1 2018-02-02 Julia Koval <julia.koval@intel.com>
3 * config.gcc: Add -march=icelake.
4 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
5 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
6 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
7 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
8 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
9 (processor_target_table): Add icelake.
10 (ix86_option_override_internal): Handle new PTAs.
11 (get_builtin_code_for_version): Handle icelake.
12 (M_INTEL_COREI7_ICELAKE): New.
13 (fold_builtin_cpu): Handle icelake.
14 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
15 * doc/invoke.texi: Add -march=icelake.
17 2018-02-02 Julia Koval <julia.koval@intel.com>
19 * config/i386/i386.c (ix86_option_override_internal): Change flags type
21 * wide-int-bitmask.h: New.
23 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
26 * config/i386/i386.md: Replace Pmode with word_mode in
27 builtin_setjmp_setup and builtin_longjmp to support x32.
29 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
33 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
35 (rs6000_supported_cpu_names): New static variable.
36 (linux_cpu_translation_table): Likewise.
37 (elf_platform) <cpu>: Define new static variable and use it.
38 Translate kernel AT_PLATFORM name to canonical name if needed.
39 Error if platform name is unknown.
41 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
44 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
46 2018-02-01 Jeff Law <law@redhat.com>
49 * config/i386/i386.c (release_scratch_register_on_entry): Add new
50 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
51 the scratch if RELEASE_VIA_POP is false.
52 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
53 If we have to save a temporary register, decrement SIZE appropriately.
54 Pass new arguments to release_scratch_register_on_entry.
55 (ix86_adjust_stack_and_probe): Likewise.
56 (ix86_emit_probe_stack_range): Pass new arguments to
57 release_scratch_register_on_entry.
59 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
61 PR rtl-optimization/84157
62 * combine.c (change_zero_ext): Use REG_P predicate in
63 front of HARD_REGISTER_P predicate.
65 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
67 * config/avr/avr.c (avr_option_override): Move disabling of
68 -fdelete-null-pointer-checks to...
69 * common/config/avr/avr-common.c (avr_option_optimization_table):
72 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
74 PR tree-optimization/81635
75 * tree-data-ref.c (split_constant_offset_1): For types that
76 wrap on overflow, try to use range info to prove that wrapping
79 2018-02-01 Renlin Li <renlin.li@arm.com>
82 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
84 (aarch64_register_move_cost): Likewise.
85 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
87 (REG_CLASS_NAMES): Likewise.
88 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
89 TAILCALL_ADDR_REGS. Remove IP registers.
90 * config/aarch64/aarch64.md (Ucs): Update register constraint.
92 2018-02-01 Richard Biener <rguenther@suse.de>
94 * domwalk.h (dom_walker::dom_walker): Add additional constructor
95 for specifying RPO order and allow NULL for that.
96 * domwalk.c (dom_walker::dom_walker): Likewise.
97 (dom_walker::walk): Handle NULL RPO order.
98 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
100 (rewrite_update_dom_walker): Likewise.
101 (mark_def_dom_walker): Likewise.
103 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
105 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
106 (aarch64_maybe_expand_sve_subreg_move): Declare.
107 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
108 * config/aarch64/predicates.md (aarch64_any_register_operand): New
110 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
111 that are semantically a reverse operation.
112 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
113 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
114 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
116 (aarch64_can_change_mode_class): For big-endian, forbid changes
117 between two SVE modes if they have different element sizes.
119 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
121 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
122 the TImode handling for big-endian targets.
124 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
126 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
127 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
129 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
130 Remove BSWAP handing for big-endian targets and use the form of
131 LD1RQ appropariate for the mode.
133 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
135 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
136 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
139 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
142 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
143 check for operands that need to go through aarch64_sve_reload_be.
145 2018-02-01 Jakub Jelinek <jakub@redhat.com>
147 PR tree-optimization/81661
148 PR tree-optimization/84117
149 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
150 * tree-eh.c: Include gimplify.h.
151 (find_trapping_overflow, replace_trapping_overflow,
152 rewrite_to_non_trapping_overflow): New functions.
153 * tree-vect-loop.c: Include tree-eh.h.
154 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
155 * tree-data-ref.c: Include tree-eh.h.
156 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
158 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
160 PR rtl-optimization/84123
161 * combine.c (change_zero_ext): Check if hard register satisfies
162 can_change_dest_mode before calling gen_lowpart_SUBREG.
164 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
167 * ira.c (ira_init_register_move_cost): Remove assert.
169 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
171 PR rtl-optimization/84071
172 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
173 * doc/tm.texi: Regenerate.
175 2018-01-31 Richard Biener <rguenther@suse.de>
177 PR tree-optimization/84132
178 * tree-data-ref.c (analyze_miv_subscript): Properly
179 check whether evolution_function_is_affine_multivariate_p
180 before calling gcd_of_steps_may_divide_p.
182 2018-01-31 Julia Koval <julia.koval@intel.com>
185 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
186 * config/i386/i386.md (rdpid_rex64) New.
187 (rdpid): Make 32bit only.
189 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
192 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
193 an IDENTIFIER_NODE for FUNCTION_TYPE's.
195 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
198 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
200 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
202 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
204 PR rtl-optimization/84071
205 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
206 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
208 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
210 * config/arc/arc.c (arc_handle_aux_attribute): New function.
211 (arc_attribute_table): Add 'aux' attribute.
212 (arc_in_small_data_p): Consider aux like variables.
213 (arc_is_aux_reg_p): New function.
214 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
215 (arc_get_aux_arg): New function.
216 (prepare_move_operands): Handle aux-register access.
217 (arc_handle_aux_attribute): New function.
218 * doc/extend.texi (ARC Variable attributes): Add subsection.
220 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
222 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
223 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
224 (arc_attribute_table): Add 'uncached' attribute.
225 (arc_print_operand): Print '.di' flag for uncached memory
227 (arc_in_small_data_p): Do not consider for small data the uncached
229 (arc_is_uncached_mem_p): New function.
230 * config/arc/predicates.md (compact_store_memory_operand): Check
231 for uncached memory accesses.
232 (nonvol_nonimm_operand): Likewise.
233 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
235 2018-01-31 Jakub Jelinek <jakub@redhat.com>
238 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
239 falign-loops=): Add Optimization flag.
241 2018-01-30 Jeff Law <law@redhat.com>
244 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
245 INT_REGISTERS_SAVED. Check it prior to calling
246 get_scratch_register_on_entry.
247 (ix86_adjust_stack_and_probe): Similarly.
248 (ix86_emit_probe_stack_range): Similarly.
249 (ix86_expand_prologue): Corresponding changes.
251 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
254 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
255 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
257 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
260 * lra-constraints.c (curr_insn_transform): Process AND in the
263 2018-01-30 Jakub Jelinek <jakub@redhat.com>
265 PR rtl-optimization/83986
266 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
267 dependence against last_pending_memory_flush in addition to
270 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
272 PR tree-optimization/81611
273 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
276 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
278 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
281 2018-01-30 Richard Biener <rguenther@suse.de>
282 Jakub Jelinek <jakub@redhat.com>
284 PR tree-optimization/84111
285 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
286 inner loops added during recursion, as they don't have up-to-date
289 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
292 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
293 (can_inline_edge_by_limits_p): ... here.
294 (can_early_inline_edge_p, check_callers,
295 update_caller_keys, update_callee_keys, recursive_inlining,
296 add_new_edges_to_heap, speculation_useful_p,
297 inline_small_functions,
298 inline_small_functions, flatten_function,
299 inline_to_all_callers_1): Update.
301 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
303 * profile-count.c (profile_count::combine_with_ipa_count): Handle
306 2018-01-30 Richard Biener <rguenther@suse.de>
308 PR tree-optimization/83008
309 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
310 invariant and constant vector uses in stmts when they need
313 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
316 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
317 * configure: Regenerate.
319 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
321 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
323 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
324 Use gen_rtx_REG rather than gen_lowpart.
326 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
328 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
329 rather than 0 when creating partial subregs.
331 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
333 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
336 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
339 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
340 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
341 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
342 flags. This restores the settings used before the 2017-07-24.
343 Turning off pre increment/decrement/modify allows IVOPTS to
344 optimize DF/SF loops where the index is an int.
346 2018-01-29 Richard Biener <rguenther@suse.de>
347 Kelvin Nilsen <kelvin@gcc.gnu.org>
350 * tree-vect-stmts.c (vectorizable_call): Don't call
351 targetm.vectorize_builtin_md_vectorized_function if callee is
354 2018-01-22 Carl Love <cel@us.ibm.com>
356 * doc/extend.tex: Fix typo in second arg in
357 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
359 2018-01-29 Richard Biener <rguenther@suse.de>
361 PR tree-optimization/84086
362 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
363 (flush_ssaname_freelist): When SSA names were released reset
366 2018-01-29 Richard Biener <rguenther@suse.de>
368 PR tree-optimization/84057
369 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
370 removed paths when removing edges.
372 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
374 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
375 -mfunction-return=@var{choice}.
377 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
380 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
382 (layout::print_source_line): Likewise.
383 (test_get_line_width_without_trailing_whitespace): Add test cases.
385 2018-01-27 Jakub Jelinek <jakub@redhat.com>
388 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
391 2018-01-26 Jim Wilson <jimw@sifive.com>
393 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
395 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
398 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
400 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
401 and CMP + SUB-immediate -> SUBS.
403 2018-01-26 Martin Sebor <msebor@redhat.com>
405 PR tree-optimization/83896
406 * tree-ssa-strlen.c (get_string_len): Rename...
407 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
408 Avoid assuming length is constant.
409 (handle_char_store): Use HOST_WIDE_INT for string length.
411 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
414 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
415 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
417 2018-01-26 Richard Biener <rguenther@suse.de>
419 PR rtl-optimization/84003
420 * dse.c (record_store): Only record redundant stores when
421 the earlier store aliases at least all accesses the later one does.
423 2018-01-26 Jakub Jelinek <jakub@redhat.com>
425 PR rtl-optimization/83985
426 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
427 REG_CFA_RESTORE insns.
428 (delete_unmarked_insns): Don't ignore separate shrink wrapping
429 REG_CFA_RESTORE insns here.
432 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
433 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
435 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
437 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
438 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
439 (arc_init): Likewise.
440 (arc_override_options): Likewise.
441 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
443 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
445 * config/arc/arc.h (TARGET_DBNZ): Define.
446 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
447 properly set the tune attribute.
448 (dbnz): Use TARGET_DBNZ guard.
449 * config/arc/arc.opt (mtune): Add core3 option.
451 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
453 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
454 recognize new pic like addresses.
455 (arc_delegitimize_address): Clean up.
457 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
459 * config/arc/arc-arches.def: Option mrf16 valid for all
461 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
462 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
463 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
464 * config/arc/arc-tables.opt: Regenerate.
465 * config/arc/arc.c (arc_conditional_register_usage): Handle
466 reduced register file case.
467 (arc_file_start): Set must have build attributes.
468 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
470 * config/arc/arc.opt (mrf16): Add new option.
471 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
472 * config/arc/genmultilib.awk: Handle new mrf16 option.
473 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
474 * config/arc/t-multilib: Regenerate.
475 * doc/invoke.texi (ARC Options): Document mrf16 option.
477 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
479 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
480 * config/arc/arc.c (arc_handle_secure_attribute): New function.
481 (arc_attribute_table): Add 'secure_call' attribute.
482 (arc_print_operand): Print secure call operand.
483 (arc_function_ok_for_sibcall): Don't optimize tail calls when
485 (arc_is_secure_call_p): New function. * config/arc/arc.md
486 (call_i): Add support for sjli instruction.
487 (call_value_i): Likewise.
488 * config/arc/constraints.md (Csc): New constraint.
490 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
491 John Eric Martin <John.Martin@emmicro-us.com>
493 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
494 * config/arc/arc.c (_arc_jli_section): New struct.
495 (arc_jli_section): New type.
496 (rc_jli_sections): New static variable.
497 (arc_handle_jli_attribute): New function.
498 (arc_attribute_table): Add jli_always and jli_fixed attribute.
499 (arc_file_end): New function.
500 (TARGET_ASM_FILE_END): Define.
501 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
502 (arc_add_jli_section): New function.
503 (jli_call_scan): Likewise.
504 (arc_reorg): Call jli_call_scan.
505 (arc_output_addsi): Remove 'S' from printing asm operand.
506 (arc_is_jli_call_p): New function.
507 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
509 (movhi_insn): Likewise.
510 (movsi_insn): Likewise.
511 (movsi_set_cc_insn): Likewise.
512 (loadqi_update): Likewise.
513 (load_zeroextendqisi_update): Likewise.
514 (load_signextendqisi_update): Likewise.
515 (loadhi_update): Likewise.
516 (load_zeroextendhisi_update): Likewise.
517 (load_signextendhisi_update): Likewise.
518 (loadsi_update): Likewise.
519 (loadsf_update): Likewise.
520 (movsicc_insn): Likewise.
521 (bset_insn): Likewise.
522 (bxor_insn): Likewise.
523 (bclr_insn): Likewise.
524 (bmsk_insn): Likewise.
525 (bicsi3_insn): Likewise.
526 (cmpsi_cc_c_insn): Likewise.
527 (movsi_ne): Likewise.
528 (movsi_cond_exec): Likewise.
529 (clrsbsi2): Likewise.
540 (call_i): Remove 'S' asm letter, add jli instruction.
541 (call_value_i): Likewise.
542 * config/arc/arc.op (mjli-always): New option.
543 * config/arc/constraints.md (Cji): New constraint.
544 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
546 (subsf3_fpx): Likewise.
547 (mulsf3_fpx): Likewise.
548 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
550 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
552 * doc/invoke.texi (ARC): Document mjli-always option.
554 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
556 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
557 and use incw and decw where possible
558 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
560 2018-01-26 Richard Biener <rguenther@suse.de>
562 PR tree-optimization/81082
563 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
564 association if it requires casting to unsigned.
565 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
566 from fold_plusminus_mult_expr to catch important cases late when
567 range info is available.
569 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
571 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
572 * configure.ac (hidden_linkonce): New test.
573 * configure: Regenerate.
574 * config.in: Regenerate.
576 2018-01-26 Julia Koval <julia.koval@intel.com>
578 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
579 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
580 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
581 _mm_mask_bitshuffle_epi64_mask): Fix type.
582 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
583 USI_FTYPE_V4DI_V4DI_USI): Remove.
584 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
585 __builtin_ia32_vpshufbitqmb256_mask,
586 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
587 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
588 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
590 2018-01-26 Alan Modra <amodra@gmail.com>
593 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
594 UNSPEC_VBPERMQ. Sort other unspecs.
596 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
598 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
600 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
603 * predict.c (drop_profile): Do not push/pop cfun; update also
605 (handle_missing_profiles): Fix logic looking for zero profiles.
607 2018-01-25 Jakub Jelinek <jakub@redhat.com>
610 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
611 on functions with #pragma omp declare simd or functions with simd
613 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
614 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
615 Remove trailing \n from warning_at calls.
617 2018-01-25 Tom de Vries <tom@codesourcery.com>
620 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
621 for neutered workers.
623 2018-01-24 Joseph Myers <joseph@codesourcery.com>
626 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
627 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
629 2018-01-24 Jeff Law <law@redhat.com>
632 * i386.c (get_probe_interval): Move to earlier point.
633 (ix86_compute_frame_layout): If -fstack-clash-protection and
634 the frame is larger than the probe interval, then use pushes
635 to save registers rather than reg->mem moves.
636 (ix86_expand_prologue): Remove conditional for int_registers_saved
639 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
642 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
643 min/max for never referenced object.
645 2018-01-24 Jakub Jelinek <jakub@redhat.com>
648 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
650 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
651 attributes from DECL_ATTRIBUTES (decl) without affecting
652 DECL_ATTRIBUTES (current_function_decl).
653 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
654 functions with non-NULL DECL_ABSTRACT_ORIGIN.
656 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
658 PR tree-optimization/83979
659 * fold-const.c (fold_comparison): Use constant_boolean_node
660 instead of boolean_{true,false}_node.
662 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
664 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
667 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
669 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
670 Simplify the clause that sets the length attribute.
671 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
672 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
673 clause that sets the length attribute.
674 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
676 2018-01-24 Tom de Vries <tom@codesourcery.com>
679 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
680 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
681 Add strict parameter.
682 (prevent_branch_around_nothing): Insert dummy insn between branch to
683 label and label with no ptx insn inbetween.
684 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
686 2018-01-24 Tom de Vries <tom@codesourcery.com>
689 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
690 for neutered threads in warp.
691 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
693 2018-01-24 Richard Biener <rguenther@suse.de>
695 PR tree-optimization/83176
696 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
699 2018-01-24 Richard Biener <rguenther@suse.de>
701 PR tree-optimization/82819
702 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
703 code generating pluses that are no-ops in the target precision.
705 2018-01-24 Richard Biener <rguenther@suse.de>
708 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
710 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
712 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
713 to merge probabilities.
714 * predict.c (probably_never_executed): Also mark as cold functions
715 with global 0 profile and guessed local profile.
716 * profile-count.c (profile_probability::combine_with_count): New
718 * profile-count.h (profile_probability::operator*,
719 profile_probability::operator*=, profile_probability::operator/,
720 profile_probability::operator/=): Reduce precision to adjusted
721 and set value to guessed on contradictory divisions.
722 (profile_probability::combine_with_freq): Remove.
723 (profile_probability::combine_wiht_count): Declare.
724 (profile_count::force_nonzero):: Set to adjusted.
725 (profile_count::probability_in):: Set quality to adjusted.
726 * tree-ssa-tail-merge.c (replace_block_by): Use
729 2018-01-23 Andrew Waterman <andrew@sifive.com>
730 Jim Wilson <jimw@sifive.com>
732 * config/riscv/riscv.c (riscv_stack_boundary): New.
733 (riscv_option_override): Set riscv_stack_boundary. Handle
734 riscv_preferred_stack_boundary_arg.
735 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
736 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
737 (STACK_BOUNDARY): Set to riscv_stack_boundary.
738 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
739 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
740 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
742 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
745 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
746 of struct ix86_frame.
747 (ix86_expand_epilogue): Likewise. Add a local variable for
748 the reg_save_offset field in struct ix86_frame.
750 2018-01-23 Bin Cheng <bin.cheng@arm.com>
752 PR tree-optimization/82604
753 * tree-loop-distribution.c (enum partition_kind): New enum item
754 PKIND_PARTIAL_MEMSET.
755 (partition_builtin_p): Support above new enum item.
756 (generate_code_for_partition): Ditto.
757 (compute_access_range): Differentiate cases that equality can be
758 proven at all loops, the innermost loops or no loops.
759 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
760 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
761 (finalize_partitions, distribute_loop): Don't fuse partition of
762 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
763 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
766 2018-01-23 Martin Liska <mliska@suse.cz>
768 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
769 order to ignore the predictor.
770 (PRED_POLYMORPHIC_CALL): Likewise.
771 (PRED_RECURSIVE_CALL): Likewise.
773 2018-01-23 Martin Liska <mliska@suse.cz>
775 * tree-profile.c (tree_profiling): Print function header to
776 aware reader which function we are working on.
777 * value-prof.c (gimple_find_values_to_profile): Do not print
778 not interesting value histograms.
780 2018-01-23 Martin Liska <mliska@suse.cz>
782 * profile-count.h (enum profile_quality): Add
783 profile_uninitialized as the first value. Do not number values
784 as they are zero based.
785 (profile_count::verify): Update sanity check.
786 (profile_probability::verify): Likewise.
788 2018-01-23 Nathan Sidwell <nathan@acm.org>
790 * doc/invoke.texi (ffor-scope): Deprecate.
792 2018-01-23 David Malcolm <dmalcolm@redhat.com>
794 PR tree-optimization/83510
795 * domwalk.c (set_all_edges_as_executable): New function.
796 (dom_walker::dom_walker): Convert bool param
797 "skip_unreachable_blocks" to enum reachability. Move setup of
798 edge flags to set_all_edges_as_executable and only do it when
799 reachability is REACHABLE_BLOCKS.
800 * domwalk.h (enum dom_walker::reachability): New enum.
801 (dom_walker::dom_walker): Convert bool param
802 "skip_unreachable_blocks" to enum reachability.
803 (set_all_edges_as_executable): New decl.
804 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
805 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
807 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
808 but converting true to REACHABLE_BLOCKS.
809 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
811 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
812 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
813 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
815 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
816 if check_all_array_refs will be called.
818 2018-01-23 David Malcolm <dmalcolm@redhat.com>
820 * tree.c (selftest::test_location_wrappers): Add more test
823 2018-01-23 David Malcolm <dmalcolm@redhat.com>
825 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
826 (selftest::test_bit_in_range): Likewise.
