1 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
3 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove irrelevant options.
5 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
7 * config/rs6000/rs6000.c (rs6000_option_override_internal):
8 Display warning message for -mno-speculate-indirect-jumps.
10 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
12 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
14 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
16 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
18 PR tree-optimization/84225
19 * tree-eh.c (find_trapping_overflow): Only call
20 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
22 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
25 * config/i386/i386.c: Reimplement the check of possible options
26 -mibt/-mshstk conbination. Change error messages.
27 * doc/invoke.texi: Fix a typo: remove extra '='.
29 2018-02-06 Marek Polacek <polacek@redhat.com>
31 PR tree-optimization/84228
32 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
34 2018-02-06 Tamar Christina <tamar.christina@arm.com>
37 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
38 emitted arch directives.
39 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
40 __ARM_FEATURE_COPROC before changing architectures.
42 2018-02-06 Richard Biener <rguenther@suse.de>
44 * config/i386/i386.c (print_reg): Fix typo.
45 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
47 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
49 * configure: Regenerate.
51 2018-02-05 Martin Sebor <msebor@redhat.com>
53 PR tree-optimization/83369
54 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
57 2018-02-05 Martin Liska <mliska@suse.cz>
59 * doc/invoke.texi: Cherry-pick upstream r323995.
61 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
63 * ira.c (ira_init_register_move_cost): Adjust comment.
65 2018-02-05 Martin Liska <mliska@suse.cz>
68 * doc/gcov.texi: Fix typo in documentation.
70 2018-02-05 Martin Liska <mliska@suse.cz>
73 * doc/gcov.texi: Document necessity of --dynamic-list-data when
74 using dlopen functionality.
76 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
78 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
79 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
80 _mm_maskz_range_ss, _mm_mask_range_round_ss,
81 _mm_maskz_range_round_ss): New intrinsics.
82 (__builtin_ia32_rangesd128_round)
83 (__builtin_ia32_rangess128_round): Remove.
84 (__builtin_ia32_rangesd128_mask_round,
85 __builtin_ia32_rangess128_mask_round): New builtins.
86 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
87 __builtin_ia32_rangess128_round): Remove.
88 (__builtin_ia32_rangesd128_mask_round,
89 __builtin_ia32_rangess128_mask_round): New builtins.
90 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
91 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
92 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
93 "<round_saeonly_constraint>")): Changed to ...
94 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
95 "<round_saeonly_scalar_constraint>")): ... this.
96 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
97 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
98 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
99 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
100 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
103 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
105 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
107 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
108 Remove all values except native, 8540 and 8548.
110 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
112 * config/i386/i386.c (ix86_output_function_return): Pass
113 INVALID_REGNUM, instead of -1, as invalid register number to
114 indirect_thunk_name and output_indirect_thunk.
116 2018-02-02 Julia Koval <julia.koval@intel.com>
118 * config.gcc: Add -march=icelake.
119 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
120 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
121 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
122 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
123 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
124 (processor_target_table): Add icelake.
125 (ix86_option_override_internal): Handle new PTAs.
126 (get_builtin_code_for_version): Handle icelake.
127 (M_INTEL_COREI7_ICELAKE): New.
128 (fold_builtin_cpu): Handle icelake.
129 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
130 * doc/invoke.texi: Add -march=icelake.
132 2018-02-02 Julia Koval <julia.koval@intel.com>
134 * config/i386/i386.c (ix86_option_override_internal): Change flags type
136 * wide-int-bitmask.h: New.
138 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
141 * config/i386/i386.md: Replace Pmode with word_mode in
142 builtin_setjmp_setup and builtin_longjmp to support x32.
144 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
148 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
150 (rs6000_supported_cpu_names): New static variable.
151 (linux_cpu_translation_table): Likewise.
152 (elf_platform) <cpu>: Define new static variable and use it.
153 Translate kernel AT_PLATFORM name to canonical name if needed.
154 Error if platform name is unknown.
156 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
159 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
161 2018-02-01 Jeff Law <law@redhat.com>
164 * config/i386/i386.c (release_scratch_register_on_entry): Add new
165 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
166 the scratch if RELEASE_VIA_POP is false.
167 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
168 If we have to save a temporary register, decrement SIZE appropriately.
169 Pass new arguments to release_scratch_register_on_entry.
170 (ix86_adjust_stack_and_probe): Likewise.
171 (ix86_emit_probe_stack_range): Pass new arguments to
172 release_scratch_register_on_entry.
174 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
176 PR rtl-optimization/84157
177 * combine.c (change_zero_ext): Use REG_P predicate in
178 front of HARD_REGISTER_P predicate.
180 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
182 * config/avr/avr.c (avr_option_override): Move disabling of
183 -fdelete-null-pointer-checks to...
184 * common/config/avr/avr-common.c (avr_option_optimization_table):
187 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
189 PR tree-optimization/81635
190 * tree-data-ref.c (split_constant_offset_1): For types that
191 wrap on overflow, try to use range info to prove that wrapping
194 2018-02-01 Renlin Li <renlin.li@arm.com>
197 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
199 (aarch64_register_move_cost): Likewise.
200 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
202 (REG_CLASS_NAMES): Likewise.
203 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
204 TAILCALL_ADDR_REGS. Remove IP registers.
205 * config/aarch64/aarch64.md (Ucs): Update register constraint.
207 2018-02-01 Richard Biener <rguenther@suse.de>
209 * domwalk.h (dom_walker::dom_walker): Add additional constructor
210 for specifying RPO order and allow NULL for that.
211 * domwalk.c (dom_walker::dom_walker): Likewise.
212 (dom_walker::walk): Handle NULL RPO order.
213 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
215 (rewrite_update_dom_walker): Likewise.
216 (mark_def_dom_walker): Likewise.
218 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
220 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
221 (aarch64_maybe_expand_sve_subreg_move): Declare.
222 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
223 * config/aarch64/predicates.md (aarch64_any_register_operand): New
225 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
226 that are semantically a reverse operation.
227 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
228 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
229 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
231 (aarch64_can_change_mode_class): For big-endian, forbid changes
232 between two SVE modes if they have different element sizes.
234 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
236 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
237 the TImode handling for big-endian targets.
239 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
241 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
242 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
244 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
245 Remove BSWAP handing for big-endian targets and use the form of
246 LD1RQ appropariate for the mode.
248 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
250 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
251 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
254 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
257 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
258 check for operands that need to go through aarch64_sve_reload_be.
260 2018-02-01 Jakub Jelinek <jakub@redhat.com>
262 PR tree-optimization/81661
263 PR tree-optimization/84117
264 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
265 * tree-eh.c: Include gimplify.h.
266 (find_trapping_overflow, replace_trapping_overflow,
267 rewrite_to_non_trapping_overflow): New functions.
268 * tree-vect-loop.c: Include tree-eh.h.
269 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
270 * tree-data-ref.c: Include tree-eh.h.
271 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
273 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
275 PR rtl-optimization/84123
276 * combine.c (change_zero_ext): Check if hard register satisfies
277 can_change_dest_mode before calling gen_lowpart_SUBREG.
279 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
282 * ira.c (ira_init_register_move_cost): Remove assert.
284 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
286 PR rtl-optimization/84071
287 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
288 * doc/tm.texi: Regenerate.
290 2018-01-31 Richard Biener <rguenther@suse.de>
292 PR tree-optimization/84132
293 * tree-data-ref.c (analyze_miv_subscript): Properly
294 check whether evolution_function_is_affine_multivariate_p
295 before calling gcd_of_steps_may_divide_p.
297 2018-01-31 Julia Koval <julia.koval@intel.com>
300 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
301 * config/i386/i386.md (rdpid_rex64) New.
302 (rdpid): Make 32bit only.
304 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
307 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
308 an IDENTIFIER_NODE for FUNCTION_TYPE's.
310 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
313 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
315 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
317 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
319 PR rtl-optimization/84071
320 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
321 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
323 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
325 * config/arc/arc.c (arc_handle_aux_attribute): New function.
326 (arc_attribute_table): Add 'aux' attribute.
327 (arc_in_small_data_p): Consider aux like variables.
328 (arc_is_aux_reg_p): New function.
329 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
330 (arc_get_aux_arg): New function.
331 (prepare_move_operands): Handle aux-register access.
332 (arc_handle_aux_attribute): New function.
333 * doc/extend.texi (ARC Variable attributes): Add subsection.
335 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
337 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
338 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
339 (arc_attribute_table): Add 'uncached' attribute.
340 (arc_print_operand): Print '.di' flag for uncached memory
342 (arc_in_small_data_p): Do not consider for small data the uncached
344 (arc_is_uncached_mem_p): New function.
345 * config/arc/predicates.md (compact_store_memory_operand): Check
346 for uncached memory accesses.
347 (nonvol_nonimm_operand): Likewise.
348 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
350 2018-01-31 Jakub Jelinek <jakub@redhat.com>
353 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
354 falign-loops=): Add Optimization flag.
356 2018-01-30 Jeff Law <law@redhat.com>
359 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
360 INT_REGISTERS_SAVED. Check it prior to calling
361 get_scratch_register_on_entry.
362 (ix86_adjust_stack_and_probe): Similarly.
363 (ix86_emit_probe_stack_range): Similarly.
364 (ix86_expand_prologue): Corresponding changes.
366 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
369 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
370 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
372 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
375 * lra-constraints.c (curr_insn_transform): Process AND in the
378 2018-01-30 Jakub Jelinek <jakub@redhat.com>
380 PR rtl-optimization/83986
381 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
382 dependence against last_pending_memory_flush in addition to
385 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
387 PR tree-optimization/81611
388 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
391 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
393 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
396 2018-01-30 Richard Biener <rguenther@suse.de>
397 Jakub Jelinek <jakub@redhat.com>
399 PR tree-optimization/84111
400 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
401 inner loops added during recursion, as they don't have up-to-date
404 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
407 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
408 (can_inline_edge_by_limits_p): ... here.
409 (can_early_inline_edge_p, check_callers,
410 update_caller_keys, update_callee_keys, recursive_inlining,
411 add_new_edges_to_heap, speculation_useful_p,
412 inline_small_functions,
413 inline_small_functions, flatten_function,
414 inline_to_all_callers_1): Update.
416 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
418 * profile-count.c (profile_count::combine_with_ipa_count): Handle
421 2018-01-30 Richard Biener <rguenther@suse.de>
423 PR tree-optimization/83008
424 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
425 invariant and constant vector uses in stmts when they need
428 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
431 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
432 * configure: Regenerate.
434 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
436 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
438 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
439 Use gen_rtx_REG rather than gen_lowpart.
441 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
443 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
444 rather than 0 when creating partial subregs.
446 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
448 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
451 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
454 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
455 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
456 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
457 flags. This restores the settings used before the 2017-07-24.
458 Turning off pre increment/decrement/modify allows IVOPTS to
459 optimize DF/SF loops where the index is an int.
461 2018-01-29 Richard Biener <rguenther@suse.de>
462 Kelvin Nilsen <kelvin@gcc.gnu.org>
465 * tree-vect-stmts.c (vectorizable_call): Don't call
466 targetm.vectorize_builtin_md_vectorized_function if callee is
469 2018-01-22 Carl Love <cel@us.ibm.com>
471 * doc/extend.tex: Fix typo in second arg in
472 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
474 2018-01-29 Richard Biener <rguenther@suse.de>
476 PR tree-optimization/84086
477 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
478 (flush_ssaname_freelist): When SSA names were released reset
481 2018-01-29 Richard Biener <rguenther@suse.de>
483 PR tree-optimization/84057
484 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
485 removed paths when removing edges.
487 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
489 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
490 -mfunction-return=@var{choice}.
492 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
495 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
497 (layout::print_source_line): Likewise.
498 (test_get_line_width_without_trailing_whitespace): Add test cases.
500 2018-01-27 Jakub Jelinek <jakub@redhat.com>
503 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
506 2018-01-26 Jim Wilson <jimw@sifive.com>
508 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
510 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
513 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
515 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
516 and CMP + SUB-immediate -> SUBS.
518 2018-01-26 Martin Sebor <msebor@redhat.com>
520 PR tree-optimization/83896
521 * tree-ssa-strlen.c (get_string_len): Rename...
522 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
523 Avoid assuming length is constant.
524 (handle_char_store): Use HOST_WIDE_INT for string length.
526 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
529 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
530 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
532 2018-01-26 Richard Biener <rguenther@suse.de>
534 PR rtl-optimization/84003
535 * dse.c (record_store): Only record redundant stores when
536 the earlier store aliases at least all accesses the later one does.
538 2018-01-26 Jakub Jelinek <jakub@redhat.com>
540 PR rtl-optimization/83985
541 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
542 REG_CFA_RESTORE insns.
543 (delete_unmarked_insns): Don't ignore separate shrink wrapping
544 REG_CFA_RESTORE insns here.
547 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
548 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
550 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
552 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
553 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
554 (arc_init): Likewise.
555 (arc_override_options): Likewise.
556 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
558 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
560 * config/arc/arc.h (TARGET_DBNZ): Define.
561 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
562 properly set the tune attribute.
563 (dbnz): Use TARGET_DBNZ guard.
564 * config/arc/arc.opt (mtune): Add core3 option.
566 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
568 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
569 recognize new pic like addresses.
570 (arc_delegitimize_address): Clean up.
572 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
574 * config/arc/arc-arches.def: Option mrf16 valid for all
576 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
577 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
578 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
579 * config/arc/arc-tables.opt: Regenerate.
580 * config/arc/arc.c (arc_conditional_register_usage): Handle
581 reduced register file case.
582 (arc_file_start): Set must have build attributes.
583 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
585 * config/arc/arc.opt (mrf16): Add new option.
586 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
587 * config/arc/genmultilib.awk: Handle new mrf16 option.
588 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
589 * config/arc/t-multilib: Regenerate.
590 * doc/invoke.texi (ARC Options): Document mrf16 option.
592 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
594 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
595 * config/arc/arc.c (arc_handle_secure_attribute): New function.
596 (arc_attribute_table): Add 'secure_call' attribute.
597 (arc_print_operand): Print secure call operand.
598 (arc_function_ok_for_sibcall): Don't optimize tail calls when
600 (arc_is_secure_call_p): New function. * config/arc/arc.md
601 (call_i): Add support for sjli instruction.
602 (call_value_i): Likewise.
603 * config/arc/constraints.md (Csc): New constraint.
605 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
606 John Eric Martin <John.Martin@emmicro-us.com>
608 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
609 * config/arc/arc.c (_arc_jli_section): New struct.
610 (arc_jli_section): New type.
611 (rc_jli_sections): New static variable.
612 (arc_handle_jli_attribute): New function.
613 (arc_attribute_table): Add jli_always and jli_fixed attribute.
614 (arc_file_end): New function.
615 (TARGET_ASM_FILE_END): Define.
616 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
617 (arc_add_jli_section): New function.
618 (jli_call_scan): Likewise.
619 (arc_reorg): Call jli_call_scan.
620 (arc_output_addsi): Remove 'S' from printing asm operand.
621 (arc_is_jli_call_p): New function.
622 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
624 (movhi_insn): Likewise.
625 (movsi_insn): Likewise.
626 (movsi_set_cc_insn): Likewise.
627 (loadqi_update): Likewise.
628 (load_zeroextendqisi_update): Likewise.
629 (load_signextendqisi_update): Likewise.
630 (loadhi_update): Likewise.
631 (load_zeroextendhisi_update): Likewise.
632 (load_signextendhisi_update): Likewise.
633 (loadsi_update): Likewise.
634 (loadsf_update): Likewise.
635 (movsicc_insn): Likewise.
636 (bset_insn): Likewise.
637 (bxor_insn): Likewise.
638 (bclr_insn): Likewise.
639 (bmsk_insn): Likewise.
640 (bicsi3_insn): Likewise.
641 (cmpsi_cc_c_insn): Likewise.
642 (movsi_ne): Likewise.
643 (movsi_cond_exec): Likewise.
644 (clrsbsi2): Likewise.
655 (call_i): Remove 'S' asm letter, add jli instruction.
656 (call_value_i): Likewise.
657 * config/arc/arc.op (mjli-always): New option.
658 * config/arc/constraints.md (Cji): New constraint.
659 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
661 (subsf3_fpx): Likewise.
662 (mulsf3_fpx): Likewise.
663 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
665 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
667 * doc/invoke.texi (ARC): Document mjli-always option.
669 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
671 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
672 and use incw and decw where possible
673 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
675 2018-01-26 Richard Biener <rguenther@suse.de>
677 PR tree-optimization/81082
678 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
679 association if it requires casting to unsigned.
680 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
681 from fold_plusminus_mult_expr to catch important cases late when
682 range info is available.
684 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
686 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
687 * configure.ac (hidden_linkonce): New test.
688 * configure: Regenerate.
689 * config.in: Regenerate.
691 2018-01-26 Julia Koval <julia.koval@intel.com>
693 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
694 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
695 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
696 _mm_mask_bitshuffle_epi64_mask): Fix type.
697 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
698 USI_FTYPE_V4DI_V4DI_USI): Remove.
699 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
700 __builtin_ia32_vpshufbitqmb256_mask,
701 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
702 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
703 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
705 2018-01-26 Alan Modra <amodra@gmail.com>
708 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
709 UNSPEC_VBPERMQ. Sort other unspecs.
711 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
713 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
715 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
718 * predict.c (drop_profile): Do not push/pop cfun; update also
720 (handle_missing_profiles): Fix logic looking for zero profiles.
722 2018-01-25 Jakub Jelinek <jakub@redhat.com>
725 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
726 on functions with #pragma omp declare simd or functions with simd
728 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
729 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
730 Remove trailing \n from warning_at calls.
732 2018-01-25 Tom de Vries <tom@codesourcery.com>
735 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
736 for neutered workers.
738 2018-01-24 Joseph Myers <joseph@codesourcery.com>
741 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
742 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
744 2018-01-24 Jeff Law <law@redhat.com>
747 * i386.c (get_probe_interval): Move to earlier point.
748 (ix86_compute_frame_layout): If -fstack-clash-protection and
749 the frame is larger than the probe interval, then use pushes
750 to save registers rather than reg->mem moves.
751 (ix86_expand_prologue): Remove conditional for int_registers_saved
754 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
757 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
758 min/max for never referenced object.
760 2018-01-24 Jakub Jelinek <jakub@redhat.com>
763 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
765 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
766 attributes from DECL_ATTRIBUTES (decl) without affecting
767 DECL_ATTRIBUTES (current_function_decl).
768 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
769 functions with non-NULL DECL_ABSTRACT_ORIGIN.
771 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
773 PR tree-optimization/83979
774 * fold-const.c (fold_comparison): Use constant_boolean_node
775 instead of boolean_{true,false}_node.
777 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
779 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
782 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
784 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
785 Simplify the clause that sets the length attribute.
786 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
787 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
788 clause that sets the length attribute.
789 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
791 2018-01-24 Tom de Vries <tom@codesourcery.com>
794 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
795 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
796 Add strict parameter.
797 (prevent_branch_around_nothing): Insert dummy insn between branch to
798 label and label with no ptx insn inbetween.
799 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
801 2018-01-24 Tom de Vries <tom@codesourcery.com>
804 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
805 for neutered threads in warp.
806 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
808 2018-01-24 Richard Biener <rguenther@suse.de>
810 PR tree-optimization/83176
811 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
814 2018-01-24 Richard Biener <rguenther@suse.de>
816 PR tree-optimization/82819
817 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
818 code generating pluses that are no-ops in the target precision.
820 2018-01-24 Richard Biener <rguenther@suse.de>
823 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
825 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
827 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
828 to merge probabilities.
829 * predict.c (probably_never_executed): Also mark as cold functions
830 with global 0 profile and guessed local profile.
831 * profile-count.c (profile_probability::combine_with_count): New
833 * profile-count.h (profile_probability::operator*,
834 profile_probability::operator*=, profile_probability::operator/,
835 profile_probability::operator/=): Reduce precision to adjusted
836 and set value to guessed on contradictory divisions.
837 (profile_probability::combine_with_freq): Remove.
838 (profile_probability::combine_wiht_count): Declare.
839 (profile_count::force_nonzero):: Set to adjusted.
840 (profile_count::probability_in):: Set quality to adjusted.
841 * tree-ssa-tail-merge.c (replace_block_by): Use
844 2018-01-23 Andrew Waterman <andrew@sifive.com>
845 Jim Wilson <jimw@sifive.com>
847 * config/riscv/riscv.c (riscv_stack_boundary): New.
848 (riscv_option_override): Set riscv_stack_boundary. Handle
849 riscv_preferred_stack_boundary_arg.
850 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
851 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
852 (STACK_BOUNDARY): Set to riscv_stack_boundary.
853 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
854 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
855 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
857 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
860 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
861 of struct ix86_frame.
862 (ix86_expand_epilogue): Likewise. Add a local variable for
863 the reg_save_offset field in struct ix86_frame.
865 2018-01-23 Bin Cheng <bin.cheng@arm.com>
867 PR tree-optimization/82604
868 * tree-loop-distribution.c (enum partition_kind): New enum item
869 PKIND_PARTIAL_MEMSET.
870 (partition_builtin_p): Support above new enum item.
871 (generate_code_for_partition): Ditto.
872 (compute_access_range): Differentiate cases that equality can be
873 proven at all loops, the innermost loops or no loops.
874 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
875 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
876 (finalize_partitions, distribute_loop): Don't fuse partition of
877 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
878 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
881 2018-01-23 Martin Liska <mliska@suse.cz>
883 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
884 order to ignore the predictor.
885 (PRED_POLYMORPHIC_CALL): Likewise.
886 (PRED_RECURSIVE_CALL): Likewise.
888 2018-01-23 Martin Liska <mliska@suse.cz>
890 * tree-profile.c (tree_profiling): Print function header to
891 aware reader which function we are working on.
