayward <alan.hayward@arm.com>
5476 David Sherwood <david.sherwood@arm.com>
5478 * tree.h (build_index_vector): Declare.
5479 * tree.c (build_index_vector): New function.
5480 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5481 of units as polynomial, forcibly converting it to a constant if
5482 vectorizable_reduction has already enforced the condition.
5483 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5484 to create a {1,2,3,...} vector.
5485 (vectorizable_reduction): Treat the number of units as polynomial.
5486 Choose vectype_in based on the largest scalar element size rather
5487 than the smallest number of units. Enforce the restrictions
5490 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5491 Alan Hayward <alan.hayward@arm.com>
5492 David Sherwood <david.sherwood@arm.com>
5494 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5495 number of units as polynomial.
5497 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5498 Alan Hayward <alan.hayward@arm.com>
5499 David Sherwood <david.sherwood@arm.com>
5501 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5502 * target.def (autovectorize_vector_sizes): Return the vector sizes
5503 by pointer, using vector_sizes rather than a bitmask.
5504 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5505 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5506 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5508 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5509 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5510 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5511 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5512 * omp-general.c (omp_max_vf): Likewise.
5513 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5514 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5515 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5516 * tree-vect-slp.c (vect_slp_bb): Likewise.
5517 * doc/tm.texi: Regenerate.
5518 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5520 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5521 the vector size as a poly_uint64 rather than an unsigned int.
5522 (current_vector_size): Change from an unsigned int to a poly_uint64.
5523 (get_vectype_for_scalar_type): Update accordingly.
5524 * tree.h (build_truth_vector_type): Take the size and number of
5525 units as a poly_uint64 rather than an unsigned int.
5526 (build_vector_type): Add a temporary overload that takes
5527 the number of units as a poly_uint64 rather than an unsigned int.
5528 * tree.c (make_vector_type): Likewise.
5529 (build_truth_vector_type): Take the number of units as a poly_uint64
5530 rather than an unsigned int.
5532 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5533 Alan Hayward <alan.hayward@arm.com>
5534 David Sherwood <david.sherwood@arm.com>
5536 * target.def (get_mask_mode): Take the number of units and length
5537 as poly_uint64s rather than unsigned ints.
5538 * targhooks.h (default_get_mask_mode): Update accordingly.
5539 * targhooks.c (default_get_mask_mode): Likewise.
5540 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5541 * doc/tm.texi: Regenerate.
5543 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5544 Alan Hayward <alan.hayward@arm.com>
5545 David Sherwood <david.sherwood@arm.com>
5547 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5548 * omp-general.c (omp_max_vf): Likewise.
5549 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5550 (expand_omp_simd): Handle polynomial safelen.
5551 * omp-low.c (omplow_simd_context): Add a default constructor.
5552 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5553 (lower_rec_simd_input_clauses): Update accordingly.
5554 (lower_rec_input_clauses): Likewise.
5556 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5557 Alan Hayward <alan.hayward@arm.com>
5558 David Sherwood <david.sherwood@arm.com>
5560 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5561 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5562 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5563 (vect_analyze_slp_cost): Likewise.
5564 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5565 (vect_model_load_cost): Likewise.
5567 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5568 Alan Hayward <alan.hayward@arm.com>
5569 David Sherwood <david.sherwood@arm.com>
5571 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5572 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5573 from an unsigned int * to a poly_uint64_pod *.
5574 (calculate_unrolling_factor): New function.
5575 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5577 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5578 Alan Hayward <alan.hayward@arm.com>
5579 David Sherwood <david.sherwood@arm.com>
5581 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5582 from an unsigned int to a poly_uint64.
5583 (_loop_vec_info::slp_unrolling_factor): Likewise.
5584 (_loop_vec_info::vectorization_factor): Change from an int
5586 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5587 (vect_get_num_vectors): New function.
5588 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5589 (vect_get_num_copies): Use vect_get_num_vectors.
5590 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5591 to an unsigned int *.
5592 (vect_analyze_data_refs): Change min_vf from an int * to a
5594 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5595 than an unsigned HOST_WIDE_INT.
5596 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5597 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5598 to an unsigned int *.
5599 (vect_analyze_data_ref_dependences): Likewise.
5600 (vect_compute_data_ref_alignment): Handle polynomial vf.