828 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
831 * doc/sourcebuild.texi (vect_float): Say that the selector
832 only describes the situation when -funsafe-math-optimizations is on.
833 (vect_float_strict): Document.
835 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
837 PR tree-optimization/83965
838 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
839 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
840 instead of checking only for a reduction.
841 (vect_recog_widen_sum_pattern): Likewise.
843 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
845 * predict.c (probably_never_executed): Only use precise profile info.
846 (compute_function_frequency): Skip after inlining hack since we now
847 have quality checking.
849 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
851 * profile-count.h (profile_probability::very_unlikely,
852 profile_probability::unlikely, profile_probability::even): Set
853 precision to guessed.
855 2018-01-23 Richard Biener <rguenther@suse.de>
857 PR tree-optimization/83963
858 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
859 Properly terminate dominator walk when crossing the exit edge not
860 when visiting its source block.
862 2018-01-23 Jakub Jelinek <jakub@redhat.com>
865 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
866 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
868 2018-01-22 Jakub Jelinek <jakub@redhat.com>
870 PR tree-optimization/83957
871 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
872 semicolon after for body surrounded by braces.
874 PR tree-optimization/83081
875 * profile-count.h (profile_probability::split): New method.
876 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
877 Use profile_probability::split.
878 (do_compare_rtx_and_jump): Fix adjustment of probabilities
879 when splitting a single conditional jump into 2.
881 2018-01-22 David Malcolm <dmalcolm@redhat.com>
883 PR tree-optimization/69452
884 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
887 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
889 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
890 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
891 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
893 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
895 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
896 * config/rl78/rl78.md: New define_expand "movdi"
897 * config/rl78/rl78.c: New function definition rl78_split_movdi
899 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
902 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
904 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
905 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
906 128-bit to produce an UNSPEC move to get the double word with the
907 signbit and then a shift directly to do signbit.
908 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
909 implementation with a new version that just does either a direct
910 move or a regular move. Move memory interface to separate insns.
911 Move insns so they are next to the expander.
912 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
913 with signbit move. Split big and little endian case.
914 (signbit<mode>2_dm_mem_le): Likewise.
915 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
916 (signbit<mode>2_dm2): Likewise.
918 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
920 * config/rl78/rl78.md: New define_expand "anddi3".
922 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
924 * config/rl78/rl78.md: New define_expand "umindi3".
926 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
928 * config/rl78/rl78.md: New define_expand "smindi3".
930 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
932 * config/rl78/rl78.md: New define_expand "smaxdi3".
934 2018-01-22 Carl Love <cel@us.ibm.com>
936 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
937 LVX_V1TI): Add macro expansion.
938 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
939 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
940 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
941 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
942 Change check to determine if the instruction is a byte reversing
943 entry. Fix typo in comment.
944 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
945 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
946 Add def_builtin calls for new builtins.
947 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
948 Add define_insn expansion.
950 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
952 * config/rl78/rl78.md: New define_expand "umaxdi3".
954 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
956 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
957 for non-QImode registers
959 2018-01-22 Richard Biener <rguenther@suse.de>
961 PR tree-optimization/83963
962 * graphite-scop-detection.c (scop_detection::get_sese): Delay
963 including the loop exit block.
964 (scop_detection::merge_sese): Likewise.
965 (scop_detection::add_scop): Do it here instead.
967 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
969 * doc/sourcebuild.texi (arm_softfloat): Document.
971 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
974 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
975 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
976 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
978 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
979 David Edelsohn <dje.gcc@gmail.com>
982 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
983 Change "crset eq" to "crset 2".
984 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
985 (*call_indirect_aix<mode>_nospec): Likewise.
986 (*call_value_indirect_aix<mode>_nospec): Likewise.
987 (*call_indirect_elfv2<mode>_nospec): Likewise.
988 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
989 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
990 change assembly output from . to $.
991 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
992 (indirect_jump<mode>_nospec): Change assembly output from . to $.
993 (*tablejump<mode>_internal1_nospec): Likewise.
995 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
998 * config/sh/sh_optimize_sett_clrt.cc:
999 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1001 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1003 PR tree-optimization/83940
1004 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1005 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1006 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1007 use it to pass back the definition type.
1008 (vect_check_store_rhs): Likewise rhs_dt_out.
1009 (vect_build_gather_load_calls): Add a mask_dt argument and use
1010 it instead of a call to vect_is_simple_use.
1011 (vectorizable_store): Update calls to vect_check_load_store_mask
1012 and vect_check_store_rhs. Use the dt returned by the latter instead
1013 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1014 instead of calls to vect_is_simple_use. Pass the scalar rather
1015 than the vector operand to vect_is_simple_use when handling
1016 second and subsequent copies of an rhs value.
1017 (vectorizable_load): Update calls to vect_check_load_store_mask
1018 and vect_build_gather_load_calls. Use the cached mask_dt and
1019 gs_info.offset_dt instead of calls to vect_is_simple_use.
1021 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1024 * tree-emutls.c: Include gimplify.h.
1025 (lower_emutls_2): New function.
1026 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1027 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1028 it before further processing.
1031 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1032 UINTVAL (trueop1) instead of INTVAL (op1).
1034 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1038 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1039 INCOMING_FRAME_SP_OFFSET if not defined.
1040 (scan_trace): Add ENTRY argument. If true and
1041 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1042 emit a note to adjust the CFA offset.
1043 (create_cfi_notes): Adjust scan_trace callers.
1044 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1045 INCOMING_FRAME_SP_OFFSET in the CIE.
1046 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1047 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1049 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1050 * doc/tm.texi: Regenerated.
1052 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1054 PR rtl-optimization/83147
1055 * lra-constraints.c (remove_inheritance_pseudos): Use
1056 lra_substitute_pseudo_within_insn.
1058 2018-01-19 Tom de Vries <tom@codesourcery.com>
1059 Cesar Philippidis <cesar@codesourcery.com>
1062 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1064 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1067 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1068 spaces for function labels.
1070 2018-01-19 Martin Liska <mliska@suse.cz>
1072 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1073 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1074 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1075 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1076 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1077 (PRED_CONST_RETURN): Change from 69 to 65.
1078 (PRED_NULL_RETURN): Change from 91 to 71.
1079 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1080 (PRED_LOOP_GUARD): Change from 66 to 73.
1082 2018-01-19 Martin Liska <mliska@suse.cz>
1084 * predict.c (predict_insn_def): Add new assert.
1085 (struct branch_predictor): Change type to signed integer.
1086 (test_prediction_value_range): Amend test to cover
1088 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1089 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1090 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1091 (PRED_LOOP_IV_COMPARE): Likewise.
1092 * predict.h (PROB_UNINITIALIZED): Define new constant.
1094 2018-01-19 Martin Liska <mliska@suse.cz>
1096 * predict.c (dump_prediction): Add new format for
1097 analyze_brprob.py script which is enabled with -details
1099 * profile-count.h (precise_p): New function.
1101 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1103 PR tree-optimization/83922
1104 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1105 there are no statements that need masking.
1106 (vect_active_double_reduction_p): New function.
1107 (vect_analyze_loop_operations): Use it when handling phis that
1108 are not in the loop header.
1110 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1112 PR tree-optimization/83914
1113 * tree-vect-loop.c (vectorizable_induction): Don't convert
1114 init_expr or apply the peeling adjustment for inductions
1115 that are nested within the vectorized loop.
1117 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1119 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1122 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1126 * function.h (gimplify_parameters): Add gimple_seq * argument.
1127 * function.c: Include gimple.h and options.h.
1128 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1129 for the added local temporaries if needed.
1130 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1131 if there are any parameter cleanups, wrap whole body into a
1132 try/finally with the cleanups.
1134 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1137 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1138 Use GET_MODE_CLASS for scalar floating point.
1140 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1144 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1145 Fix call of call_cgraph_insertion_hooks.
1147 2018-01-18 Martin Sebor <msebor@redhat.com>
1149 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1151 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1154 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1157 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1160 * common.opt: (-ffile-prefix-map): New option.
1161 * opts.c (common_handle_option): Defer it.
1162 * opts-global.c (handle_common_deferred_options): Handle it.
1163 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1164 * file-prefix-map.h: New file.
1165 (remap_debug_filename, add_debug_prefix_map): ...here.
1166 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1167 * final.c (debug_prefix_map, add_debug_prefix_map
1168 remap_debug_filename): Move to...
1169 * file-prefix-map.c: New file.
1170 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1171 generalize, get rid of alloca(), use strrchr() instead of strchr().
1172 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1173 Implement in terms of add_prefix_map().
1174 (remap_macro_filename, remap_debug_filename): Implement in term of
1176 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1177 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1178 * dbxout.c: Include file-prefix-map.h.
1179 * varasm.c: Likewise.
1180 * vmsdbgout.c: Likewise.
1181 * xcoffout.c: Likewise.
1182 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1183 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1184 * doc/invoke.texi (-ffile-prefix-map): Document.
1185 (-fdebug-prefix-map): Update description.
1187 2018-01-18 Martin Liska <mliska@suse.cz>
1189 * config/i386/i386.c (indirect_thunk_name): Document that also
1191 (output_indirect_thunk): Document why both instructions
1192 (pause and lfence) are generated.
1194 2018-01-18 Richard Biener <rguenther@suse.de>
1196 PR tree-optimization/83887
1197 * graphite-scop-detection.c
1198 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1199 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1200 (scop_detection::merge_sese): Re-implement with a flood-fill
1201 algorithm that properly finds a SESE region if it exists.
1203 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1206 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1207 pointer_diff optimizations use view_convert instead of convert.
1209 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1211 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1212 Generate different code for -mno-speculate-indirect-jumps.
1213 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1214 (*call_indirect_aix<mode>): Disable for
1215 -mno-speculate-indirect-jumps.
1216 (*call_indirect_aix<mode>_nospec): New define_insn.
1217 (*call_value_indirect_aix<mode>): Disable for
1218 -mno-speculate-indirect-jumps.
1219 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1220 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1221 -mno-speculate-indirect-jumps.
1222 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1224 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1226 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1227 long double type, set the flags for noting the default long double
1228 type, even if we don't pass or return a long double type.
1230 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1233 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1236 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1238 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1239 support for merge[hl].
1240 (fold_mergehl_helper): New helper function.
1241 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1242 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1243 (altivec_vmrglw_direct): Add xxmrglw insn.
1245 2018-01-17 Andrew Waterman <andrew@sifive.com>
1247 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1248 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1250 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1253 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1254 call the lto_location_cache before reading the
1255 DECL_SOURCE_LOCATION of the types.
1257 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1258 Richard Sandiford <richard.sandiford@linaro.org>
1260 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1261 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1262 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1263 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1264 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1266 * config/aarch64/constraints.md (aarch64_movti_operand):
1268 * config/aarch64/predicates.md (Uti): Add new constraint.
1270 2018-01-17 Carl Love <cel@us.ibm.com>
1271 * config/rs6000/vsx.md (define_expand xl_len_r,
1272 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1273 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1275 (define_expand, define_insn): Move the shift left from the
1276 define_insn to the define_expand for lxvl and stxvl instructions.
1277 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1278 and XL_LEN_R definitions to PURE.
1280 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1282 * config/i386/i386.c (indirect_thunk_name): Declare regno
1283 as unsigned int. Compare regno with INVALID_REGNUM.
1284 (output_indirect_thunk): Ditto.
1285 (output_indirect_thunk_function): Ditto.
1286 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1287 in the call to output_indirect_thunk_function.
1289 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1292 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1293 rather than the size of inner_type to determine the stack slot size
1294 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1296 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1299 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1302 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1304 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1305 endian Linux systems to optionally enable multilibs for selecting
1306 the long double type if the user configured an explicit type.
1307 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1308 have no long double multilibs if not defined.
1309 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1310 warn if the user used -mabi={ieee,ibm}longdouble and we built
1311 multilibs for long double.
1312 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1313 appropriate multilib option.
1314 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1316 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1317 for building long double multilibs.
1318 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1320 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1322 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1325 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1327 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1330 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1333 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1336 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1338 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1339 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1341 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1343 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1344 different rtl trees depending on TARGET_64BIT.
1345 (rs6000_gen_lvx): Likewise.
1347 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1349 * config/visium/visium.md (nop): Tweak comment.
1350 (hazard_nop): Likewise.
1352 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1354 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1355 -mspeculate-indirect-jumps.
1356 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1357 for -mno-speculate-indirect-jumps.
1358 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1359 (*call_value_indirect_elfv2<mode>): Disable for
1360 -mno-speculate-indirect-jumps.
1361 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1362 (indirect_jump): Emit different RTL for
1363 -mno-speculate-indirect-jumps.
1364 (*indirect_jump<mode>): Disable for
1365 -mno-speculate-indirect-jumps.
1366 (*indirect_jump<mode>_nospec): New define_insn.
1367 (tablejump): Emit different RTL for
1368 -mno-speculate-indirect-jumps.
1369 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1370 (tablejumpsi_nospec): New define_expand.
1371 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1372 (tablejumpdi_nospec): New define_expand.
1373 (*tablejump<mode>_internal1): Disable for
1374 -mno-speculate-indirect-jumps.
1375 (*tablejump<mode>_internal1_nospec): New define_insn.
1376 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1379 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1381 * caller-save.c (insert_save): Drop unnecessary parameter. All
1384 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1385 Richard Biener <rguenth@suse.de>
1388 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1389 return early, inline manually is_gimple_sizepos. Make sure if we
1390 call gimplify_expr we don't end up with a gimple constant.
1391 * tree.c (variably_modified_type_p): Don't return true for
1392 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1393 * gimplify.h (is_gimple_sizepos): Remove.
1395 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1397 PR tree-optimization/83857
1398 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1399 vectorizable_live_operation for pure SLP statements.
1400 (vectorizable_live_operation): Handle PHIs.
1402 2018-01-16 Richard Biener <rguenther@suse.de>
1404 PR tree-optimization/83867
1405 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1406 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1408 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1411 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1412 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1413 If off is not INTEGER_CST, issue a may not be aligned warning
1414 rather than isn't aligned. Use isn%'t rather than isn't.
1415 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1417 <case MULT_EXPR>: Improve the case when bottom and one of the
1418 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1419 operand, in that case check if the other operand is multiple of
1420 bottom divided by the INTEGER_CST operand.
1422 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1425 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1426 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1427 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1428 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1429 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1430 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1431 (pa_function_arg_size): New function.
1433 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1435 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1436 in a separate statement.
1438 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1440 PR tree-optimization/83847
1441 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1442 group gathers and scatters.
1444 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1446 PR rtl-optimization/86620
1447 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1449 PR rtl-optimization/83213
1450 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1451 to last if both are JUMP_INSNs.
1453 PR tree-optimization/83843
1454 * gimple-ssa-store-merging.c
1455 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1456 store_immediate_info for bswap/nop orig_stores.
1458 2018-01-15 Andrew Waterman <andrew@sifive.com>
1460 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1462 <UDIV>: Increase cost if !TARGET_DIV.
1464 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1466 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1467 (define_attr "cr_logical_3op"): New.
1468 (cceq_ior_compare): Adjust.
1469 (cceq_ior_compare_complement): Adjust.
1470 (*cceq_rev_compare): Adjust.
1471 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1472 (is_cracked_insn): Adjust.
1473 (insn_must_be_first_in_group): Adjust.
1474 * config/rs6000/40x.md: Adjust.
1475 * config/rs6000/440.md: Adjust.
1476 * config/rs6000/476.md: Adjust.
1477 * config/rs6000/601.md: Adjust.
1478 * config/rs6000/603.md: Adjust.
1479 * config/rs6000/6xx.md: Adjust.
1480 * config/rs6000/7450.md: Adjust.
1481 * config/rs6000/7xx.md: Adjust.
1482 * config/rs6000/8540.md: Adjust.
1483 * config/rs6000/cell.md: Adjust.
1484 * config/rs6000/e300c2c3.md: Adjust.
1485 * config/rs6000/e500mc.md: Adjust.
1486 * config/rs6000/e500mc64.md: Adjust.
1487 * config/rs6000/e5500.md: Adjust.
1488 * config/rs6000/e6500.md: Adjust.
1489 * config/rs6000/mpc.md: Adjust.
1490 * config/rs6000/power4.md: Adjust.
1491 * config/rs6000/power5.md: Adjust.
1492 * config/rs6000/power6.md: Adjust.
1493 * config/rs6000/power7.md: Adjust.
1494 * config/rs6000/power8.md: Adjust.
1495 * config/rs6000/power9.md: Adjust.
1496 * config/rs6000/rs64.md: Adjust.
1497 * config/rs6000/titan.md: Adjust.
1499 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1501 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1502 ix86_indirect_branch_register logic.
1504 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1506 * config/i386/constraints.md (Bs): Update
1507 ix86_indirect_branch_register check. Don't check
1508 ix86_indirect_branch_register with GOT_memory_operand.
1510 * config/i386/predicates.md (GOT_memory_operand): Don't check
1511 ix86_indirect_branch_register here.
1512 (GOT32_symbol_operand): Likewise.
1514 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1516 * config/i386/predicates.md (constant_call_address_operand):
1517 Rewrite ix86_indirect_branch_register logic.
1518 (sibcall_insn_operand): Likewise.
1520 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1522 * config/i386/constraints.md (Bs): Replace
1523 ix86_indirect_branch_thunk_register with
1524 ix86_indirect_branch_register.
1526 * config/i386/i386.md (indirect_jump): Likewise.
1527 (tablejump): Likewise.
1528 (*sibcall_memory): Likewise.
1529 (*sibcall_value_memory): Likewise.
1530 Peepholes of indirect call and jump via memory: Likewise.
1531 * config/i386/i386.opt: Likewise.
1532 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1533 (GOT_memory_operand): Likewise.
1534 (call_insn_operand): Likewise.
1535 (sibcall_insn_operand): Likewise.
1536 (GOT32_symbol_operand): Likewise.
1538 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1541 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1542 type rather than type addr's type points to.
1543 (expand_omp_atomic_mutex): Likewise.
1544 (expand_omp_atomic): Likewise.
1546 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1549 * config/i386/i386.c (output_indirect_thunk_function): Use
1550 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1551 for __x86_return_thunk.
1553 2018-01-15 Richard Biener <rguenther@suse.de>
1556 * expmed.c (extract_bit_field_1): Fix typo.
1558 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1561 * config/arm/iterators.md (VF): New mode iterator.
1562 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1563 Remove integer-related logic from pattern.
1564 (neon_vabd<mode>_3): Likewise.
1566 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1569 * common.opt (fstrict-overflow): No longer an alias.
1570 (fwrapv-pointer): New option.
1571 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1572 also for pointer types based on flag_wrapv_pointer.
1573 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1574 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1575 opts->x_flag_wrapv got set.
1576 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1577 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1578 POINTER_TYPE_OVERFLOW_UNDEFINED.
1579 * match.pd: Likewise in address comparison pattern.
1580 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1582 2018-01-15 Richard Biener <rguenther@suse.de>
1585 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1586 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1587 Reset type names to their identifier if their TYPE_DECL doesn't
1588 have linkage (and thus is used for ODR and devirt).
1589 (save_debug_info_for_decl): Remove.
1590 (save_debug_info_for_type): Likewise.
1591 (add_tree_to_fld_list): Adjust.
1592 * tree-pretty-print.c (dump_generic_node): Make dumping of
1593 type names more robust.
1595 2018-01-15 Richard Biener <rguenther@suse.de>
1597 * BASE-VER: Bump to 8.0.1.
1599 2018-01-14 Martin Sebor <msebor@redhat.com>
1602 * builtins.c (check_access): Avoid warning when the no-warning bit
1605 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1607 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1608 * ira-color (allocno_hard_regs_compare): Likewise.
1610 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1613 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1614 Use .pushsection/.popsection.
1616 2018-01-14 Martin Sebor <msebor@redhat.com>
1619 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1621 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1623 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1624 entry from extra_headers.
1625 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1626 extra_headers, make the list bitwise identical to the i?86-*-* one.
1628 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1630 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1631 -mcmodel=large with -mindirect-branch=thunk,
1632 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1633 -mfunction-return=thunk-extern.
1634 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1635 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1636 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1638 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1640 * config/i386/i386.c (print_reg): Print the name of the full
1641 integer register without '%'.
1642 (ix86_print_operand): Handle 'V'.
1643 * doc/extend.texi: Document 'V' modifier.
1645 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1647 * config/i386/constraints.md (Bs): Disallow memory operand for
1648 -mindirect-branch-register.
1650 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1651 (GOT_memory_operand): Likewise.
1652 (call_insn_operand): Likewise.
1653 (sibcall_insn_operand): Likewise.
1654 (GOT32_symbol_operand): Likewise.
1655 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1656 for -mindirect-branch-register.