892 * value-prof.c (gimple_find_values_to_profile): Do not print
893 not interesting value histograms.
895 2018-01-23 Martin Liska <mliska@suse.cz>
897 * profile-count.h (enum profile_quality): Add
898 profile_uninitialized as the first value. Do not number values
899 as they are zero based.
900 (profile_count::verify): Update sanity check.
901 (profile_probability::verify): Likewise.
903 2018-01-23 Nathan Sidwell <nathan@acm.org>
905 * doc/invoke.texi (ffor-scope): Deprecate.
907 2018-01-23 David Malcolm <dmalcolm@redhat.com>
909 PR tree-optimization/83510
910 * domwalk.c (set_all_edges_as_executable): New function.
911 (dom_walker::dom_walker): Convert bool param
912 "skip_unreachable_blocks" to enum reachability. Move setup of
913 edge flags to set_all_edges_as_executable and only do it when
914 reachability is REACHABLE_BLOCKS.
915 * domwalk.h (enum dom_walker::reachability): New enum.
916 (dom_walker::dom_walker): Convert bool param
917 "skip_unreachable_blocks" to enum reachability.
918 (set_all_edges_as_executable): New decl.
919 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
920 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
922 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
923 but converting true to REACHABLE_BLOCKS.
924 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
926 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
927 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
928 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
930 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
931 if check_all_array_refs will be called.
933 2018-01-23 David Malcolm <dmalcolm@redhat.com>
935 * tree.c (selftest::test_location_wrappers): Add more test
938 2018-01-23 David Malcolm <dmalcolm@redhat.com>
940 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
941 (selftest::test_bit_in_range): Likewise.
943 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
946 * doc/sourcebuild.texi (vect_float): Say that the selector
947 only describes the situation when -funsafe-math-optimizations is on.
948 (vect_float_strict): Document.
950 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
952 PR tree-optimization/83965
953 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
954 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
955 instead of checking only for a reduction.
956 (vect_recog_widen_sum_pattern): Likewise.
958 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
960 * predict.c (probably_never_executed): Only use precise profile info.
961 (compute_function_frequency): Skip after inlining hack since we now
962 have quality checking.
964 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
966 * profile-count.h (profile_probability::very_unlikely,
967 profile_probability::unlikely, profile_probability::even): Set
968 precision to guessed.
970 2018-01-23 Richard Biener <rguenther@suse.de>
972 PR tree-optimization/83963
973 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
974 Properly terminate dominator walk when crossing the exit edge not
975 when visiting its source block.
977 2018-01-23 Jakub Jelinek <jakub@redhat.com>
980 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
981 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
983 2018-01-22 Jakub Jelinek <jakub@redhat.com>
985 PR tree-optimization/83957
986 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
987 semicolon after for body surrounded by braces.
989 PR tree-optimization/83081
990 * profile-count.h (profile_probability::split): New method.
991 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
992 Use profile_probability::split.
993 (do_compare_rtx_and_jump): Fix adjustment of probabilities
994 when splitting a single conditional jump into 2.
996 2018-01-22 David Malcolm <dmalcolm@redhat.com>
998 PR tree-optimization/69452
999 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1002 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1004 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1005 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1006 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1008 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1010 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1011 * config/rl78/rl78.md: New define_expand "movdi"
1012 * config/rl78/rl78.c: New function definition rl78_split_movdi
1014 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1017 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1019 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1020 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1021 128-bit to produce an UNSPEC move to get the double word with the
1022 signbit and then a shift directly to do signbit.
1023 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1024 implementation with a new version that just does either a direct
1025 move or a regular move. Move memory interface to separate insns.
1026 Move insns so they are next to the expander.
1027 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1028 with signbit move. Split big and little endian case.
1029 (signbit<mode>2_dm_mem_le): Likewise.
1030 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1031 (signbit<mode>2_dm2): Likewise.
1033 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1035 * config/rl78/rl78.md: New define_expand "anddi3".
1037 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1039 * config/rl78/rl78.md: New define_expand "umindi3".
1041 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1043 * config/rl78/rl78.md: New define_expand "smindi3".
1045 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1047 * config/rl78/rl78.md: New define_expand "smaxdi3".
1049 2018-01-22 Carl Love <cel@us.ibm.com>
1051 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1052 LVX_V1TI): Add macro expansion.
1053 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1054 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1055 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1056 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1057 Change check to determine if the instruction is a byte reversing
1058 entry. Fix typo in comment.
1059 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1060 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1061 Add def_builtin calls for new builtins.
1062 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1063 Add define_insn expansion.
1065 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1067 * config/rl78/rl78.md: New define_expand "umaxdi3".
1069 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1071 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1072 for non-QImode registers
1074 2018-01-22 Richard Biener <rguenther@suse.de>
1076 PR tree-optimization/83963
1077 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1078 including the loop exit block.
1079 (scop_detection::merge_sese): Likewise.
1080 (scop_detection::add_scop): Do it here instead.
1082 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1084 * doc/sourcebuild.texi (arm_softfloat): Document.
1086 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1089 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1090 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1091 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1093 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1094 David Edelsohn <dje.gcc@gmail.com>
1097 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1098 Change "crset eq" to "crset 2".
1099 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1100 (*call_indirect_aix<mode>_nospec): Likewise.
1101 (*call_value_indirect_aix<mode>_nospec): Likewise.
1102 (*call_indirect_elfv2<mode>_nospec): Likewise.
1103 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1104 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1105 change assembly output from . to $.
1106 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1107 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1108 (*tablejump<mode>_internal1_nospec): Likewise.
1110 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1113 * config/sh/sh_optimize_sett_clrt.cc:
1114 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1116 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1118 PR tree-optimization/83940
1119 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1120 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1121 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1122 use it to pass back the definition type.
1123 (vect_check_store_rhs): Likewise rhs_dt_out.
1124 (vect_build_gather_load_calls): Add a mask_dt argument and use
1125 it instead of a call to vect_is_simple_use.
1126 (vectorizable_store): Update calls to vect_check_load_store_mask
1127 and vect_check_store_rhs. Use the dt returned by the latter instead
1128 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1129 instead of calls to vect_is_simple_use. Pass the scalar rather
1130 than the vector operand to vect_is_simple_use when handling
1131 second and subsequent copies of an rhs value.
1132 (vectorizable_load): Update calls to vect_check_load_store_mask
1133 and vect_build_gather_load_calls. Use the cached mask_dt and
1134 gs_info.offset_dt instead of calls to vect_is_simple_use.
1136 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1139 * tree-emutls.c: Include gimplify.h.
1140 (lower_emutls_2): New function.
1141 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1142 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1143 it before further processing.
1146 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1147 UINTVAL (trueop1) instead of INTVAL (op1).
1149 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1153 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1154 INCOMING_FRAME_SP_OFFSET if not defined.
1155 (scan_trace): Add ENTRY argument. If true and
1156 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1157 emit a note to adjust the CFA offset.
1158 (create_cfi_notes): Adjust scan_trace callers.
1159 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1160 INCOMING_FRAME_SP_OFFSET in the CIE.
1161 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1162 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1164 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1165 * doc/tm.texi: Regenerated.
1167 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1169 PR rtl-optimization/83147
1170 * lra-constraints.c (remove_inheritance_pseudos): Use
1171 lra_substitute_pseudo_within_insn.
1173 2018-01-19 Tom de Vries <tom@codesourcery.com>
1174 Cesar Philippidis <cesar@codesourcery.com>
1177 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1179 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1182 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1183 spaces for function labels.
1185 2018-01-19 Martin Liska <mliska@suse.cz>
1187 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1188 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1189 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1190 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1191 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1192 (PRED_CONST_RETURN): Change from 69 to 65.
1193 (PRED_NULL_RETURN): Change from 91 to 71.
1194 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1195 (PRED_LOOP_GUARD): Change from 66 to 73.
1197 2018-01-19 Martin Liska <mliska@suse.cz>
1199 * predict.c (predict_insn_def): Add new assert.
1200 (struct branch_predictor): Change type to signed integer.
1201 (test_prediction_value_range): Amend test to cover
1203 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1204 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1205 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1206 (PRED_LOOP_IV_COMPARE): Likewise.
1207 * predict.h (PROB_UNINITIALIZED): Define new constant.
1209 2018-01-19 Martin Liska <mliska@suse.cz>
1211 * predict.c (dump_prediction): Add new format for
1212 analyze_brprob.py script which is enabled with -details
1214 * profile-count.h (precise_p): New function.
1216 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1218 PR tree-optimization/83922
1219 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1220 there are no statements that need masking.
1221 (vect_active_double_reduction_p): New function.
1222 (vect_analyze_loop_operations): Use it when handling phis that
1223 are not in the loop header.
1225 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1227 PR tree-optimization/83914
1228 * tree-vect-loop.c (vectorizable_induction): Don't convert
1229 init_expr or apply the peeling adjustment for inductions
1230 that are nested within the vectorized loop.
1232 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1234 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1237 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1241 * function.h (gimplify_parameters): Add gimple_seq * argument.
1242 * function.c: Include gimple.h and options.h.
1243 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1244 for the added local temporaries if needed.
1245 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1246 if there are any parameter cleanups, wrap whole body into a
1247 try/finally with the cleanups.
1249 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1252 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1253 Use GET_MODE_CLASS for scalar floating point.
1255 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1259 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1260 Fix call of call_cgraph_insertion_hooks.
1262 2018-01-18 Martin Sebor <msebor@redhat.com>
1264 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1266 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1269 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1272 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1275 * common.opt: (-ffile-prefix-map): New option.
1276 * opts.c (common_handle_option): Defer it.
1277 * opts-global.c (handle_common_deferred_options): Handle it.
1278 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1279 * file-prefix-map.h: New file.
1280 (remap_debug_filename, add_debug_prefix_map): ...here.
1281 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1282 * final.c (debug_prefix_map, add_debug_prefix_map
1283 remap_debug_filename): Move to...
1284 * file-prefix-map.c: New file.
1285 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1286 generalize, get rid of alloca(), use strrchr() instead of strchr().
1287 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1288 Implement in terms of add_prefix_map().
1289 (remap_macro_filename, remap_debug_filename): Implement in term of
1291 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1292 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1293 * dbxout.c: Include file-prefix-map.h.
1294 * varasm.c: Likewise.
1295 * vmsdbgout.c: Likewise.
1296 * xcoffout.c: Likewise.
1297 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1298 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1299 * doc/invoke.texi (-ffile-prefix-map): Document.
1300 (-fdebug-prefix-map): Update description.
1302 2018-01-18 Martin Liska <mliska@suse.cz>
1304 * config/i386/i386.c (indirect_thunk_name): Document that also
1306 (output_indirect_thunk): Document why both instructions
1307 (pause and lfence) are generated.
1309 2018-01-18 Richard Biener <rguenther@suse.de>
1311 PR tree-optimization/83887
1312 * graphite-scop-detection.c
1313 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1314 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1315 (scop_detection::merge_sese): Re-implement with a flood-fill
1316 algorithm that properly finds a SESE region if it exists.
1318 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1321 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1322 pointer_diff optimizations use view_convert instead of convert.
1324 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1326 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1327 Generate different code for -mno-speculate-indirect-jumps.
1328 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1329 (*call_indirect_aix<mode>): Disable for
1330 -mno-speculate-indirect-jumps.
1331 (*call_indirect_aix<mode>_nospec): New define_insn.
1332 (*call_value_indirect_aix<mode>): Disable for
1333 -mno-speculate-indirect-jumps.
1334 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1335 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1336 -mno-speculate-indirect-jumps.
1337 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1339 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1341 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1342 long double type, set the flags for noting the default long double
1343 type, even if we don't pass or return a long double type.
1345 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1348 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1351 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1353 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1354 support for merge[hl].
1355 (fold_mergehl_helper): New helper function.
1356 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1357 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1358 (altivec_vmrglw_direct): Add xxmrglw insn.
1360 2018-01-17 Andrew Waterman <andrew@sifive.com>
1362 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1363 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1365 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1368 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1369 call the lto_location_cache before reading the
1370 DECL_SOURCE_LOCATION of the types.
1372 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1373 Richard Sandiford <richard.sandiford@linaro.org>
1375 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1376 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1377 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1378 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1379 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1381 * config/aarch64/constraints.md (aarch64_movti_operand):
1383 * config/aarch64/predicates.md (Uti): Add new constraint.
1385 2018-01-17 Carl Love <cel@us.ibm.com>
1386 * config/rs6000/vsx.md (define_expand xl_len_r,
1387 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1388 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1390 (define_expand, define_insn): Move the shift left from the
1391 define_insn to the define_expand for lxvl and stxvl instructions.
1392 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1393 and XL_LEN_R definitions to PURE.
1395 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1397 * config/i386/i386.c (indirect_thunk_name): Declare regno
1398 as unsigned int. Compare regno with INVALID_REGNUM.
1399 (output_indirect_thunk): Ditto.
1400 (output_indirect_thunk_function): Ditto.
1401 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1402 in the call to output_indirect_thunk_function.
1404 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1407 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1408 rather than the size of inner_type to determine the stack slot size
1409 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1411 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1414 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1417 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1419 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1420 endian Linux systems to optionally enable multilibs for selecting
1421 the long double type if the user configured an explicit type.
1422 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1423 have no long double multilibs if not defined.
1424 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1425 warn if the user used -mabi={ieee,ibm}longdouble and we built
1426 multilibs for long double.
1427 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1428 appropriate multilib option.
1429 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1431 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1432 for building long double multilibs.
1433 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1435 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1437 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1440 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1442 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1445 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1448 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1451 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1453 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1454 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1456 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1458 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1459 different rtl trees depending on TARGET_64BIT.
1460 (rs6000_gen_lvx): Likewise.
1462 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1464 * config/visium/visium.md (nop): Tweak comment.
1465 (hazard_nop): Likewise.
1467 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1469 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1470 -mspeculate-indirect-jumps.
1471 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1472 for -mno-speculate-indirect-jumps.
1473 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1474 (*call_value_indirect_elfv2<mode>): Disable for
1475 -mno-speculate-indirect-jumps.
1476 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1477 (indirect_jump): Emit different RTL for
1478 -mno-speculate-indirect-jumps.
1479 (*indirect_jump<mode>): Disable for
1480 -mno-speculate-indirect-jumps.
1481 (*indirect_jump<mode>_nospec): New define_insn.
1482 (tablejump): Emit different RTL for
1483 -mno-speculate-indirect-jumps.
1484 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1485 (tablejumpsi_nospec): New define_expand.
1486 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1487 (tablejumpdi_nospec): New define_expand.
1488 (*tablejump<mode>_internal1): Disable for
1489 -mno-speculate-indirect-jumps.
1490 (*tablejump<mode>_internal1_nospec): New define_insn.
1491 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1494 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1496 * caller-save.c (insert_save): Drop unnecessary parameter. All
1499 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1500 Richard Biener <rguenth@suse.de>
1503 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1504 return early, inline manually is_gimple_sizepos. Make sure if we
1505 call gimplify_expr we don't end up with a gimple constant.
1506 * tree.c (variably_modified_type_p): Don't return true for
1507 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1508 * gimplify.h (is_gimple_sizepos): Remove.
1510 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1512 PR tree-optimization/83857
1513 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1514 vectorizable_live_operation for pure SLP statements.
1515 (vectorizable_live_operation): Handle PHIs.
1517 2018-01-16 Richard Biener <rguenther@suse.de>
1519 PR tree-optimization/83867
1520 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1521 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1523 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1526 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1527 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1528 If off is not INTEGER_CST, issue a may not be aligned warning
1529 rather than isn't aligned. Use isn%'t rather than isn't.
1530 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1532 <case MULT_EXPR>: Improve the case when bottom and one of the
1533 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1534 operand, in that case check if the other operand is multiple of
1535 bottom divided by the INTEGER_CST operand.
1537 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1540 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1541 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1542 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1543 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1544 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1545 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1546 (pa_function_arg_size): New function.
1548 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1550 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1551 in a separate statement.
1553 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1555 PR tree-optimization/83847
1556 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1557 group gathers and scatters.
1559 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1561 PR rtl-optimization/86620
1562 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1564 PR rtl-optimization/83213
1565 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1566 to last if both are JUMP_INSNs.
1568 PR tree-optimization/83843
1569 * gimple-ssa-store-merging.c
1570 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1571 store_immediate_info for bswap/nop orig_stores.
1573 2018-01-15 Andrew Waterman <andrew@sifive.com>
1575 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1577 <UDIV>: Increase cost if !TARGET_DIV.
1579 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1581 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1582 (define_attr "cr_logical_3op"): New.
1583 (cceq_ior_compare): Adjust.
1584 (cceq_ior_compare_complement): Adjust.
1585 (*cceq_rev_compare): Adjust.
1586 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1587 (is_cracked_insn): Adjust.
1588 (insn_must_be_first_in_group): Adjust.
1589 * config/rs6000/40x.md: Adjust.
1590 * config/rs6000/440.md: Adjust.
1591 * config/rs6000/476.md: Adjust.
1592 * config/rs6000/601.md: Adjust.
1593 * config/rs6000/603.md: Adjust.
1594 * config/rs6000/6xx.md: Adjust.
1595 * config/rs6000/7450.md: Adjust.
1596 * config/rs6000/7xx.md: Adjust.
1597 * config/rs6000/8540.md: Adjust.
1598 * config/rs6000/cell.md: Adjust.
1599 * config/rs6000/e300c2c3.md: Adjust.
1600 * config/rs6000/e500mc.md: Adjust.
1601 * config/rs6000/e500mc64.md: Adjust.
1602 * config/rs6000/e5500.md: Adjust.
1603 * config/rs6000/e6500.md: Adjust.
1604 * config/rs6000/mpc.md: Adjust.
1605 * config/rs6000/power4.md: Adjust.
1606 * config/rs6000/power5.md: Adjust.
1607 * config/rs6000/power6.md: Adjust.
1608 * config/rs6000/power7.md: Adjust.
1609 * config/rs6000/power8.md: Adjust.
1610 * config/rs6000/power9.md: Adjust.
1611 * config/rs6000/rs64.md: Adjust.
1612 * config/rs6000/titan.md: Adjust.
1614 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1616 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1617 ix86_indirect_branch_register logic.
1619 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1621 * config/i386/constraints.md (Bs): Update
1622 ix86_indirect_branch_register check. Don't check
1623 ix86_indirect_branch_register with GOT_memory_operand.
1625 * config/i386/predicates.md (GOT_memory_operand): Don't check
1626 ix86_indirect_branch_register here.
1627 (GOT32_symbol_operand): Likewise.
1629 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1631 * config/i386/predicates.md (constant_call_address_operand):
1632 Rewrite ix86_indirect_branch_register logic.
1633 (sibcall_insn_operand): Likewise.
1635 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1637 * config/i386/constraints.md (Bs): Replace
1638 ix86_indirect_branch_thunk_register with
1639 ix86_indirect_branch_register.
1641 * config/i386/i386.md (indirect_jump): Likewise.
1642 (tablejump): Likewise.
1643 (*sibcall_memory): Likewise.
1644 (*sibcall_value_memory): Likewise.
1645 Peepholes of indirect call and jump via memory: Likewise.
1646 * config/i386/i386.opt: Likewise.
1647 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1648 (GOT_memory_operand): Likewise.
1649 (call_insn_operand): Likewise.
1650 (sibcall_insn_operand): Likewise.
1651 (GOT32_symbol_operand): Likewise.
1653 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1656 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1657 type rather than type addr's type points to.
1658 (expand_omp_atomic_mutex): Likewise.
1659 (expand_omp_atomic): Likewise.
1661 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1664 * config/i386/i386.c (output_indirect_thunk_function): Use
1665 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1666 for __x86_return_thunk.
1668 2018-01-15 Richard Biener <rguenther@suse.de>
1671 * expmed.c (extract_bit_field_1): Fix typo.
1673 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1676 * config/arm/iterators.md (VF): New mode iterator.
1677 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1678 Remove integer-related logic from pattern.
1679 (neon_vabd<mode>_3): Likewise.
1681 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1684 * common.opt (fstrict-overflow): No longer an alias.
1685 (fwrapv-pointer): New option.
1686 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1687 also for pointer types based on flag_wrapv_pointer.
1688 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1689 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1690 opts->x_flag_wrapv got set.
1691 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1692 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1693 POINTER_TYPE_OVERFLOW_UNDEFINED.
1694 * match.pd: Likewise in address comparison pattern.
1695 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1697 2018-01-15 Richard Biener <rguenther@suse.de>
1700 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1701 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1702 Reset type names to their identifier if their TYPE_DECL doesn't
1703 have linkage (and thus is used for ODR and devirt).
1704 (save_debug_info_for_decl): Remove.
1705 (save_debug_info_for_type): Likewise.
1706 (add_tree_to_fld_list): Adjust.
1707 * tree-pretty-print.c (dump_generic_node): Make dumping of
1708 type names more robust.
1710 2018-01-15 Richard Biener <rguenther@suse.de>
1712 * BASE-VER: Bump to 8.0.1.
1714 2018-01-14 Martin Sebor <msebor@redhat.com>
1717 * builtins.c (check_access): Avoid warning when the no-warning bit
1720 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1722 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1723 * ira-color (allocno_hard_regs_compare): Likewise.
1725 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1728 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1729 Use .pushsection/.popsection.
1731 2018-01-14 Martin Sebor <msebor@redhat.com>
1734 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1736 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1738 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1739 entry from extra_headers.
1740 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1741 extra_headers, make the list bitwise identical to the i?86-*-* one.
1743 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1745 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1746 -mcmodel=large with -mindirect-branch=thunk,
1747 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1748 -mfunction-return=thunk-extern.
1749 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1750 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1751 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1753 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1755 * config/i386/i386.c (print_reg): Print the name of the full
1756 integer register without '%'.