5601 (vect_enhance_data_refs_alignment): Likewise.
5602 (vect_prune_runtime_alias_test_list): Likewise.
5603 (vect_shift_permute_load_chain): Likewise.
5604 (vect_supportable_dr_alignment): Likewise.
5605 (dependence_distance_ge_vf): Take the vectorization factor as a
5606 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5607 (vect_analyze_data_refs): Change min_vf from an int * to a
5609 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5610 vfm1 as a poly_uint64 rather than an int. Make the same change
5611 for the returned bound_scalar.
5612 (vect_gen_vector_loop_niters): Handle polynomial vf.
5613 (vect_do_peeling): Likewise. Update call to
5614 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5615 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5617 * tree-vect-loop.c (vect_determine_vectorization_factor)
5618 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5619 (vect_get_known_peeling_cost): Likewise.
5620 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5621 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5622 (vect_transform_loop): Likewise. Use the lowest possible VF when
5623 updating the upper bounds of the loop.
5624 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5626 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5627 polynomial unroll factors.
5628 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5629 (vect_make_slp_decision): Likewise.
5630 (vect_supported_load_permutation_p): Likewise, and polynomial
5632 (vect_analyze_slp_cost): Handle polynomial vf.
5633 (vect_slp_analyze_node_operations): Likewise.
5634 (vect_slp_analyze_bb_1): Likewise.
5635 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5636 than an unsigned HOST_WIDE_INT.
5637 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5638 (vectorizable_load): Handle polynomial vf.
5639 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5641 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5643 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5644 Alan Hayward <alan.hayward@arm.com>
5645 David Sherwood <david.sherwood@arm.com>
5647 * match.pd: Handle bit operations involving three constants
5648 and try to fold one pair.
5650 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5652 * tree-vect-loop-manip.c: Include gimple-fold.h.
5653 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5654 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5655 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5656 Add a path that uses a step of VF instead of 1, but disable it
5658 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5659 and niters_no_overflow parameters. Update calls to
5660 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5661 Create a new SSA name if the latter choses to use a ste other
5662 than zero, and return it via niters_vector_mult_vf_var.
5663 * tree-vect-loop.c (vect_transform_loop): Update calls to
5664 vect_do_peeling, vect_gen_vector_loop_niters and
5665 slpeel_make_loop_iterate_ntimes.
5666 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5667 (vect_gen_vector_loop_niters): Update declarations after above changes.
5669 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5671 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5672 128-bit round to integer instructions.
5673 (ceil<mode>2): Likewise.
5674 (btrunc<mode>2): Likewise.
5675 (round<mode>2): Likewise.
5677 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5679 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5680 unaligned VSX load/store on P8/P9.
5681 (expand_block_clear): Allow the use of unaligned VSX
5682 load/store on P8/P9.
5684 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5686 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5688 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5689 swap associated with both a load and a store.
5691 2018-01-02 Andrew Waterman <andrew@sifive.com>
5693 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5694 * config/riscv/riscv.md (clear_cache): Use it.
5696 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5698 * web.c: Remove out-of-date comment.
5700 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5702 * expr.c (fixup_args_size_notes): Check that any existing
5703 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5704 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5705 (emit_single_push_insn): ...here.
5707 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5709 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5710 (const_vector_encoded_nelts): New function.
5711 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5712 (const_vector_int_elt, const_vector_elt): Declare.
5713 * emit-rtl.c (const_vector_int_elt_1): New function.
5714 (const_vector_elt): Likewise.
5715 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5716 of CONST_VECTOR_ELT.
5718 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5720 * expr.c: Include rtx-vector-builder.h.
5721 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5722 directly on the tree encoding.
5723 (const_vector_from_tree): Likewise.
5724 * optabs.c: Include rtx-vector-builder.h.
5725 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5726 sequence of "u" values.
5727 * vec-perm-indices.c: Include rtx-vector-builder.h.
5728 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5729 directly on the vec_perm_indices encoding.
5731 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5733 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5734 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5735 * rtx-vector-builder.h: New file.
5736 * rtx-vector-builder.c: Likewise.
5737 * rtl.h (rtx_def::u2): Add a const_vector field.
5738 (CONST_VECTOR_NPATTERNS): New macro.