1657 (tablejump): Likewise.
1658 (*sibcall_memory): Likewise.
1659 (*sibcall_value_memory): Likewise.
1660 Disallow peepholes of indirect call and jump via memory for
1661 -mindirect-branch-register.
1662 (*call_pop): Replace m with Bw.
1663 (*call_value_pop): Likewise.
1664 (*sibcall_pop_memory): Replace m with Bs.
1665 * config/i386/i386.opt (mindirect-branch-register): New option.
1666 * doc/invoke.texi: Document -mindirect-branch-register option.
1668 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1670 * config/i386/i386-protos.h (ix86_output_function_return): New.
1671 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1672 set function_return_type.
1673 (indirect_thunk_name): Add ret_p to indicate thunk for function
1675 (output_indirect_thunk_function): Pass false to
1676 indirect_thunk_name.
1677 (ix86_output_indirect_branch_via_reg): Likewise.
1678 (ix86_output_indirect_branch_via_push): Likewise.
1679 (output_indirect_thunk_function): Create alias for function
1680 return thunk if regno < 0.
1681 (ix86_output_function_return): New function.
1682 (ix86_handle_fndecl_attribute): Handle function_return.
1683 (ix86_attribute_table): Add function_return.
1684 * config/i386/i386.h (machine_function): Add
1685 function_return_type.
1686 * config/i386/i386.md (simple_return_internal): Use
1687 ix86_output_function_return.
1688 (simple_return_internal_long): Likewise.
1689 * config/i386/i386.opt (mfunction-return=): New option.
1690 (indirect_branch): Mention -mfunction-return=.
1691 * doc/extend.texi: Document function_return function attribute.
1692 * doc/invoke.texi: Document -mfunction-return= option.
1694 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1696 * config/i386/i386-opts.h (indirect_branch): New.
1697 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1698 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1699 with local indirect jump when converting indirect call and jump.
1700 (ix86_set_indirect_branch_type): New.
1701 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1702 (indirectlabelno): New.
1703 (indirect_thunk_needed): Likewise.
1704 (indirect_thunk_bnd_needed): Likewise.
1705 (indirect_thunks_used): Likewise.
1706 (indirect_thunks_bnd_used): Likewise.
1707 (INDIRECT_LABEL): Likewise.
1708 (indirect_thunk_name): Likewise.
1709 (output_indirect_thunk): Likewise.
1710 (output_indirect_thunk_function): Likewise.
1711 (ix86_output_indirect_branch_via_reg): Likewise.
1712 (ix86_output_indirect_branch_via_push): Likewise.
1713 (ix86_output_indirect_branch): Likewise.
1714 (ix86_output_indirect_jmp): Likewise.
1715 (ix86_code_end): Call output_indirect_thunk_function if needed.
1716 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1718 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1719 (ix86_attribute_table): Add indirect_branch.
1720 * config/i386/i386.h (machine_function): Add indirect_branch_type
1721 and has_local_indirect_jump.
1722 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1724 (tablejump): Likewise.
1725 (*indirect_jump): Use ix86_output_indirect_jmp.
1726 (*tablejump_1): Likewise.
1727 (simple_return_indirect_internal): Likewise.
1728 * config/i386/i386.opt (mindirect-branch=): New option.
1729 (indirect_branch): New.
1732 (thunk-inline): Likewise.
1733 (thunk-extern): Likewise.
1734 * doc/extend.texi: Document indirect_branch function attribute.
1735 * doc/invoke.texi: Document -mindirect-branch= option.
1737 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1740 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1742 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1744 * ipa-inline.c (want_inline_small_function_p): Return false if
1745 inlining has already failed with CIF_FINAL_ERROR.
1746 (update_caller_keys): Call want_inline_small_function_p before
1748 (update_callee_keys): Likewise.
1750 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1752 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1754 (rs6000_quadword_masked_address_p): Likewise.
1755 (quad_aligned_load_p): Likewise.
1756 (quad_aligned_store_p): Likewise.
1757 (const_load_sequence_p): Add comment to describe the outer-most loop.
1758 (mimic_memory_attributes_and_flags): New function.
1759 (rs6000_gen_stvx): Likewise.
1760 (replace_swapped_aligned_store): Likewise.
1761 (rs6000_gen_lvx): Likewise.
1762 (replace_swapped_aligned_load): Likewise.
1763 (replace_swapped_load_constant): Capitalize argument name in
1764 comment describing this function.
1765 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1766 and stores that access quad-word aligned addresses and replace
1767 with stvx or lvx instructions when appropriate.
1768 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1769 New function prototype.
1770 (rs6000_quadword_masked_address_p): Likewise.
1771 (rs6000_gen_lvx): Likewise.
1772 (rs6000_gen_stvx): Likewise.
1773 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1774 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1775 when memory address is aligned.
1776 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1777 this split to select lvx instruction when memory address is aligned.
1778 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1779 instruction when memory address is aligned.
1780 (*vsx_le_perm_load_v16qi): Likewise.
1781 (four unnamed splitters): Modify to select the stvx instruction
1782 when memory is aligned.
1784 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1786 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1787 which appears in the queue multiple times.
1789 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1790 Alan Hayward <alan.hayward@arm.com>
1791 David Sherwood <david.sherwood@arm.com>
1793 * tree-vectorizer.h (vec_lower_bound): New structure.
1794 (_loop_vec_info): Add check_nonzero and lower_bounds.
1795 (LOOP_VINFO_CHECK_NONZERO): New macro.
1796 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1797 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1798 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1799 fields. Make seg_len the distance travelled, not including the
1801 (dr_direction_indicator): Declare.
1802 (dr_zero_step_indicator): Likewise.
1803 (dr_known_forward_stride_p): Likewise.
1804 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1806 (runtime_alias_check_p): Allow runtime alias checks with
1808 (operator ==): Compare access_size and align.
1809 (prune_runtime_alias_test_list): Rework for new distinction between
1810 the access_size and seg_len.
1811 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1813 (get_segment_min_max): New function.
1814 (create_intersect_range_checks): Use it.
1815 (dr_step_indicator): New function.
1816 (dr_direction_indicator): Likewise.
1817 (dr_zero_step_indicator): Likewise.
1818 (dr_known_forward_stride_p): Likewise.
1819 * tree-loop-distribution.c (data_ref_segment_size): Return
1820 DR_STEP * (niters - 1).
1821 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1823 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1824 (vect_preserves_scalar_order_p): New function, split out from...
1825 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1826 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1827 (vect_vfa_access_size): New function.
1828 (vect_vfa_align): Likewise.
1829 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1830 (dump_lower_bound): New function.
1831 (vect_check_lower_bound): Likewise.
1832 (vect_small_gap_p): Likewise.
1833 (vectorizable_with_step_bound_p): Likewise.
1834 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1835 depencies if the vectorization factor is 1. Convert the checks
1836 for nonzero steps into checks on the bounds of DR_STEP. Try using
1837 a bunds check for variable steps if the minimum required step is
1838 relatively small. Update calls to the dr_with_seg_len
1839 constructor and to vect_compile_time_alias.
1840 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1842 (vect_loop_versioning): Call it.
1843 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1845 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1847 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1848 Alan Hayward <alan.hayward@arm.com>
1849 David Sherwood <david.sherwood@arm.com>
1851 * doc/sourcebuild.texi (vect_scatter_store): Document.
1852 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1854 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1856 * genopinit.c (main): Add supports_vec_scatter_store and
1857 supports_vec_scatter_store_cached to target_optabs.
1858 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1859 IFN_MASK_SCATTER_STORE.
1860 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1862 * internal-fn.h (internal_store_fn_p): Declare.
1863 (internal_fn_stored_value_index): Likewise.
1864 * internal-fn.c (scatter_store_direct): New macro.
1865 (expand_scatter_store_optab_fn): New function.
1866 (direct_scatter_store_optab_supported_p): New macro.
1867 (internal_store_fn_p): New function.
1868 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1869 IFN_MASK_SCATTER_STORE.
1870 (internal_fn_mask_index): Likewise.
1871 (internal_fn_stored_value_index): New function.
1872 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1874 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1875 * optabs-query.c (supports_vec_scatter_store_p): New function.
1876 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1877 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1878 true for scatter stores.
1879 (vect_gather_scatter_fn_p): Handle scatter stores too.
1880 (vect_check_gather_scatter): Consider using scatter stores if
1881 supports_vec_scatter_store_p.
1882 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1884 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1885 internal_fn_stored_value_index.
1886 (check_load_store_masking): Handle scatter stores too.
1887 (vect_get_store_rhs): Make public.
1888 (vectorizable_call): Use internal_store_fn_p.
1889 (vectorizable_store): Handle scatter store internal functions.
1890 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1891 when deciding whether the end of the group has been reached.
1892 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1893 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1894 (mask_scatter_store<mode>): New insns.
1896 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1897 Alan Hayward <alan.hayward@arm.com>
1898 David Sherwood <david.sherwood@arm.com>
1900 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1901 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1902 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1904 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1905 Use vect_truncate_gather_scatter_offset if we can't treat the
1906 operation as a normal gather load or scatter store.
1907 (get_group_load_store_type): Take the gather_scatter_info
1908 as argument. Try using a gather load or scatter store for
1909 single-element groups.
1910 (get_load_store_type): Update calls to get_group_load_store_type
1911 and vect_use_strided_gather_scatters_p.
1913 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1914 Alan Hayward <alan.hayward@arm.com>
1915 David Sherwood <david.sherwood@arm.com>
1917 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1918 optional tree argument.
1919 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1921 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1922 but continue to use the current value as a fallback.
1923 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1924 to compare the updates.
1925 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1926 (get_load_store_type): Use it when handling a strided access.
1927 (vect_get_strided_load_store_ops): New function.
1928 (vect_get_data_ptr_increment): Likewise.
1929 (vectorizable_load): Handle strided gather loads. Always pass
1930 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1932 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1933 Alan Hayward <alan.hayward@arm.com>
1934 David Sherwood <david.sherwood@arm.com>
1936 * doc/md.texi (gather_load@var{m}): Document.
1937 (mask_gather_load@var{m}): Likewise.
1938 * genopinit.c (main): Add supports_vec_gather_load and
1939 supports_vec_gather_load_cached to target_optabs.
1940 * optabs-tree.c (init_tree_optimization_optabs): Use
1941 ggc_cleared_alloc to allocate target_optabs.
1942 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1943 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1945 * internal-fn.h (internal_load_fn_p): Declare.
1946 (internal_gather_scatter_fn_p): Likewise.
1947 (internal_fn_mask_index): Likewise.
1948 (internal_gather_scatter_fn_supported_p): Likewise.
1949 * internal-fn.c (gather_load_direct): New macro.
1950 (expand_gather_load_optab_fn): New function.
1951 (direct_gather_load_optab_supported_p): New macro.
1952 (direct_internal_fn_optab): New function.
1953 (internal_load_fn_p): Likewise.
1954 (internal_gather_scatter_fn_p): Likewise.
1955 (internal_fn_mask_index): Likewise.
1956 (internal_gather_scatter_fn_supported_p): Likewise.
1957 * optabs-query.c (supports_at_least_one_mode_p): New function.
1958 (supports_vec_gather_load_p): Likewise.
1959 * optabs-query.h (supports_vec_gather_load_p): Declare.
1960 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1961 and memory_type field.
1962 (NUM_PATTERNS): Bump to 15.
1963 * tree-vect-data-refs.c: Include internal-fn.h.
1964 (vect_gather_scatter_fn_p): New function.
1965 (vect_describe_gather_scatter_call): Likewise.
1966 (vect_check_gather_scatter): Try using internal functions for
1967 gather loads. Recognize existing calls to a gather load function.
1968 (vect_analyze_data_refs): Consider using gather loads if
1969 supports_vec_gather_load_p.
1970 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1971 (vect_get_gather_scatter_offset_type): Likewise.
1972 (vect_convert_mask_for_vectype): Likewise.
1973 (vect_add_conversion_to_patterm): Likewise.
1974 (vect_try_gather_scatter_pattern): Likewise.
1975 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1976 (vect_vect_recog_func_ptrs): Add it.
1977 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1978 internal_fn_mask_index and internal_gather_scatter_fn_p.
1979 (check_load_store_masking): Take the gather_scatter_info as an
1980 argument and handle gather loads.
1981 (vect_get_gather_scatter_ops): New function.
1982 (vectorizable_call): Check internal_load_fn_p.
1983 (vectorizable_load): Likewise. Handle gather load internal
1985 (vectorizable_store): Update call to check_load_store_masking.
1986 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1987 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1988 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1989 (aarch64_gather_scale_operand_d): New predicates.
1990 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1991 (mask_gather_load<mode>): New insns.
1993 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1994 Alan Hayward <alan.hayward@arm.com>
1995 David Sherwood <david.sherwood@arm.com>
1997 * optabs.def (fold_left_plus_optab): New optab.
1998 * doc/md.texi (fold_left_plus_@var{m}): Document.
1999 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2000 * internal-fn.c (fold_left_direct): Define.
2001 (expand_fold_left_optab_fn): Likewise.
2002 (direct_fold_left_optab_supported_p): Likewise.
2003 * fold-const-call.c (fold_const_fold_left): New function.
2004 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2005 * tree-parloops.c (valid_reduction_p): New function.
2006 (gather_scalar_reductions): Use it.
2007 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2008 (vect_finish_replace_stmt): Declare.
2009 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2010 (needs_fold_left_reduction_p): New function, split out from...
2011 (vect_is_simple_reduction): ...here. Accept reductions that
2012 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2013 (vect_force_simple_reduction): Also store the reduction type in
2014 the assignment's STMT_VINFO_REDUC_TYPE.
2015 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2016 (merge_with_identity): New function.
2017 (vect_expand_fold_left): Likewise.
2018 (vectorize_fold_left_reduction): Likewise.
2019 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2020 scalar phi in place for it. Check for target support and reject
2021 cases that would reassociate the operation. Defer the transform
2022 phase to vectorize_fold_left_reduction.
2023 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2024 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2025 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2027 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2029 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2030 call to ifc_temp_var.
2032 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2033 Alan Hayward <alan.hayward@arm.com>
2034 David Sherwood <david.sherwood@arm.com>
2036 * target.def (legitimize_address_displacement): Take the original
2037 offset as a poly_int.
2038 * targhooks.h (default_legitimize_address_displacement): Update
2040 * targhooks.c (default_legitimize_address_displacement): Likewise.
2041 * doc/tm.texi: Regenerate.
2042 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2043 as an argument, moving assert of ad->disp == ad->disp_term to...
2044 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2045 Try calling targetm.legitimize_address_displacement before expanding
2046 the address rather than afterwards, and adjust for the new interface.
2047 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2048 Match the new hook interface. Handle SVE addresses.
2049 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2052 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2054 * Makefile.in (OBJS): Add early-remat.o.
2055 * target.def (select_early_remat_modes): New hook.
2056 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2057 * doc/tm.texi: Regenerate.
2058 * targhooks.h (default_select_early_remat_modes): Declare.
2059 * targhooks.c (default_select_early_remat_modes): New function.
2060 * timevar.def (TV_EARLY_REMAT): New timevar.
2061 * passes.def (pass_early_remat): New pass.
2062 * tree-pass.h (make_pass_early_remat): Declare.
2063 * early-remat.c: New file.
2064 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2066 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2068 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2069 Alan Hayward <alan.hayward@arm.com>
2070 David Sherwood <david.sherwood@arm.com>
2072 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2073 vfm1 with a bound_epilog parameter.
2074 (vect_do_peeling): Update calls accordingly, and move the prologue
2075 call earlier in the function. Treat the base bound_epilog as 0 for
2076 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2077 this base when peeling for gaps.
2078 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2079 with fully-masked loops.
2080 (vect_estimate_min_profitable_iters): Handle the single peeled
2081 iteration in that case.
2083 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2084 Alan Hayward <alan.hayward@arm.com>
2085 David Sherwood <david.sherwood@arm.com>
2087 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2088 single-element interleaving even if the size is not a power of 2.
2089 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2090 accesses for single-element interleaving if the group size is
2093 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2094 Alan Hayward <alan.hayward@arm.com>
2095 David Sherwood <david.sherwood@arm.com>
2097 * doc/md.texi (fold_extract_last_@var{m}): Document.
2098 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2099 * optabs.def (fold_extract_last_optab): New optab.
2100 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2101 * internal-fn.c (fold_extract_direct): New macro.
2102 (expand_fold_extract_optab_fn): Likewise.
2103 (direct_fold_extract_optab_supported_p): Likewise.
2104 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2105 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2106 EXTRACT_LAST_REDUCTION.
2107 (get_initial_def_for_reduction): Do not create an initial vector
2108 for EXTRACT_LAST_REDUCTION reductions.
2109 (vectorizable_reduction): Leave the scalar phi in place for
2110 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2111 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2112 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2113 transform phase to vectorizable_condition.
2114 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2116 (vect_finish_stmt_generation): ...here.
2117 (vect_finish_replace_stmt): New function.
2118 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2119 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2121 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2123 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2124 Alan Hayward <alan.hayward@arm.com>
2125 David Sherwood <david.sherwood@arm.com>
2127 * doc/md.texi (extract_last_@var{m}): Document.
2128 * optabs.def (extract_last_optab): New optab.
2129 * internal-fn.def (EXTRACT_LAST): New internal function.
2130 * internal-fn.c (cond_unary_direct): New macro.
2131 (expand_cond_unary_optab_fn): Likewise.
2132 (direct_cond_unary_optab_supported_p): Likewise.
2133 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2134 loops using EXTRACT_LAST.
2135 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2136 (extract_last_<mode>): ...this optab.
2137 (vec_extract<mode><Vel>): Update accordingly.
2139 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2140 Alan Hayward <alan.hayward@arm.com>
2141 David Sherwood <david.sherwood@arm.com>
2143 * target.def (empty_mask_is_expensive): New hook.
2144 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2145 * doc/tm.texi: Regenerate.
2146 * targhooks.h (default_empty_mask_is_expensive): Declare.
2147 * targhooks.c (default_empty_mask_is_expensive): New function.
2148 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2149 if the target says that empty masks are expensive.
2150 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2152 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2154 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2155 Alan Hayward <alan.hayward@arm.com>
2156 David Sherwood <david.sherwood@arm.com>
2158 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2159 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2160 (vect_use_loop_mask_for_alignment_p): New function.
2161 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2162 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2163 niters_skip argument. Make sure that the first niters_skip elements
2164 of the first iteration are inactive.
2165 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2166 Update call to vect_set_loop_masks_directly.
2167 (get_misalign_in_elems): New function, split out from...
2168 (vect_gen_prolog_loop_niters): ...here.
2169 (vect_update_init_of_dr): Take a code argument that specifies whether
2170 the adjustment should be added or subtracted.
2171 (vect_update_init_of_drs): Likewise.
2172 (vect_prepare_for_masked_peels): New function.
2173 (vect_do_peeling): Skip prologue peeling if we're using a mask
2174 instead. Update call to vect_update_inits_of_drs.
2175 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2177 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2178 alignment. Do not include the number of peeled iterations in
2179 the minimum threshold in that case.
2180 (vectorizable_induction): Adjust the start value down by
2181 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2182 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2183 Take the number of skipped iterations into account when calculating
2185 * tree-vect-stmts.c (vect_gen_while_not): New function.
2187 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2188 Alan Hayward <alan.hayward@arm.com>
2189 David Sherwood <david.sherwood@arm.com>
2191 * doc/sourcebuild.texi (vect_fully_masked): Document.
2192 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2194 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2196 (vect_analyze_loop_2): ...here. Don't check the vectorization
2197 factor against the number of loop iterations if the loop is
2200 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2201 Alan Hayward <alan.hayward@arm.com>
2202 David Sherwood <david.sherwood@arm.com>
2204 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2205 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2206 (dump_groups): Update accordingly.
2207 (iv_use::mem_type): New member variable.
2208 (address_p): New function.
2209 (record_use): Add a mem_type argument and initialize the new
2211 (record_group_use): Add a mem_type argument. Use address_p.
2212 Remove obsolete null checks of base_object. Update call to record_use.
2213 (find_interesting_uses_op): Update call to record_group_use.
2214 (find_interesting_uses_cond): Likewise.
2215 (find_interesting_uses_address): Likewise.
2216 (get_mem_type_for_internal_fn): New function.
2217 (find_address_like_use): Likewise.
2218 (find_interesting_uses_stmt): Try find_address_like_use before
2219 calling find_interesting_uses_op.
2220 (addr_offset_valid_p): Use the iv mem_type field as the type
2221 of the addressed memory.
2222 (add_autoinc_candidates): Likewise.
2223 (get_address_cost): Likewise.
2224 (split_small_address_groups_p): Use address_p.
2225 (split_address_groups): Likewise.