1757 (ix86_print_operand): Handle 'V'.
1758 * doc/extend.texi: Document 'V' modifier.
1760 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1762 * config/i386/constraints.md (Bs): Disallow memory operand for
1763 -mindirect-branch-register.
1765 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1766 (GOT_memory_operand): Likewise.
1767 (call_insn_operand): Likewise.
1768 (sibcall_insn_operand): Likewise.
1769 (GOT32_symbol_operand): Likewise.
1770 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1771 for -mindirect-branch-register.
1772 (tablejump): Likewise.
1773 (*sibcall_memory): Likewise.
1774 (*sibcall_value_memory): Likewise.
1775 Disallow peepholes of indirect call and jump via memory for
1776 -mindirect-branch-register.
1777 (*call_pop): Replace m with Bw.
1778 (*call_value_pop): Likewise.
1779 (*sibcall_pop_memory): Replace m with Bs.
1780 * config/i386/i386.opt (mindirect-branch-register): New option.
1781 * doc/invoke.texi: Document -mindirect-branch-register option.
1783 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1785 * config/i386/i386-protos.h (ix86_output_function_return): New.
1786 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1787 set function_return_type.
1788 (indirect_thunk_name): Add ret_p to indicate thunk for function
1790 (output_indirect_thunk_function): Pass false to
1791 indirect_thunk_name.
1792 (ix86_output_indirect_branch_via_reg): Likewise.
1793 (ix86_output_indirect_branch_via_push): Likewise.
1794 (output_indirect_thunk_function): Create alias for function
1795 return thunk if regno < 0.
1796 (ix86_output_function_return): New function.
1797 (ix86_handle_fndecl_attribute): Handle function_return.
1798 (ix86_attribute_table): Add function_return.
1799 * config/i386/i386.h (machine_function): Add
1800 function_return_type.
1801 * config/i386/i386.md (simple_return_internal): Use
1802 ix86_output_function_return.
1803 (simple_return_internal_long): Likewise.
1804 * config/i386/i386.opt (mfunction-return=): New option.
1805 (indirect_branch): Mention -mfunction-return=.
1806 * doc/extend.texi: Document function_return function attribute.
1807 * doc/invoke.texi: Document -mfunction-return= option.
1809 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1811 * config/i386/i386-opts.h (indirect_branch): New.
1812 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1813 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1814 with local indirect jump when converting indirect call and jump.
1815 (ix86_set_indirect_branch_type): New.
1816 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1817 (indirectlabelno): New.
1818 (indirect_thunk_needed): Likewise.
1819 (indirect_thunk_bnd_needed): Likewise.
1820 (indirect_thunks_used): Likewise.
1821 (indirect_thunks_bnd_used): Likewise.
1822 (INDIRECT_LABEL): Likewise.
1823 (indirect_thunk_name): Likewise.
1824 (output_indirect_thunk): Likewise.
1825 (output_indirect_thunk_function): Likewise.
1826 (ix86_output_indirect_branch_via_reg): Likewise.
1827 (ix86_output_indirect_branch_via_push): Likewise.
1828 (ix86_output_indirect_branch): Likewise.
1829 (ix86_output_indirect_jmp): Likewise.
1830 (ix86_code_end): Call output_indirect_thunk_function if needed.
1831 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1833 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1834 (ix86_attribute_table): Add indirect_branch.
1835 * config/i386/i386.h (machine_function): Add indirect_branch_type
1836 and has_local_indirect_jump.
1837 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1839 (tablejump): Likewise.
1840 (*indirect_jump): Use ix86_output_indirect_jmp.
1841 (*tablejump_1): Likewise.
1842 (simple_return_indirect_internal): Likewise.
1843 * config/i386/i386.opt (mindirect-branch=): New option.
1844 (indirect_branch): New.
1847 (thunk-inline): Likewise.
1848 (thunk-extern): Likewise.
1849 * doc/extend.texi: Document indirect_branch function attribute.
1850 * doc/invoke.texi: Document -mindirect-branch= option.
1852 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1855 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1857 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1859 * ipa-inline.c (want_inline_small_function_p): Return false if
1860 inlining has already failed with CIF_FINAL_ERROR.
1861 (update_caller_keys): Call want_inline_small_function_p before
1863 (update_callee_keys): Likewise.
1865 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1867 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1869 (rs6000_quadword_masked_address_p): Likewise.
1870 (quad_aligned_load_p): Likewise.
1871 (quad_aligned_store_p): Likewise.
1872 (const_load_sequence_p): Add comment to describe the outer-most loop.
1873 (mimic_memory_attributes_and_flags): New function.
1874 (rs6000_gen_stvx): Likewise.
1875 (replace_swapped_aligned_store): Likewise.
1876 (rs6000_gen_lvx): Likewise.
1877 (replace_swapped_aligned_load): Likewise.
1878 (replace_swapped_load_constant): Capitalize argument name in
1879 comment describing this function.
1880 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1881 and stores that access quad-word aligned addresses and replace
1882 with stvx or lvx instructions when appropriate.
1883 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1884 New function prototype.
1885 (rs6000_quadword_masked_address_p): Likewise.
1886 (rs6000_gen_lvx): Likewise.
1887 (rs6000_gen_stvx): Likewise.
1888 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1889 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1890 when memory address is aligned.
1891 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1892 this split to select lvx instruction when memory address is aligned.
1893 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1894 instruction when memory address is aligned.
1895 (*vsx_le_perm_load_v16qi): Likewise.
1896 (four unnamed splitters): Modify to select the stvx instruction
1897 when memory is aligned.
1899 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1901 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1902 which appears in the queue multiple times.
1904 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1905 Alan Hayward <alan.hayward@arm.com>
1906 David Sherwood <david.sherwood@arm.com>
1908 * tree-vectorizer.h (vec_lower_bound): New structure.
1909 (_loop_vec_info): Add check_nonzero and lower_bounds.
1910 (LOOP_VINFO_CHECK_NONZERO): New macro.
1911 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1912 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1913 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1914 fields. Make seg_len the distance travelled, not including the
1916 (dr_direction_indicator): Declare.
1917 (dr_zero_step_indicator): Likewise.
1918 (dr_known_forward_stride_p): Likewise.
1919 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1921 (runtime_alias_check_p): Allow runtime alias checks with
1923 (operator ==): Compare access_size and align.
1924 (prune_runtime_alias_test_list): Rework for new distinction between
1925 the access_size and seg_len.
1926 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1928 (get_segment_min_max): New function.
1929 (create_intersect_range_checks): Use it.
1930 (dr_step_indicator): New function.
1931 (dr_direction_indicator): Likewise.
1932 (dr_zero_step_indicator): Likewise.
1933 (dr_known_forward_stride_p): Likewise.
1934 * tree-loop-distribution.c (data_ref_segment_size): Return
1935 DR_STEP * (niters - 1).
1936 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1938 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1939 (vect_preserves_scalar_order_p): New function, split out from...
1940 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1941 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1942 (vect_vfa_access_size): New function.
1943 (vect_vfa_align): Likewise.
1944 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1945 (dump_lower_bound): New function.
1946 (vect_check_lower_bound): Likewise.
1947 (vect_small_gap_p): Likewise.
1948 (vectorizable_with_step_bound_p): Likewise.
1949 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1950 depencies if the vectorization factor is 1. Convert the checks
1951 for nonzero steps into checks on the bounds of DR_STEP. Try using
1952 a bunds check for variable steps if the minimum required step is
1953 relatively small. Update calls to the dr_with_seg_len
1954 constructor and to vect_compile_time_alias.
1955 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1957 (vect_loop_versioning): Call it.
1958 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1960 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1962 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1963 Alan Hayward <alan.hayward@arm.com>
1964 David Sherwood <david.sherwood@arm.com>
1966 * doc/sourcebuild.texi (vect_scatter_store): Document.
1967 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1969 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1971 * genopinit.c (main): Add supports_vec_scatter_store and
1972 supports_vec_scatter_store_cached to target_optabs.
1973 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1974 IFN_MASK_SCATTER_STORE.
1975 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1977 * internal-fn.h (internal_store_fn_p): Declare.
1978 (internal_fn_stored_value_index): Likewise.
1979 * internal-fn.c (scatter_store_direct): New macro.
1980 (expand_scatter_store_optab_fn): New function.
1981 (direct_scatter_store_optab_supported_p): New macro.
1982 (internal_store_fn_p): New function.
1983 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1984 IFN_MASK_SCATTER_STORE.
1985 (internal_fn_mask_index): Likewise.
1986 (internal_fn_stored_value_index): New function.
1987 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1989 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1990 * optabs-query.c (supports_vec_scatter_store_p): New function.
1991 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1992 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1993 true for scatter stores.
1994 (vect_gather_scatter_fn_p): Handle scatter stores too.
1995 (vect_check_gather_scatter): Consider using scatter stores if
1996 supports_vec_scatter_store_p.
1997 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1999 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2000 internal_fn_stored_value_index.
2001 (check_load_store_masking): Handle scatter stores too.
2002 (vect_get_store_rhs): Make public.
2003 (vectorizable_call): Use internal_store_fn_p.
2004 (vectorizable_store): Handle scatter store internal functions.
2005 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2006 when deciding whether the end of the group has been reached.
2007 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2008 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2009 (mask_scatter_store<mode>): New insns.
2011 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2012 Alan Hayward <alan.hayward@arm.com>
2013 David Sherwood <david.sherwood@arm.com>
2015 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2016 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2017 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2019 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2020 Use vect_truncate_gather_scatter_offset if we can't treat the
2021 operation as a normal gather load or scatter store.
2022 (get_group_load_store_type): Take the gather_scatter_info
2023 as argument. Try using a gather load or scatter store for
2024 single-element groups.
2025 (get_load_store_type): Update calls to get_group_load_store_type
2026 and vect_use_strided_gather_scatters_p.
2028 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2029 Alan Hayward <alan.hayward@arm.com>
2030 David Sherwood <david.sherwood@arm.com>
2032 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2033 optional tree argument.
2034 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2036 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2037 but continue to use the current value as a fallback.
2038 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2039 to compare the updates.
2040 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2041 (get_load_store_type): Use it when handling a strided access.
2042 (vect_get_strided_load_store_ops): New function.
2043 (vect_get_data_ptr_increment): Likewise.
2044 (vectorizable_load): Handle strided gather loads. Always pass
2045 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2047 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2048 Alan Hayward <alan.hayward@arm.com>
2049 David Sherwood <david.sherwood@arm.com>
2051 * doc/md.texi (gather_load@var{m}): Document.
2052 (mask_gather_load@var{m}): Likewise.
2053 * genopinit.c (main): Add supports_vec_gather_load and
2054 supports_vec_gather_load_cached to target_optabs.
2055 * optabs-tree.c (init_tree_optimization_optabs): Use
2056 ggc_cleared_alloc to allocate target_optabs.
2057 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2058 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2060 * internal-fn.h (internal_load_fn_p): Declare.
2061 (internal_gather_scatter_fn_p): Likewise.
2062 (internal_fn_mask_index): Likewise.
2063 (internal_gather_scatter_fn_supported_p): Likewise.
2064 * internal-fn.c (gather_load_direct): New macro.
2065 (expand_gather_load_optab_fn): New function.
2066 (direct_gather_load_optab_supported_p): New macro.
2067 (direct_internal_fn_optab): New function.
2068 (internal_load_fn_p): Likewise.
2069 (internal_gather_scatter_fn_p): Likewise.
2070 (internal_fn_mask_index): Likewise.
2071 (internal_gather_scatter_fn_supported_p): Likewise.
2072 * optabs-query.c (supports_at_least_one_mode_p): New function.
2073 (supports_vec_gather_load_p): Likewise.
2074 * optabs-query.h (supports_vec_gather_load_p): Declare.
2075 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2076 and memory_type field.
2077 (NUM_PATTERNS): Bump to 15.
2078 * tree-vect-data-refs.c: Include internal-fn.h.
2079 (vect_gather_scatter_fn_p): New function.
2080 (vect_describe_gather_scatter_call): Likewise.
2081 (vect_check_gather_scatter): Try using internal functions for
2082 gather loads. Recognize existing calls to a gather load function.
2083 (vect_analyze_data_refs): Consider using gather loads if
2084 supports_vec_gather_load_p.
2085 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2086 (vect_get_gather_scatter_offset_type): Likewise.
2087 (vect_convert_mask_for_vectype): Likewise.
2088 (vect_add_conversion_to_patterm): Likewise.
2089 (vect_try_gather_scatter_pattern): Likewise.
2090 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2091 (vect_vect_recog_func_ptrs): Add it.
2092 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2093 internal_fn_mask_index and internal_gather_scatter_fn_p.
2094 (check_load_store_masking): Take the gather_scatter_info as an
2095 argument and handle gather loads.
2096 (vect_get_gather_scatter_ops): New function.
2097 (vectorizable_call): Check internal_load_fn_p.
2098 (vectorizable_load): Likewise. Handle gather load internal
2100 (vectorizable_store): Update call to check_load_store_masking.
2101 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2102 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2103 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2104 (aarch64_gather_scale_operand_d): New predicates.
2105 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2106 (mask_gather_load<mode>): New insns.
2108 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2109 Alan Hayward <alan.hayward@arm.com>
2110 David Sherwood <david.sherwood@arm.com>
2112 * optabs.def (fold_left_plus_optab): New optab.
2113 * doc/md.texi (fold_left_plus_@var{m}): Document.
2114 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2115 * internal-fn.c (fold_left_direct): Define.
2116 (expand_fold_left_optab_fn): Likewise.
2117 (direct_fold_left_optab_supported_p): Likewise.
2118 * fold-const-call.c (fold_const_fold_left): New function.
2119 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2120 * tree-parloops.c (valid_reduction_p): New function.
2121 (gather_scalar_reductions): Use it.
2122 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2123 (vect_finish_replace_stmt): Declare.
2124 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2125 (needs_fold_left_reduction_p): New function, split out from...
2126 (vect_is_simple_reduction): ...here. Accept reductions that
2127 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2128 (vect_force_simple_reduction): Also store the reduction type in
2129 the assignment's STMT_VINFO_REDUC_TYPE.
2130 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2131 (merge_with_identity): New function.
2132 (vect_expand_fold_left): Likewise.
2133 (vectorize_fold_left_reduction): Likewise.
2134 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2135 scalar phi in place for it. Check for target support and reject
2136 cases that would reassociate the operation. Defer the transform
2137 phase to vectorize_fold_left_reduction.
2138 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2139 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2140 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2142 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2144 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2145 call to ifc_temp_var.
2147 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2148 Alan Hayward <alan.hayward@arm.com>
2149 David Sherwood <david.sherwood@arm.com>
2151 * target.def (legitimize_address_displacement): Take the original
2152 offset as a poly_int.
2153 * targhooks.h (default_legitimize_address_displacement): Update
2155 * targhooks.c (default_legitimize_address_displacement): Likewise.
2156 * doc/tm.texi: Regenerate.
2157 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2158 as an argument, moving assert of ad->disp == ad->disp_term to...
2159 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2160 Try calling targetm.legitimize_address_displacement before expanding
2161 the address rather than afterwards, and adjust for the new interface.
2162 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2163 Match the new hook interface. Handle SVE addresses.
2164 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2167 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2169 * Makefile.in (OBJS): Add early-remat.o.
2170 * target.def (select_early_remat_modes): New hook.
2171 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2172 * doc/tm.texi: Regenerate.
2173 * targhooks.h (default_select_early_remat_modes): Declare.
2174 * targhooks.c (default_select_early_remat_modes): New function.
2175 * timevar.def (TV_EARLY_REMAT): New timevar.
2176 * passes.def (pass_early_remat): New pass.
2177 * tree-pass.h (make_pass_early_remat): Declare.
2178 * early-remat.c: New file.
2179 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2181 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2183 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2184 Alan Hayward <alan.hayward@arm.com>
2185 David Sherwood <david.sherwood@arm.com>
2187 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2188 vfm1 with a bound_epilog parameter.
2189 (vect_do_peeling): Update calls accordingly, and move the prologue
2190 call earlier in the function. Treat the base bound_epilog as 0 for
2191 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2192 this base when peeling for gaps.
2193 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2194 with fully-masked loops.
2195 (vect_estimate_min_profitable_iters): Handle the single peeled
2196 iteration in that case.
2198 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2199 Alan Hayward <alan.hayward@arm.com>
2200 David Sherwood <david.sherwood@arm.com>
2202 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2203 single-element interleaving even if the size is not a power of 2.
2204 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2205 accesses for single-element interleaving if the group size is
2208 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2209 Alan Hayward <alan.hayward@arm.com>
2210 David Sherwood <david.sherwood@arm.com>
2212 * doc/md.texi (fold_extract_last_@var{m}): Document.
2213 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2214 * optabs.def (fold_extract_last_optab): New optab.
2215 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2216 * internal-fn.c (fold_extract_direct): New macro.
2217 (expand_fold_extract_optab_fn): Likewise.
2218 (direct_fold_extract_optab_supported_p): Likewise.
2219 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2220 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2221 EXTRACT_LAST_REDUCTION.
2222 (get_initial_def_for_reduction): Do not create an initial vector
2223 for EXTRACT_LAST_REDUCTION reductions.
2224 (vectorizable_reduction): Leave the scalar phi in place for
2225 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2226 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2227 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2228 transform phase to vectorizable_condition.
2229 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2231 (vect_finish_stmt_generation): ...here.
2232 (vect_finish_replace_stmt): New function.
2233 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2234 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2236 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2238 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2239 Alan Hayward <alan.hayward@arm.com>
2240 David Sherwood <david.sherwood@arm.com>
2242 * doc/md.texi (extract_last_@var{m}): Document.
2243 * optabs.def (extract_last_optab): New optab.
2244 * internal-fn.def (EXTRACT_LAST): New internal function.
2245 * internal-fn.c (cond_unary_direct): New macro.
2246 (expand_cond_unary_optab_fn): Likewise.
2247 (direct_cond_unary_optab_supported_p): Likewise.
2248 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2249 loops using EXTRACT_LAST.
2250 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2251 (extract_last_<mode>): ...this optab.
2252 (vec_extract<mode><Vel>): Update accordingly.
2254 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2255 Alan Hayward <alan.hayward@arm.com>
2256 David Sherwood <david.sherwood@arm.com>
2258 * target.def (empty_mask_is_expensive): New hook.
2259 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2260 * doc/tm.texi: Regenerate.
2261 * targhooks.h (default_empty_mask_is_expensive): Declare.
2262 * targhooks.c (default_empty_mask_is_expensive): New function.
2263 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2264 if the target says that empty masks are expensive.
2265 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2267 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2269 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2270 Alan Hayward <alan.hayward@arm.com>
2271 David Sherwood <david.sherwood@arm.com>
2273 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2274 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2275 (vect_use_loop_mask_for_alignment_p): New function.
2276 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2277 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2278 niters_skip argument. Make sure that the first niters_skip elements
2279 of the first iteration are inactive.
2280 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2281 Update call to vect_set_loop_masks_directly.
2282 (get_misalign_in_elems): New function, split out from...
2283 (vect_gen_prolog_loop_niters): ...here.
2284 (vect_update_init_of_dr): Take a code argument that specifies whether
2285 the adjustment should be added or subtracted.
2286 (vect_update_init_of_drs): Likewise.
2287 (vect_prepare_for_masked_peels): New function.
2288 (vect_do_peeling): Skip prologue peeling if we're using a mask
2289 instead. Update call to vect_update_inits_of_drs.
2290 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2292 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2293 alignment. Do not include the number of peeled iterations in
2294 the minimum threshold in that case.
2295 (vectorizable_induction): Adjust the start value down by
2296 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2297 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2298 Take the number of skipped iterations into account when calculating
2300 * tree-vect-stmts.c (vect_gen_while_not): New function.
2302 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2303 Alan Hayward <alan.hayward@arm.com>
2304 David Sherwood <david.sherwood@arm.com>
2306 * doc/sourcebuild.texi (vect_fully_masked): Document.
2307 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2309 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2311 (vect_analyze_loop_2): ...here. Don't check the vectorization
2312 factor against the number of loop iterations if the loop is
2315 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2316 Alan Hayward <alan.hayward@arm.com>
2317 David Sherwood <david.sherwood@arm.com>
2319 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2320 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2321 (dump_groups): Update accordingly.
2322 (iv_use::mem_type): New member variable.
2323 (address_p): New function.
2324 (record_use): Add a mem_type argument and initialize the new
2326 (record_group_use): Add a mem_type argument. Use address_p.
2327 Remove obsolete null checks of base_object. Update call to record_use.
2328 (find_interesting_uses_op): Update call to record_group_use.
2329 (find_interesting_uses_cond): Likewise.
2330 (find_interesting_uses_address): Likewise.
2331 (get_mem_type_for_internal_fn): New function.
2332 (find_address_like_use): Likewise.
2333 (find_interesting_uses_stmt): Try find_address_like_use before
2334 calling find_interesting_uses_op.
2335 (addr_offset_valid_p): Use the iv mem_type field as the type
2336 of the addressed memory.
2337 (add_autoinc_candidates): Likewise.
2338 (get_address_cost): Likewise.
2339 (split_small_address_groups_p): Use address_p.
2340 (split_address_groups): Likewise.
2341 (add_iv_candidate_for_use): Likewise.
2342 (autoinc_possible_for_pair): Likewise.
2343 (rewrite_groups): Likewise.
2344 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2345 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2346 (get_alias_ptr_type_for_ptr_address): New function.
2347 (rewrite_use_address): Rewrite address uses in calls that were
2348 identified by find_address_like_use.
2350 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2351 Alan Hayward <alan.hayward@arm.com>
2352 David Sherwood <david.sherwood@arm.com>
2354 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2356 * gimple-expr.h (is_gimple_addressable: Likewise.