5739 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5740 (CONST_VECTOR_DUPLICATE_P): Likewise.
5741 (CONST_VECTOR_STEPPED_P): Likewise.
5742 (CONST_VECTOR_ENCODED_ELT): Likewise.
5743 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5744 (unwrap_const_vec_duplicate): Likewise.
5745 (const_vec_series_p): Check for a non-duplicated vector encoding.
5746 Say that the function only returns true for integer vectors.
5747 * emit-rtl.c: Include rtx-vector-builder.h.
5748 (gen_const_vec_duplicate_1): Delete.
5749 (gen_const_vector): Call gen_const_vec_duplicate instead of
5750 gen_const_vec_duplicate_1.
5751 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5752 (gen_const_vec_duplicate): Use rtx_vector_builder.
5753 (gen_const_vec_series): Likewise.
5754 (gen_rtx_CONST_VECTOR): Likewise.
5755 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5756 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5757 Build a new vector rather than modifying a CONST_VECTOR in-place.
5758 (handle_special_swappables): Update call accordingly.
5759 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5760 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5761 Build a new vector rather than modifying a CONST_VECTOR in-place.
5762 (handle_special_swappables): Update call accordingly.
5764 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5766 * simplify-rtx.c (simplify_const_binary_operation): Use
5767 CONST_VECTOR_ELT instead of XVECEXP.
5769 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5771 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5772 the selector elements to be different from the data elements
5773 if the selector is a VECTOR_CST.
5774 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5775 ssizetype for the selector.
5777 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5779 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5780 before testing each element individually.
5781 * tree-vect-generic.c (lower_vec_perm): Likewise.
5783 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5785 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5786 * selftest-run-tests.c (selftest::run_tests): Call it.
5787 * vector-builder.h (vector_builder::operator ==): New function.
5788 (vector_builder::operator !=): Likewise.
5789 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5790 (vec_perm_indices::all_from_input_p): New function.
5791 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5792 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5793 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5794 instead of reading the VECTOR_CST directly. Detect whether both
5795 vector inputs are the same before constructing the vec_perm_indices,
5796 and update the number of inputs argument accordingly. Use the
5797 utility functions added above. Only construct sel2 if we need to.
5799 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5801 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5802 the broadcast of the low byte.
5803 (expand_mult_highpart): Use an explicit encoding for the permutes.
5804 * optabs-query.c (can_mult_highpart_p): Likewise.
5805 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5806 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5807 (vectorizable_bswap): Likewise.
5808 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5809 explicit encoding for the power-of-2 permutes.
5810 (vect_permute_store_chain): Likewise.
5811 (vect_grouped_load_supported): Likewise.
5812 (vect_permute_load_chain): Likewise.
5814 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5816 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5817 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5818 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5819 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5820 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5821 (vect_gen_perm_mask_any): Likewise.
5823 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5825 * int-vector-builder.h: New file.
5826 * vec-perm-indices.h: Include int-vector-builder.h.
5827 (vec_perm_indices): Redefine as an int_vector_builder.
5828 (auto_vec_perm_indices): Delete.
5829 (vec_perm_builder): Redefine as a stand-alone class.
5830 (vec_perm_indices::vec_perm_indices): New function.
5831 (vec_perm_indices::clamp): Likewise.
5832 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5833 (vec_perm_indices::new_vector): New function.
5834 (vec_perm_indices::new_expanded_vector): Update for new
5835 vec_perm_indices class.
5836 (vec_perm_indices::rotate_inputs): New function.
5837 (vec_perm_indices::all_in_range_p): Operate directly on the
5838 encoded form, without computing elided elements.
5839 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5840 encoding. Update for new vec_perm_indices class.
5841 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5842 the given vec_perm_builder.
5843 (expand_vec_perm_var): Update vec_perm_builder constructor.
5844 (expand_mult_highpart): Use vec_perm_builder instead of
5845 auto_vec_perm_indices.
5846 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5847 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5848 or double series encoding as appropriate.
5849 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5850 vec_perm_indices instead of auto_vec_perm_indices.
5851 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5852 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5853 (vect_permute_store_chain): Likewise.
5854 (vect_grouped_load_supported): Likewise.
5855 (vect_permute_load_chain): Likewise.
5856 (vect_shift_permute_load_chain): Likewise.