2226 (add_iv_candidate_for_use): Likewise.
2227 (autoinc_possible_for_pair): Likewise.
2228 (rewrite_groups): Likewise.
2229 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2230 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2231 (get_alias_ptr_type_for_ptr_address): New function.
2232 (rewrite_use_address): Rewrite address uses in calls that were
2233 identified by find_address_like_use.
2235 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2236 Alan Hayward <alan.hayward@arm.com>
2237 David Sherwood <david.sherwood@arm.com>
2239 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2241 * gimple-expr.h (is_gimple_addressable: Likewise.
2242 * gimple-expr.c (is_gimple_address): Likewise.
2243 * internal-fn.c (expand_call_mem_ref): New function.
2244 (expand_mask_load_optab_fn): Use it.
2245 (expand_mask_store_optab_fn): Likewise.
2247 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2248 Alan Hayward <alan.hayward@arm.com>
2249 David Sherwood <david.sherwood@arm.com>
2251 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2252 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2253 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2254 (cond_umax@var{mode}): Document.
2255 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2256 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2257 (cond_umin_optab, cond_umax_optab): New optabs.
2258 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2259 (COND_IOR, COND_XOR): New internal functions.
2260 * internal-fn.h (get_conditional_internal_fn): Declare.
2261 * internal-fn.c (cond_binary_direct): New macro.
2262 (expand_cond_binary_optab_fn): Likewise.
2263 (direct_cond_binary_optab_supported_p): Likewise.
2264 (get_conditional_internal_fn): New function.
2265 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2266 Cope with reduction statements that are vectorized as calls rather
2268 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2269 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2270 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2271 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2272 (UNSPEC_COND_EOR): New unspecs.
2273 (optab): Add mappings for them.
2274 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2275 (sve_int_op, sve_fp_op): New int attributes.
2277 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2278 Alan Hayward <alan.hayward@arm.com>
2279 David Sherwood <david.sherwood@arm.com>
2281 * optabs.def (while_ult_optab): New optab.
2282 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2283 * internal-fn.def (WHILE_ULT): New internal function.
2284 * internal-fn.h (direct_internal_fn_supported_p): New override
2285 that takes two types as argument.
2286 * internal-fn.c (while_direct): New macro.
2287 (expand_while_optab_fn): New function.
2288 (convert_optab_supported_p): Likewise.
2289 (direct_while_optab_supported_p): New macro.
2290 * wide-int.h (wi::udiv_ceil): New function.
2291 * tree-vectorizer.h (rgroup_masks): New structure.
2292 (vec_loop_masks): New typedef.
2293 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2295 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2296 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2297 (vect_max_vf): New function.
2298 (slpeel_make_loop_iterate_ntimes): Delete.
2299 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2300 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2301 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2302 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2303 internal-fn.h, stor-layout.h and optabs-query.h.
2304 (vect_set_loop_mask): New function.
2305 (add_preheader_seq): Likewise.
2306 (add_header_seq): Likewise.
2307 (interleave_supported_p): Likewise.
2308 (vect_maybe_permute_loop_masks): Likewise.
2309 (vect_set_loop_masks_directly): Likewise.
2310 (vect_set_loop_condition_masked): Likewise.
2311 (vect_set_loop_condition_unmasked): New function, split out from
2312 slpeel_make_loop_iterate_ntimes.
2313 (slpeel_make_loop_iterate_ntimes): Rename to..
2314 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2315 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2316 (vect_do_peeling): Update call accordingly.
2317 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2319 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2320 mask_compare_type, can_fully_mask_p and fully_masked_p.
2321 (release_vec_loop_masks): New function.
2322 (_loop_vec_info): Use it to free the loop masks.
2323 (can_produce_all_loop_masks_p): New function.
2324 (vect_get_max_nscalars_per_iter): Likewise.
2325 (vect_verify_full_masking): Likewise.
2326 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2327 retries, and free the mask rgroups before retrying. Check loop-wide
2328 reasons for disallowing fully-masked loops. Make the final decision
2329 about whether use a fully-masked loop or not.
2330 (vect_estimate_min_profitable_iters): Do not assume that peeling
2331 for the number of iterations will be needed for fully-masked loops.
2332 (vectorizable_reduction): Disable fully-masked loops.
2333 (vectorizable_live_operation): Likewise.
2334 (vect_halve_mask_nunits): New function.
2335 (vect_double_mask_nunits): Likewise.
2336 (vect_record_loop_mask): Likewise.
2337 (vect_get_loop_mask): Likewise.
2338 (vect_transform_loop): Handle the case in which the final loop
2339 iteration might handle a partial vector. Call vect_set_loop_condition
2340 instead of slpeel_make_loop_iterate_ntimes.
2341 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2342 (check_load_store_masking): New function.
2343 (prepare_load_store_mask): Likewise.
2344 (vectorizable_store): Handle fully-masked loops.
2345 (vectorizable_load): Likewise.
2346 (supportable_widening_operation): Use vect_halve_mask_nunits for
2348 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2349 (vect_gen_while): New function.
2350 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2351 (aarch64_uqdec<mode>): New insn.
2353 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2354 Alan Hayward <alan.hayward@arm.com>
2355 David Sherwood <david.sherwood@arm.com>
2357 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2358 (reduc_xor_scal_optab): New optabs.
2359 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2360 (reduc_xor_scal_@var{m}): Document.
2361 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2362 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2364 * fold-const-call.c (fold_const_call): Handle them.
2365 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2366 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2367 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2368 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2369 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2370 (UNSPEC_XORV): New unspecs.
2371 (optab): Add entries for them.
2372 (BITWISEV): New int iterator.
2373 (bit_reduc_op): New int attributes.
2375 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2376 Alan Hayward <alan.hayward@arm.com>
2377 David Sherwood <david.sherwood@arm.com>
2379 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2380 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2381 * optabs.def (vec_shl_insert_optab): New optab.
2382 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2383 (duplicate_and_interleave): Likewise.
2384 * tree-vect-loop.c: Include internal-fn.h.
2385 (neutral_op_for_slp_reduction): New function, split out from
2386 get_initial_defs_for_reduction.
2387 (get_initial_def_for_reduction): Handle option 2 for variable-length
2388 vectors by loading the neutral value into a vector and then shifting
2389 the initial value into element 0.
2390 (get_initial_defs_for_reduction): Replace the code argument with
2391 the neutral value calculated by neutral_op_for_slp_reduction.
2392 Use gimple_build_vector for constant-length vectors.
2393 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2394 but the first group_size elements have a neutral value.
2395 Use duplicate_and_interleave otherwise.
2396 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2397 Update call to get_initial_defs_for_reduction. Handle SLP
2398 reductions for variable-length vectors by creating one vector
2399 result for each scalar result, with the elements associated
2400 with other scalar results stubbed out with the neutral value.
2401 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2402 Require IFN_VEC_SHL_INSERT for double reductions on
2403 variable-length vectors, or SLP reductions that have
2404 a neutral value. Require can_duplicate_and_interleave_p
2405 support for variable-length unchained SLP reductions if there
2406 is no neutral value, such as for MIN/MAX reductions. Also require
2407 the number of vector elements to be a multiple of the number of
2408 SLP statements when doing variable-length unchained SLP reductions.
2409 Update call to vect_create_epilog_for_reduction.
2410 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2411 and remove initial values.
2412 (duplicate_and_interleave): Make public.
2413 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2414 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2416 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2417 Alan Hayward <alan.hayward@arm.com>
2418 David Sherwood <david.sherwood@arm.com>
2420 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2421 (can_duplicate_and_interleave_p): New function.
2422 (vect_get_and_check_slp_defs): Take the vector of statements
2423 rather than just the current one. Remove excess parentheses.
2424 Restriction rejectinon of vect_constant_def and vect_external_def
2425 for variable-length vectors to boolean types, or types for which
2426 can_duplicate_and_interleave_p is false.
2427 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2428 (duplicate_and_interleave): New function.
2429 (vect_get_constant_vectors): Use gimple_build_vector for
2430 constant-length vectors and suitable variable-length constant
2431 vectors. Use duplicate_and_interleave for other variable-length
2432 vectors. Don't defer the update when inserting new statements.
2434 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2435 Alan Hayward <alan.hayward@arm.com>
2436 David Sherwood <david.sherwood@arm.com>
2438 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2439 min_profitable_iters doesn't go negative.
2441 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2442 Alan Hayward <alan.hayward@arm.com>
2443 David Sherwood <david.sherwood@arm.com>
2445 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2446 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2447 * optabs.def (vec_mask_load_lanes_optab): New optab.
2448 (vec_mask_store_lanes_optab): Likewise.
2449 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2450 (MASK_STORE_LANES): Likewise.
2451 * internal-fn.c (mask_load_lanes_direct): New macro.
2452 (mask_store_lanes_direct): Likewise.
2453 (expand_mask_load_optab_fn): Handle masked operations.
2454 (expand_mask_load_lanes_optab_fn): New macro.
2455 (expand_mask_store_optab_fn): Handle masked operations.
2456 (expand_mask_store_lanes_optab_fn): New macro.
2457 (direct_mask_load_lanes_optab_supported_p): Likewise.
2458 (direct_mask_store_lanes_optab_supported_p): Likewise.
2459 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2461 (vect_load_lanes_supported): Likewise.
2462 * tree-vect-data-refs.c (strip_conversion): New function.
2463 (can_group_stmts_p): Likewise.
2464 (vect_analyze_data_ref_accesses): Use it instead of checking
2465 for a pair of assignments.
2466 (vect_store_lanes_supported): Take a masked_p parameter.
2467 (vect_load_lanes_supported): Likewise.
2468 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2469 vect_store_lanes_supported and vect_load_lanes_supported.
2470 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2471 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2472 parameter. Don't allow gaps for masked accesses.
2473 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2474 and vect_load_lanes_supported.
2475 (get_load_store_type): Take a masked_p parameter and update
2476 call to get_group_load_store_type.
2477 (vectorizable_store): Update call to get_load_store_type.
2478 Handle IFN_MASK_STORE_LANES.
2479 (vectorizable_load): Update call to get_load_store_type.
2480 Handle IFN_MASK_LOAD_LANES.
2482 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2483 Alan Hayward <alan.hayward@arm.com>
2484 David Sherwood <david.sherwood@arm.com>
2486 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2488 * config/aarch64/aarch64-protos.h
2489 (aarch64_sve_struct_memory_operand_p): Declare.
2490 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2491 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2492 (VPRED, vpred): Handle SVE structure modes.
2493 * config/aarch64/constraints.md (Utx): New constraint.
2494 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2495 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2496 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2497 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2498 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2499 structure modes. Split into pieces after RA.
2500 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2501 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2503 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2504 SVE structure modes.
2505 (aarch64_classify_address): Likewise.
2506 (sizetochar): Move earlier in file.
2507 (aarch64_print_operand): Handle SVE register lists.
2508 (aarch64_array_mode): New function.
2509 (aarch64_sve_struct_memory_operand_p): Likewise.
2510 (TARGET_ARRAY_MODE): Redefine.
2512 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2513 Alan Hayward <alan.hayward@arm.com>
2514 David Sherwood <david.sherwood@arm.com>
2516 * target.def (array_mode): New target hook.
2517 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2518 * doc/tm.texi: Regenerate.
2519 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2520 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2521 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2523 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2526 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2527 Alan Hayward <alan.hayward@arm.com>
2528 David Sherwood <david.sherwood@arm.com>
2530 * fold-const.c (fold_binary_loc): Check the argument types
2531 rather than the result type when testing for a vector operation.
2533 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2535 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2536 * doc/tm.texi: Regenerate.
2538 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2539 Alan Hayward <alan.hayward@arm.com>
2540 David Sherwood <david.sherwood@arm.com>
2542 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2543 (sve): Document new AArch64 extension.
2544 * doc/md.texi (w): Extend the description of the AArch64
2545 constraint to include SVE vectors.
2546 (Upl, Upa): Document new AArch64 predicate constraints.
2547 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2549 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2550 (msve-vector-bits=): New option.
2551 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2552 SVE when these are disabled.
2553 (sve): New extension.
2554 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2555 modes. Adjust their number of units based on aarch64_sve_vg.
2556 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2557 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2558 aarch64_addr_query_type.
2559 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2560 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2561 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2562 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2563 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2564 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2565 (aarch64_simd_imm_zero_p): Delete.
2566 (aarch64_check_zero_based_sve_index_immediate): Declare.
2567 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2568 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2569 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2570 (aarch64_sve_float_mul_immediate_p): Likewise.
2571 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2573 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2574 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2575 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2576 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2577 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2578 (aarch64_regmode_natural_size): Likewise.
2579 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2580 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2582 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2583 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2584 for VG and the SVE predicate registers.
2585 (V_ALIASES): Add a "z"-prefixed alias.
2586 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2587 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2588 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2589 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2590 (REG_CLASS_NAMES): Add entries for them.
2591 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2592 and the predicate registers.
2593 (aarch64_sve_vg): Declare.
2594 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2595 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2596 (REGMODE_NATURAL_SIZE): Define.
2597 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2599 * config/aarch64/aarch64.c: Include cfgrtl.h.
2600 (simd_immediate_info): Add a constructor for series vectors,
2601 and an associated step field.
2602 (aarch64_sve_vg): New variable.
2603 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2604 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2605 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2606 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2607 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2608 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2609 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2610 (aarch64_get_mask_mode): New functions.
2611 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2612 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2613 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2614 predicate modes and predicate registers. Explicitly restrict
2615 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2616 to store a vector mode if it is recognized by
2617 aarch64_classify_vector_mode.
2618 (aarch64_regmode_natural_size): New function.
2619 (aarch64_hard_regno_caller_save_mode): Return the original mode
2621 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2622 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2623 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2624 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2626 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2627 does not overlap dest if the function is frame-related. Handle
2629 (aarch64_split_add_offset): New function.
2630 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2631 them aarch64_add_offset.
2632 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2633 and update call to aarch64_sub_sp.
2634 (aarch64_add_cfa_expression): New function.
2635 (aarch64_expand_prologue): Pass extra temporary registers to the
2636 functions above. Handle the case in which we need to emit new
2637 DW_CFA_expressions for registers that were originally saved
2638 relative to the stack pointer, but now have to be expressed
2639 relative to the frame pointer.
2640 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2642 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2643 IP0 and IP1 values for SVE frames.
2644 (aarch64_expand_vec_series): New function.
2645 (aarch64_expand_sve_widened_duplicate): Likewise.
2646 (aarch64_expand_sve_const_vector): Likewise.
2647 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2648 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2649 into the register, rather than emitting a SET directly.
2650 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2651 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2652 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2653 (offset_9bit_signed_scaled_p): New functions.
2654 (aarch64_replicate_bitmask_imm): New function.
2655 (aarch64_bitmask_imm): Use it.
2656 (aarch64_cannot_force_const_mem): Reject expressions involving
2657 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2658 (aarch64_classify_index): Handle SVE indices, by requiring
2659 a plain register index with a scale that matches the element size.
2660 (aarch64_classify_address): Handle SVE addresses. Assert that
2661 the mode of the address is VOIDmode or an integer mode.
2662 Update call to aarch64_classify_symbol.
2663 (aarch64_classify_symbolic_expression): Update call to
2664 aarch64_classify_symbol.
2665 (aarch64_const_vec_all_in_range_p): New function.
2666 (aarch64_print_vector_float_operand): Likewise.
2667 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2668 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2669 and the FP immediates 1.0 and 0.5.
2670 (aarch64_print_address_internal): Handle SVE addresses.
2671 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2672 (aarch64_regno_regclass): Handle predicate registers.
2673 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2675 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2676 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2677 (aarch64_convert_sve_vector_bits): New function.
2678 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2679 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2681 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2682 Handle SVE vector and predicate modes. Accept VL-based constants
2683 that need only one temporary register, and VL offsets that require
2684 no temporary registers.
2685 (aarch64_conditional_register_usage): Mark the predicate registers
2686 as fixed if SVE isn't available.
2687 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2688 Return true for SVE vector and predicate modes.
2689 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2690 rather than an unsigned int. Handle SVE modes.
2691 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2693 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2695 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2696 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2697 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2698 (aarch64_sve_float_mul_immediate_p): New functions.
2699 (aarch64_sve_valid_immediate): New function.
2700 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2701 Explicitly reject structure modes. Check for INDEX constants.
2702 Handle PTRUE and PFALSE constants.
2703 (aarch64_check_zero_based_sve_index_immediate): New function.
2704 (aarch64_simd_imm_zero_p): Delete.
2705 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2706 vector modes. Accept constants in the range of CNT[BHWD].
2707 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2708 ask for an Advanced SIMD mode.
2709 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2710 (aarch64_simd_vector_alignment): Handle SVE predicates.
2711 (aarch64_vectorize_preferred_vector_alignment): New function.
2712 (aarch64_simd_vector_alignment_reachable): Use it instead of
2714 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2715 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2717 (MAX_VECT_LEN): Delete.
2718 (expand_vec_perm_d): Add a vec_flags field.
2719 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2720 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2721 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2723 (aarch64_evpc_rev): Rename to...
2724 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2725 (aarch64_evpc_rev_global): New function.
2726 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2727 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2729 (aarch64_evpc_sve_tbl): New function.
2730 (aarch64_expand_vec_perm_const_1): Update after rename of
2731 aarch64_evpc_rev. Handle SVE permutes too, trying
2732 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2733 than aarch64_evpc_tbl.
2734 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2735 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2736 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2737 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2738 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2739 (aarch64_expand_sve_vcond): New functions.
2740 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2741 of aarch64_vector_mode_p.
2742 (aarch64_dwarf_poly_indeterminate_value): New function.
2743 (aarch64_compute_pressure_classes): Likewise.
2744 (aarch64_can_change_mode_class): Likewise.
2745 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2746 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2747 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2748 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2749 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2750 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2751 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2752 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2754 (Dn, Dl, Dr): Accept const as well as const_vector.
2755 (Dz): Likewise. Compare against CONST0_RTX.
2756 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2757 of "vector" where appropriate.
2758 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2759 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2760 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2761 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2762 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2763 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2764 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2765 (v_int_equiv): Extend to SVE modes.
2766 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2768 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2769 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2770 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2771 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2772 (SVE_COND_FP_CMP): New int iterators.
2773 (perm_hilo): Handle the new unpack unspecs.
2774 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2776 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2777 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2778 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2779 (aarch64_equality_operator, aarch64_constant_vector_operand)
2780 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2781 (aarch64_sve_nonimmediate_operand): Likewise.
2782 (aarch64_sve_general_operand): Likewise.
2783 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2784 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2785 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2786 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2787 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2788 (aarch64_sve_float_arith_immediate): Likewise.
2789 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2790 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2791 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2792 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2793 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2794 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2795 (aarch64_sve_float_arith_operand): Likewise.
2796 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2797 (aarch64_sve_float_mul_operand): Likewise.
2798 (aarch64_sve_vec_perm_operand): Likewise.
2799 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2800 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2801 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2802 as well as const_vector.
2803 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2804 in file. Use CONST0_RTX and CONSTM1_RTX.
2805 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2806 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2807 Use aarch64_simd_imm_zero.
2808 * config/aarch64/aarch64-sve.md: New file.
2809 * config/aarch64/aarch64.md: Include it.
2810 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2811 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2812 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2813 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2814 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2815 (sve): New attribute.
2816 (enabled): Disable instructions with the sve attribute unless
2818 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2819 aarch64_expand_mov_immediate.
2820 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2821 CNT[BHSD] immediates.
2822 (movti): Split CONST_POLY_INT moves into two halves.
2823 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2824 Split additions that need a temporary here if the destination
2825 is the stack pointer.
2826 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2827 (*add<mode>3_poly_1): New instruction.
2828 (set_clobber_cc): New expander.
2830 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2832 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2833 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2834 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2835 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2836 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2837 Change innermode from fixed_mode_size to machine_mode.
2838 (simplify_subreg): Update call accordingly. Handle a constant-sized
2839 subreg of a variable-length CONST_VECTOR.
2841 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2842 Alan Hayward <alan.hayward@arm.com>
2843 David Sherwood <david.sherwood@arm.com>
2845 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2846 (add_offset_to_base): New function, split out from...
2847 (create_mem_ref): ...here. When handling a scale other than 1,
2848 check first whether the address is valid without the offset.
2849 Add it into the base if so, leaving the index and scale as-is.
2851 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2854 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2855 fold_for_warn before checking if arg2 is INTEGER_CST.
2857 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2859 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2860 (store_multiple_operation): Delete.
2861 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2862 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2863 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2864 guarded by TARGET_STRING.
2865 (rs6000_output_load_multiple): Delete.
2866 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2867 OPTION_MASK_STRING / TARGET_STRING handling.