2357 * gimple-expr.c (is_gimple_address): Likewise.
2358 * internal-fn.c (expand_call_mem_ref): New function.
2359 (expand_mask_load_optab_fn): Use it.
2360 (expand_mask_store_optab_fn): Likewise.
2362 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2363 Alan Hayward <alan.hayward@arm.com>
2364 David Sherwood <david.sherwood@arm.com>
2366 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2367 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2368 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2369 (cond_umax@var{mode}): Document.
2370 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2371 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2372 (cond_umin_optab, cond_umax_optab): New optabs.
2373 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2374 (COND_IOR, COND_XOR): New internal functions.
2375 * internal-fn.h (get_conditional_internal_fn): Declare.
2376 * internal-fn.c (cond_binary_direct): New macro.
2377 (expand_cond_binary_optab_fn): Likewise.
2378 (direct_cond_binary_optab_supported_p): Likewise.
2379 (get_conditional_internal_fn): New function.
2380 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2381 Cope with reduction statements that are vectorized as calls rather
2383 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2384 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2385 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2386 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2387 (UNSPEC_COND_EOR): New unspecs.
2388 (optab): Add mappings for them.
2389 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2390 (sve_int_op, sve_fp_op): New int attributes.
2392 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2393 Alan Hayward <alan.hayward@arm.com>
2394 David Sherwood <david.sherwood@arm.com>
2396 * optabs.def (while_ult_optab): New optab.
2397 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2398 * internal-fn.def (WHILE_ULT): New internal function.
2399 * internal-fn.h (direct_internal_fn_supported_p): New override
2400 that takes two types as argument.
2401 * internal-fn.c (while_direct): New macro.
2402 (expand_while_optab_fn): New function.
2403 (convert_optab_supported_p): Likewise.
2404 (direct_while_optab_supported_p): New macro.
2405 * wide-int.h (wi::udiv_ceil): New function.
2406 * tree-vectorizer.h (rgroup_masks): New structure.
2407 (vec_loop_masks): New typedef.
2408 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2410 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2411 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2412 (vect_max_vf): New function.
2413 (slpeel_make_loop_iterate_ntimes): Delete.
2414 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2415 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2416 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2417 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2418 internal-fn.h, stor-layout.h and optabs-query.h.
2419 (vect_set_loop_mask): New function.
2420 (add_preheader_seq): Likewise.
2421 (add_header_seq): Likewise.
2422 (interleave_supported_p): Likewise.
2423 (vect_maybe_permute_loop_masks): Likewise.
2424 (vect_set_loop_masks_directly): Likewise.
2425 (vect_set_loop_condition_masked): Likewise.
2426 (vect_set_loop_condition_unmasked): New function, split out from
2427 slpeel_make_loop_iterate_ntimes.
2428 (slpeel_make_loop_iterate_ntimes): Rename to..
2429 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2430 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2431 (vect_do_peeling): Update call accordingly.
2432 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2434 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2435 mask_compare_type, can_fully_mask_p and fully_masked_p.
2436 (release_vec_loop_masks): New function.
2437 (_loop_vec_info): Use it to free the loop masks.
2438 (can_produce_all_loop_masks_p): New function.
2439 (vect_get_max_nscalars_per_iter): Likewise.
2440 (vect_verify_full_masking): Likewise.
2441 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2442 retries, and free the mask rgroups before retrying. Check loop-wide
2443 reasons for disallowing fully-masked loops. Make the final decision
2444 about whether use a fully-masked loop or not.
2445 (vect_estimate_min_profitable_iters): Do not assume that peeling
2446 for the number of iterations will be needed for fully-masked loops.
2447 (vectorizable_reduction): Disable fully-masked loops.
2448 (vectorizable_live_operation): Likewise.
2449 (vect_halve_mask_nunits): New function.
2450 (vect_double_mask_nunits): Likewise.
2451 (vect_record_loop_mask): Likewise.
2452 (vect_get_loop_mask): Likewise.
2453 (vect_transform_loop): Handle the case in which the final loop
2454 iteration might handle a partial vector. Call vect_set_loop_condition
2455 instead of slpeel_make_loop_iterate_ntimes.
2456 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2457 (check_load_store_masking): New function.
2458 (prepare_load_store_mask): Likewise.
2459 (vectorizable_store): Handle fully-masked loops.
2460 (vectorizable_load): Likewise.
2461 (supportable_widening_operation): Use vect_halve_mask_nunits for
2463 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2464 (vect_gen_while): New function.
2465 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2466 (aarch64_uqdec<mode>): New insn.
2468 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2469 Alan Hayward <alan.hayward@arm.com>
2470 David Sherwood <david.sherwood@arm.com>
2472 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2473 (reduc_xor_scal_optab): New optabs.
2474 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2475 (reduc_xor_scal_@var{m}): Document.
2476 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2477 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2479 * fold-const-call.c (fold_const_call): Handle them.
2480 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2481 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2482 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2483 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2484 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2485 (UNSPEC_XORV): New unspecs.
2486 (optab): Add entries for them.
2487 (BITWISEV): New int iterator.
2488 (bit_reduc_op): New int attributes.
2490 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2491 Alan Hayward <alan.hayward@arm.com>
2492 David Sherwood <david.sherwood@arm.com>
2494 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2495 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2496 * optabs.def (vec_shl_insert_optab): New optab.
2497 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2498 (duplicate_and_interleave): Likewise.
2499 * tree-vect-loop.c: Include internal-fn.h.
2500 (neutral_op_for_slp_reduction): New function, split out from
2501 get_initial_defs_for_reduction.
2502 (get_initial_def_for_reduction): Handle option 2 for variable-length
2503 vectors by loading the neutral value into a vector and then shifting
2504 the initial value into element 0.
2505 (get_initial_defs_for_reduction): Replace the code argument with
2506 the neutral value calculated by neutral_op_for_slp_reduction.
2507 Use gimple_build_vector for constant-length vectors.
2508 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2509 but the first group_size elements have a neutral value.
2510 Use duplicate_and_interleave otherwise.
2511 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2512 Update call to get_initial_defs_for_reduction. Handle SLP
2513 reductions for variable-length vectors by creating one vector
2514 result for each scalar result, with the elements associated
2515 with other scalar results stubbed out with the neutral value.
2516 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2517 Require IFN_VEC_SHL_INSERT for double reductions on
2518 variable-length vectors, or SLP reductions that have
2519 a neutral value. Require can_duplicate_and_interleave_p
2520 support for variable-length unchained SLP reductions if there
2521 is no neutral value, such as for MIN/MAX reductions. Also require
2522 the number of vector elements to be a multiple of the number of
2523 SLP statements when doing variable-length unchained SLP reductions.
2524 Update call to vect_create_epilog_for_reduction.
2525 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2526 and remove initial values.
2527 (duplicate_and_interleave): Make public.
2528 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2529 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2531 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2532 Alan Hayward <alan.hayward@arm.com>
2533 David Sherwood <david.sherwood@arm.com>
2535 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2536 (can_duplicate_and_interleave_p): New function.
2537 (vect_get_and_check_slp_defs): Take the vector of statements
2538 rather than just the current one. Remove excess parentheses.
2539 Restriction rejectinon of vect_constant_def and vect_external_def
2540 for variable-length vectors to boolean types, or types for which
2541 can_duplicate_and_interleave_p is false.
2542 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2543 (duplicate_and_interleave): New function.
2544 (vect_get_constant_vectors): Use gimple_build_vector for
2545 constant-length vectors and suitable variable-length constant
2546 vectors. Use duplicate_and_interleave for other variable-length
2547 vectors. Don't defer the update when inserting new statements.
2549 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2550 Alan Hayward <alan.hayward@arm.com>
2551 David Sherwood <david.sherwood@arm.com>
2553 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2554 min_profitable_iters doesn't go negative.
2556 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2557 Alan Hayward <alan.hayward@arm.com>
2558 David Sherwood <david.sherwood@arm.com>
2560 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2561 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2562 * optabs.def (vec_mask_load_lanes_optab): New optab.
2563 (vec_mask_store_lanes_optab): Likewise.
2564 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2565 (MASK_STORE_LANES): Likewise.
2566 * internal-fn.c (mask_load_lanes_direct): New macro.
2567 (mask_store_lanes_direct): Likewise.
2568 (expand_mask_load_optab_fn): Handle masked operations.
2569 (expand_mask_load_lanes_optab_fn): New macro.
2570 (expand_mask_store_optab_fn): Handle masked operations.
2571 (expand_mask_store_lanes_optab_fn): New macro.
2572 (direct_mask_load_lanes_optab_supported_p): Likewise.
2573 (direct_mask_store_lanes_optab_supported_p): Likewise.
2574 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2576 (vect_load_lanes_supported): Likewise.
2577 * tree-vect-data-refs.c (strip_conversion): New function.
2578 (can_group_stmts_p): Likewise.
2579 (vect_analyze_data_ref_accesses): Use it instead of checking
2580 for a pair of assignments.
2581 (vect_store_lanes_supported): Take a masked_p parameter.
2582 (vect_load_lanes_supported): Likewise.
2583 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2584 vect_store_lanes_supported and vect_load_lanes_supported.
2585 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2586 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2587 parameter. Don't allow gaps for masked accesses.
2588 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2589 and vect_load_lanes_supported.
2590 (get_load_store_type): Take a masked_p parameter and update
2591 call to get_group_load_store_type.
2592 (vectorizable_store): Update call to get_load_store_type.
2593 Handle IFN_MASK_STORE_LANES.
2594 (vectorizable_load): Update call to get_load_store_type.
2595 Handle IFN_MASK_LOAD_LANES.
2597 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2598 Alan Hayward <alan.hayward@arm.com>
2599 David Sherwood <david.sherwood@arm.com>
2601 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2603 * config/aarch64/aarch64-protos.h
2604 (aarch64_sve_struct_memory_operand_p): Declare.
2605 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2606 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2607 (VPRED, vpred): Handle SVE structure modes.
2608 * config/aarch64/constraints.md (Utx): New constraint.
2609 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2610 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2611 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2612 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2613 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2614 structure modes. Split into pieces after RA.
2615 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2616 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2618 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2619 SVE structure modes.
2620 (aarch64_classify_address): Likewise.
2621 (sizetochar): Move earlier in file.
2622 (aarch64_print_operand): Handle SVE register lists.
2623 (aarch64_array_mode): New function.
2624 (aarch64_sve_struct_memory_operand_p): Likewise.
2625 (TARGET_ARRAY_MODE): Redefine.
2627 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2628 Alan Hayward <alan.hayward@arm.com>
2629 David Sherwood <david.sherwood@arm.com>
2631 * target.def (array_mode): New target hook.
2632 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2633 * doc/tm.texi: Regenerate.
2634 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2635 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2636 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2638 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2641 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2642 Alan Hayward <alan.hayward@arm.com>
2643 David Sherwood <david.sherwood@arm.com>
2645 * fold-const.c (fold_binary_loc): Check the argument types
2646 rather than the result type when testing for a vector operation.
2648 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2650 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2651 * doc/tm.texi: Regenerate.
2653 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2654 Alan Hayward <alan.hayward@arm.com>
2655 David Sherwood <david.sherwood@arm.com>
2657 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2658 (sve): Document new AArch64 extension.
2659 * doc/md.texi (w): Extend the description of the AArch64
2660 constraint to include SVE vectors.
2661 (Upl, Upa): Document new AArch64 predicate constraints.
2662 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2664 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2665 (msve-vector-bits=): New option.
2666 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2667 SVE when these are disabled.
2668 (sve): New extension.
2669 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2670 modes. Adjust their number of units based on aarch64_sve_vg.
2671 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2672 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2673 aarch64_addr_query_type.
2674 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2675 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2676 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2677 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2678 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2679 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2680 (aarch64_simd_imm_zero_p): Delete.
2681 (aarch64_check_zero_based_sve_index_immediate): Declare.
2682 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2683 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2684 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2685 (aarch64_sve_float_mul_immediate_p): Likewise.
2686 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2688 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2689 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2690 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2691 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2692 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2693 (aarch64_regmode_natural_size): Likewise.
2694 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2695 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2697 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2698 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2699 for VG and the SVE predicate registers.
2700 (V_ALIASES): Add a "z"-prefixed alias.
2701 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2702 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2703 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2704 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2705 (REG_CLASS_NAMES): Add entries for them.
2706 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2707 and the predicate registers.
2708 (aarch64_sve_vg): Declare.
2709 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2710 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2711 (REGMODE_NATURAL_SIZE): Define.
2712 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2714 * config/aarch64/aarch64.c: Include cfgrtl.h.
2715 (simd_immediate_info): Add a constructor for series vectors,
2716 and an associated step field.
2717 (aarch64_sve_vg): New variable.
2718 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2719 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2720 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2721 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2722 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2723 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2724 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2725 (aarch64_get_mask_mode): New functions.
2726 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2727 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2728 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2729 predicate modes and predicate registers. Explicitly restrict
2730 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2731 to store a vector mode if it is recognized by
2732 aarch64_classify_vector_mode.
2733 (aarch64_regmode_natural_size): New function.
2734 (aarch64_hard_regno_caller_save_mode): Return the original mode
2736 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2737 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2738 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2739 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2741 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2742 does not overlap dest if the function is frame-related. Handle
2744 (aarch64_split_add_offset): New function.
2745 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2746 them aarch64_add_offset.
2747 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2748 and update call to aarch64_sub_sp.
2749 (aarch64_add_cfa_expression): New function.
2750 (aarch64_expand_prologue): Pass extra temporary registers to the
2751 functions above. Handle the case in which we need to emit new
2752 DW_CFA_expressions for registers that were originally saved
2753 relative to the stack pointer, but now have to be expressed
2754 relative to the frame pointer.
2755 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2757 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2758 IP0 and IP1 values for SVE frames.
2759 (aarch64_expand_vec_series): New function.
2760 (aarch64_expand_sve_widened_duplicate): Likewise.
2761 (aarch64_expand_sve_const_vector): Likewise.
2762 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2763 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2764 into the register, rather than emitting a SET directly.
2765 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2766 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2767 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2768 (offset_9bit_signed_scaled_p): New functions.
2769 (aarch64_replicate_bitmask_imm): New function.
2770 (aarch64_bitmask_imm): Use it.
2771 (aarch64_cannot_force_const_mem): Reject expressions involving
2772 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2773 (aarch64_classify_index): Handle SVE indices, by requiring
2774 a plain register index with a scale that matches the element size.
2775 (aarch64_classify_address): Handle SVE addresses. Assert that
2776 the mode of the address is VOIDmode or an integer mode.
2777 Update call to aarch64_classify_symbol.
2778 (aarch64_classify_symbolic_expression): Update call to
2779 aarch64_classify_symbol.
2780 (aarch64_const_vec_all_in_range_p): New function.
2781 (aarch64_print_vector_float_operand): Likewise.
2782 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2783 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2784 and the FP immediates 1.0 and 0.5.
2785 (aarch64_print_address_internal): Handle SVE addresses.
2786 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2787 (aarch64_regno_regclass): Handle predicate registers.
2788 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2790 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2791 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2792 (aarch64_convert_sve_vector_bits): New function.
2793 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2794 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2796 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2797 Handle SVE vector and predicate modes. Accept VL-based constants
2798 that need only one temporary register, and VL offsets that require
2799 no temporary registers.
2800 (aarch64_conditional_register_usage): Mark the predicate registers
2801 as fixed if SVE isn't available.
2802 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2803 Return true for SVE vector and predicate modes.
2804 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2805 rather than an unsigned int. Handle SVE modes.
2806 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2808 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2810 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2811 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2812 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2813 (aarch64_sve_float_mul_immediate_p): New functions.
2814 (aarch64_sve_valid_immediate): New function.
2815 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2816 Explicitly reject structure modes. Check for INDEX constants.
2817 Handle PTRUE and PFALSE constants.
2818 (aarch64_check_zero_based_sve_index_immediate): New function.
2819 (aarch64_simd_imm_zero_p): Delete.
2820 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2821 vector modes. Accept constants in the range of CNT[BHWD].
2822 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2823 ask for an Advanced SIMD mode.
2824 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2825 (aarch64_simd_vector_alignment): Handle SVE predicates.
2826 (aarch64_vectorize_preferred_vector_alignment): New function.
2827 (aarch64_simd_vector_alignment_reachable): Use it instead of
2829 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2830 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2832 (MAX_VECT_LEN): Delete.
2833 (expand_vec_perm_d): Add a vec_flags field.
2834 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2835 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2836 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2838 (aarch64_evpc_rev): Rename to...
2839 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2840 (aarch64_evpc_rev_global): New function.
2841 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2842 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2844 (aarch64_evpc_sve_tbl): New function.
2845 (aarch64_expand_vec_perm_const_1): Update after rename of
2846 aarch64_evpc_rev. Handle SVE permutes too, trying
2847 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2848 than aarch64_evpc_tbl.
2849 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2850 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2851 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2852 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2853 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2854 (aarch64_expand_sve_vcond): New functions.
2855 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2856 of aarch64_vector_mode_p.
2857 (aarch64_dwarf_poly_indeterminate_value): New function.
2858 (aarch64_compute_pressure_classes): Likewise.
2859 (aarch64_can_change_mode_class): Likewise.
2860 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2861 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2862 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2863 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2864 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2865 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2866 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2867 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2869 (Dn, Dl, Dr): Accept const as well as const_vector.
2870 (Dz): Likewise. Compare against CONST0_RTX.
2871 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2872 of "vector" where appropriate.
2873 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2874 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2875 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2876 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2877 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2878 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2879 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2880 (v_int_equiv): Extend to SVE modes.
2881 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2883 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2884 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2885 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2886 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2887 (SVE_COND_FP_CMP): New int iterators.
2888 (perm_hilo): Handle the new unpack unspecs.
2889 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2891 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2892 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2893 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2894 (aarch64_equality_operator, aarch64_constant_vector_operand)
2895 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2896 (aarch64_sve_nonimmediate_operand): Likewise.
2897 (aarch64_sve_general_operand): Likewise.
2898 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2899 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2900 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2901 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2902 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2903 (aarch64_sve_float_arith_immediate): Likewise.
2904 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2905 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2906 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2907 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2908 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2909 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2910 (aarch64_sve_float_arith_operand): Likewise.
2911 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2912 (aarch64_sve_float_mul_operand): Likewise.
2913 (aarch64_sve_vec_perm_operand): Likewise.
2914 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2915 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2916 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2917 as well as const_vector.
2918 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2919 in file. Use CONST0_RTX and CONSTM1_RTX.
2920 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2921 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2922 Use aarch64_simd_imm_zero.
2923 * config/aarch64/aarch64-sve.md: New file.
2924 * config/aarch64/aarch64.md: Include it.
2925 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2926 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2927 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2928 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2929 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2930 (sve): New attribute.
2931 (enabled): Disable instructions with the sve attribute unless
2933 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2934 aarch64_expand_mov_immediate.
2935 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2936 CNT[BHSD] immediates.
2937 (movti): Split CONST_POLY_INT moves into two halves.
2938 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2939 Split additions that need a temporary here if the destination
2940 is the stack pointer.
2941 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2942 (*add<mode>3_poly_1): New instruction.
2943 (set_clobber_cc): New expander.
2945 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2947 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2948 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2949 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2950 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2951 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2952 Change innermode from fixed_mode_size to machine_mode.
2953 (simplify_subreg): Update call accordingly. Handle a constant-sized
2954 subreg of a variable-length CONST_VECTOR.
2956 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2957 Alan Hayward <alan.hayward@arm.com>
2958 David Sherwood <david.sherwood@arm.com>
2960 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2961 (add_offset_to_base): New function, split out from...
2962 (create_mem_ref): ...here. When handling a scale other than 1,
2963 check first whether the address is valid without the offset.
2964 Add it into the base if so, leaving the index and scale as-is.
2966 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2969 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2970 fold_for_warn before checking if arg2 is INTEGER_CST.
2972 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2974 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2975 (store_multiple_operation): Delete.
2976 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2977 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2978 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2979 guarded by TARGET_STRING.
2980 (rs6000_output_load_multiple): Delete.
2981 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2982 OPTION_MASK_STRING / TARGET_STRING handling.
2983 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2984 (const rs6000_opt_masks) <"string">: Change mask to 0.
2985 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2986 (MASK_STRING): Delete.
2987 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2989 (load_multiple): Delete.
2996 (store_multiple): Delete.
3003 (movmemsi_8reg): Delete.
3004 (corresponding unnamed define_insn): Delete.
3005 (movmemsi_6reg): Delete.
3006 (corresponding unnamed define_insn): Delete.
3007 (movmemsi_4reg): Delete.
3008 (corresponding unnamed define_insn): Delete.
3009 (movmemsi_2reg): Delete.
3010 (corresponding unnamed define_insn): Delete.
3011 (movmemsi_1reg): Delete.
3012 (corresponding unnamed define_insn): Delete.
3013 * config/rs6000/rs6000.opt (mno-string): New.
3014 (mstring): Replace by deprecation warning stub.
3015 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3017 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3019 * regrename.c (regrename_do_replace): If replacing the same
3020 reg multiple times, try to reuse last created gen_raw_REG.
3023 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3024 main to workaround a bug in GDB.
3026 2018-01-12 Tom de Vries <tom@codesourcery.com>
3029 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3031 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3033 PR rtl-optimization/80481
3034 * ira-color.c (get_cap_member): New function.
3035 (allocnos_conflict_by_live_ranges_p): Use it.
3036 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3037 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3039 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3042 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3043 (*saddl_se_1): Ditto.
3045 (*ssubl_se_1): Ditto.
3047 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3049 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3050 rather than wi::to_widest for DR_INITs.
3051 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3052 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3053 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3055 (vect_analyze_group_access_1): Note that here.