5857 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5858 (vect_transform_slp_perm_load): Likewise.
5859 (vect_schedule_slp_instance): Likewise.
5860 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5861 (vectorizable_mask_load_store): Likewise.
5862 (vectorizable_bswap): Likewise.
5863 (vectorizable_store): Likewise.
5864 (vectorizable_load): Likewise.
5865 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5866 vec_perm_indices instead of auto_vec_perm_indices. Use
5867 tree_to_vec_perm_builder to read the vector from a tree.
5868 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5869 vec_perm_builder instead of a vec_perm_indices.
5870 (have_whole_vector_shift): Use vec_perm_builder and
5871 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5872 truncation to calc_vec_perm_mask_for_shift.
5873 (vect_create_epilog_for_reduction): Likewise.
5874 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5875 from auto_vec_perm_indices to vec_perm_indices.
5876 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5877 instead of changing individual elements.
5878 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5879 the vector in d.perm.
5880 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5881 from auto_vec_perm_indices to vec_perm_indices.
5882 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5883 instead of changing individual elements.
5884 (arm_vectorize_vec_perm_const): Use new_vector to install
5885 the vector in d.perm.
5886 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5887 Update vec_perm_builder constructor.
5888 (rs6000_expand_interleave): Likewise.
5889 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5890 (rs6000_expand_interleave): Likewise.
5892 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5894 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5895 to qimode could truncate the indices.
5896 * optabs.c (expand_vec_perm_var): Likewise.
5898 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5900 * Makefile.in (OBJS): Add vec-perm-indices.o.
5901 * vec-perm-indices.h: New file.
5902 * vec-perm-indices.c: Likewise.
5903 * target.h (vec_perm_indices): Replace with a forward class
5905 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5906 * optabs.h: Include vec-perm-indices.h.
5907 (expand_vec_perm): Delete.
5908 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5909 (expand_vec_perm_const): Declare.
5910 * target.def (vec_perm_const_ok): Replace with...
5911 (vec_perm_const): ...this new hook.
5912 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5913 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5914 * doc/tm.texi: Regenerate.
5915 * optabs.def (vec_perm_const): Delete.
5916 * doc/md.texi (vec_perm_const): Likewise.
5917 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5918 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5919 expand_vec_perm for constant permutation vectors. Assert that
5920 the mode of variable permutation vectors is the integer equivalent
5921 of the mode that is being permuted.
5922 * optabs-query.h (selector_fits_mode_p): Declare.
5923 * optabs-query.c: Include vec-perm-indices.h.
5924 (selector_fits_mode_p): New function.
5925 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5926 is defined, instead of checking whether the vec_perm_const_optab
5927 exists. Use targetm.vectorize.vec_perm_const instead of
5928 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5929 fit in the vector mode before using a variable permute.
5930 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5931 vec_perm_indices instead of an rtx.
5932 (expand_vec_perm): Replace with...
5933 (expand_vec_perm_const): ...this new function. Take the selector
5934 as a vec_perm_indices rather than an rtx. Also take the mode of
5935 the selector. Update call to shift_amt_for_vec_perm_mask.
5936 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5937 Use vec_perm_indices::new_expanded_vector to expand the original
5938 selector into bytes. Check whether the indices fit in the vector
5939 mode before using a variable permute.
5940 (expand_vec_perm_var): Make global.
5941 (expand_mult_highpart): Use expand_vec_perm_const.
5942 * fold-const.c: Includes vec-perm-indices.h.
5943 * tree-ssa-forwprop.c: Likewise.
5944 * tree-vect-data-refs.c: Likewise.
5945 * tree-vect-generic.c: Likewise.
5946 * tree-vect-loop.c: Likewise.
5947 * tree-vect-slp.c: Likewise.
5948 * tree-vect-stmts.c: Likewise.
5949 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5951 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5952 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5953 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5954 (aarch64_vectorize_vec_perm_const): ...this new function.
5955 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5956 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5957 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5958 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5959 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5960 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5961 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5963 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5964 check for NEON modes.
5965 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5966 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5967 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5968 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5970 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5971 the old VEC_PERM_CONST conditions.
5972 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5973 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5974 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5975 (ia64_vectorize_vec_perm_const_ok): Merge into...