2868 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2869 (const rs6000_opt_masks) <"string">: Change mask to 0.
2870 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2871 (MASK_STRING): Delete.
2872 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2874 (load_multiple): Delete.
2881 (store_multiple): Delete.
2888 (movmemsi_8reg): Delete.
2889 (corresponding unnamed define_insn): Delete.
2890 (movmemsi_6reg): Delete.
2891 (corresponding unnamed define_insn): Delete.
2892 (movmemsi_4reg): Delete.
2893 (corresponding unnamed define_insn): Delete.
2894 (movmemsi_2reg): Delete.
2895 (corresponding unnamed define_insn): Delete.
2896 (movmemsi_1reg): Delete.
2897 (corresponding unnamed define_insn): Delete.
2898 * config/rs6000/rs6000.opt (mno-string): New.
2899 (mstring): Replace by deprecation warning stub.
2900 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2902 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2904 * regrename.c (regrename_do_replace): If replacing the same
2905 reg multiple times, try to reuse last created gen_raw_REG.
2908 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2909 main to workaround a bug in GDB.
2911 2018-01-12 Tom de Vries <tom@codesourcery.com>
2914 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2916 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2918 PR rtl-optimization/80481
2919 * ira-color.c (get_cap_member): New function.
2920 (allocnos_conflict_by_live_ranges_p): Use it.
2921 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2922 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2924 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2927 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2928 (*saddl_se_1): Ditto.
2930 (*ssubl_se_1): Ditto.
2932 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2934 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2935 rather than wi::to_widest for DR_INITs.
2936 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2937 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2938 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2940 (vect_analyze_group_access_1): Note that here.
2942 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2944 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2945 polynomial type sizes.
2947 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2949 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2950 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2951 (gimple_add_tmp_var): Likewise.
2953 2018-01-12 Martin Liska <mliska@suse.cz>
2955 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2956 (gimple_alloc_sizes): Likewise.
2957 (dump_gimple_statistics): Use PRIu64 in printf format.
2958 * gimple.h: Change uint64_t to int.
2960 2018-01-12 Martin Liska <mliska@suse.cz>
2962 * tree-core.h: Use uint64_t instead of int.
2963 * tree.c (tree_node_counts): Likewise.
2964 (tree_node_sizes): Likewise.
2965 (dump_tree_statistics): Use PRIu64 in printf format.
2967 2018-01-12 Martin Liska <mliska@suse.cz>
2969 * Makefile.in: As qsort_chk is implemented in vec.c, add
2970 vec.o to linkage of gencfn-macros.
2971 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2972 passing the info to record_node_allocation_statistics.
2973 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2975 * ggc-common.c (struct ggc_usage): Add operator== and use
2976 it in operator< and compare function.
2977 * mem-stats.h (struct mem_usage): Likewise.
2978 * vec.c (struct vec_usage): Remove operator< and compare
2979 function. Can be simply inherited.
2981 2018-01-12 Martin Jambor <mjambor@suse.cz>
2984 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2985 * tree-ssa-math-opts.c: Include domwalk.h.
2986 (convert_mult_to_fma_1): New function.
2987 (fma_transformation_info): New type.
2988 (fma_deferring_state): Likewise.
2989 (cancel_fma_deferring): New function.
2990 (result_of_phi): Likewise.
2991 (last_fma_candidate_feeds_initial_phi): Likewise.
2992 (convert_mult_to_fma): Added deferring logic, split actual
2993 transformation to convert_mult_to_fma_1.
2994 (math_opts_dom_walker): New type.
2995 (math_opts_dom_walker::after_dom_children): New method, body moved
2996 here from pass_optimize_widening_mul::execute, added deferring logic
2998 (pass_optimize_widening_mul::execute): Moved most of code to
2999 math_opts_dom_walker::after_dom_children.
3000 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3001 * config/i386/i386.c (ix86_option_override_internal): Added
3002 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3004 2018-01-12 Richard Biener <rguenther@suse.de>
3007 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3008 inline instance vars.
3010 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3013 * config/rx/rx.c (rx_is_restricted_memory_address):
3016 2018-01-12 Richard Biener <rguenther@suse.de>
3018 PR tree-optimization/80846
3019 * target.def (split_reduction): New target hook.
3020 * targhooks.c (default_split_reduction): New function.
3021 * targhooks.h (default_split_reduction): Declare.
3022 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3023 target requests first reduce vectors by combining low and high
3025 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3026 (get_vectype_for_scalar_type_and_size): Export.
3027 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3028 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3029 * doc/tm.texi: Regenerate.
3030 * config/i386/i386.c (ix86_split_reduction): Implement
3031 TARGET_VECTORIZE_SPLIT_REDUCTION.
3033 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3036 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3037 in PIC mode except for TARGET_VXWORKS_RTP.
3038 * config/sparc/sparc.c: Include cfgrtl.h.
3039 (TARGET_INIT_PIC_REG): Define.
3040 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3041 (sparc_pic_register_p): New predicate.
3042 (sparc_legitimate_address_p): Use it.
3043 (sparc_legitimize_pic_address): Likewise.
3044 (sparc_delegitimize_address): Likewise.
3045 (sparc_mode_dependent_address_p): Likewise.
3046 (gen_load_pcrel_sym): Remove 4th parameter.
3047 (load_got_register): Adjust call to above. Remove obsolete stuff.
3048 (sparc_expand_prologue): Do not call load_got_register here.
3049 (sparc_flat_expand_prologue): Likewise.
3050 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3051 (sparc_use_pseudo_pic_reg): New function.
3052 (sparc_init_pic_reg): Likewise.
3053 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3054 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3056 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3058 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3059 Add item for branch_cost.
3061 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3063 PR rtl-optimization/83565
3064 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3065 not extend the result to a larger mode for rotate operations.
3066 (num_sign_bit_copies1): Likewise.
3068 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3071 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3073 Use values-Xc.o for -pedantic.
3074 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3076 2018-01-12 Martin Liska <mliska@suse.cz>
3079 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3081 (possible_polymorphic_call_targets): Use it.
3082 (ipa_devirt): Likewise.
3084 2018-01-12 Martin Liska <mliska@suse.cz>
3086 * profile-count.h (enum profile_quality): Use 0 as invalid
3087 enum value of profile_quality.
3089 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3091 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3092 -mext-string options.
3094 2018-01-12 Richard Biener <rguenther@suse.de>
3096 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3097 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3098 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3100 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3102 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3104 * configure.ac (--with-long-double-format): Add support for the
3105 configuration option to change the default long double format on
3107 * config.gcc (powerpc*-linux*-*): Likewise.
3108 * configure: Regenerate.
3109 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3110 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3111 used without modification.
3113 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3115 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3116 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3117 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3118 MISC_BUILTIN_SPEC_BARRIER.
3119 (rs6000_init_builtins): Likewise.
3120 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3122 (speculation_barrier): New define_insn.
3123 * doc/extend.texi: Document __builtin_speculation_barrier.
3125 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3128 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3129 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3130 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3132 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3133 integral modes instead of "ss" and "sd".
3134 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3135 vectors with 32-bit and 64-bit elements.
3136 (vecdupssescalarmodesuffix): New mode attribute.
3137 (vec_dup<mode>): Use it.
3139 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3142 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3143 frame if argument is passed on stack.
3145 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3148 * ree.c (combine_reaching_defs): Optimize also
3149 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3150 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3152 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3155 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3157 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3160 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3161 after they are computed.
3163 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3165 PR tree-optimization/83695
3166 * gimple-loop-linterchange.cc
3167 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3168 reset cached scev information after interchange.
3169 (pass_linterchange::execute): Remove call to scev_reset_htab.
3171 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3173 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3174 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3175 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3176 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3177 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3178 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3179 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3180 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3181 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3182 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3183 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3184 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3185 (V_lane_reg): Likewise.
3186 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3188 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3189 (vfmal_lane_low<mode>_intrinsic,
3190 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3191 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3192 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3193 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3194 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3195 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3197 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3199 * config/arm/arm-cpus.in (fp16fml): New feature.
3200 (ALL_SIMD): Add fp16fml.
3201 (armv8.2-a): Add fp16fml as an option.
3202 (armv8.3-a): Likewise.
3203 (armv8.4-a): Add fp16fml as part of fp16.
3204 * config/arm/arm.h (TARGET_FP16FML): Define.
3205 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3207 * config/arm/arm-modes.def (V2HF): Define.
3208 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3209 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3210 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3211 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3212 vfmsl_low, vfmsl_high): New set of builtins.
3213 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3214 (vfml_op): New code attribute.
3215 (VFMLHALVES): New int iterator.
3216 (VFML, VFMLSEL): New mode attributes.
3217 (V_reg): Define mapping for V2HF.
3218 (V_hi, V_lo): New mode attributes.
3219 (VF_constraint): Likewise.
3220 (vfml_half, vfml_half_selector): New int attributes.
3221 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3223 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3224 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3226 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3227 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3228 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3229 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3231 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3232 Document new effective target and option set.
3234 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3236 * config/arm/arm-cpus.in (armv8_4): New feature.
3237 (ARMv8_4a): New fgroup.
3238 (armv8.4-a): New arch.
3239 * config/arm/arm-tables.opt: Regenerate.
3240 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3241 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3242 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3243 Add matching rules for -march=armv8.4-a and extensions.
3244 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3246 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3249 * config/rx/rx.md (BW): New mode attribute.
3250 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3252 2018-01-11 Richard Biener <rguenther@suse.de>
3254 PR tree-optimization/83435
3255 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3256 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3257 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3259 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3260 Alan Hayward <alan.hayward@arm.com>
3261 David Sherwood <david.sherwood@arm.com>
3263 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3265 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3266 (aarch64_print_address_internal): Use it to check for a zero offset.
3268 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3269 Alan Hayward <alan.hayward@arm.com>
3270 David Sherwood <david.sherwood@arm.com>
3272 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3273 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3274 Return a poly_int64 rather than a HOST_WIDE_INT.
3275 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3276 rather than a HOST_WIDE_INT.
3277 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3278 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3279 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3280 final_offset from HOST_WIDE_INT to poly_int64.
3281 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3282 to_constant when getting the number of units in an Advanced SIMD
3284 (aarch64_builtin_vectorized_function): Check for a constant number
3286 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3288 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3289 attribute instead of GET_MODE_NUNITS.
3290 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3291 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3292 GET_MODE_SIZE for fixed-size registers.
3293 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3294 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3295 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3296 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3297 (aarch64_print_operand, aarch64_print_address_internal)
3298 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3299 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3300 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3301 Handle polynomial GET_MODE_SIZE.
3302 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3303 wider than SImode without modification.
3304 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3305 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3306 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3307 passing and returning SVE modes.
3308 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3309 rather than GEN_INT.
3310 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3311 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3312 (aarch64_allocate_and_probe_stack_space): Likewise.
3313 (aarch64_layout_frame): Cope with polynomial offsets.
3314 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3315 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3317 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3318 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3319 poly_int64 rather than a HOST_WIDE_INT.
3320 (aarch64_get_separate_components, aarch64_process_components)
3321 (aarch64_expand_prologue, aarch64_expand_epilogue)
3322 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3323 (aarch64_anchor_offset): New function, split out from...
3324 (aarch64_legitimize_address): ...here.
3325 (aarch64_builtin_vectorization_cost): Handle polynomial
3326 TYPE_VECTOR_SUBPARTS.
3327 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3329 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3330 number of elements from the PARALLEL rather than the mode.
3331 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3332 rather than GET_MODE_BITSIZE.
3333 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3334 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3335 (aarch64_expand_vec_perm_const_1): Handle polynomial
3336 d->perm.length () and d->perm elements.
3337 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3338 Apply to_constant to d->perm elements.
3339 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3340 polynomial CONST_VECTOR_NUNITS.
3341 (aarch64_move_pointer): Take amount as a poly_int64 rather
3343 (aarch64_progress_pointer): Avoid temporary variable.
3344 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3345 the mode attribute instead of GET_MODE.
3347 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3348 Alan Hayward <alan.hayward@arm.com>
3349 David Sherwood <david.sherwood@arm.com>
3351 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3352 x exists before using it.
3353 (aarch64_add_constant_internal): Rename to...
3354 (aarch64_add_offset_1): ...this. Replace regnum with separate
3355 src and dest rtxes. Handle the case in which they're different,
3356 including when the offset is zero. Replace scratchreg with an rtx.
3357 Use 2 additions if there is no spare register into which we can
3358 move a 16-bit constant.
3359 (aarch64_add_constant): Delete.
3360 (aarch64_add_offset): Replace reg with separate src and dest
3361 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3362 Use aarch64_add_offset_1.
3363 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3364 an rtx rather than an int. Take the delta as a poly_int64
3365 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3366 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3367 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3368 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3369 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3371 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3372 aarch64_add_constant.
3374 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3376 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3377 Use scalar_float_mode.
3379 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3381 * config/aarch64/aarch64-simd.md
3382 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3383 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3384 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3385 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3386 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3387 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3388 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3389 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3390 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3391 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3393 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3396 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3397 targ_options->x_arm_arch_string is non NULL.
3399 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3401 * config/aarch64/aarch64.h
3402 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3404 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3407 * expmed.c (emit_store_flag_force): Swap if const op0
3408 and change VOIDmode to mode of op0.
3410 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3412 PR rtl-optimization/83761
3413 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3414 than bytes to mode_for_size.
3416 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3419 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3420 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3423 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3426 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3427 when in layout mode.
3428 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3429 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3432 2018-01-10 Michael Collison <michael.collison@arm.com>
3434 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3435 * config/aarch64/aarch64-option-extension.def: Add
3436 AARCH64_OPT_EXTENSION of 'fp16fml'.
3437 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3438 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3439 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3440 * config/aarch64/constraints.md (Ui7): New constraint.
3441 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3442 (VFMLA_SEL_W): Ditto.
3445 (VFMLA16_LOW): New int iterator.
3446 (VFMLA16_HIGH): Ditto.
3447 (UNSPEC_FMLAL): New unspec.
3448 (UNSPEC_FMLSL): Ditto.
3449 (UNSPEC_FMLAL2): Ditto.
3450 (UNSPEC_FMLSL2): Ditto.
3451 (f16mac): New code attribute.
3452 * config/aarch64/aarch64-simd-builtins.def
3453 (aarch64_fmlal_lowv2sf): Ditto.
3454 (aarch64_fmlsl_lowv2sf): Ditto.
3455 (aarch64_fmlalq_lowv4sf): Ditto.
3456 (aarch64_fmlslq_lowv4sf): Ditto.
3457 (aarch64_fmlal_highv2sf): Ditto.
3458 (aarch64_fmlsl_highv2sf): Ditto.
3459 (aarch64_fmlalq_highv4sf): Ditto.
3460 (aarch64_fmlslq_highv4sf): Ditto.
3461 (aarch64_fmlal_lane_lowv2sf): Ditto.
3462 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3463 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3464 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3465 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3466 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3467 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3468 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3469 (aarch64_fmlal_lane_highv2sf): Ditto.
3470 (aarch64_fmlsl_lane_highv2sf): Ditto.
3471 (aarch64_fmlal_laneq_highv2sf): Ditto.
3472 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3473 (aarch64_fmlalq_lane_highv4sf): Ditto.
3474 (aarch64_fmlsl_lane_highv4sf): Ditto.
3475 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3476 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3477 * config/aarch64/aarch64-simd.md:
3478 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3479 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3480 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3481 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3482 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3483 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3484 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3485 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3486 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3487 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3488 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3489 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3490 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3491 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3492 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3493 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3494 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3495 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3496 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3497 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3498 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3499 (vfmlsl_low_u32): Ditto.
3500 (vfmlalq_low_u32): Ditto.
3501 (vfmlslq_low_u32): Ditto.
3502 (vfmlal_high_u32): Ditto.
3503 (vfmlsl_high_u32): Ditto.
3504 (vfmlalq_high_u32): Ditto.
3505 (vfmlslq_high_u32): Ditto.
3506 (vfmlal_lane_low_u32): Ditto.
3507 (vfmlsl_lane_low_u32): Ditto.
3508 (vfmlal_laneq_low_u32): Ditto.
3509 (vfmlsl_laneq_low_u32): Ditto.
3510 (vfmlalq_lane_low_u32): Ditto.
3511 (vfmlslq_lane_low_u32): Ditto.
3512 (vfmlalq_laneq_low_u32): Ditto.
3513 (vfmlslq_laneq_low_u32): Ditto.
3514 (vfmlal_lane_high_u32): Ditto.
3515 (vfmlsl_lane_high_u32): Ditto.
3516 (vfmlal_laneq_high_u32): Ditto.
3517 (vfmlsl_laneq_high_u32): Ditto.
3518 (vfmlalq_lane_high_u32): Ditto.
3519 (vfmlslq_lane_high_u32): Ditto.
3520 (vfmlalq_laneq_high_u32): Ditto.
3521 (vfmlslq_laneq_high_u32): Ditto.
3522 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3523 (AARCH64_FL_FOR_ARCH8_4): New.
3524 (AARCH64_ISA_F16FML): New ISA flag.
3525 (TARGET_F16FML): New feature flag for fp16fml.
3526 (doc/invoke.texi): Document new fp16fml option.
3528 2018-01-10 Michael Collison <michael.collison@arm.com>
3530 * config/aarch64/aarch64-builtins.c:
3531 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3532 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3533 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3534 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3535 (AARCH64_ISA_SHA3): New ISA flag.
3536 (TARGET_SHA3): New feature flag for sha3.
3537 * config/aarch64/iterators.md (sha512_op): New int attribute.
3538 (CRYPTO_SHA512): New int iterator.
3539 (UNSPEC_SHA512H): New unspec.
3540 (UNSPEC_SHA512H2): Ditto.
3541 (UNSPEC_SHA512SU0): Ditto.
3542 (UNSPEC_SHA512SU1): Ditto.
3543 * config/aarch64/aarch64-simd-builtins.def
3544 (aarch64_crypto_sha512hqv2di): New builtin.
3545 (aarch64_crypto_sha512h2qv2di): Ditto.
3546 (aarch64_crypto_sha512su0qv2di): Ditto.
3547 (aarch64_crypto_sha512su1qv2di): Ditto.
3548 (aarch64_eor3qv8hi): Ditto.
3549 (aarch64_rax1qv2di): Ditto.
3550 (aarch64_xarqv2di): Ditto.
3551 (aarch64_bcaxqv8hi): Ditto.
3552 * config/aarch64/aarch64-simd.md:
3553 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3554 (aarch64_crypto_sha512su0qv2di): Ditto.
3555 (aarch64_crypto_sha512su1qv2di): Ditto.
3556 (aarch64_eor3qv8hi): Ditto.
3557 (aarch64_rax1qv2di): Ditto.
3558 (aarch64_xarqv2di): Ditto.
3559 (aarch64_bcaxqv8hi): Ditto.
3560 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3561 (vsha512h2q_u64): Ditto.
3562 (vsha512su0q_u64): Ditto.
3563 (vsha512su1q_u64): Ditto.
3564 (veor3q_u16): Ditto.
3565 (vrax1q_u64): Ditto.
3567 (vbcaxq_u16): Ditto.
3568 * config/arm/types.md (crypto_sha512): New type attribute.
3569 (crypto_sha3): Ditto.
3570 (doc/invoke.texi): Document new sha3 option.
3572 2018-01-10 Michael Collison <michael.collison@arm.com>
3574 * config/aarch64/aarch64-builtins.c:
3575 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3576 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3577 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3578 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3579 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3580 (AARCH64_ISA_SM4): New ISA flag.
3581 (TARGET_SM4): New feature flag for sm4.
3582 * config/aarch64/aarch64-simd-builtins.def
3583 (aarch64_sm3ss1qv4si): Ditto.
3584 (aarch64_sm3tt1aq4si): Ditto.
3585 (aarch64_sm3tt1bq4si): Ditto.
3586 (aarch64_sm3tt2aq4si): Ditto.
3587 (aarch64_sm3tt2bq4si): Ditto.
3588 (aarch64_sm3partw1qv4si): Ditto.
3589 (aarch64_sm3partw2qv4si): Ditto.
3590 (aarch64_sm4eqv4si): Ditto.
3591 (aarch64_sm4ekeyqv4si): Ditto.
3592 * config/aarch64/aarch64-simd.md:
3593 (aarch64_sm3ss1qv4si): Ditto.
3594 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3595 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3596 (aarch64_sm4eqv4si): Ditto.
3597 (aarch64_sm4ekeyqv4si): Ditto.
3598 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3599 (sm3part_op): Ditto.
3600 (CRYPTO_SM3TT): Ditto.
3601 (CRYPTO_SM3PART): Ditto.
3602 (UNSPEC_SM3SS1): New unspec.
3603 (UNSPEC_SM3TT1A): Ditto.