3057 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3059 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3060 polynomial type sizes.
3062 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3064 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3065 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3066 (gimple_add_tmp_var): Likewise.
3068 2018-01-12 Martin Liska <mliska@suse.cz>
3070 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3071 (gimple_alloc_sizes): Likewise.
3072 (dump_gimple_statistics): Use PRIu64 in printf format.
3073 * gimple.h: Change uint64_t to int.
3075 2018-01-12 Martin Liska <mliska@suse.cz>
3077 * tree-core.h: Use uint64_t instead of int.
3078 * tree.c (tree_node_counts): Likewise.
3079 (tree_node_sizes): Likewise.
3080 (dump_tree_statistics): Use PRIu64 in printf format.
3082 2018-01-12 Martin Liska <mliska@suse.cz>
3084 * Makefile.in: As qsort_chk is implemented in vec.c, add
3085 vec.o to linkage of gencfn-macros.
3086 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3087 passing the info to record_node_allocation_statistics.
3088 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3090 * ggc-common.c (struct ggc_usage): Add operator== and use
3091 it in operator< and compare function.
3092 * mem-stats.h (struct mem_usage): Likewise.
3093 * vec.c (struct vec_usage): Remove operator< and compare
3094 function. Can be simply inherited.
3096 2018-01-12 Martin Jambor <mjambor@suse.cz>
3099 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3100 * tree-ssa-math-opts.c: Include domwalk.h.
3101 (convert_mult_to_fma_1): New function.
3102 (fma_transformation_info): New type.
3103 (fma_deferring_state): Likewise.
3104 (cancel_fma_deferring): New function.
3105 (result_of_phi): Likewise.
3106 (last_fma_candidate_feeds_initial_phi): Likewise.
3107 (convert_mult_to_fma): Added deferring logic, split actual
3108 transformation to convert_mult_to_fma_1.
3109 (math_opts_dom_walker): New type.
3110 (math_opts_dom_walker::after_dom_children): New method, body moved
3111 here from pass_optimize_widening_mul::execute, added deferring logic
3113 (pass_optimize_widening_mul::execute): Moved most of code to
3114 math_opts_dom_walker::after_dom_children.
3115 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3116 * config/i386/i386.c (ix86_option_override_internal): Added
3117 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3119 2018-01-12 Richard Biener <rguenther@suse.de>
3122 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3123 inline instance vars.
3125 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3128 * config/rx/rx.c (rx_is_restricted_memory_address):
3131 2018-01-12 Richard Biener <rguenther@suse.de>
3133 PR tree-optimization/80846
3134 * target.def (split_reduction): New target hook.
3135 * targhooks.c (default_split_reduction): New function.
3136 * targhooks.h (default_split_reduction): Declare.
3137 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3138 target requests first reduce vectors by combining low and high
3140 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3141 (get_vectype_for_scalar_type_and_size): Export.
3142 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3143 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3144 * doc/tm.texi: Regenerate.
3145 * config/i386/i386.c (ix86_split_reduction): Implement
3146 TARGET_VECTORIZE_SPLIT_REDUCTION.
3148 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3151 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3152 in PIC mode except for TARGET_VXWORKS_RTP.
3153 * config/sparc/sparc.c: Include cfgrtl.h.
3154 (TARGET_INIT_PIC_REG): Define.
3155 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3156 (sparc_pic_register_p): New predicate.
3157 (sparc_legitimate_address_p): Use it.
3158 (sparc_legitimize_pic_address): Likewise.
3159 (sparc_delegitimize_address): Likewise.
3160 (sparc_mode_dependent_address_p): Likewise.
3161 (gen_load_pcrel_sym): Remove 4th parameter.
3162 (load_got_register): Adjust call to above. Remove obsolete stuff.
3163 (sparc_expand_prologue): Do not call load_got_register here.
3164 (sparc_flat_expand_prologue): Likewise.
3165 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3166 (sparc_use_pseudo_pic_reg): New function.
3167 (sparc_init_pic_reg): Likewise.
3168 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3169 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3171 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3173 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3174 Add item for branch_cost.
3176 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3178 PR rtl-optimization/83565
3179 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3180 not extend the result to a larger mode for rotate operations.
3181 (num_sign_bit_copies1): Likewise.
3183 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3186 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3188 Use values-Xc.o for -pedantic.
3189 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3191 2018-01-12 Martin Liska <mliska@suse.cz>
3194 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3196 (possible_polymorphic_call_targets): Use it.
3197 (ipa_devirt): Likewise.
3199 2018-01-12 Martin Liska <mliska@suse.cz>
3201 * profile-count.h (enum profile_quality): Use 0 as invalid
3202 enum value of profile_quality.
3204 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3206 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3207 -mext-string options.
3209 2018-01-12 Richard Biener <rguenther@suse.de>
3211 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3212 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3213 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3215 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3217 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3219 * configure.ac (--with-long-double-format): Add support for the
3220 configuration option to change the default long double format on
3222 * config.gcc (powerpc*-linux*-*): Likewise.
3223 * configure: Regenerate.
3224 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3225 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3226 used without modification.
3228 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3230 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3231 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3232 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3233 MISC_BUILTIN_SPEC_BARRIER.
3234 (rs6000_init_builtins): Likewise.
3235 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3237 (speculation_barrier): New define_insn.
3238 * doc/extend.texi: Document __builtin_speculation_barrier.
3240 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3243 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3244 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3245 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3247 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3248 integral modes instead of "ss" and "sd".
3249 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3250 vectors with 32-bit and 64-bit elements.
3251 (vecdupssescalarmodesuffix): New mode attribute.
3252 (vec_dup<mode>): Use it.
3254 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3257 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3258 frame if argument is passed on stack.
3260 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3263 * ree.c (combine_reaching_defs): Optimize also
3264 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3265 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3267 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3270 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3272 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3275 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3276 after they are computed.
3278 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3280 PR tree-optimization/83695
3281 * gimple-loop-linterchange.cc
3282 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3283 reset cached scev information after interchange.
3284 (pass_linterchange::execute): Remove call to scev_reset_htab.
3286 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3288 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3289 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3290 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3291 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3292 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3293 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3294 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3295 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3296 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3297 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3298 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3299 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3300 (V_lane_reg): Likewise.
3301 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3303 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3304 (vfmal_lane_low<mode>_intrinsic,
3305 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3306 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3307 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3308 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3309 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3310 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3312 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3314 * config/arm/arm-cpus.in (fp16fml): New feature.
3315 (ALL_SIMD): Add fp16fml.
3316 (armv8.2-a): Add fp16fml as an option.
3317 (armv8.3-a): Likewise.
3318 (armv8.4-a): Add fp16fml as part of fp16.
3319 * config/arm/arm.h (TARGET_FP16FML): Define.
3320 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3322 * config/arm/arm-modes.def (V2HF): Define.
3323 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3324 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3325 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3326 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3327 vfmsl_low, vfmsl_high): New set of builtins.
3328 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3329 (vfml_op): New code attribute.
3330 (VFMLHALVES): New int iterator.
3331 (VFML, VFMLSEL): New mode attributes.
3332 (V_reg): Define mapping for V2HF.
3333 (V_hi, V_lo): New mode attributes.
3334 (VF_constraint): Likewise.
3335 (vfml_half, vfml_half_selector): New int attributes.
3336 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3338 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3339 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3341 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3342 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3343 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3344 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3346 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3347 Document new effective target and option set.
3349 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3351 * config/arm/arm-cpus.in (armv8_4): New feature.
3352 (ARMv8_4a): New fgroup.
3353 (armv8.4-a): New arch.
3354 * config/arm/arm-tables.opt: Regenerate.
3355 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3356 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3357 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3358 Add matching rules for -march=armv8.4-a and extensions.
3359 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3361 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3364 * config/rx/rx.md (BW): New mode attribute.
3365 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3367 2018-01-11 Richard Biener <rguenther@suse.de>
3369 PR tree-optimization/83435
3370 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3371 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3372 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3374 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3375 Alan Hayward <alan.hayward@arm.com>
3376 David Sherwood <david.sherwood@arm.com>
3378 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3380 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3381 (aarch64_print_address_internal): Use it to check for a zero offset.
3383 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3384 Alan Hayward <alan.hayward@arm.com>
3385 David Sherwood <david.sherwood@arm.com>
3387 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3388 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3389 Return a poly_int64 rather than a HOST_WIDE_INT.
3390 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3391 rather than a HOST_WIDE_INT.
3392 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3393 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3394 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3395 final_offset from HOST_WIDE_INT to poly_int64.
3396 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3397 to_constant when getting the number of units in an Advanced SIMD
3399 (aarch64_builtin_vectorized_function): Check for a constant number
3401 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3403 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3404 attribute instead of GET_MODE_NUNITS.
3405 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3406 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3407 GET_MODE_SIZE for fixed-size registers.
3408 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3409 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3410 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3411 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3412 (aarch64_print_operand, aarch64_print_address_internal)
3413 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3414 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3415 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3416 Handle polynomial GET_MODE_SIZE.
3417 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3418 wider than SImode without modification.
3419 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3420 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3421 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3422 passing and returning SVE modes.
3423 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3424 rather than GEN_INT.
3425 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3426 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3427 (aarch64_allocate_and_probe_stack_space): Likewise.
3428 (aarch64_layout_frame): Cope with polynomial offsets.
3429 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3430 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3432 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3433 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3434 poly_int64 rather than a HOST_WIDE_INT.
3435 (aarch64_get_separate_components, aarch64_process_components)
3436 (aarch64_expand_prologue, aarch64_expand_epilogue)
3437 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3438 (aarch64_anchor_offset): New function, split out from...
3439 (aarch64_legitimize_address): ...here.
3440 (aarch64_builtin_vectorization_cost): Handle polynomial
3441 TYPE_VECTOR_SUBPARTS.
3442 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3444 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3445 number of elements from the PARALLEL rather than the mode.
3446 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3447 rather than GET_MODE_BITSIZE.
3448 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3449 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3450 (aarch64_expand_vec_perm_const_1): Handle polynomial
3451 d->perm.length () and d->perm elements.
3452 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3453 Apply to_constant to d->perm elements.
3454 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3455 polynomial CONST_VECTOR_NUNITS.
3456 (aarch64_move_pointer): Take amount as a poly_int64 rather
3458 (aarch64_progress_pointer): Avoid temporary variable.
3459 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3460 the mode attribute instead of GET_MODE.
3462 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3463 Alan Hayward <alan.hayward@arm.com>
3464 David Sherwood <david.sherwood@arm.com>
3466 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3467 x exists before using it.
3468 (aarch64_add_constant_internal): Rename to...
3469 (aarch64_add_offset_1): ...this. Replace regnum with separate
3470 src and dest rtxes. Handle the case in which they're different,
3471 including when the offset is zero. Replace scratchreg with an rtx.
3472 Use 2 additions if there is no spare register into which we can
3473 move a 16-bit constant.
3474 (aarch64_add_constant): Delete.
3475 (aarch64_add_offset): Replace reg with separate src and dest
3476 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3477 Use aarch64_add_offset_1.
3478 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3479 an rtx rather than an int. Take the delta as a poly_int64
3480 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3481 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3482 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3483 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3484 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3486 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3487 aarch64_add_constant.
3489 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3491 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3492 Use scalar_float_mode.
3494 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3496 * config/aarch64/aarch64-simd.md
3497 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3498 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3499 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3500 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3501 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3502 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3503 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3504 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3505 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3506 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3508 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3511 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3512 targ_options->x_arm_arch_string is non NULL.
3514 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3516 * config/aarch64/aarch64.h
3517 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3519 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3522 * expmed.c (emit_store_flag_force): Swap if const op0
3523 and change VOIDmode to mode of op0.
3525 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3527 PR rtl-optimization/83761
3528 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3529 than bytes to mode_for_size.
3531 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3534 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3535 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3538 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3541 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3542 when in layout mode.
3543 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3544 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3547 2018-01-10 Michael Collison <michael.collison@arm.com>
3549 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3550 * config/aarch64/aarch64-option-extension.def: Add
3551 AARCH64_OPT_EXTENSION of 'fp16fml'.
3552 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3553 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3554 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3555 * config/aarch64/constraints.md (Ui7): New constraint.
3556 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3557 (VFMLA_SEL_W): Ditto.
3560 (VFMLA16_LOW): New int iterator.
3561 (VFMLA16_HIGH): Ditto.
3562 (UNSPEC_FMLAL): New unspec.
3563 (UNSPEC_FMLSL): Ditto.
3564 (UNSPEC_FMLAL2): Ditto.
3565 (UNSPEC_FMLSL2): Ditto.
3566 (f16mac): New code attribute.
3567 * config/aarch64/aarch64-simd-builtins.def
3568 (aarch64_fmlal_lowv2sf): Ditto.
3569 (aarch64_fmlsl_lowv2sf): Ditto.
3570 (aarch64_fmlalq_lowv4sf): Ditto.
3571 (aarch64_fmlslq_lowv4sf): Ditto.
3572 (aarch64_fmlal_highv2sf): Ditto.
3573 (aarch64_fmlsl_highv2sf): Ditto.
3574 (aarch64_fmlalq_highv4sf): Ditto.
3575 (aarch64_fmlslq_highv4sf): Ditto.
3576 (aarch64_fmlal_lane_lowv2sf): Ditto.
3577 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3578 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3579 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3580 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3581 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3582 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3583 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3584 (aarch64_fmlal_lane_highv2sf): Ditto.
3585 (aarch64_fmlsl_lane_highv2sf): Ditto.
3586 (aarch64_fmlal_laneq_highv2sf): Ditto.
3587 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3588 (aarch64_fmlalq_lane_highv4sf): Ditto.
3589 (aarch64_fmlsl_lane_highv4sf): Ditto.
3590 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3591 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3592 * config/aarch64/aarch64-simd.md:
3593 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3594 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3595 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3596 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3597 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3598 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3599 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3600 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3601 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3602 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3603 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3604 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3605 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3606 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3607 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3608 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3609 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3610 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3611 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3612 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3613 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3614 (vfmlsl_low_u32): Ditto.
3615 (vfmlalq_low_u32): Ditto.
3616 (vfmlslq_low_u32): Ditto.
3617 (vfmlal_high_u32): Ditto.
3618 (vfmlsl_high_u32): Ditto.
3619 (vfmlalq_high_u32): Ditto.
3620 (vfmlslq_high_u32): Ditto.
3621 (vfmlal_lane_low_u32): Ditto.
3622 (vfmlsl_lane_low_u32): Ditto.
3623 (vfmlal_laneq_low_u32): Ditto.
3624 (vfmlsl_laneq_low_u32): Ditto.
3625 (vfmlalq_lane_low_u32): Ditto.
3626 (vfmlslq_lane_low_u32): Ditto.
3627 (vfmlalq_laneq_low_u32): Ditto.
3628 (vfmlslq_laneq_low_u32): Ditto.
3629 (vfmlal_lane_high_u32): Ditto.
3630 (vfmlsl_lane_high_u32): Ditto.
3631 (vfmlal_laneq_high_u32): Ditto.
3632 (vfmlsl_laneq_high_u32): Ditto.
3633 (vfmlalq_lane_high_u32): Ditto.
3634 (vfmlslq_lane_high_u32): Ditto.
3635 (vfmlalq_laneq_high_u32): Ditto.
3636 (vfmlslq_laneq_high_u32): Ditto.
3637 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3638 (AARCH64_FL_FOR_ARCH8_4): New.
3639 (AARCH64_ISA_F16FML): New ISA flag.
3640 (TARGET_F16FML): New feature flag for fp16fml.
3641 (doc/invoke.texi): Document new fp16fml option.
3643 2018-01-10 Michael Collison <michael.collison@arm.com>
3645 * config/aarch64/aarch64-builtins.c:
3646 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3647 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3648 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3649 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3650 (AARCH64_ISA_SHA3): New ISA flag.
3651 (TARGET_SHA3): New feature flag for sha3.
3652 * config/aarch64/iterators.md (sha512_op): New int attribute.
3653 (CRYPTO_SHA512): New int iterator.
3654 (UNSPEC_SHA512H): New unspec.
3655 (UNSPEC_SHA512H2): Ditto.
3656 (UNSPEC_SHA512SU0): Ditto.
3657 (UNSPEC_SHA512SU1): Ditto.
3658 * config/aarch64/aarch64-simd-builtins.def
3659 (aarch64_crypto_sha512hqv2di): New builtin.
3660 (aarch64_crypto_sha512h2qv2di): Ditto.
3661 (aarch64_crypto_sha512su0qv2di): Ditto.
3662 (aarch64_crypto_sha512su1qv2di): Ditto.
3663 (aarch64_eor3qv8hi): Ditto.
3664 (aarch64_rax1qv2di): Ditto.
3665 (aarch64_xarqv2di): Ditto.
3666 (aarch64_bcaxqv8hi): Ditto.
3667 * config/aarch64/aarch64-simd.md:
3668 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3669 (aarch64_crypto_sha512su0qv2di): Ditto.
3670 (aarch64_crypto_sha512su1qv2di): Ditto.
3671 (aarch64_eor3qv8hi): Ditto.
3672 (aarch64_rax1qv2di): Ditto.
3673 (aarch64_xarqv2di): Ditto.
3674 (aarch64_bcaxqv8hi): Ditto.
3675 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3676 (vsha512h2q_u64): Ditto.
3677 (vsha512su0q_u64): Ditto.
3678 (vsha512su1q_u64): Ditto.
3679 (veor3q_u16): Ditto.
3680 (vrax1q_u64): Ditto.
3682 (vbcaxq_u16): Ditto.
3683 * config/arm/types.md (crypto_sha512): New type attribute.
3684 (crypto_sha3): Ditto.
3685 (doc/invoke.texi): Document new sha3 option.
3687 2018-01-10 Michael Collison <michael.collison@arm.com>
3689 * config/aarch64/aarch64-builtins.c:
3690 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3691 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3692 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3693 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3694 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3695 (AARCH64_ISA_SM4): New ISA flag.
3696 (TARGET_SM4): New feature flag for sm4.
3697 * config/aarch64/aarch64-simd-builtins.def
3698 (aarch64_sm3ss1qv4si): Ditto.
3699 (aarch64_sm3tt1aq4si): Ditto.
3700 (aarch64_sm3tt1bq4si): Ditto.
3701 (aarch64_sm3tt2aq4si): Ditto.
3702 (aarch64_sm3tt2bq4si): Ditto.
3703 (aarch64_sm3partw1qv4si): Ditto.
3704 (aarch64_sm3partw2qv4si): Ditto.
3705 (aarch64_sm4eqv4si): Ditto.
3706 (aarch64_sm4ekeyqv4si): Ditto.
3707 * config/aarch64/aarch64-simd.md:
3708 (aarch64_sm3ss1qv4si): Ditto.
3709 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3710 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3711 (aarch64_sm4eqv4si): Ditto.
3712 (aarch64_sm4ekeyqv4si): Ditto.
3713 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3714 (sm3part_op): Ditto.
3715 (CRYPTO_SM3TT): Ditto.
3716 (CRYPTO_SM3PART): Ditto.
3717 (UNSPEC_SM3SS1): New unspec.
3718 (UNSPEC_SM3TT1A): Ditto.
3719 (UNSPEC_SM3TT1B): Ditto.
3720 (UNSPEC_SM3TT2A): Ditto.
3721 (UNSPEC_SM3TT2B): Ditto.
3722 (UNSPEC_SM3PARTW1): Ditto.
3723 (UNSPEC_SM3PARTW2): Ditto.
3724 (UNSPEC_SM4E): Ditto.
3725 (UNSPEC_SM4EKEY): Ditto.
3726 * config/aarch64/constraints.md (Ui2): New constraint.
3727 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3728 * config/arm/types.md (crypto_sm3): New type attribute.
3729 (crypto_sm4): Ditto.
3730 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3731 (vsm3tt1aq_u32): Ditto.
3732 (vsm3tt1bq_u32): Ditto.
3733 (vsm3tt2aq_u32): Ditto.
3734 (vsm3tt2bq_u32): Ditto.
3735 (vsm3partw1q_u32): Ditto.
3736 (vsm3partw2q_u32): Ditto.
3737 (vsm4eq_u32): Ditto.
3738 (vsm4ekeyq_u32): Ditto.
3739 (doc/invoke.texi): Document new sm4 option.
3741 2018-01-10 Michael Collison <michael.collison@arm.com>
3743 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3744 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3745 (AARCH64_FL_FOR_ARCH8_4): New.
3746 (AARCH64_FL_V8_4): New flag.
3747 (doc/invoke.texi): Document new armv8.4-a option.
3749 2018-01-10 Michael Collison <michael.collison@arm.com>
3751 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3752 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3753 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3754 * config/aarch64/aarch64-option-extension.def: Add
3755 AARCH64_OPT_EXTENSION of 'sha2'.
3756 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3757 (crypto): Disable sha2 and aes if crypto disabled.
3758 (crypto): Enable aes and sha2 if enabled.
3759 (simd): Disable sha2 and aes if simd disabled.
3760 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3762 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3763 (TARGET_SHA2): New feature flag for sha2.
3764 (TARGET_AES): New feature flag for aes.
3765 * config/aarch64/aarch64-simd.md:
3766 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3767 conditional on TARGET_AES.
3768 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3769 (aarch64_crypto_sha1hsi): Make pattern conditional
3771 (aarch64_crypto_sha1hv4si): Ditto.
3772 (aarch64_be_crypto_sha1hv4si): Ditto.
3773 (aarch64_crypto_sha1su1v4si): Ditto.
3774 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3775 (aarch64_crypto_sha1su0v4si): Ditto.
3776 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3777 (aarch64_crypto_sha256su0v4si): Ditto.
3778 (aarch64_crypto_sha256su1v4si): Ditto.