5976 (ia64_vectorize_vec_perm_const): ...this new function.
5977 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5978 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5979 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5980 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5981 * config/mips/mips.c (mips_expand_vec_perm_const)
5982 (mips_vectorize_vec_perm_const_ok): Merge into...
5983 (mips_vectorize_vec_perm_const): ...this new function.
5984 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5985 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5986 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5987 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5988 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5989 (rs6000_expand_vec_perm_const): Delete.
5990 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5992 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5993 (altivec_expand_vec_perm_const_le): Take each operand individually.
5994 Operate on constant selectors rather than rtxes.
5995 (altivec_expand_vec_perm_const): Likewise. Update call to
5996 altivec_expand_vec_perm_const_le.
5997 (rs6000_expand_vec_perm_const): Delete.
5998 (rs6000_vectorize_vec_perm_const_ok): Delete.
5999 (rs6000_vectorize_vec_perm_const): New function.
6000 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6001 an element count and rtx array.
6002 (rs6000_expand_extract_even): Update call accordingly.
6003 (rs6000_expand_interleave): Likewise.
6004 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6005 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6006 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6007 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6008 (rs6000_expand_vec_perm_const): Delete.
6009 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6010 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6011 (altivec_expand_vec_perm_const_le): Take each operand individually.
6012 Operate on constant selectors rather than rtxes.
6013 (altivec_expand_vec_perm_const): Likewise. Update call to
6014 altivec_expand_vec_perm_const_le.
6015 (rs6000_expand_vec_perm_const): Delete.
6016 (rs6000_vectorize_vec_perm_const_ok): Delete.
6017 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6018 reference to the SPE evmerge intructions.
6019 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6020 an element count and rtx array.
6021 (rs6000_expand_extract_even): Update call accordingly.
6022 (rs6000_expand_interleave): Likewise.
6023 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6024 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6026 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6028 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6030 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6031 vector mode and that that mode matches the mode of the data
6033 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6034 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6035 directly using expand_vec_perm_1 when forcing selectors into
6037 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6039 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6041 * optabs-query.h (can_vec_perm_p): Delete.
6042 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6043 * optabs-query.c (can_vec_perm_p): Split into...
6044 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6045 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6046 particular selector is valid.
6047 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6048 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6049 (vect_grouped_load_supported): Likewise.
6050 (vect_shift_permute_load_chain): Likewise.
6051 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6052 (vect_transform_slp_perm_load): Likewise.
6053 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6054 (vectorizable_bswap): Likewise.
6055 (vect_gen_perm_mask_checked): Likewise.
6056 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6057 implementations of variable permutation vectors into account
6058 when deciding which selector to use.
6059 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6060 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6061 with a false third argument.
6062 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6063 to test whether the constant selector is valid and can_vec_perm_var_p
6064 to test whether a variable selector is valid.
6066 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6068 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6069 * optabs-query.c (can_vec_perm_p): Likewise.
6070 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6071 instead of vec_perm_indices.
6072 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6073 (vect_gen_perm_mask_checked): Likewise,
6074 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6075 (vect_gen_perm_mask_checked): Likewise,
6077 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6079 * optabs-query.h (qimode_for_vec_perm): Declare.
6080 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6081 (qimode_for_vec_perm): ...this new function.
6082 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6084 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6086 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6087 does not have a conditional at the top.
6089 2018-01-02 Richard Biener <rguenther@suse.de>
6091 * ipa-inline.c (big_speedup_p): Fix expression.
6093 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6096 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6099 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6103 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6104 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6105 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6106 cond_taken_branch_cost 3->4.
6108 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6110 PR tree-optimization/83581
6111 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6112 TODO_cleanup_cfg if any changes have been made.
6115 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6116 convert_modes if target mode has the right side, but different mode
6120 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6121 last argument when extracting from CONCAT. If either from_real or
6122 from_imag is NULL, use expansion through memory. If result is not
6123 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6124 the parts directly to inner mode, if even that fails, use expansion
6128 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6129 check for bswap in mode rather than HImode and use that in expand_unop
6132 Copyright (C) 2018 Free Software Foundation, Inc.
6134 Copying and distribution of this file, with or without modification,
6135 are permitted in any medium without royalty provided the copyright
6136 notice and this notice are preserved.