3604 (UNSPEC_SM3TT1B): Ditto.
3605 (UNSPEC_SM3TT2A): Ditto.
3606 (UNSPEC_SM3TT2B): Ditto.
3607 (UNSPEC_SM3PARTW1): Ditto.
3608 (UNSPEC_SM3PARTW2): Ditto.
3609 (UNSPEC_SM4E): Ditto.
3610 (UNSPEC_SM4EKEY): Ditto.
3611 * config/aarch64/constraints.md (Ui2): New constraint.
3612 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3613 * config/arm/types.md (crypto_sm3): New type attribute.
3614 (crypto_sm4): Ditto.
3615 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3616 (vsm3tt1aq_u32): Ditto.
3617 (vsm3tt1bq_u32): Ditto.
3618 (vsm3tt2aq_u32): Ditto.
3619 (vsm3tt2bq_u32): Ditto.
3620 (vsm3partw1q_u32): Ditto.
3621 (vsm3partw2q_u32): Ditto.
3622 (vsm4eq_u32): Ditto.
3623 (vsm4ekeyq_u32): Ditto.
3624 (doc/invoke.texi): Document new sm4 option.
3626 2018-01-10 Michael Collison <michael.collison@arm.com>
3628 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3629 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3630 (AARCH64_FL_FOR_ARCH8_4): New.
3631 (AARCH64_FL_V8_4): New flag.
3632 (doc/invoke.texi): Document new armv8.4-a option.
3634 2018-01-10 Michael Collison <michael.collison@arm.com>
3636 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3637 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3638 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3639 * config/aarch64/aarch64-option-extension.def: Add
3640 AARCH64_OPT_EXTENSION of 'sha2'.
3641 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3642 (crypto): Disable sha2 and aes if crypto disabled.
3643 (crypto): Enable aes and sha2 if enabled.
3644 (simd): Disable sha2 and aes if simd disabled.
3645 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3647 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3648 (TARGET_SHA2): New feature flag for sha2.
3649 (TARGET_AES): New feature flag for aes.
3650 * config/aarch64/aarch64-simd.md:
3651 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3652 conditional on TARGET_AES.
3653 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3654 (aarch64_crypto_sha1hsi): Make pattern conditional
3656 (aarch64_crypto_sha1hv4si): Ditto.
3657 (aarch64_be_crypto_sha1hv4si): Ditto.
3658 (aarch64_crypto_sha1su1v4si): Ditto.
3659 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3660 (aarch64_crypto_sha1su0v4si): Ditto.
3661 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3662 (aarch64_crypto_sha256su0v4si): Ditto.
3663 (aarch64_crypto_sha256su1v4si): Ditto.
3664 (doc/invoke.texi): Document new aes and sha2 options.
3666 2018-01-10 Martin Sebor <msebor@redhat.com>
3668 PR tree-optimization/83781
3669 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3672 2018-01-11 Martin Sebor <msebor@gmail.com>
3673 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3675 PR tree-optimization/83501
3676 PR tree-optimization/81703
3678 * tree-ssa-strlen.c (get_string_cst): Rename...
3679 (get_string_len): ...to this. Handle global constants.
3680 (handle_char_store): Adjust.
3682 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3683 Jim Wilson <jimw@sifive.com>
3685 * config/riscv/riscv-protos.h (riscv_output_return): New.
3686 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3687 (riscv_attribute_table, riscv_output_return),
3688 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3689 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3690 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3692 (riscv_expand_prologue): Add early return for naked function.
3693 (riscv_expand_epilogue): Likewise.
3694 (riscv_function_ok_for_sibcall): Return false for naked function.
3695 (riscv_set_current_function): New.
3696 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3697 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3698 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3699 * doc/extend.texi (RISC-V Function Attributes): New.
3701 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3703 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3704 check for 128-bit long double before checking TCmode.
3705 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3706 128-bit long doubles before checking TFmode or TCmode.
3707 (FLOAT128_IBM_P): Likewise.
3709 2018-01-10 Martin Sebor <msebor@redhat.com>
3711 PR tree-optimization/83671
3712 * builtins.c (c_strlen): Unconditionally return zero for the empty
3714 Use -Warray-bounds for warnings.
3715 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3716 for non-constant array indices with COMPONENT_REF, arrays of
3717 arrays, and pointers to arrays.
3718 (gimple_fold_builtin_strlen): Determine and set length range for
3719 non-constant character arrays.
3721 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3724 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3727 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3729 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3731 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3734 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3735 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3736 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3737 indexed_or_indirect_operand predicate.
3738 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3739 (*vsx_le_perm_load_v8hi): Likewise.
3740 (*vsx_le_perm_load_v16qi): Likewise.
3741 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3742 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3743 (*vsx_le_perm_store_v8hi): Likewise.
3744 (*vsx_le_perm_store_v16qi): Likewise.
3745 (eight unnamed splitters): Likewise.
3747 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3749 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3750 * config/rs6000/emmintrin.h: Likewise.
3751 * config/rs6000/mmintrin.h: Likewise.
3752 * config/rs6000/xmmintrin.h: Likewise.
3754 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3757 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3759 * tree.c (tree_nop_conversion): Return true for location wrapper
3761 (maybe_wrap_with_location): New function.
3762 (selftest::check_strip_nops): New function.
3763 (selftest::test_location_wrappers): New function.
3764 (selftest::tree_c_tests): Call it.
3765 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3766 (maybe_wrap_with_location): New decl.
3767 (EXPR_LOCATION_WRAPPER_P): New macro.
3768 (location_wrapper_p): New inline function.
3769 (tree_strip_any_location_wrapper): New inline function.
3771 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3774 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3775 stack_realign_offset for the largest alignment of stack slot
3777 (ix86_find_max_used_stack_alignment): New function.
3778 (ix86_finalize_stack_frame_flags): Use it. Set
3779 max_used_stack_alignment if we don't realign stack.
3780 * config/i386/i386.h (machine_function): Add
3781 max_used_stack_alignment.
3783 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3785 * config/arm/arm.opt (-mbranch-cost): New option.
3786 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3789 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3792 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3793 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3795 2018-01-10 Richard Biener <rguenther@suse.de>
3798 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3799 early out so it also covers the case where we have a non-NULL
3802 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3804 PR tree-optimization/83753
3805 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3806 for non-strided grouped accesses if the number of elements is 1.
3808 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3811 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3812 * i386.h (TARGET_USE_GATHER): Define.
3813 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3815 2018-01-10 Martin Liska <mliska@suse.cz>
3818 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3819 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3821 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3822 CLEANUP_NO_PARTITIONING is not set.
3824 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3826 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3827 for vectors, as a partial revert of r254296.
3828 * rtl.h (const_vec_p): Delete.
3829 (const_vec_duplicate_p): Don't test for vector CONSTs.
3830 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3831 * expmed.c (make_tree): Likewise.
3834 * common.md (E, F): Use CONSTANT_P instead of checking for
3836 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3837 checking for CONST_VECTOR.
3839 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3842 * predict.c (force_edge_cold): Handle in more sane way edges
3845 2018-01-09 Carl Love <cel@us.ibm.com>
3847 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3849 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3850 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3851 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3852 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3853 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3854 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3855 * config/rs6000/rs6000-protos.h: Add extern defition for
3856 rs6000_generate_float2_double_code.
3857 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3859 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3860 (float2_v2df): Add define_expand.
3862 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3865 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3866 op_mode in the force_to_mode call.
3868 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3870 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3871 instead of checking each element individually.
3872 (aarch64_evpc_uzp): Likewise.
3873 (aarch64_evpc_zip): Likewise.
3874 (aarch64_evpc_ext): Likewise.
3875 (aarch64_evpc_rev): Likewise.
3876 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3877 instead of checking each element individually. Return true without
3879 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3880 whether all selected elements come from the same input, instead of
3881 checking each element individually. Remove calls to gen_rtx_REG,
3882 start_sequence and end_sequence and instead assert that no rtl is
3885 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3887 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3888 order of HIGH and CONST checks.
3890 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3892 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3893 if the destination isn't an SSA_NAME.
3895 2018-01-09 Richard Biener <rguenther@suse.de>
3897 PR tree-optimization/83668
3898 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3900 (canonicalize_loop_form): ... here, renamed from ...
3901 (canonicalize_loop_closed_ssa_form): ... this and amended to
3902 swap successor edges for loop exit blocks to make us use
3903 the RPO order we need for initial schedule generation.
3905 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3907 PR tree-optimization/64811
3908 * match.pd: When optimizing comparisons with Inf, avoid
3909 introducing or losing exceptions from comparisons with NaN.
3911 2018-01-09 Martin Liska <mliska@suse.cz>
3914 * asan.c (shadow_mem_size): Add gcc_assert.
3916 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3918 Don't save registers in main().
3921 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3922 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3923 * config/avr/avr.c (avr_set_current_function): Don't error if
3924 naked, OS_task or OS_main are specified at the same time.
3925 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3927 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3929 * common/config/avr/avr-common.c (avr_option_optimization_table):
3930 Switch on -mmain-is-OS_task for optimizing compilations.
3932 2018-01-09 Richard Biener <rguenther@suse.de>
3934 PR tree-optimization/83572
3935 * graphite.c: Include cfganal.h.
3936 (graphite_transform_loops): Connect infinite loops to exit
3937 and remove fake edges at the end.
3939 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3941 * ipa-inline.c (edge_badness): Revert accidental checkin.
3943 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3946 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3947 symbols; not inline clones.
3949 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3952 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3953 hard registers. Formatting fixes.
3955 PR preprocessor/83722
3956 * gcc.c (try_generate_repro): Pass
3957 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3958 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3961 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3962 Kito Cheng <kito.cheng@gmail.com>
3964 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3965 (riscv_leaf_function_p): Delete.
3966 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3968 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3970 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3972 (do_ifelse): New function.
3973 (do_isel): New function.
3974 (do_sub3): New function.
3975 (do_add3): New function.
3976 (do_load_mask_compare): New function.
3977 (do_overlap_load_compare): New function.
3978 (expand_compare_loop): New function.
3979 (expand_block_compare): Call expand_compare_loop() when appropriate.
3980 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3982 (-mblock-compare-inline-loop-limit): New option.
3984 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3987 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3988 Reverse order of second and third operands in first alternative.
3989 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3990 of first and second elements in UNSPEC_VPERMR vector.
3991 (altivec_expand_vec_perm_le): Likewise.
3993 2018-01-08 Jeff Law <law@redhat.com>
3995 PR rtl-optimizatin/81308
3996 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3997 (process_switch): If group_case_labels makes a change, then set
3999 (pass_convert_switch::execute): If a switch is converted, then
4000 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4002 PR rtl-optimization/81308
4003 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4006 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4008 PR target/83663 - Revert r255946
4009 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4010 generation for cases where splatting a value is not useful.
4011 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4012 across a vec_duplicate and a paradoxical subreg forming a vector
4013 mode to a vec_concat.
4015 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4017 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4018 -march=armv8.3-a variants.
4019 * config/arm/t-multilib: Likewise.
4020 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4022 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4024 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4026 (cceq_ior_compare_complement): Give it a name so I can use it, and
4027 change boolean_or_operator predicate to boolean_operator so it can
4028 be used to generate a crand.
4029 (eqne): New code iterator.
4030 (bd/bd_neg): New code_attrs.
4031 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4032 a single define_insn.
4033 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4034 decrement (bdnzt/bdnzf/bdzt/bdzf).
4035 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4036 with the new names of the branch decrement patterns, and added the
4037 names of the branch decrement conditional patterns.
4039 2018-01-08 Richard Biener <rguenther@suse.de>
4041 PR tree-optimization/83563
4042 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4045 2018-01-08 Richard Biener <rguenther@suse.de>
4048 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4050 2018-01-08 Richard Biener <rguenther@suse.de>
4052 PR tree-optimization/83685
4053 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4054 references to abnormals.
4056 2018-01-08 Richard Biener <rguenther@suse.de>
4059 * dwarf2out.c (output_indirect_strings): Handle empty
4060 skeleton_debug_str_hash.
4061 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4063 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4065 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4066 (emit_store_direct): Likewise.
4067 (arc_trampoline_adjust_address): Likewise.
4068 (arc_asm_trampoline_template): New function.
4069 (arc_initialize_trampoline): Use asm_trampoline_template.
4070 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4071 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4072 * config/arc/arc.md (flush_icache): Delete pattern.
4074 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4076 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4077 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4080 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4083 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4084 by not USED_FOR_TARGET.
4085 (make_pass_resolve_sw_modes): Likewise.
4087 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4089 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4092 2018-01-08 Richard Biener <rguenther@suse.de>
4095 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4097 2018-01-08 Richard Biener <rguenther@suse.de>
4100 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4102 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4105 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4106 basic blocks with a small number of successors.
4107 (convert_control_dep_chain_into_preds): Improve handling of
4109 (dump_predicates): Split apart into...
4110 (dump_pred_chain): ...here...
4111 (dump_pred_info): ...and here.
4112 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4113 (can_chain_union_be_invalidated_p): Improve check for invalidation
4115 (uninit_uses_cannot_happen): Avoid unnecessary if
4116 convert_control_dep_chain_into_preds yielded nothing.
4118 2018-01-06 Martin Sebor <msebor@redhat.com>
4120 PR tree-optimization/83640
4121 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4122 subtracting negative offset from size.
4123 (builtin_access::overlap): Adjust offset bounds of the access to fall
4124 within the size of the object if possible.
4126 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4128 PR rtl-optimization/83699
4129 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4130 extract_bit_field_as_subreg to cases in which the extracted
4131 value is also a vector.
4133 * lra-constraints.c (process_alt_operands): Test for the equivalence
4134 substitutions when detecting a possible reload cycle.
4136 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4139 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4140 by default if flag_selective_schedling{,2}. Formatting fixes.
4142 PR rtl-optimization/83682
4143 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4144 if it has non-VECTOR_MODE element mode.
4145 (vec_duplicate_p): Likewise.
4148 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4149 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4151 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4154 * config/i386/i386-builtin.def
4155 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4156 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4157 Require also OPTION_MASK_ISA_AVX512F in addition to
4158 OPTION_MASK_ISA_GFNI.
4159 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4160 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4161 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4162 to OPTION_MASK_ISA_GFNI.
4163 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4164 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4165 OPTION_MASK_ISA_AVX512BW.
4166 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4167 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4168 addition to OPTION_MASK_ISA_GFNI.
4169 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4170 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4171 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4172 to OPTION_MASK_ISA_GFNI.
4173 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4174 a requirement for all ISAs rather than any of them with a few
4176 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4178 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4179 bitmasks to be enabled with 3 exceptions, instead of requiring any
4180 enabled ISA with lots of exceptions.
4181 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4182 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4183 Change avx512bw in isa attribute to avx512f.
4184 * config/i386/sgxintrin.h: Add license boilerplate.
4185 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4186 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4187 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4188 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4190 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4191 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4192 temporarily sse2 rather than sse if not enabled already.
4195 * config/i386/sse.md (VI248_VLBW): Rename to ...
4196 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4197 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4198 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4199 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4200 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4201 mode iterator instead of VI248_VLBW.
4203 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4205 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4206 (record_modified): Skip clobbers; add debug output.
4207 (param_change_prob): Use sreal frequencies.
4209 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4211 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4212 punt for user-aligned variables.
4214 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4216 * tree-chrec.c (chrec_contains_symbols): Return true for
4219 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4222 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4223 of (x|y) == x for BICS pattern.
4225 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4227 PR tree-optimization/83605
4228 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4229 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4232 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4234 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4235 * config/epiphany/rtems.h: New file.
4237 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4238 Uros Bizjak <ubizjak@gmail.com>
4241 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4242 QIreg_operand instead of register_operand predicate.
4243 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4244 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4245 comments instead of -fmitigate[-_]rop.
4247 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4250 * cgraphunit.c (symbol_table::compile): Switch to text_section
4251 before calling assembly_start debug hook.
4252 * run-rtl-passes.c (run_rtl_passes): Likewise.
4255 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4257 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4258 range_int_cst_p rather than !symbolic_range_p before calling
4259 extract_range_from_multiplicative_op_1.
4261 2018-01-04 Jeff Law <law@redhat.com>
4263 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4264 redundant test in assertion.
4266 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4268 * doc/rtl.texi: Document machine_mode wrapper classes.
4270 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4272 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4275 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4277 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4278 the VEC_PERM_EXPR fold to fail.
4280 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4283 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4284 to switched_sections.
4286 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4289 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4292 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4295 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4296 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4298 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4301 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4302 is BLKmode and bitpos not zero or mode change is needed.
4304 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4307 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4310 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4313 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4314 instead of MULT rtx. Update all corresponding splitters.
4316 (*ssub<modesuffix>): Ditto.
4318 (*cmp_sadd_di): Update split patterns.
4319 (*cmp_sadd_si): Ditto.
4320 (*cmp_sadd_sidi): Ditto.
4321 (*cmp_ssub_di): Ditto.
4322 (*cmp_ssub_si): Ditto.
4323 (*cmp_ssub_sidi): Ditto.
4324 * config/alpha/predicates.md (const23_operand): New predicate.
4325 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4326 Look for ASHIFT, not MULT inner operand.
4327 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4329 2018-01-04 Martin Liska <mliska@suse.cz>
4331 PR gcov-profile/83669
4332 * gcov.c (output_intermediate_file): Add version to intermediate
4334 * doc/gcov.texi: Document new field 'version' in intermediate
4335 file format. Fix location of '-k' option of gcov command.
4337 2018-01-04 Martin Liska <mliska@suse.cz>
4340 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4342 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4344 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4346 2018-01-03 Martin Sebor <msebor@redhat.com>
4348 PR tree-optimization/83655
4349 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4350 checking calls with invalid arguments.
4352 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4354 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4355 (vectorizable_mask_load_store): Delete.
4356 (vectorizable_call): Return false for masked loads and stores.
4357 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4358 instead of gimple_assign_rhs1.
4359 (vectorizable_load): Handle IFN_MASK_LOAD.
4360 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4362 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4364 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4366 (vectorizable_mask_load_store): ...here.
4367 (vectorizable_load): ...and here.
4369 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4371 * tree-vect-stmts.c (vect_build_all_ones_mask)
4372 (vect_build_zero_merge_argument): New functions, split out from...
4373 (vectorizable_load): ...here.
4375 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4377 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4379 (vectorizable_mask_load_store): ...here.
4380 (vectorizable_store): ...and here.
4382 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4384 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4386 (vectorizable_mask_load_store): ...here.
4388 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4390 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4391 (vect_model_store_cost): Take a vec_load_store_type instead of a
4393 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4394 (vect_model_store_cost): Take a vec_load_store_type instead of a
4396 (vectorizable_mask_load_store): Update accordingly.
4397 (vectorizable_store): Likewise.
4398 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4400 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4402 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4403 IFN_MASK_LOAD calls here rather than...
4404 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4406 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4407 Alan Hayward <alan.hayward@arm.com>
4408 David Sherwood <david.sherwood@arm.com>
4410 * expmed.c (extract_bit_field_1): For vector extracts,
4411 fall back to extract_bit_field_as_subreg if vec_extract
4414 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4415 Alan Hayward <alan.hayward@arm.com>
4416 David Sherwood <david.sherwood@arm.com>
4418 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4419 they are variable or constant sized.
4420 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4421 slots for constant-sized data.
4423 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4424 Alan Hayward <alan.hayward@arm.com>
4425 David Sherwood <david.sherwood@arm.com>
4427 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4428 handling COND_EXPRs with boolean comparisons, try to find a better
4429 basis for the mask type than the boolean itself.
4431 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4433 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4434 is calculated and how it can be overridden.
4435 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4436 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4438 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4441 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4442 Alan Hayward <alan.hayward@arm.com>
4443 David Sherwood <david.sherwood@arm.com>
4445 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4446 Remove the mode argument.
4447 (aarch64_simd_valid_immediate): Remove the mode and inverse
4449 * config/aarch64/iterators.md (bitsize): New iterator.
4450 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4451 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4452 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4453 aarch64_simd_valid_immediate.
4454 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4455 (aarch64_reg_or_bic_imm): Likewise.
4456 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4457 with an insn_type enum and msl with a modifier_type enum.
4458 Replace element_width with a scalar_mode. Change the shift
4459 to unsigned int. Add constructors for scalar_float_mode and
4460 scalar_int_mode elements.
4461 (aarch64_vect_float_const_representable_p): Delete.
4462 (aarch64_can_const_movi_rtx_p)
4463 (aarch64_simd_scalar_immediate_valid_for_move)
4464 (aarch64_simd_make_constant): Update call to
4465 aarch64_simd_valid_immediate.
4466 (aarch64_advsimd_valid_immediate_hs): New function.