3779 (doc/invoke.texi): Document new aes and sha2 options.
3781 2018-01-10 Martin Sebor <msebor@redhat.com>
3783 PR tree-optimization/83781
3784 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3787 2018-01-11 Martin Sebor <msebor@gmail.com>
3788 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3790 PR tree-optimization/83501
3791 PR tree-optimization/81703
3793 * tree-ssa-strlen.c (get_string_cst): Rename...
3794 (get_string_len): ...to this. Handle global constants.
3795 (handle_char_store): Adjust.
3797 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3798 Jim Wilson <jimw@sifive.com>
3800 * config/riscv/riscv-protos.h (riscv_output_return): New.
3801 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3802 (riscv_attribute_table, riscv_output_return),
3803 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3804 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3805 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3807 (riscv_expand_prologue): Add early return for naked function.
3808 (riscv_expand_epilogue): Likewise.
3809 (riscv_function_ok_for_sibcall): Return false for naked function.
3810 (riscv_set_current_function): New.
3811 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3812 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3813 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3814 * doc/extend.texi (RISC-V Function Attributes): New.
3816 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3818 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3819 check for 128-bit long double before checking TCmode.
3820 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3821 128-bit long doubles before checking TFmode or TCmode.
3822 (FLOAT128_IBM_P): Likewise.
3824 2018-01-10 Martin Sebor <msebor@redhat.com>
3826 PR tree-optimization/83671
3827 * builtins.c (c_strlen): Unconditionally return zero for the empty
3829 Use -Warray-bounds for warnings.
3830 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3831 for non-constant array indices with COMPONENT_REF, arrays of
3832 arrays, and pointers to arrays.
3833 (gimple_fold_builtin_strlen): Determine and set length range for
3834 non-constant character arrays.
3836 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3839 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3842 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3844 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3846 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3849 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3850 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3851 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3852 indexed_or_indirect_operand predicate.
3853 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3854 (*vsx_le_perm_load_v8hi): Likewise.
3855 (*vsx_le_perm_load_v16qi): Likewise.
3856 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3857 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3858 (*vsx_le_perm_store_v8hi): Likewise.
3859 (*vsx_le_perm_store_v16qi): Likewise.
3860 (eight unnamed splitters): Likewise.
3862 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3864 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3865 * config/rs6000/emmintrin.h: Likewise.
3866 * config/rs6000/mmintrin.h: Likewise.
3867 * config/rs6000/xmmintrin.h: Likewise.
3869 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3872 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3874 * tree.c (tree_nop_conversion): Return true for location wrapper
3876 (maybe_wrap_with_location): New function.
3877 (selftest::check_strip_nops): New function.
3878 (selftest::test_location_wrappers): New function.
3879 (selftest::tree_c_tests): Call it.
3880 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3881 (maybe_wrap_with_location): New decl.
3882 (EXPR_LOCATION_WRAPPER_P): New macro.
3883 (location_wrapper_p): New inline function.
3884 (tree_strip_any_location_wrapper): New inline function.
3886 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3889 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3890 stack_realign_offset for the largest alignment of stack slot
3892 (ix86_find_max_used_stack_alignment): New function.
3893 (ix86_finalize_stack_frame_flags): Use it. Set
3894 max_used_stack_alignment if we don't realign stack.
3895 * config/i386/i386.h (machine_function): Add
3896 max_used_stack_alignment.
3898 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3900 * config/arm/arm.opt (-mbranch-cost): New option.
3901 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3904 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3907 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3908 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3910 2018-01-10 Richard Biener <rguenther@suse.de>
3913 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3914 early out so it also covers the case where we have a non-NULL
3917 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3919 PR tree-optimization/83753
3920 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3921 for non-strided grouped accesses if the number of elements is 1.
3923 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3926 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3927 * i386.h (TARGET_USE_GATHER): Define.
3928 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3930 2018-01-10 Martin Liska <mliska@suse.cz>
3933 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3934 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3936 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3937 CLEANUP_NO_PARTITIONING is not set.
3939 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3941 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3942 for vectors, as a partial revert of r254296.
3943 * rtl.h (const_vec_p): Delete.
3944 (const_vec_duplicate_p): Don't test for vector CONSTs.
3945 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3946 * expmed.c (make_tree): Likewise.
3949 * common.md (E, F): Use CONSTANT_P instead of checking for
3951 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3952 checking for CONST_VECTOR.
3954 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3957 * predict.c (force_edge_cold): Handle in more sane way edges
3960 2018-01-09 Carl Love <cel@us.ibm.com>
3962 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3964 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3965 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3966 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3967 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3968 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3969 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3970 * config/rs6000/rs6000-protos.h: Add extern defition for
3971 rs6000_generate_float2_double_code.
3972 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3974 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3975 (float2_v2df): Add define_expand.
3977 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3980 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3981 op_mode in the force_to_mode call.
3983 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3985 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3986 instead of checking each element individually.
3987 (aarch64_evpc_uzp): Likewise.
3988 (aarch64_evpc_zip): Likewise.
3989 (aarch64_evpc_ext): Likewise.
3990 (aarch64_evpc_rev): Likewise.
3991 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3992 instead of checking each element individually. Return true without
3994 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3995 whether all selected elements come from the same input, instead of
3996 checking each element individually. Remove calls to gen_rtx_REG,
3997 start_sequence and end_sequence and instead assert that no rtl is
4000 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4002 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4003 order of HIGH and CONST checks.
4005 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4007 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4008 if the destination isn't an SSA_NAME.
4010 2018-01-09 Richard Biener <rguenther@suse.de>
4012 PR tree-optimization/83668
4013 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4015 (canonicalize_loop_form): ... here, renamed from ...
4016 (canonicalize_loop_closed_ssa_form): ... this and amended to
4017 swap successor edges for loop exit blocks to make us use
4018 the RPO order we need for initial schedule generation.
4020 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4022 PR tree-optimization/64811
4023 * match.pd: When optimizing comparisons with Inf, avoid
4024 introducing or losing exceptions from comparisons with NaN.
4026 2018-01-09 Martin Liska <mliska@suse.cz>
4029 * asan.c (shadow_mem_size): Add gcc_assert.
4031 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4033 Don't save registers in main().
4036 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4037 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4038 * config/avr/avr.c (avr_set_current_function): Don't error if
4039 naked, OS_task or OS_main are specified at the same time.
4040 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4042 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4044 * common/config/avr/avr-common.c (avr_option_optimization_table):
4045 Switch on -mmain-is-OS_task for optimizing compilations.
4047 2018-01-09 Richard Biener <rguenther@suse.de>
4049 PR tree-optimization/83572
4050 * graphite.c: Include cfganal.h.
4051 (graphite_transform_loops): Connect infinite loops to exit
4052 and remove fake edges at the end.
4054 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4056 * ipa-inline.c (edge_badness): Revert accidental checkin.
4058 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4061 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4062 symbols; not inline clones.
4064 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4067 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4068 hard registers. Formatting fixes.
4070 PR preprocessor/83722
4071 * gcc.c (try_generate_repro): Pass
4072 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4073 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4076 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4077 Kito Cheng <kito.cheng@gmail.com>
4079 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4080 (riscv_leaf_function_p): Delete.
4081 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4083 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4085 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4087 (do_ifelse): New function.
4088 (do_isel): New function.
4089 (do_sub3): New function.
4090 (do_add3): New function.
4091 (do_load_mask_compare): New function.
4092 (do_overlap_load_compare): New function.
4093 (expand_compare_loop): New function.
4094 (expand_block_compare): Call expand_compare_loop() when appropriate.
4095 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4097 (-mblock-compare-inline-loop-limit): New option.
4099 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4102 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4103 Reverse order of second and third operands in first alternative.
4104 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4105 of first and second elements in UNSPEC_VPERMR vector.
4106 (altivec_expand_vec_perm_le): Likewise.
4108 2018-01-08 Jeff Law <law@redhat.com>
4110 PR rtl-optimizatin/81308
4111 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4112 (process_switch): If group_case_labels makes a change, then set
4114 (pass_convert_switch::execute): If a switch is converted, then
4115 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4117 PR rtl-optimization/81308
4118 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4121 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4123 PR target/83663 - Revert r255946
4124 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4125 generation for cases where splatting a value is not useful.
4126 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4127 across a vec_duplicate and a paradoxical subreg forming a vector
4128 mode to a vec_concat.
4130 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4132 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4133 -march=armv8.3-a variants.
4134 * config/arm/t-multilib: Likewise.
4135 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4137 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4139 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4141 (cceq_ior_compare_complement): Give it a name so I can use it, and
4142 change boolean_or_operator predicate to boolean_operator so it can
4143 be used to generate a crand.
4144 (eqne): New code iterator.
4145 (bd/bd_neg): New code_attrs.
4146 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4147 a single define_insn.
4148 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4149 decrement (bdnzt/bdnzf/bdzt/bdzf).
4150 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4151 with the new names of the branch decrement patterns, and added the
4152 names of the branch decrement conditional patterns.
4154 2018-01-08 Richard Biener <rguenther@suse.de>
4156 PR tree-optimization/83563
4157 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4160 2018-01-08 Richard Biener <rguenther@suse.de>
4163 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4165 2018-01-08 Richard Biener <rguenther@suse.de>
4167 PR tree-optimization/83685
4168 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4169 references to abnormals.
4171 2018-01-08 Richard Biener <rguenther@suse.de>
4174 * dwarf2out.c (output_indirect_strings): Handle empty
4175 skeleton_debug_str_hash.
4176 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4178 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4180 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4181 (emit_store_direct): Likewise.
4182 (arc_trampoline_adjust_address): Likewise.
4183 (arc_asm_trampoline_template): New function.
4184 (arc_initialize_trampoline): Use asm_trampoline_template.
4185 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4186 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4187 * config/arc/arc.md (flush_icache): Delete pattern.
4189 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4191 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4192 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4195 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4198 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4199 by not USED_FOR_TARGET.
4200 (make_pass_resolve_sw_modes): Likewise.
4202 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4204 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4207 2018-01-08 Richard Biener <rguenther@suse.de>
4210 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4212 2018-01-08 Richard Biener <rguenther@suse.de>
4215 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4217 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4220 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4221 basic blocks with a small number of successors.
4222 (convert_control_dep_chain_into_preds): Improve handling of
4224 (dump_predicates): Split apart into...
4225 (dump_pred_chain): ...here...
4226 (dump_pred_info): ...and here.
4227 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4228 (can_chain_union_be_invalidated_p): Improve check for invalidation
4230 (uninit_uses_cannot_happen): Avoid unnecessary if
4231 convert_control_dep_chain_into_preds yielded nothing.
4233 2018-01-06 Martin Sebor <msebor@redhat.com>
4235 PR tree-optimization/83640
4236 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4237 subtracting negative offset from size.
4238 (builtin_access::overlap): Adjust offset bounds of the access to fall
4239 within the size of the object if possible.
4241 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4243 PR rtl-optimization/83699
4244 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4245 extract_bit_field_as_subreg to cases in which the extracted
4246 value is also a vector.
4248 * lra-constraints.c (process_alt_operands): Test for the equivalence
4249 substitutions when detecting a possible reload cycle.
4251 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4254 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4255 by default if flag_selective_schedling{,2}. Formatting fixes.
4257 PR rtl-optimization/83682
4258 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4259 if it has non-VECTOR_MODE element mode.
4260 (vec_duplicate_p): Likewise.
4263 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4264 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4266 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4269 * config/i386/i386-builtin.def
4270 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4271 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4272 Require also OPTION_MASK_ISA_AVX512F in addition to
4273 OPTION_MASK_ISA_GFNI.
4274 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4275 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4276 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4277 to OPTION_MASK_ISA_GFNI.
4278 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4279 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4280 OPTION_MASK_ISA_AVX512BW.
4281 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4282 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4283 addition to OPTION_MASK_ISA_GFNI.
4284 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4285 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4286 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4287 to OPTION_MASK_ISA_GFNI.
4288 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4289 a requirement for all ISAs rather than any of them with a few
4291 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4293 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4294 bitmasks to be enabled with 3 exceptions, instead of requiring any
4295 enabled ISA with lots of exceptions.
4296 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4297 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4298 Change avx512bw in isa attribute to avx512f.
4299 * config/i386/sgxintrin.h: Add license boilerplate.
4300 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4301 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4302 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4303 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4305 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4306 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4307 temporarily sse2 rather than sse if not enabled already.
4310 * config/i386/sse.md (VI248_VLBW): Rename to ...
4311 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4312 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4313 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4314 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4315 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4316 mode iterator instead of VI248_VLBW.
4318 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4320 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4321 (record_modified): Skip clobbers; add debug output.
4322 (param_change_prob): Use sreal frequencies.
4324 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4326 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4327 punt for user-aligned variables.
4329 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4331 * tree-chrec.c (chrec_contains_symbols): Return true for
4334 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4337 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4338 of (x|y) == x for BICS pattern.
4340 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4342 PR tree-optimization/83605
4343 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4344 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4347 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4349 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4350 * config/epiphany/rtems.h: New file.
4352 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4353 Uros Bizjak <ubizjak@gmail.com>
4356 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4357 QIreg_operand instead of register_operand predicate.
4358 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4359 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4360 comments instead of -fmitigate[-_]rop.
4362 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4365 * cgraphunit.c (symbol_table::compile): Switch to text_section
4366 before calling assembly_start debug hook.
4367 * run-rtl-passes.c (run_rtl_passes): Likewise.
4370 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4372 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4373 range_int_cst_p rather than !symbolic_range_p before calling
4374 extract_range_from_multiplicative_op_1.
4376 2018-01-04 Jeff Law <law@redhat.com>
4378 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4379 redundant test in assertion.
4381 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4383 * doc/rtl.texi: Document machine_mode wrapper classes.
4385 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4387 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4390 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4392 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4393 the VEC_PERM_EXPR fold to fail.
4395 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4398 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4399 to switched_sections.
4401 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4404 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4407 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4410 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4411 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4413 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4416 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4417 is BLKmode and bitpos not zero or mode change is needed.
4419 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4422 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4425 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4428 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4429 instead of MULT rtx. Update all corresponding splitters.
4431 (*ssub<modesuffix>): Ditto.
4433 (*cmp_sadd_di): Update split patterns.
4434 (*cmp_sadd_si): Ditto.
4435 (*cmp_sadd_sidi): Ditto.
4436 (*cmp_ssub_di): Ditto.
4437 (*cmp_ssub_si): Ditto.
4438 (*cmp_ssub_sidi): Ditto.
4439 * config/alpha/predicates.md (const23_operand): New predicate.
4440 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4441 Look for ASHIFT, not MULT inner operand.
4442 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4444 2018-01-04 Martin Liska <mliska@suse.cz>
4446 PR gcov-profile/83669
4447 * gcov.c (output_intermediate_file): Add version to intermediate
4449 * doc/gcov.texi: Document new field 'version' in intermediate
4450 file format. Fix location of '-k' option of gcov command.
4452 2018-01-04 Martin Liska <mliska@suse.cz>
4455 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4457 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4459 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4461 2018-01-03 Martin Sebor <msebor@redhat.com>
4463 PR tree-optimization/83655
4464 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4465 checking calls with invalid arguments.
4467 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4469 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4470 (vectorizable_mask_load_store): Delete.
4471 (vectorizable_call): Return false for masked loads and stores.
4472 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4473 instead of gimple_assign_rhs1.
4474 (vectorizable_load): Handle IFN_MASK_LOAD.
4475 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4477 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4479 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4481 (vectorizable_mask_load_store): ...here.
4482 (vectorizable_load): ...and here.
4484 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4486 * tree-vect-stmts.c (vect_build_all_ones_mask)
4487 (vect_build_zero_merge_argument): New functions, split out from...
4488 (vectorizable_load): ...here.
4490 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4492 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4494 (vectorizable_mask_load_store): ...here.
4495 (vectorizable_store): ...and here.
4497 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4499 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4501 (vectorizable_mask_load_store): ...here.
4503 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4505 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4506 (vect_model_store_cost): Take a vec_load_store_type instead of a
4508 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4509 (vect_model_store_cost): Take a vec_load_store_type instead of a
4511 (vectorizable_mask_load_store): Update accordingly.
4512 (vectorizable_store): Likewise.
4513 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4515 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4517 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4518 IFN_MASK_LOAD calls here rather than...
4519 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4521 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4522 Alan Hayward <alan.hayward@arm.com>
4523 David Sherwood <david.sherwood@arm.com>
4525 * expmed.c (extract_bit_field_1): For vector extracts,
4526 fall back to extract_bit_field_as_subreg if vec_extract
4529 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4530 Alan Hayward <alan.hayward@arm.com>
4531 David Sherwood <david.sherwood@arm.com>
4533 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4534 they are variable or constant sized.
4535 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4536 slots for constant-sized data.
4538 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4539 Alan Hayward <alan.hayward@arm.com>
4540 David Sherwood <david.sherwood@arm.com>
4542 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4543 handling COND_EXPRs with boolean comparisons, try to find a better
4544 basis for the mask type than the boolean itself.
4546 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4548 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4549 is calculated and how it can be overridden.
4550 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4551 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4553 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4556 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4557 Alan Hayward <alan.hayward@arm.com>
4558 David Sherwood <david.sherwood@arm.com>
4560 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4561 Remove the mode argument.
4562 (aarch64_simd_valid_immediate): Remove the mode and inverse
4564 * config/aarch64/iterators.md (bitsize): New iterator.
4565 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4566 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4567 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4568 aarch64_simd_valid_immediate.
4569 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4570 (aarch64_reg_or_bic_imm): Likewise.
4571 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4572 with an insn_type enum and msl with a modifier_type enum.
4573 Replace element_width with a scalar_mode. Change the shift
4574 to unsigned int. Add constructors for scalar_float_mode and
4575 scalar_int_mode elements.
4576 (aarch64_vect_float_const_representable_p): Delete.
4577 (aarch64_can_const_movi_rtx_p)
4578 (aarch64_simd_scalar_immediate_valid_for_move)
4579 (aarch64_simd_make_constant): Update call to
4580 aarch64_simd_valid_immediate.
4581 (aarch64_advsimd_valid_immediate_hs): New function.
4582 (aarch64_advsimd_valid_immediate): Likewise.
4583 (aarch64_simd_valid_immediate): Remove mode and inverse
4584 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4585 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4586 and aarch64_float_const_representable_p on the result.
4587 (aarch64_output_simd_mov_immediate): Remove mode argument.
4588 Update call to aarch64_simd_valid_immediate and use of
4589 simd_immediate_info.
4590 (aarch64_output_scalar_simd_mov_immediate): Update call
4593 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4594 Alan Hayward <alan.hayward@arm.com>
4595 David Sherwood <david.sherwood@arm.com>
4597 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4598 (mode_nunits): Likewise CONST_MODE_NUNITS.
4599 * machmode.def (ADJUST_NUNITS): Document.
4600 * genmodes.c (mode_data::need_nunits_adj): New field.
4601 (blank_mode): Update accordingly.
4602 (adj_nunits): New variable.
4603 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4605 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4606 listed in adj_nunits.
4607 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4608 listed in adj_nunits. Don't emit case statements for such modes.
4609 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4610 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4611 nothing if adj_nunits is nonnull.
4612 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4613 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4614 (emit_mode_fbit): Update use of print_maybe_const_decl.
4615 (emit_move_size): Likewise. Treat the array as non-const
4617 (emit_mode_adjustments): Handle adj_nunits.
4619 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4621 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4622 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4623 (VECTOR_MODES): Use it.
4624 (make_vector_modes): Take the prefix as an argument.
4626 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4627 Alan Hayward <alan.hayward@arm.com>
4628 David Sherwood <david.sherwood@arm.com>
4630 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4631 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4632 for MODE_VECTOR_BOOL.
4633 * machmode.def (VECTOR_BOOL_MODE): Document.
4634 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4635 (make_vector_bool_mode): New function.
4636 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4638 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4639 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4641 * stor-layout.c (int_mode_for_mode): Likewise.
4642 * tree.c (build_vector_type_for_mode): Likewise.
4643 * varasm.c (output_constant_pool_2): Likewise.
4644 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4645 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4646 for MODE_VECTOR_BOOL.
4647 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4648 of mode class checks.
4649 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4650 instead of a list of mode class checks.
4651 (expand_vector_scalar_condition): Likewise.
4652 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4654 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4655 Alan Hayward <alan.hayward@arm.com>
4656 David Sherwood <david.sherwood@arm.com>
4658 * machmode.h (mode_size): Change from unsigned short to
4660 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4661 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4662 or if measurement_type is not polynomial.
4663 (fixed_size_mode::includes_p): Check for constant-sized modes.
4664 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4665 return a poly_uint16 rather than an unsigned short.
4666 (emit_mode_size): Change the type of mode_size from unsigned short
4667 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4668 (emit_mode_adjustments): Cope with polynomial vector sizes.
4669 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4671 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4673 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4674 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4675 * caller-save.c (setup_save_areas): Likewise.
4676 (replace_reg_with_saved_mem): Likewise.
4677 * calls.c (emit_library_call_value_1): Likewise.
4678 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4679 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4680 (gen_lowpart_for_combine): Likewise.
4681 * convert.c (convert_to_integer_1): Likewise.
4682 * cse.c (equiv_constant, cse_insn): Likewise.
4683 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4684 (cselib_subst_to_values): Likewise.
4685 * dce.c (word_dce_process_block): Likewise.
4686 * df-problems.c (df_word_lr_mark_ref): Likewise.
4687 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4688 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4689 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4690 (rtl_for_decl_location): Likewise.
4691 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4692 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4693 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4694 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4695 (expand_expr_real_1): Likewise.
4696 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4697 (pad_below): Likewise.
4698 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4699 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4700 * ira.c (get_subreg_tracking_sizes): Likewise.
4701 * ira-build.c (ira_create_allocno_objects): Likewise.