4467 (aarch64_advsimd_valid_immediate): Likewise.
4468 (aarch64_simd_valid_immediate): Remove mode and inverse
4469 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4470 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4471 and aarch64_float_const_representable_p on the result.
4472 (aarch64_output_simd_mov_immediate): Remove mode argument.
4473 Update call to aarch64_simd_valid_immediate and use of
4474 simd_immediate_info.
4475 (aarch64_output_scalar_simd_mov_immediate): Update call
4478 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4479 Alan Hayward <alan.hayward@arm.com>
4480 David Sherwood <david.sherwood@arm.com>
4482 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4483 (mode_nunits): Likewise CONST_MODE_NUNITS.
4484 * machmode.def (ADJUST_NUNITS): Document.
4485 * genmodes.c (mode_data::need_nunits_adj): New field.
4486 (blank_mode): Update accordingly.
4487 (adj_nunits): New variable.
4488 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4490 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4491 listed in adj_nunits.
4492 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4493 listed in adj_nunits. Don't emit case statements for such modes.
4494 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4495 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4496 nothing if adj_nunits is nonnull.
4497 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4498 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4499 (emit_mode_fbit): Update use of print_maybe_const_decl.
4500 (emit_move_size): Likewise. Treat the array as non-const
4502 (emit_mode_adjustments): Handle adj_nunits.
4504 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4506 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4507 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4508 (VECTOR_MODES): Use it.
4509 (make_vector_modes): Take the prefix as an argument.
4511 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4512 Alan Hayward <alan.hayward@arm.com>
4513 David Sherwood <david.sherwood@arm.com>
4515 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4516 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4517 for MODE_VECTOR_BOOL.
4518 * machmode.def (VECTOR_BOOL_MODE): Document.
4519 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4520 (make_vector_bool_mode): New function.
4521 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4523 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4524 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4526 * stor-layout.c (int_mode_for_mode): Likewise.
4527 * tree.c (build_vector_type_for_mode): Likewise.
4528 * varasm.c (output_constant_pool_2): Likewise.
4529 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4530 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4531 for MODE_VECTOR_BOOL.
4532 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4533 of mode class checks.
4534 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4535 instead of a list of mode class checks.
4536 (expand_vector_scalar_condition): Likewise.
4537 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4539 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4540 Alan Hayward <alan.hayward@arm.com>
4541 David Sherwood <david.sherwood@arm.com>
4543 * machmode.h (mode_size): Change from unsigned short to
4545 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4546 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4547 or if measurement_type is not polynomial.
4548 (fixed_size_mode::includes_p): Check for constant-sized modes.
4549 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4550 return a poly_uint16 rather than an unsigned short.
4551 (emit_mode_size): Change the type of mode_size from unsigned short
4552 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4553 (emit_mode_adjustments): Cope with polynomial vector sizes.
4554 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4556 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4558 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4559 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4560 * caller-save.c (setup_save_areas): Likewise.
4561 (replace_reg_with_saved_mem): Likewise.
4562 * calls.c (emit_library_call_value_1): Likewise.
4563 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4564 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4565 (gen_lowpart_for_combine): Likewise.
4566 * convert.c (convert_to_integer_1): Likewise.
4567 * cse.c (equiv_constant, cse_insn): Likewise.
4568 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4569 (cselib_subst_to_values): Likewise.
4570 * dce.c (word_dce_process_block): Likewise.
4571 * df-problems.c (df_word_lr_mark_ref): Likewise.
4572 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4573 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4574 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4575 (rtl_for_decl_location): Likewise.
4576 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4577 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4578 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4579 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4580 (expand_expr_real_1): Likewise.
4581 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4582 (pad_below): Likewise.
4583 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4584 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4585 * ira.c (get_subreg_tracking_sizes): Likewise.
4586 * ira-build.c (ira_create_allocno_objects): Likewise.
4587 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4588 (ira_sort_regnos_for_alter_reg): Likewise.
4589 * ira-costs.c (record_operand_costs): Likewise.
4590 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4591 (resolve_simple_move): Likewise.
4592 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4593 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4594 (lra_constraints): Likewise.
4595 (CONST_POOL_OK_P): Reject variable-sized modes.
4596 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4597 (add_pseudo_to_slot, lra_spill): Likewise.
4598 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4599 * optabs-query.c (get_best_extraction_insn): Likewise.
4600 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4601 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4602 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4603 * recog.c (offsettable_address_addr_space_p): Likewise.
4604 * regcprop.c (maybe_mode_change): Likewise.
4605 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4606 * regrename.c (build_def_use): Likewise.
4607 * regstat.c (dump_reg_info): Likewise.
4608 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4609 (find_reloads, find_reloads_subreg_address): Likewise.
4610 * reload1.c (eliminate_regs_1): Likewise.
4611 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4612 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4613 (simplify_binary_operation_1, simplify_subreg): Likewise.
4614 * targhooks.c (default_function_arg_padding): Likewise.
4615 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4616 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4617 (verify_gimple_assign_ternary): Likewise.
4618 * tree-inline.c (estimate_move_cost): Likewise.
4619 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4620 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4621 (get_address_cost_ainc): Likewise.
4622 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4623 (vect_supportable_dr_alignment): Likewise.
4624 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4625 (vectorizable_reduction): Likewise.
4626 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4627 (vectorizable_operation, vectorizable_load): Likewise.
4628 * tree.c (build_same_sized_truth_vector_type): Likewise.
4629 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4630 * var-tracking.c (emit_note_insn_var_location): Likewise.
4631 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4632 (ADDR_VEC_ALIGN): Likewise.
4634 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4635 Alan Hayward <alan.hayward@arm.com>
4636 David Sherwood <david.sherwood@arm.com>
4638 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4640 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4641 or if measurement_type is polynomial.
4642 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4643 * combine.c (make_extraction): Likewise.
4644 * dse.c (find_shift_sequence): Likewise.
4645 * dwarf2out.c (mem_loc_descriptor): Likewise.
4646 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4647 (extract_bit_field, extract_low_bits): Likewise.
4648 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4649 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4650 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4651 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4652 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4653 * reload.c (find_reloads): Likewise.
4654 * reload1.c (alter_reg): Likewise.
4655 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4656 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4657 * tree-if-conv.c (predicate_mem_writes): Likewise.
4658 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4659 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4660 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4661 * valtrack.c (dead_debug_insert_temp): Likewise.
4662 * varasm.c (mergeable_constant_section): Likewise.
4663 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4665 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4666 Alan Hayward <alan.hayward@arm.com>
4667 David Sherwood <david.sherwood@arm.com>
4669 * expr.c (expand_assignment): Cope with polynomial mode sizes
4670 when assigning to a CONCAT.
4672 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4673 Alan Hayward <alan.hayward@arm.com>
4674 David Sherwood <david.sherwood@arm.com>
4676 * machmode.h (mode_precision): Change from unsigned short to
4678 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4680 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4681 or if measurement_type is not polynomial.
4682 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4683 in which the mode is already known to be a scalar_int_mode.
4684 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4685 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4687 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4688 for GET_MODE_PRECISION.
4689 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4690 for GET_MODE_PRECISION.
4691 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4693 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4694 (expand_field_assignment, make_extraction): Likewise.
4695 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4696 (get_last_value): Likewise.
4697 * convert.c (convert_to_integer_1): Likewise.
4698 * cse.c (cse_insn): Likewise.
4699 * expr.c (expand_expr_real_1): Likewise.
4700 * lra-constraints.c (simplify_operand_subreg): Likewise.
4701 * optabs-query.c (can_atomic_load_p): Likewise.
4702 * optabs.c (expand_atomic_load): Likewise.
4703 (expand_atomic_store): Likewise.
4704 * ree.c (combine_reaching_defs): Likewise.
4705 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4706 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4707 * tree.h (type_has_mode_precision_p): Likewise.
4708 * ubsan.c (instrument_si_overflow): Likewise.
4710 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4711 Alan Hayward <alan.hayward@arm.com>
4712 David Sherwood <david.sherwood@arm.com>
4714 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4715 polynomial numbers of units.
4716 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4717 (valid_vector_subparts_p): New function.
4718 (build_vector_type): Remove temporary shim and take the number
4719 of units as a poly_uint64 rather than an int.
4720 (build_opaque_vector_type): Take the number of units as a
4721 poly_uint64 rather than an int.
4722 * tree.c (build_vector_from_ctor): Handle polynomial
4723 TYPE_VECTOR_SUBPARTS.
4724 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4725 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4726 (build_vector_from_val): If the number of units is variable,
4727 use build_vec_duplicate_cst for constant operands and
4728 VEC_DUPLICATE_EXPR otherwise.
4729 (make_vector_type): Remove temporary is_constant ().
4730 (build_vector_type, build_opaque_vector_type): Take the number of
4731 units as a poly_uint64 rather than an int.
4732 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4734 * cfgexpand.c (expand_debug_expr): Likewise.
4735 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4736 (store_constructor, expand_expr_real_1): Likewise.
4737 (const_scalar_mask_from_tree): Likewise.
4738 * fold-const-call.c (fold_const_reduction): Likewise.
4739 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4740 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4741 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4742 (fold_relational_const): Likewise.
4743 (native_interpret_vector): Likewise. Change the size from an
4744 int to an unsigned int.
4745 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4746 TYPE_VECTOR_SUBPARTS.
4747 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4748 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4749 duplicating a non-constant operand into a variable-length vector.
4750 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4751 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4752 * ipa-icf.c (sem_variable::equals): Likewise.
4753 * match.pd: Likewise.
4754 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4755 * print-tree.c (print_node): Likewise.
4756 * stor-layout.c (layout_type): Likewise.
4757 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4758 * tree-cfg.c (verify_gimple_comparison): Likewise.
4759 (verify_gimple_assign_binary): Likewise.
4760 (verify_gimple_assign_ternary): Likewise.
4761 (verify_gimple_assign_single): Likewise.
4762 * tree-pretty-print.c (dump_generic_node): Likewise.
4763 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4764 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4765 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4766 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4767 (vect_shift_permute_load_chain): Likewise.
4768 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4769 (expand_vector_condition, optimize_vector_constructor): Likewise.
4770 (lower_vec_perm, get_compute_type): Likewise.
4771 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4772 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4773 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4774 (vect_recog_mask_conversion_pattern): Likewise.
4775 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4776 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4777 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4778 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4779 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4780 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4781 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4782 (supportable_widening_operation): Likewise.
4783 (supportable_narrowing_operation): Likewise.
4784 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4786 * varasm.c (output_constant): Likewise.
4788 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4789 Alan Hayward <alan.hayward@arm.com>
4790 David Sherwood <david.sherwood@arm.com>
4792 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4793 so that both the length == 3 and length != 3 cases set up their
4794 own permute vectors. Add comments explaining why we know the
4795 number of elements is constant.
4796 (vect_permute_load_chain): Likewise.
4798 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4799 Alan Hayward <alan.hayward@arm.com>
4800 David Sherwood <david.sherwood@arm.com>
4802 * machmode.h (mode_nunits): Change from unsigned char to
4804 (ONLY_FIXED_SIZE_MODES): New macro.
4805 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4806 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4807 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4809 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4810 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4811 or if measurement_type is not polynomial.
4812 * genmodes.c (ZERO_COEFFS): New macro.
4813 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4815 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4816 Use ZERO_COEFFS when emitting initializers.
4817 * data-streamer.h (bp_pack_poly_value): New function.
4818 (bp_unpack_poly_value): Likewise.
4819 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4820 for GET_MODE_NUNITS.
4821 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4822 for GET_MODE_NUNITS.
4823 * tree.c (make_vector_type): Remove temporary shim and make
4824 the real function take the number of units as a poly_uint64
4826 (build_vector_type_for_mode): Handle polynomial nunits.
4827 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4828 * emit-rtl.c (const_vec_series_p_1): Likewise.
4829 (gen_rtx_CONST_VECTOR): Likewise.
4830 * fold-const.c (test_vec_duplicate_folding): Likewise.
4831 * genrecog.c (validate_pattern): Likewise.
4832 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4833 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4834 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4835 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4836 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4837 * rtlanal.c (subreg_get_info): Likewise.
4838 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4839 (vect_grouped_load_supported): Likewise.
4840 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4841 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4842 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4843 (simplify_const_unary_operation, simplify_binary_operation_1)
4844 (simplify_const_binary_operation, simplify_ternary_operation)
4845 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4846 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4847 instead of CONST_VECTOR_NUNITS.
4848 * varasm.c (output_constant_pool_2): Likewise.
4849 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4850 explicit-encoded elements in the XVEC for variable-length vectors.
4852 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4854 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4856 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4857 Alan Hayward <alan.hayward@arm.com>
4858 David Sherwood <david.sherwood@arm.com>
4860 * coretypes.h (fixed_size_mode): Declare.
4861 (fixed_size_mode_pod): New typedef.
4862 * builtins.h (target_builtins::x_apply_args_mode)
4863 (target_builtins::x_apply_result_mode): Change type to
4864 fixed_size_mode_pod.
4865 * builtins.c (apply_args_size, apply_result_size, result_vector)
4866 (expand_builtin_apply_args_1, expand_builtin_apply)
4867 (expand_builtin_return): Update accordingly.
4869 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4871 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4872 * cselib.c (cselib_hash_rtx): Likewise.
4873 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4874 CONST_VECTOR encoding.
4876 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4877 Jeff Law <law@redhat.com>
4880 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4881 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4882 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4883 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4886 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4887 explicitly probe *sp in a noreturn function if there were any callee
4888 register saves or frame pointer is needed.
4890 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4893 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4894 BLKmode for ternary, binary or unary expressions.
4897 * var-tracking.c (delete_vta_debug_insn): New inline function.
4898 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4899 insns from get_insns () to NULL instead of each bb separately.
4900 Use delete_vta_debug_insn. No longer static.
4901 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4902 delete_vta_debug_insns callers.
4903 * rtl.h (delete_vta_debug_insns): Declare.
4904 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4905 instead of variable_tracking_main.
4907 2018-01-03 Martin Sebor <msebor@redhat.com>
4909 PR tree-optimization/83603
4910 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4911 arguments past the endof the argument list in functions declared
4912 without a prototype.
4913 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4914 Avoid checking when arguments are null.
4916 2018-01-03 Martin Sebor <msebor@redhat.com>
4919 * doc/extend.texi (attribute const): Fix a typo.
4920 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4921 issuing -Wsuggest-attribute for void functions.
4923 2018-01-03 Martin Sebor <msebor@redhat.com>
4925 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4926 offset_int::from instead of wide_int::to_shwi.
4927 (maybe_diag_overlap): Remove assertion.
4928 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4929 * gimple-ssa-sprintf.c (format_directive): Same.
4930 (parse_directive): Same.
4931 (sprintf_dom_walker::compute_format_length): Same.
4932 (try_substitute_return_value): Same.
4934 2018-01-03 Jeff Law <law@redhat.com>
4937 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4938 non-constant residual for zero at runtime and avoid probing in
4939 that case. Reorganize code for trailing problem to mirror handling
4942 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4944 PR tree-optimization/83501
4945 * tree-ssa-strlen.c (get_string_cst): New.
4946 (handle_char_store): Call get_string_cst.
4948 2018-01-03 Martin Liska <mliska@suse.cz>
4950 PR tree-optimization/83593
4951 * tree-ssa-strlen.c: Include tree-cfg.h.
4952 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4953 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4954 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4956 (strlen_dom_walker::before_dom_children): Call
4957 gimple_purge_dead_eh_edges. Dump tranformation with details
4959 (strlen_dom_walker::before_dom_children): Update call by adding
4960 new argument cleanup_eh.
4961 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4963 2018-01-03 Martin Liska <mliska@suse.cz>
4966 * cif-code.def (VARIADIC_THUNK): New enum value.
4967 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4970 2018-01-03 Jan Beulich <jbeulich@suse.com>
4972 * sse.md (mov<mode>_internal): Tighten condition for when to use
4973 vmovdqu<ssescalarsize> for TI and OI modes.
4975 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4977 Update copyright years.
4979 2018-01-03 Martin Liska <mliska@suse.cz>
4982 * ipa-visibility.c (function_and_variable_visibility): Skip
4983 functions with noipa attribure.
4985 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4987 * gcc.c (process_command): Update copyright notice dates.
4988 * gcov-dump.c (print_version): Ditto.
4989 * gcov.c (print_version): Ditto.
4990 * gcov-tool.c (print_version): Ditto.
4991 * gengtype.c (create_file): Ditto.
4992 * doc/cpp.texi: Bump @copying's copyright year.
4993 * doc/cppinternals.texi: Ditto.
4994 * doc/gcc.texi: Ditto.
4995 * doc/gccint.texi: Ditto.
4996 * doc/gcov.texi: Ditto.
4997 * doc/install.texi: Ditto.
4998 * doc/invoke.texi: Ditto.
5000 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5002 * vector-builder.h (vector_builder::m_full_nelts): Change from
5003 unsigned int to poly_uint64.
5004 (vector_builder::full_nelts): Update prototype accordingly.
5005 (vector_builder::new_vector): Likewise.
5006 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5007 (vector_builder::operator ==): Likewise.
5008 (vector_builder::finalize): Likewise.
5009 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5010 Take the number of elements as a poly_uint64 rather than an
5012 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5013 from unsigned int to poly_uint64.
5014 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5015 (vec_perm_indices::new_vector): Likewise.
5016 (vec_perm_indices::length): Likewise.
5017 (vec_perm_indices::nelts_per_input): Likewise.
5018 (vec_perm_indices::input_nelts): Likewise.
5019 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5020 number of elements per input as a poly_uint64 rather than an
5021 unsigned int. Use the original encoding for variable-length
5022 vectors, rather than clamping each individual element.
5023 For the second and subsequent elements in each pattern,
5024 clamp the step and base before clamping their sum.
5025 (vec_perm_indices::series_p): Handle polynomial element counts.
5026 (vec_perm_indices::all_in_range_p): Likewise.
5027 (vec_perm_indices_to_tree): Likewise.
5028 (vec_perm_indices_to_rtx): Likewise.
5029 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5030 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5031 (tree_vector_builder::new_binary_operation): Handle polynomial
5032 element counts. Return false if we need to know the number
5033 of elements at compile time.
5034 * fold-const.c (fold_vec_perm): Punt if the number of elements
5035 isn't known at compile time.
5037 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5039 * vec-perm-indices.h (vec_perm_builder): Change element type
5040 from HOST_WIDE_INT to poly_int64.
5041 (vec_perm_indices::element_type): Update accordingly.
5042 (vec_perm_indices::clamp): Handle polynomial element_types.
5043 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5044 (vec_perm_indices::all_in_range_p): Likewise.
5045 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5047 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5048 polynomial vec_perm_indices element types.
5049 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5050 * fold-const.c (fold_vec_perm): Likewise.
5051 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5052 * tree-vect-generic.c (lower_vec_perm): Likewise.
5053 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5054 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5055 element type to HOST_WIDE_INT.
5057 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5058 Alan Hayward <alan.hayward@arm.com>
5059 David Sherwood <david.sherwood@arm.com>
5061 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5062 rather than an int. Use plus_constant.
5063 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5064 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5066 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5067 Alan Hayward <alan.hayward@arm.com>
5068 David Sherwood <david.sherwood@arm.com>
5070 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5071 a HOST_WIDE_INT to a poly_int64.
5073 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5074 Alan Hayward <alan.hayward@arm.com>
5075 David Sherwood <david.sherwood@arm.com>
5077 * calls.c (load_register_parameters): Cope with polynomial
5078 mode sizes. Require a constant size for BLKmode parameters
5079 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5080 forces a parameter to be padded at the lsb end in order to
5081 fill a complete number of words, require the parameter size
5082 to be ordered wrt UNITS_PER_WORD.
5084 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5085 Alan Hayward <alan.hayward@arm.com>
5086 David Sherwood <david.sherwood@arm.com>
5088 * reload1.c (spill_stack_slot_width): Change element type
5089 from unsigned int to poly_uint64_pod.
5090 (alter_reg): Treat mode sizes as polynomial.
5092 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5093 Alan Hayward <alan.hayward@arm.com>
5094 David Sherwood <david.sherwood@arm.com>
5096 * reload.c (complex_word_subreg_p): New function.
5097 (reload_inner_reg_of_subreg, push_reload): Use it.
5099 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5100 Alan Hayward <alan.hayward@arm.com>
5101 David Sherwood <david.sherwood@arm.com>
5103 * lra-constraints.c (process_alt_operands): Reject matched
5104 operands whose sizes aren't ordered.
5105 (match_reload): Refer to this check here.
5107 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5108 Alan Hayward <alan.hayward@arm.com>
5109 David Sherwood <david.sherwood@arm.com>
5111 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5112 that the mode size is in the set {1, 2, 4, 8, 16}.