4702 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4703 (ira_sort_regnos_for_alter_reg): Likewise.
4704 * ira-costs.c (record_operand_costs): Likewise.
4705 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4706 (resolve_simple_move): Likewise.
4707 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4708 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4709 (lra_constraints): Likewise.
4710 (CONST_POOL_OK_P): Reject variable-sized modes.
4711 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4712 (add_pseudo_to_slot, lra_spill): Likewise.
4713 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4714 * optabs-query.c (get_best_extraction_insn): Likewise.
4715 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4716 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4717 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4718 * recog.c (offsettable_address_addr_space_p): Likewise.
4719 * regcprop.c (maybe_mode_change): Likewise.
4720 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4721 * regrename.c (build_def_use): Likewise.
4722 * regstat.c (dump_reg_info): Likewise.
4723 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4724 (find_reloads, find_reloads_subreg_address): Likewise.
4725 * reload1.c (eliminate_regs_1): Likewise.
4726 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4727 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4728 (simplify_binary_operation_1, simplify_subreg): Likewise.
4729 * targhooks.c (default_function_arg_padding): Likewise.
4730 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4731 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4732 (verify_gimple_assign_ternary): Likewise.
4733 * tree-inline.c (estimate_move_cost): Likewise.
4734 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4735 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4736 (get_address_cost_ainc): Likewise.
4737 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4738 (vect_supportable_dr_alignment): Likewise.
4739 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4740 (vectorizable_reduction): Likewise.
4741 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4742 (vectorizable_operation, vectorizable_load): Likewise.
4743 * tree.c (build_same_sized_truth_vector_type): Likewise.
4744 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4745 * var-tracking.c (emit_note_insn_var_location): Likewise.
4746 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4747 (ADDR_VEC_ALIGN): Likewise.
4749 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4750 Alan Hayward <alan.hayward@arm.com>
4751 David Sherwood <david.sherwood@arm.com>
4753 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4755 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4756 or if measurement_type is polynomial.
4757 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4758 * combine.c (make_extraction): Likewise.
4759 * dse.c (find_shift_sequence): Likewise.
4760 * dwarf2out.c (mem_loc_descriptor): Likewise.
4761 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4762 (extract_bit_field, extract_low_bits): Likewise.
4763 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4764 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4765 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4766 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4767 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4768 * reload.c (find_reloads): Likewise.
4769 * reload1.c (alter_reg): Likewise.
4770 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4771 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4772 * tree-if-conv.c (predicate_mem_writes): Likewise.
4773 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4774 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4775 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4776 * valtrack.c (dead_debug_insert_temp): Likewise.
4777 * varasm.c (mergeable_constant_section): Likewise.
4778 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4780 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4781 Alan Hayward <alan.hayward@arm.com>
4782 David Sherwood <david.sherwood@arm.com>
4784 * expr.c (expand_assignment): Cope with polynomial mode sizes
4785 when assigning to a CONCAT.
4787 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4788 Alan Hayward <alan.hayward@arm.com>
4789 David Sherwood <david.sherwood@arm.com>
4791 * machmode.h (mode_precision): Change from unsigned short to
4793 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4795 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4796 or if measurement_type is not polynomial.
4797 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4798 in which the mode is already known to be a scalar_int_mode.
4799 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4800 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4802 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4803 for GET_MODE_PRECISION.
4804 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4805 for GET_MODE_PRECISION.
4806 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4808 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4809 (expand_field_assignment, make_extraction): Likewise.
4810 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4811 (get_last_value): Likewise.
4812 * convert.c (convert_to_integer_1): Likewise.
4813 * cse.c (cse_insn): Likewise.
4814 * expr.c (expand_expr_real_1): Likewise.
4815 * lra-constraints.c (simplify_operand_subreg): Likewise.
4816 * optabs-query.c (can_atomic_load_p): Likewise.
4817 * optabs.c (expand_atomic_load): Likewise.
4818 (expand_atomic_store): Likewise.
4819 * ree.c (combine_reaching_defs): Likewise.
4820 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4821 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4822 * tree.h (type_has_mode_precision_p): Likewise.
4823 * ubsan.c (instrument_si_overflow): Likewise.
4825 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4826 Alan Hayward <alan.hayward@arm.com>
4827 David Sherwood <david.sherwood@arm.com>
4829 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4830 polynomial numbers of units.
4831 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4832 (valid_vector_subparts_p): New function.
4833 (build_vector_type): Remove temporary shim and take the number
4834 of units as a poly_uint64 rather than an int.
4835 (build_opaque_vector_type): Take the number of units as a
4836 poly_uint64 rather than an int.
4837 * tree.c (build_vector_from_ctor): Handle polynomial
4838 TYPE_VECTOR_SUBPARTS.
4839 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4840 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4841 (build_vector_from_val): If the number of units is variable,
4842 use build_vec_duplicate_cst for constant operands and
4843 VEC_DUPLICATE_EXPR otherwise.
4844 (make_vector_type): Remove temporary is_constant ().
4845 (build_vector_type, build_opaque_vector_type): Take the number of
4846 units as a poly_uint64 rather than an int.
4847 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4849 * cfgexpand.c (expand_debug_expr): Likewise.
4850 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4851 (store_constructor, expand_expr_real_1): Likewise.
4852 (const_scalar_mask_from_tree): Likewise.
4853 * fold-const-call.c (fold_const_reduction): Likewise.
4854 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4855 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4856 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4857 (fold_relational_const): Likewise.
4858 (native_interpret_vector): Likewise. Change the size from an
4859 int to an unsigned int.
4860 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4861 TYPE_VECTOR_SUBPARTS.
4862 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4863 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4864 duplicating a non-constant operand into a variable-length vector.
4865 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4866 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4867 * ipa-icf.c (sem_variable::equals): Likewise.
4868 * match.pd: Likewise.
4869 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4870 * print-tree.c (print_node): Likewise.
4871 * stor-layout.c (layout_type): Likewise.
4872 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4873 * tree-cfg.c (verify_gimple_comparison): Likewise.
4874 (verify_gimple_assign_binary): Likewise.
4875 (verify_gimple_assign_ternary): Likewise.
4876 (verify_gimple_assign_single): Likewise.
4877 * tree-pretty-print.c (dump_generic_node): Likewise.
4878 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4879 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4880 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4881 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4882 (vect_shift_permute_load_chain): Likewise.
4883 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4884 (expand_vector_condition, optimize_vector_constructor): Likewise.
4885 (lower_vec_perm, get_compute_type): Likewise.
4886 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4887 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4888 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4889 (vect_recog_mask_conversion_pattern): Likewise.
4890 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4891 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4892 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4893 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4894 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4895 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4896 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4897 (supportable_widening_operation): Likewise.
4898 (supportable_narrowing_operation): Likewise.
4899 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4901 * varasm.c (output_constant): Likewise.
4903 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4904 Alan Hayward <alan.hayward@arm.com>
4905 David Sherwood <david.sherwood@arm.com>
4907 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4908 so that both the length == 3 and length != 3 cases set up their
4909 own permute vectors. Add comments explaining why we know the
4910 number of elements is constant.
4911 (vect_permute_load_chain): Likewise.
4913 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4914 Alan Hayward <alan.hayward@arm.com>
4915 David Sherwood <david.sherwood@arm.com>
4917 * machmode.h (mode_nunits): Change from unsigned char to
4919 (ONLY_FIXED_SIZE_MODES): New macro.
4920 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4921 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4922 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4924 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4925 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4926 or if measurement_type is not polynomial.
4927 * genmodes.c (ZERO_COEFFS): New macro.
4928 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4930 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4931 Use ZERO_COEFFS when emitting initializers.
4932 * data-streamer.h (bp_pack_poly_value): New function.
4933 (bp_unpack_poly_value): Likewise.
4934 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4935 for GET_MODE_NUNITS.
4936 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4937 for GET_MODE_NUNITS.
4938 * tree.c (make_vector_type): Remove temporary shim and make
4939 the real function take the number of units as a poly_uint64
4941 (build_vector_type_for_mode): Handle polynomial nunits.
4942 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4943 * emit-rtl.c (const_vec_series_p_1): Likewise.
4944 (gen_rtx_CONST_VECTOR): Likewise.
4945 * fold-const.c (test_vec_duplicate_folding): Likewise.
4946 * genrecog.c (validate_pattern): Likewise.
4947 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4948 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4949 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4950 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4951 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4952 * rtlanal.c (subreg_get_info): Likewise.
4953 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4954 (vect_grouped_load_supported): Likewise.
4955 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4956 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4957 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4958 (simplify_const_unary_operation, simplify_binary_operation_1)
4959 (simplify_const_binary_operation, simplify_ternary_operation)
4960 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4961 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4962 instead of CONST_VECTOR_NUNITS.
4963 * varasm.c (output_constant_pool_2): Likewise.
4964 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4965 explicit-encoded elements in the XVEC for variable-length vectors.
4967 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4969 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4971 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4972 Alan Hayward <alan.hayward@arm.com>
4973 David Sherwood <david.sherwood@arm.com>
4975 * coretypes.h (fixed_size_mode): Declare.
4976 (fixed_size_mode_pod): New typedef.
4977 * builtins.h (target_builtins::x_apply_args_mode)
4978 (target_builtins::x_apply_result_mode): Change type to
4979 fixed_size_mode_pod.
4980 * builtins.c (apply_args_size, apply_result_size, result_vector)
4981 (expand_builtin_apply_args_1, expand_builtin_apply)
4982 (expand_builtin_return): Update accordingly.
4984 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4986 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4987 * cselib.c (cselib_hash_rtx): Likewise.
4988 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4989 CONST_VECTOR encoding.
4991 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4992 Jeff Law <law@redhat.com>
4995 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4996 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4997 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4998 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5001 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5002 explicitly probe *sp in a noreturn function if there were any callee
5003 register saves or frame pointer is needed.
5005 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5008 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5009 BLKmode for ternary, binary or unary expressions.
5012 * var-tracking.c (delete_vta_debug_insn): New inline function.
5013 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5014 insns from get_insns () to NULL instead of each bb separately.
5015 Use delete_vta_debug_insn. No longer static.
5016 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5017 delete_vta_debug_insns callers.
5018 * rtl.h (delete_vta_debug_insns): Declare.
5019 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5020 instead of variable_tracking_main.
5022 2018-01-03 Martin Sebor <msebor@redhat.com>
5024 PR tree-optimization/83603
5025 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5026 arguments past the endof the argument list in functions declared
5027 without a prototype.
5028 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5029 Avoid checking when arguments are null.
5031 2018-01-03 Martin Sebor <msebor@redhat.com>
5034 * doc/extend.texi (attribute const): Fix a typo.
5035 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5036 issuing -Wsuggest-attribute for void functions.
5038 2018-01-03 Martin Sebor <msebor@redhat.com>
5040 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5041 offset_int::from instead of wide_int::to_shwi.
5042 (maybe_diag_overlap): Remove assertion.
5043 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5044 * gimple-ssa-sprintf.c (format_directive): Same.
5045 (parse_directive): Same.
5046 (sprintf_dom_walker::compute_format_length): Same.
5047 (try_substitute_return_value): Same.
5049 2018-01-03 Jeff Law <law@redhat.com>
5052 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5053 non-constant residual for zero at runtime and avoid probing in
5054 that case. Reorganize code for trailing problem to mirror handling
5057 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5059 PR tree-optimization/83501
5060 * tree-ssa-strlen.c (get_string_cst): New.
5061 (handle_char_store): Call get_string_cst.
5063 2018-01-03 Martin Liska <mliska@suse.cz>
5065 PR tree-optimization/83593
5066 * tree-ssa-strlen.c: Include tree-cfg.h.
5067 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5068 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5069 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5071 (strlen_dom_walker::before_dom_children): Call
5072 gimple_purge_dead_eh_edges. Dump tranformation with details
5074 (strlen_dom_walker::before_dom_children): Update call by adding
5075 new argument cleanup_eh.
5076 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5078 2018-01-03 Martin Liska <mliska@suse.cz>
5081 * cif-code.def (VARIADIC_THUNK): New enum value.
5082 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5085 2018-01-03 Jan Beulich <jbeulich@suse.com>
5087 * sse.md (mov<mode>_internal): Tighten condition for when to use
5088 vmovdqu<ssescalarsize> for TI and OI modes.
5090 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5092 Update copyright years.
5094 2018-01-03 Martin Liska <mliska@suse.cz>
5097 * ipa-visibility.c (function_and_variable_visibility): Skip
5098 functions with noipa attribure.
5100 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5102 * gcc.c (process_command): Update copyright notice dates.
5103 * gcov-dump.c (print_version): Ditto.
5104 * gcov.c (print_version): Ditto.
5105 * gcov-tool.c (print_version): Ditto.
5106 * gengtype.c (create_file): Ditto.
5107 * doc/cpp.texi: Bump @copying's copyright year.
5108 * doc/cppinternals.texi: Ditto.
5109 * doc/gcc.texi: Ditto.
5110 * doc/gccint.texi: Ditto.
5111 * doc/gcov.texi: Ditto.
5112 * doc/install.texi: Ditto.
5113 * doc/invoke.texi: Ditto.
5115 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5117 * vector-builder.h (vector_builder::m_full_nelts): Change from
5118 unsigned int to poly_uint64.
5119 (vector_builder::full_nelts): Update prototype accordingly.
5120 (vector_builder::new_vector): Likewise.
5121 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5122 (vector_builder::operator ==): Likewise.
5123 (vector_builder::finalize): Likewise.
5124 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5125 Take the number of elements as a poly_uint64 rather than an
5127 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5128 from unsigned int to poly_uint64.
5129 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5130 (vec_perm_indices::new_vector): Likewise.
5131 (vec_perm_indices::length): Likewise.
5132 (vec_perm_indices::nelts_per_input): Likewise.
5133 (vec_perm_indices::input_nelts): Likewise.
5134 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5135 number of elements per input as a poly_uint64 rather than an
5136 unsigned int. Use the original encoding for variable-length
5137 vectors, rather than clamping each individual element.
5138 For the second and subsequent elements in each pattern,
5139 clamp the step and base before clamping their sum.
5140 (vec_perm_indices::series_p): Handle polynomial element counts.
5141 (vec_perm_indices::all_in_range_p): Likewise.
5142 (vec_perm_indices_to_tree): Likewise.
5143 (vec_perm_indices_to_rtx): Likewise.
5144 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5145 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5146 (tree_vector_builder::new_binary_operation): Handle polynomial
5147 element counts. Return false if we need to know the number
5148 of elements at compile time.
5149 * fold-const.c (fold_vec_perm): Punt if the number of elements
5150 isn't known at compile time.
5152 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5154 * vec-perm-indices.h (vec_perm_builder): Change element type
5155 from HOST_WIDE_INT to poly_int64.
5156 (vec_perm_indices::element_type): Update accordingly.
5157 (vec_perm_indices::clamp): Handle polynomial element_types.
5158 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5159 (vec_perm_indices::all_in_range_p): Likewise.
5160 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5162 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5163 polynomial vec_perm_indices element types.
5164 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5165 * fold-const.c (fold_vec_perm): Likewise.
5166 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5167 * tree-vect-generic.c (lower_vec_perm): Likewise.
5168 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5169 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5170 element type to HOST_WIDE_INT.
5172 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5173 Alan Hayward <alan.hayward@arm.com>
5174 David Sherwood <david.sherwood@arm.com>
5176 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5177 rather than an int. Use plus_constant.
5178 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5179 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5181 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5182 Alan Hayward <alan.hayward@arm.com>
5183 David Sherwood <david.sherwood@arm.com>
5185 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5186 a HOST_WIDE_INT to a poly_int64.
5188 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5189 Alan Hayward <alan.hayward@arm.com>
5190 David Sherwood <david.sherwood@arm.com>
5192 * calls.c (load_register_parameters): Cope with polynomial
5193 mode sizes. Require a constant size for BLKmode parameters
5194 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5195 forces a parameter to be padded at the lsb end in order to
5196 fill a complete number of words, require the parameter size
5197 to be ordered wrt UNITS_PER_WORD.
5199 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5200 Alan Hayward <alan.hayward@arm.com>
5201 David Sherwood <david.sherwood@arm.com>
5203 * reload1.c (spill_stack_slot_width): Change element type
5204 from unsigned int to poly_uint64_pod.
5205 (alter_reg): Treat mode sizes as polynomial.
5207 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5208 Alan Hayward <alan.hayward@arm.com>
5209 David Sherwood <david.sherwood@arm.com>
5211 * reload.c (complex_word_subreg_p): New function.
5212 (reload_inner_reg_of_subreg, push_reload): Use it.
5214 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5215 Alan Hayward <alan.hayward@arm.com>
5216 David Sherwood <david.sherwood@arm.com>
5218 * lra-constraints.c (process_alt_operands): Reject matched
5219 operands whose sizes aren't ordered.
5220 (match_reload): Refer to this check here.
5222 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5223 Alan Hayward <alan.hayward@arm.com>
5224 David Sherwood <david.sherwood@arm.com>
5226 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5227 that the mode size is in the set {1, 2, 4, 8, 16}.
5229 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5230 Alan Hayward <alan.hayward@arm.com>
5231 David Sherwood <david.sherwood@arm.com>
5233 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5234 Use plus_constant instead of gen_rtx_PLUS.
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 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5241 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5242 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5243 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5244 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5245 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5246 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5247 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5248 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5249 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5251 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5252 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5253 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5254 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5255 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5256 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5257 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5258 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5259 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5260 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5262 * expr.c (emit_move_resolve_push): Treat the input and result
5263 of PUSH_ROUNDING as a poly_int64.
5264 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5265 (emit_push_insn): Likewise.
5266 * lra-eliminations.c (mark_not_eliminable): Likewise.
5267 * recog.c (push_operand): Likewise.
5268 * reload1.c (elimination_effects): Likewise.
5269 * rtlanal.c (nonzero_bits1): Likewise.
5270 * calls.c (store_one_arg): Likewise. Require the padding to be
5271 known at compile time.
5273 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5274 Alan Hayward <alan.hayward@arm.com>
5275 David Sherwood <david.sherwood@arm.com>
5277 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5278 Use plus_constant instead of gen_rtx_PLUS.
5280 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5281 Alan Hayward <alan.hayward@arm.com>
5282 David Sherwood <david.sherwood@arm.com>
5284 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5287 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5288 Alan Hayward <alan.hayward@arm.com>
5289 David Sherwood <david.sherwood@arm.com>
5291 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5292 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5293 via stack temporaries. Treat the mode size as polynomial too.
5295 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5296 Alan Hayward <alan.hayward@arm.com>
5297 David Sherwood <david.sherwood@arm.com>
5299 * expr.c (expand_expr_real_2): When handling conversions involving
5300 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5301 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5302 as a poly_uint64 too.
5304 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5305 Alan Hayward <alan.hayward@arm.com>
5306 David Sherwood <david.sherwood@arm.com>
5308 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5310 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5311 Alan Hayward <alan.hayward@arm.com>
5312 David Sherwood <david.sherwood@arm.com>
5314 * combine.c (can_change_dest_mode): Handle polynomial
5315 REGMODE_NATURAL_SIZE.
5316 * expmed.c (store_bit_field_1): Likewise.
5317 * expr.c (store_constructor): Likewise.
5318 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5319 and polynomial REGMODE_NATURAL_SIZE.
5320 (gen_lowpart_common): Likewise.
5321 * reginfo.c (record_subregs_of_mode): Likewise.
5322 * rtlanal.c (read_modify_subreg_p): Likewise.
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 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5329 numbers of elements.
5331 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5332 Alan Hayward <alan.hayward@arm.com>
5333 David Sherwood <david.sherwood@arm.com>
5335 * match.pd: Cope with polynomial numbers of vector elements.
5337 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5338 Alan Hayward <alan.hayward@arm.com>
5339 David Sherwood <david.sherwood@arm.com>
5341 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5342 in a POINTER_PLUS_EXPR.
5344 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5345 Alan Hayward <alan.hayward@arm.com>
5346 David Sherwood <david.sherwood@arm.com>
5348 * omp-simd-clone.c (simd_clone_subparts): New function.
5349 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5350 (ipa_simd_modify_function_body): Likewise.
5352 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5353 Alan Hayward <alan.hayward@arm.com>
5354 David Sherwood <david.sherwood@arm.com>
5356 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5357 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5358 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5359 (expand_vector_condition, vector_element): Likewise.
5360 (subparts_gt): New function.
5361 (get_compute_type): Use subparts_gt.
5362 (count_type_subparts): Delete.
5363 (expand_vector_operations_1): Use subparts_gt instead of
5364 count_type_subparts.
5366 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5367 Alan Hayward <alan.hayward@arm.com>
5368 David Sherwood <david.sherwood@arm.com>
5370 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5371 (vect_compile_time_alias): ...this new function. Do the calculation
5372 on poly_ints rather than trees.
5373 (vect_prune_runtime_alias_test_list): Update call accordingly.
5375 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5376 Alan Hayward <alan.hayward@arm.com>
5377 David Sherwood <david.sherwood@arm.com>
5379 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5381 (vect_schedule_slp_instance): Likewise.
5383 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5384 Alan Hayward <alan.hayward@arm.com>
5385 David Sherwood <david.sherwood@arm.com>
5387 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5388 constant and extern definitions for variable-length vectors.
5389 (vect_get_constant_vectors): Note that the number of units
5390 is known to be constant.
5392 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5393 Alan Hayward <alan.hayward@arm.com>
5394 David Sherwood <david.sherwood@arm.com>
5396 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5397 of units as polynomial. Choose between WIDE and NARROW based
5400 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5401 Alan Hayward <alan.hayward@arm.com>
5402 David Sherwood <david.sherwood@arm.com>
5404 * tree-vect-stmts.c (simd_clone_subparts): New function.
5405 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5407 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5408 Alan Hayward <alan.hayward@arm.com>
5409 David Sherwood <david.sherwood@arm.com>
5411 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5412 vectors as polynomial. Use build_index_vector for
5415 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5416 Alan Hayward <alan.hayward@arm.com>
5417 David Sherwood <david.sherwood@arm.com>
5419 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5420 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5421 for variable-length vectors.
5422 (vectorizable_mask_load_store): Treat the number of units as
5423 polynomial, asserting that it is constant if the condition has
5424 already been enforced.
5425 (vectorizable_store, vectorizable_load): Likewise.
5427 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5428 Alan Hayward <alan.hayward@arm.com>
5429 David Sherwood <david.sherwood@arm.com>
5431 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5432 of units as polynomial. Punt if we can't tell at compile time
5433 which vector contains the final result.
5435 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5436 Alan Hayward <alan.hayward@arm.com>
5437 David Sherwood <david.sherwood@arm.com>
5439 * tree-vect-loop.c (vectorizable_induction): Treat the number
5440 of units as polynomial. Punt on SLP inductions. Use an integer
5441 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5442 cast of such a series for variable-length floating-point
5445 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5446 Alan Hayward <alan.hayward@arm.com>
5447 David Sherwood <david.sherwood@arm.com>
5449 * tree.h (build_index_vector): Declare.
5450 * tree.c (build_index_vector): New function.
5451 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5452 of units as polynomial, forcibly converting it to a constant if
5453 vectorizable_reduction has already enforced the condition.
5454 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5455 to create a {1,2,3,...} vector.
5456 (vectorizable_reduction): Treat the number of units as polynomial.
5457 Choose vectype_in based on the largest scalar element size rather
5458 than the smallest number of units. Enforce the restrictions
5461 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5462 Alan Hayward <alan.hayward@arm.com>
5463 David Sherwood <david.sherwood@arm.com>
5465 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5466 number of units as polynomial.
5468 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5469 Alan Hayward <alan.hayward@arm.com>
5470 David Sherwood <david.sherwood@arm.com>
5472 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5473 * target.def (autovectorize_vector_sizes): Return the vector sizes
5474 by pointer, using vector_sizes rather than a bitmask.
5475 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5476 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5477 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5479 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5480 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5481 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5482 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5483 * omp-general.c (omp_max_vf): Likewise.
5484 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5485 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5486 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5487 * tree-vect-slp.c (vect_slp_bb): Likewise.
5488 * doc/tm.texi: Regenerate.
5489 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5491 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5492 the vector size as a poly_uint64 rather than an unsigned int.
5493 (current_vector_size): Change from an unsigned int to a poly_uint64.
5494 (get_vectype_for_scalar_type): Update accordingly.
5495 * tree.h (build_truth_vector_type): Take the size and number of
5496 units as a poly_uint64 rather than an unsigned int.
5497 (build_vector_type): Add a temporary overload that takes
5498 the number of units as a poly_uint64 rather than an unsigned int.
5499 * tree.c (make_vector_type): Likewise.
5500 (build_truth_vector_type): Take the number of units as a poly_uint64
5501 rather than an unsigned int.
5503 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5504 Alan Hayward <alan.hayward@arm.com>
5505 David Sherwood <david.sherwood@arm.com>
5507 * target.def (get_mask_mode): Take the number of units and length
5508 as poly_uint64s rather than unsigned ints.
5509 * targhooks.h (default_get_mask_mode): Update accordingly.
5510 * targhooks.c (default_get_mask_mode): Likewise.
5511 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5512 * doc/tm.texi: Regenerate.
5514 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5515 Alan Hayward <alan.hayward@arm.com>
5516 David Sherwood <david.sherwood@arm.com>
5518 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5519 * omp-general.c (omp_max_vf): Likewise.
5520 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5521 (expand_omp_simd): Handle polynomial safelen.
5522 * omp-low.c (omplow_simd_context): Add a default constructor.
5523 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5524 (lower_rec_simd_input_clauses): Update accordingly.
5525 (lower_rec_input_clauses): Likewise.
5527 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5528 Alan Hayward <alan.hayward@arm.com>
5529 David Sherwood <david.sherwood@arm.com>
5531 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5532 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5533 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5534 (vect_analyze_slp_cost): Likewise.
5535 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5536 (vect_model_load_cost): Likewise.
5538 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5539 Alan Hayward <alan.hayward@arm.com>
5540 David Sherwood <david.sherwood@arm.com>
5542 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5543 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5544 from an unsigned int * to a poly_uint64_pod *.
5545 (calculate_unrolling_factor): New function.
5546 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5548 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5549 Alan Hayward <alan.hayward@arm.com>
5550 David Sherwood <david.sherwood@arm.com>
5552 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5553 from an unsigned int to a poly_uint64.
5554 (_loop_vec_info::slp_unrolling_factor): Likewise.
5555 (_loop_vec_info::vectorization_factor): Change from an int
5557 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5558 (vect_get_num_vectors): New function.
5559 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5560 (vect_get_num_copies): Use vect_get_num_vectors.
5561 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5562 to an unsigned int *.
5563 (vect_analyze_data_refs): Change min_vf from an int * to a
5565 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5566 than an unsigned HOST_WIDE_INT.
5567 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5568 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5569 to an unsigned int *.
5570 (vect_analyze_data_ref_dependences): Likewise.
5571 (vect_compute_data_ref_alignment): Handle polynomial vf.
5572 (vect_enhance_data_refs_alignment): Likewise.
5573 (vect_prune_runtime_alias_test_list): Likewise.
5574 (vect_shift_permute_load_chain): Likewise.
5575 (vect_supportable_dr_alignment): Likewise.
5576 (dependence_distance_ge_vf): Take the vectorization factor as a
5577 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5578 (vect_analyze_data_refs): Change min_vf from an int * to a
5580 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5581 vfm1 as a poly_uint64 rather than an int. Make the same change
5582 for the returned bound_scalar.
5583 (vect_gen_vector_loop_niters): Handle polynomial vf.
5584 (vect_do_peeling): Likewise. Update call to
5585 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5586 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5588 * tree-vect-loop.c (vect_determine_vectorization_factor)
5589 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5590 (vect_get_known_peeling_cost): Likewise.
5591 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5592 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5593 (vect_transform_loop): Likewise. Use the lowest possible VF when
5594 updating the upper bounds of the loop.
5595 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5597 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5598 polynomial unroll factors.
5599 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5600 (vect_make_slp_decision): Likewise.
5601 (vect_supported_load_permutation_p): Likewise, and polynomial
5603 (vect_analyze_slp_cost): Handle polynomial vf.
5604 (vect_slp_analyze_node_operations): Likewise.
5605 (vect_slp_analyze_bb_1): Likewise.
5606 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5607 than an unsigned HOST_WIDE_INT.
5608 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5609 (vectorizable_load): Handle polynomial vf.
5610 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5612 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5614 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5615 Alan Hayward <alan.hayward@arm.com>
5616 David Sherwood <david.sherwood@arm.com>
5618 * match.pd: Handle bit operations involving three constants
5619 and try to fold one pair.
5621 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5623 * tree-vect-loop-manip.c: Include gimple-fold.h.
5624 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5625 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5626 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5627 Add a path that uses a step of VF instead of 1, but disable it
5629 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5630 and niters_no_overflow parameters. Update calls to
5631 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5632 Create a new SSA name if the latter choses to use a ste other
5633 than zero, and return it via niters_vector_mult_vf_var.
5634 * tree-vect-loop.c (vect_transform_loop): Update calls to
5635 vect_do_peeling, vect_gen_vector_loop_niters and
5636 slpeel_make_loop_iterate_ntimes.
5637 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5638 (vect_gen_vector_loop_niters): Update declarations after above changes.
5640 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5642 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5643 128-bit round to integer instructions.
5644 (ceil<mode>2): Likewise.
5645 (btrunc<mode>2): Likewise.
5646 (round<mode>2): Likewise.
5648 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5650 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5651 unaligned VSX load/store on P8/P9.
5652 (expand_block_clear): Allow the use of unaligned VSX
5653 load/store on P8/P9.
5655 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5657 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5659 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5660 swap associated with both a load and a store.
5662 2018-01-02 Andrew Waterman <andrew@sifive.com>
5664 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5665 * config/riscv/riscv.md (clear_cache): Use it.
5667 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5669 * web.c: Remove out-of-date comment.
5671 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5673 * expr.c (fixup_args_size_notes): Check that any existing
5674 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5675 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5676 (emit_single_push_insn): ...here.
5678 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5680 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5681 (const_vector_encoded_nelts): New function.
5682 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5683 (const_vector_int_elt, const_vector_elt): Declare.
5684 * emit-rtl.c (const_vector_int_elt_1): New function.
5685 (const_vector_elt): Likewise.
5686 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5687 of CONST_VECTOR_ELT.
5689 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5691 * expr.c: Include rtx-vector-builder.h.
5692 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5693 directly on the tree encoding.
5694 (const_vector_from_tree): Likewise.
5695 * optabs.c: Include rtx-vector-builder.h.
5696 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5697 sequence of "u" values.
5698 * vec-perm-indices.c: Include rtx-vector-builder.h.
5699 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5700 directly on the vec_perm_indices encoding.
5702 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5704 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5705 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5706 * rtx-vector-builder.h: New file.
5707 * rtx-vector-builder.c: Likewise.
5708 * rtl.h (rtx_def::u2): Add a const_vector field.
5709 (CONST_VECTOR_NPATTERNS): New macro.
5710 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5711 (CONST_VECTOR_DUPLICATE_P): Likewise.
5712 (CONST_VECTOR_STEPPED_P): Likewise.
5713 (CONST_VECTOR_ENCODED_ELT): Likewise.
5714 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5715 (unwrap_const_vec_duplicate): Likewise.
5716 (const_vec_series_p): Check for a non-duplicated vector encoding.
5717 Say that the function only returns true for integer vectors.
5718 * emit-rtl.c: Include rtx-vector-builder.h.
5719 (gen_const_vec_duplicate_1): Delete.
5720 (gen_const_vector): Call gen_const_vec_duplicate instead of
5721 gen_const_vec_duplicate_1.
5722 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5723 (gen_const_vec_duplicate): Use rtx_vector_builder.
5724 (gen_const_vec_series): Likewise.
5725 (gen_rtx_CONST_VECTOR): Likewise.
5726 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5727 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5728 Build a new vector rather than modifying a CONST_VECTOR in-place.
5729 (handle_special_swappables): Update call accordingly.
5730 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5731 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5732 Build a new vector rather than modifying a CONST_VECTOR in-place.
5733 (handle_special_swappables): Update call accordingly.
5735 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5737 * simplify-rtx.c (simplify_const_binary_operation): Use
5738 CONST_VECTOR_ELT instead of XVECEXP.
5740 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5742 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5743 the selector elements to be different from the data elements
5744 if the selector is a VECTOR_CST.
5745 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5746 ssizetype for the selector.
5748 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5750 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5751 before testing each element individually.
5752 * tree-vect-generic.c (lower_vec_perm): Likewise.
5754 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5756 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5757 * selftest-run-tests.c (selftest::run_tests): Call it.
5758 * vector-builder.h (vector_builder::operator ==): New function.
5759 (vector_builder::operator !=): Likewise.
5760 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5761 (vec_perm_indices::all_from_input_p): New function.
5762 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5763 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5764 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5765 instead of reading the VECTOR_CST directly. Detect whether both
5766 vector inputs are the same before constructing the vec_perm_indices,
5767 and update the number of inputs argument accordingly. Use the
5768 utility functions added above. Only construct sel2 if we need to.
5770 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5772 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5773 the broadcast of the low byte.
5774 (expand_mult_highpart): Use an explicit encoding for the permutes.
5775 * optabs-query.c (can_mult_highpart_p): Likewise.
5776 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5777 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5778 (vectorizable_bswap): Likewise.
5779 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5780 explicit encoding for the power-of-2 permutes.
5781 (vect_permute_store_chain): Likewise.
5782 (vect_grouped_load_supported): Likewise.
5783 (vect_permute_load_chain): Likewise.
5785 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5787 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5788 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5789 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5790 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5791 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5792 (vect_gen_perm_mask_any): Likewise.
5794 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5796 * int-vector-builder.h: New file.
5797 * vec-perm-indices.h: Include int-vector-builder.h.
5798 (vec_perm_indices): Redefine as an int_vector_builder.
5799 (auto_vec_perm_indices): Delete.
5800 (vec_perm_builder): Redefine as a stand-alone class.
5801 (vec_perm_indices::vec_perm_indices): New function.
5802 (vec_perm_indices::clamp): Likewise.
5803 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5804 (vec_perm_indices::new_vector): New function.
5805 (vec_perm_indices::new_expanded_vector): Update for new
5806 vec_perm_indices class.
5807 (vec_perm_indices::rotate_inputs): New function.
5808 (vec_perm_indices::all_in_range_p): Operate directly on the
5809 encoded form, without computing elided elements.
5810 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5811 encoding. Update for new vec_perm_indices class.
5812 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5813 the given vec_perm_builder.
5814 (expand_vec_perm_var): Update vec_perm_builder constructor.
5815 (expand_mult_highpart): Use vec_perm_builder instead of
5816 auto_vec_perm_indices.
5817 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5818 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5819 or double series encoding as appropriate.
5820 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5821 vec_perm_indices instead of auto_vec_perm_indices.
5822 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5823 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5824 (vect_permute_store_chain): Likewise.
5825 (vect_grouped_load_supported): Likewise.
5826 (vect_permute_load_chain): Likewise.
5827 (vect_shift_permute_load_chain): Likewise.
5828 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5829 (vect_transform_slp_perm_load): Likewise.
5830 (vect_schedule_slp_instance): Likewise.
5831 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5832 (vectorizable_mask_load_store): Likewise.
5833 (vectorizable_bswap): Likewise.
5834 (vectorizable_store): Likewise.
5835 (vectorizable_load): Likewise.
5836 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5837 vec_perm_indices instead of auto_vec_perm_indices. Use
5838 tree_to_vec_perm_builder to read the vector from a tree.
5839 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5840 vec_perm_builder instead of a vec_perm_indices.
5841 (have_whole_vector_shift): Use vec_perm_builder and
5842 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5843 truncation to calc_vec_perm_mask_for_shift.
5844 (vect_create_epilog_for_reduction): Likewise.
5845 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5846 from auto_vec_perm_indices to vec_perm_indices.
5847 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5848 instead of changing individual elements.
5849 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5850 the vector in d.perm.
5851 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5852 from auto_vec_perm_indices to vec_perm_indices.
5853 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5854 instead of changing individual elements.
5855 (arm_vectorize_vec_perm_const): Use new_vector to install
5856 the vector in d.perm.
5857 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5858 Update vec_perm_builder constructor.
5859 (rs6000_expand_interleave): Likewise.
5860 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5861 (rs6000_expand_interleave): Likewise.
5863 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5865 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5866 to qimode could truncate the indices.
5867 * optabs.c (expand_vec_perm_var): Likewise.
5869 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5871 * Makefile.in (OBJS): Add vec-perm-indices.o.
5872 * vec-perm-indices.h: New file.
5873 * vec-perm-indices.c: Likewise.
5874 * target.h (vec_perm_indices): Replace with a forward class
5876 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5877 * optabs.h: Include vec-perm-indices.h.
5878 (expand_vec_perm): Delete.
5879 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5880 (expand_vec_perm_const): Declare.
5881 * target.def (vec_perm_const_ok): Replace with...
5882 (vec_perm_const): ...this new hook.
5883 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5884 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5885 * doc/tm.texi: Regenerate.
5886 * optabs.def (vec_perm_const): Delete.
5887 * doc/md.texi (vec_perm_const): Likewise.
5888 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5889 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5890 expand_vec_perm for constant permutation vectors. Assert that
5891 the mode of variable permutation vectors is the integer equivalent
5892 of the mode that is being permuted.
5893 * optabs-query.h (selector_fits_mode_p): Declare.
5894 * optabs-query.c: Include vec-perm-indices.h.
5895 (selector_fits_mode_p): New function.
5896 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5897 is defined, instead of checking whether the vec_perm_const_optab
5898 exists. Use targetm.vectorize.vec_perm_const instead of
5899 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5900 fit in the vector mode before using a variable permute.
5901 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5902 vec_perm_indices instead of an rtx.
5903 (expand_vec_perm): Replace with...
5904 (expand_vec_perm_const): ...this new function. Take the selector
5905 as a vec_perm_indices rather than an rtx. Also take the mode of
5906 the selector. Update call to shift_amt_for_vec_perm_mask.
5907 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5908 Use vec_perm_indices::new_expanded_vector to expand the original
5909 selector into bytes. Check whether the indices fit in the vector
5910 mode before using a variable permute.
5911 (expand_vec_perm_var): Make global.
5912 (expand_mult_highpart): Use expand_vec_perm_const.
5913 * fold-const.c: Includes vec-perm-indices.h.
5914 * tree-ssa-forwprop.c: Likewise.
5915 * tree-vect-data-refs.c: Likewise.
5916 * tree-vect-generic.c: Likewise.
5917 * tree-vect-loop.c: Likewise.
5918 * tree-vect-slp.c: Likewise.
5919 * tree-vect-stmts.c: Likewise.
5920 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5922 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5923 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5924 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5925 (aarch64_vectorize_vec_perm_const): ...this new function.
5926 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5927 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5928 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5929 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5930 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5931 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5932 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5934 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5935 check for NEON modes.
5936 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5937 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5938 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5939 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5941 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5942 the old VEC_PERM_CONST conditions.
5943 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5944 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5945 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5946 (ia64_vectorize_vec_perm_const_ok): Merge into...
5947 (ia64_vectorize_vec_perm_const): ...this new function.
5948 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5949 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5950 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5951 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5952 * config/mips/mips.c (mips_expand_vec_perm_const)
5953 (mips_vectorize_vec_perm_const_ok): Merge into...
5954 (mips_vectorize_vec_perm_const): ...this new function.
5955 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5956 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5957 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5958 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5959 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5960 (rs6000_expand_vec_perm_const): Delete.
5961 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5963 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5964 (altivec_expand_vec_perm_const_le): Take each operand individually.
5965 Operate on constant selectors rather than rtxes.
5966 (altivec_expand_vec_perm_const): Likewise. Update call to
5967 altivec_expand_vec_perm_const_le.
5968 (rs6000_expand_vec_perm_const): Delete.
5969 (rs6000_vectorize_vec_perm_const_ok): Delete.
5970 (rs6000_vectorize_vec_perm_const): New function.
5971 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5972 an element count and rtx array.
5973 (rs6000_expand_extract_even): Update call accordingly.
5974 (rs6000_expand_interleave): Likewise.
5975 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5976 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5977 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5978 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5979 (rs6000_expand_vec_perm_const): Delete.
5980 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5981 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5982 (altivec_expand_vec_perm_const_le): Take each operand individually.
5983 Operate on constant selectors rather than rtxes.
5984 (altivec_expand_vec_perm_const): Likewise. Update call to
5985 altivec_expand_vec_perm_const_le.
5986 (rs6000_expand_vec_perm_const): Delete.
5987 (rs6000_vectorize_vec_perm_const_ok): Delete.
5988 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5989 reference to the SPE evmerge intructions.
5990 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5991 an element count and rtx array.
5992 (rs6000_expand_extract_even): Update call accordingly.
5993 (rs6000_expand_interleave): Likewise.
5994 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5995 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5997 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5999 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6001 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6002 vector mode and that that mode matches the mode of the data
6004 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6005 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6006 directly using expand_vec_perm_1 when forcing selectors into
6008 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6010 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6012 * optabs-query.h (can_vec_perm_p): Delete.
6013 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6014 * optabs-query.c (can_vec_perm_p): Split into...
6015 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6016 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6017 particular selector is valid.
6018 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6019 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6020 (vect_grouped_load_supported): Likewise.
6021 (vect_shift_permute_load_chain): Likewise.
6022 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6023 (vect_transform_slp_perm_load): Likewise.
6024 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6025 (vectorizable_bswap): Likewise.
6026 (vect_gen_perm_mask_checked): Likewise.
6027 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6028 implementations of variable permutation vectors into account
6029 when deciding which selector to use.
6030 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6031 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6032 with a false third argument.
6033 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6034 to test whether the constant selector is valid and can_vec_perm_var_p
6035 to test whether a variable selector is valid.
6037 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6039 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6040 * optabs-query.c (can_vec_perm_p): Likewise.
6041 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6042 instead of vec_perm_indices.
6043 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6044 (vect_gen_perm_mask_checked): Likewise,
6045 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6046 (vect_gen_perm_mask_checked): Likewise,
6048 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6050 * optabs-query.h (qimode_for_vec_perm): Declare.
6051 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6052 (qimode_for_vec_perm): ...this new function.
6053 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6055 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6057 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6058 does not have a conditional at the top.
6060 2018-01-02 Richard Biener <rguenther@suse.de>
6062 * ipa-inline.c (big_speedup_p): Fix expression.
6064 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6067 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6070 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6074 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6075 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6076 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6077 cond_taken_branch_cost 3->4.
6079 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6081 PR tree-optimization/83581
6082 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6083 TODO_cleanup_cfg if any changes have been made.
6086 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6087 convert_modes if target mode has the right side, but different mode
6091 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6092 last argument when extracting from CONCAT. If either from_real or
6093 from_imag is NULL, use expansion through memory. If result is not
6094 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6095 the parts directly to inner mode, if even that fails, use expansion
6099 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6100 check for bswap in mode rather than HImode and use that in expand_unop
6103 Copyright (C) 2018 Free Software Foundation, Inc.
6105 Copying and distribution of this file, with or without modification,
6106 are permitted in any medium without royalty provided the copyright
6107 notice and this notice are preserved.