5114 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5115 Alan Hayward <alan.hayward@arm.com>
5116 David Sherwood <david.sherwood@arm.com>
5118 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5119 Use plus_constant instead of gen_rtx_PLUS.
5121 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5122 Alan Hayward <alan.hayward@arm.com>
5123 David Sherwood <david.sherwood@arm.com>
5125 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5126 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5127 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5128 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5129 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5130 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5131 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5132 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5133 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5134 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5136 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5137 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5138 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5139 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5140 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5141 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5142 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5143 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5144 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5145 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5147 * expr.c (emit_move_resolve_push): Treat the input and result
5148 of PUSH_ROUNDING as a poly_int64.
5149 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5150 (emit_push_insn): Likewise.
5151 * lra-eliminations.c (mark_not_eliminable): Likewise.
5152 * recog.c (push_operand): Likewise.
5153 * reload1.c (elimination_effects): Likewise.
5154 * rtlanal.c (nonzero_bits1): Likewise.
5155 * calls.c (store_one_arg): Likewise. Require the padding to be
5156 known at compile time.
5158 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5159 Alan Hayward <alan.hayward@arm.com>
5160 David Sherwood <david.sherwood@arm.com>
5162 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5163 Use plus_constant instead of gen_rtx_PLUS.
5165 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5166 Alan Hayward <alan.hayward@arm.com>
5167 David Sherwood <david.sherwood@arm.com>
5169 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
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 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5177 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5178 via stack temporaries. Treat the mode size as polynomial too.
5180 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5181 Alan Hayward <alan.hayward@arm.com>
5182 David Sherwood <david.sherwood@arm.com>
5184 * expr.c (expand_expr_real_2): When handling conversions involving
5185 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5186 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5187 as a poly_uint64 too.
5189 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5190 Alan Hayward <alan.hayward@arm.com>
5191 David Sherwood <david.sherwood@arm.com>
5193 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5195 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5196 Alan Hayward <alan.hayward@arm.com>
5197 David Sherwood <david.sherwood@arm.com>
5199 * combine.c (can_change_dest_mode): Handle polynomial
5200 REGMODE_NATURAL_SIZE.
5201 * expmed.c (store_bit_field_1): Likewise.
5202 * expr.c (store_constructor): Likewise.
5203 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5204 and polynomial REGMODE_NATURAL_SIZE.
5205 (gen_lowpart_common): Likewise.
5206 * reginfo.c (record_subregs_of_mode): Likewise.
5207 * rtlanal.c (read_modify_subreg_p): Likewise.
5209 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5210 Alan Hayward <alan.hayward@arm.com>
5211 David Sherwood <david.sherwood@arm.com>
5213 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5214 numbers of elements.
5216 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5217 Alan Hayward <alan.hayward@arm.com>
5218 David Sherwood <david.sherwood@arm.com>
5220 * match.pd: Cope with polynomial numbers of vector elements.
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 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5227 in a POINTER_PLUS_EXPR.
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 * omp-simd-clone.c (simd_clone_subparts): New function.
5234 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5235 (ipa_simd_modify_function_body): Likewise.
5237 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5238 Alan Hayward <alan.hayward@arm.com>
5239 David Sherwood <david.sherwood@arm.com>
5241 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5242 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5243 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5244 (expand_vector_condition, vector_element): Likewise.
5245 (subparts_gt): New function.
5246 (get_compute_type): Use subparts_gt.
5247 (count_type_subparts): Delete.
5248 (expand_vector_operations_1): Use subparts_gt instead of
5249 count_type_subparts.
5251 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5252 Alan Hayward <alan.hayward@arm.com>
5253 David Sherwood <david.sherwood@arm.com>
5255 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5256 (vect_compile_time_alias): ...this new function. Do the calculation
5257 on poly_ints rather than trees.
5258 (vect_prune_runtime_alias_test_list): Update call accordingly.
5260 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5261 Alan Hayward <alan.hayward@arm.com>
5262 David Sherwood <david.sherwood@arm.com>
5264 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5266 (vect_schedule_slp_instance): Likewise.
5268 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5269 Alan Hayward <alan.hayward@arm.com>
5270 David Sherwood <david.sherwood@arm.com>
5272 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5273 constant and extern definitions for variable-length vectors.
5274 (vect_get_constant_vectors): Note that the number of units
5275 is known to be constant.
5277 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5278 Alan Hayward <alan.hayward@arm.com>
5279 David Sherwood <david.sherwood@arm.com>
5281 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5282 of units as polynomial. Choose between WIDE and NARROW based
5285 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5286 Alan Hayward <alan.hayward@arm.com>
5287 David Sherwood <david.sherwood@arm.com>
5289 * tree-vect-stmts.c (simd_clone_subparts): New function.
5290 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5292 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5293 Alan Hayward <alan.hayward@arm.com>
5294 David Sherwood <david.sherwood@arm.com>
5296 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5297 vectors as polynomial. Use build_index_vector for
5300 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5301 Alan Hayward <alan.hayward@arm.com>
5302 David Sherwood <david.sherwood@arm.com>
5304 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5305 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5306 for variable-length vectors.
5307 (vectorizable_mask_load_store): Treat the number of units as
5308 polynomial, asserting that it is constant if the condition has
5309 already been enforced.
5310 (vectorizable_store, vectorizable_load): Likewise.
5312 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5313 Alan Hayward <alan.hayward@arm.com>
5314 David Sherwood <david.sherwood@arm.com>
5316 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5317 of units as polynomial. Punt if we can't tell at compile time
5318 which vector contains the final result.
5320 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5321 Alan Hayward <alan.hayward@arm.com>
5322 David Sherwood <david.sherwood@arm.com>
5324 * tree-vect-loop.c (vectorizable_induction): Treat the number
5325 of units as polynomial. Punt on SLP inductions. Use an integer
5326 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5327 cast of such a series for variable-length floating-point
5330 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5331 Alan Hayward <alan.hayward@arm.com>
5332 David Sherwood <david.sherwood@arm.com>
5334 * tree.h (build_index_vector): Declare.
5335 * tree.c (build_index_vector): New function.
5336 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5337 of units as polynomial, forcibly converting it to a constant if
5338 vectorizable_reduction has already enforced the condition.
5339 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5340 to create a {1,2,3,...} vector.
5341 (vectorizable_reduction): Treat the number of units as polynomial.
5342 Choose vectype_in based on the largest scalar element size rather
5343 than the smallest number of units. Enforce the restrictions
5346 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5347 Alan Hayward <alan.hayward@arm.com>
5348 David Sherwood <david.sherwood@arm.com>
5350 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5351 number of units as polynomial.
5353 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5354 Alan Hayward <alan.hayward@arm.com>
5355 David Sherwood <david.sherwood@arm.com>
5357 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5358 * target.def (autovectorize_vector_sizes): Return the vector sizes
5359 by pointer, using vector_sizes rather than a bitmask.
5360 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5361 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5362 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5364 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5365 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5366 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5367 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5368 * omp-general.c (omp_max_vf): Likewise.
5369 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5370 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5371 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5372 * tree-vect-slp.c (vect_slp_bb): Likewise.
5373 * doc/tm.texi: Regenerate.
5374 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5376 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5377 the vector size as a poly_uint64 rather than an unsigned int.
5378 (current_vector_size): Change from an unsigned int to a poly_uint64.
5379 (get_vectype_for_scalar_type): Update accordingly.
5380 * tree.h (build_truth_vector_type): Take the size and number of
5381 units as a poly_uint64 rather than an unsigned int.
5382 (build_vector_type): Add a temporary overload that takes
5383 the number of units as a poly_uint64 rather than an unsigned int.
5384 * tree.c (make_vector_type): Likewise.
5385 (build_truth_vector_type): Take the number of units as a poly_uint64
5386 rather than an unsigned int.
5388 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5389 Alan Hayward <alan.hayward@arm.com>
5390 David Sherwood <david.sherwood@arm.com>
5392 * target.def (get_mask_mode): Take the number of units and length
5393 as poly_uint64s rather than unsigned ints.
5394 * targhooks.h (default_get_mask_mode): Update accordingly.
5395 * targhooks.c (default_get_mask_mode): Likewise.
5396 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5397 * doc/tm.texi: Regenerate.
5399 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5400 Alan Hayward <alan.hayward@arm.com>
5401 David Sherwood <david.sherwood@arm.com>
5403 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5404 * omp-general.c (omp_max_vf): Likewise.
5405 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5406 (expand_omp_simd): Handle polynomial safelen.
5407 * omp-low.c (omplow_simd_context): Add a default constructor.
5408 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5409 (lower_rec_simd_input_clauses): Update accordingly.
5410 (lower_rec_input_clauses): Likewise.
5412 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5413 Alan Hayward <alan.hayward@arm.com>
5414 David Sherwood <david.sherwood@arm.com>
5416 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5417 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5418 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5419 (vect_analyze_slp_cost): Likewise.
5420 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5421 (vect_model_load_cost): Likewise.
5423 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5424 Alan Hayward <alan.hayward@arm.com>
5425 David Sherwood <david.sherwood@arm.com>
5427 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5428 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5429 from an unsigned int * to a poly_uint64_pod *.
5430 (calculate_unrolling_factor): New function.
5431 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5433 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5434 Alan Hayward <alan.hayward@arm.com>
5435 David Sherwood <david.sherwood@arm.com>
5437 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5438 from an unsigned int to a poly_uint64.
5439 (_loop_vec_info::slp_unrolling_factor): Likewise.
5440 (_loop_vec_info::vectorization_factor): Change from an int
5442 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5443 (vect_get_num_vectors): New function.
5444 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5445 (vect_get_num_copies): Use vect_get_num_vectors.
5446 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5447 to an unsigned int *.
5448 (vect_analyze_data_refs): Change min_vf from an int * to a
5450 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5451 than an unsigned HOST_WIDE_INT.
5452 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5453 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5454 to an unsigned int *.
5455 (vect_analyze_data_ref_dependences): Likewise.
5456 (vect_compute_data_ref_alignment): Handle polynomial vf.
5457 (vect_enhance_data_refs_alignment): Likewise.
5458 (vect_prune_runtime_alias_test_list): Likewise.
5459 (vect_shift_permute_load_chain): Likewise.
5460 (vect_supportable_dr_alignment): Likewise.
5461 (dependence_distance_ge_vf): Take the vectorization factor as a
5462 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5463 (vect_analyze_data_refs): Change min_vf from an int * to a
5465 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5466 vfm1 as a poly_uint64 rather than an int. Make the same change
5467 for the returned bound_scalar.
5468 (vect_gen_vector_loop_niters): Handle polynomial vf.
5469 (vect_do_peeling): Likewise. Update call to
5470 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5471 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5473 * tree-vect-loop.c (vect_determine_vectorization_factor)
5474 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5475 (vect_get_known_peeling_cost): Likewise.
5476 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5477 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5478 (vect_transform_loop): Likewise. Use the lowest possible VF when
5479 updating the upper bounds of the loop.
5480 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5482 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5483 polynomial unroll factors.
5484 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5485 (vect_make_slp_decision): Likewise.
5486 (vect_supported_load_permutation_p): Likewise, and polynomial
5488 (vect_analyze_slp_cost): Handle polynomial vf.
5489 (vect_slp_analyze_node_operations): Likewise.
5490 (vect_slp_analyze_bb_1): Likewise.
5491 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5492 than an unsigned HOST_WIDE_INT.
5493 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5494 (vectorizable_load): Handle polynomial vf.
5495 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5497 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5499 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5500 Alan Hayward <alan.hayward@arm.com>
5501 David Sherwood <david.sherwood@arm.com>
5503 * match.pd: Handle bit operations involving three constants
5504 and try to fold one pair.
5506 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5508 * tree-vect-loop-manip.c: Include gimple-fold.h.
5509 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5510 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5511 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5512 Add a path that uses a step of VF instead of 1, but disable it
5514 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5515 and niters_no_overflow parameters. Update calls to
5516 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5517 Create a new SSA name if the latter choses to use a ste other
5518 than zero, and return it via niters_vector_mult_vf_var.
5519 * tree-vect-loop.c (vect_transform_loop): Update calls to
5520 vect_do_peeling, vect_gen_vector_loop_niters and
5521 slpeel_make_loop_iterate_ntimes.
5522 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5523 (vect_gen_vector_loop_niters): Update declarations after above changes.
5525 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5527 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5528 128-bit round to integer instructions.
5529 (ceil<mode>2): Likewise.
5530 (btrunc<mode>2): Likewise.
5531 (round<mode>2): Likewise.
5533 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5535 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5536 unaligned VSX load/store on P8/P9.
5537 (expand_block_clear): Allow the use of unaligned VSX
5538 load/store on P8/P9.
5540 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5542 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5544 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5545 swap associated with both a load and a store.
5547 2018-01-02 Andrew Waterman <andrew@sifive.com>
5549 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5550 * config/riscv/riscv.md (clear_cache): Use it.
5552 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5554 * web.c: Remove out-of-date comment.
5556 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5558 * expr.c (fixup_args_size_notes): Check that any existing
5559 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5560 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5561 (emit_single_push_insn): ...here.
5563 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5565 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5566 (const_vector_encoded_nelts): New function.
5567 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5568 (const_vector_int_elt, const_vector_elt): Declare.
5569 * emit-rtl.c (const_vector_int_elt_1): New function.
5570 (const_vector_elt): Likewise.
5571 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5572 of CONST_VECTOR_ELT.
5574 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5576 * expr.c: Include rtx-vector-builder.h.
5577 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5578 directly on the tree encoding.
5579 (const_vector_from_tree): Likewise.
5580 * optabs.c: Include rtx-vector-builder.h.
5581 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5582 sequence of "u" values.
5583 * vec-perm-indices.c: Include rtx-vector-builder.h.
5584 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5585 directly on the vec_perm_indices encoding.
5587 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5589 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5590 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5591 * rtx-vector-builder.h: New file.
5592 * rtx-vector-builder.c: Likewise.
5593 * rtl.h (rtx_def::u2): Add a const_vector field.
5594 (CONST_VECTOR_NPATTERNS): New macro.
5595 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5596 (CONST_VECTOR_DUPLICATE_P): Likewise.
5597 (CONST_VECTOR_STEPPED_P): Likewise.
5598 (CONST_VECTOR_ENCODED_ELT): Likewise.
5599 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5600 (unwrap_const_vec_duplicate): Likewise.
5601 (const_vec_series_p): Check for a non-duplicated vector encoding.
5602 Say that the function only returns true for integer vectors.
5603 * emit-rtl.c: Include rtx-vector-builder.h.
5604 (gen_const_vec_duplicate_1): Delete.
5605 (gen_const_vector): Call gen_const_vec_duplicate instead of
5606 gen_const_vec_duplicate_1.
5607 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5608 (gen_const_vec_duplicate): Use rtx_vector_builder.
5609 (gen_const_vec_series): Likewise.
5610 (gen_rtx_CONST_VECTOR): Likewise.
5611 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5612 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5613 Build a new vector rather than modifying a CONST_VECTOR in-place.
5614 (handle_special_swappables): Update call accordingly.
5615 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5616 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5617 Build a new vector rather than modifying a CONST_VECTOR in-place.
5618 (handle_special_swappables): Update call accordingly.
5620 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5622 * simplify-rtx.c (simplify_const_binary_operation): Use
5623 CONST_VECTOR_ELT instead of XVECEXP.
5625 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5627 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5628 the selector elements to be different from the data elements
5629 if the selector is a VECTOR_CST.
5630 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5631 ssizetype for the selector.
5633 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5635 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5636 before testing each element individually.
5637 * tree-vect-generic.c (lower_vec_perm): Likewise.
5639 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5641 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5642 * selftest-run-tests.c (selftest::run_tests): Call it.
5643 * vector-builder.h (vector_builder::operator ==): New function.
5644 (vector_builder::operator !=): Likewise.
5645 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5646 (vec_perm_indices::all_from_input_p): New function.
5647 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5648 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5649 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5650 instead of reading the VECTOR_CST directly. Detect whether both
5651 vector inputs are the same before constructing the vec_perm_indices,
5652 and update the number of inputs argument accordingly. Use the
5653 utility functions added above. Only construct sel2 if we need to.
5655 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5657 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5658 the broadcast of the low byte.
5659 (expand_mult_highpart): Use an explicit encoding for the permutes.
5660 * optabs-query.c (can_mult_highpart_p): Likewise.
5661 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5662 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5663 (vectorizable_bswap): Likewise.
5664 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5665 explicit encoding for the power-of-2 permutes.
5666 (vect_permute_store_chain): Likewise.
5667 (vect_grouped_load_supported): Likewise.
5668 (vect_permute_load_chain): Likewise.
5670 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5672 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5673 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5674 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5675 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5676 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5677 (vect_gen_perm_mask_any): Likewise.
5679 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5681 * int-vector-builder.h: New file.
5682 * vec-perm-indices.h: Include int-vector-builder.h.
5683 (vec_perm_indices): Redefine as an int_vector_builder.
5684 (auto_vec_perm_indices): Delete.
5685 (vec_perm_builder): Redefine as a stand-alone class.
5686 (vec_perm_indices::vec_perm_indices): New function.
5687 (vec_perm_indices::clamp): Likewise.
5688 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5689 (vec_perm_indices::new_vector): New function.
5690 (vec_perm_indices::new_expanded_vector): Update for new
5691 vec_perm_indices class.
5692 (vec_perm_indices::rotate_inputs): New function.
5693 (vec_perm_indices::all_in_range_p): Operate directly on the
5694 encoded form, without computing elided elements.
5695 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5696 encoding. Update for new vec_perm_indices class.
5697 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5698 the given vec_perm_builder.
5699 (expand_vec_perm_var): Update vec_perm_builder constructor.
5700 (expand_mult_highpart): Use vec_perm_builder instead of
5701 auto_vec_perm_indices.
5702 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5703 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5704 or double series encoding as appropriate.
5705 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5706 vec_perm_indices instead of auto_vec_perm_indices.
5707 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5708 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5709 (vect_permute_store_chain): Likewise.
5710 (vect_grouped_load_supported): Likewise.
5711 (vect_permute_load_chain): Likewise.
5712 (vect_shift_permute_load_chain): Likewise.
5713 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5714 (vect_transform_slp_perm_load): Likewise.
5715 (vect_schedule_slp_instance): Likewise.
5716 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5717 (vectorizable_mask_load_store): Likewise.
5718 (vectorizable_bswap): Likewise.
5719 (vectorizable_store): Likewise.
5720 (vectorizable_load): Likewise.
5721 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5722 vec_perm_indices instead of auto_vec_perm_indices. Use
5723 tree_to_vec_perm_builder to read the vector from a tree.
5724 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5725 vec_perm_builder instead of a vec_perm_indices.
5726 (have_whole_vector_shift): Use vec_perm_builder and
5727 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5728 truncation to calc_vec_perm_mask_for_shift.
5729 (vect_create_epilog_for_reduction): Likewise.
5730 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5731 from auto_vec_perm_indices to vec_perm_indices.
5732 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5733 instead of changing individual elements.
5734 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5735 the vector in d.perm.
5736 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5737 from auto_vec_perm_indices to vec_perm_indices.
5738 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5739 instead of changing individual elements.
5740 (arm_vectorize_vec_perm_const): Use new_vector to install
5741 the vector in d.perm.
5742 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5743 Update vec_perm_builder constructor.
5744 (rs6000_expand_interleave): Likewise.
5745 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5746 (rs6000_expand_interleave): Likewise.
5748 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5750 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5751 to qimode could truncate the indices.
5752 * optabs.c (expand_vec_perm_var): Likewise.
5754 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5756 * Makefile.in (OBJS): Add vec-perm-indices.o.
5757 * vec-perm-indices.h: New file.
5758 * vec-perm-indices.c: Likewise.
5759 * target.h (vec_perm_indices): Replace with a forward class
5761 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5762 * optabs.h: Include vec-perm-indices.h.
5763 (expand_vec_perm): Delete.
5764 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5765 (expand_vec_perm_const): Declare.
5766 * target.def (vec_perm_const_ok): Replace with...
5767 (vec_perm_const): ...this new hook.
5768 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5769 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5770 * doc/tm.texi: Regenerate.
5771 * optabs.def (vec_perm_const): Delete.
5772 * doc/md.texi (vec_perm_const): Likewise.
5773 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5774 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5775 expand_vec_perm for constant permutation vectors. Assert that
5776 the mode of variable permutation vectors is the integer equivalent
5777 of the mode that is being permuted.
5778 * optabs-query.h (selector_fits_mode_p): Declare.
5779 * optabs-query.c: Include vec-perm-indices.h.
5780 (selector_fits_mode_p): New function.
5781 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5782 is defined, instead of checking whether the vec_perm_const_optab
5783 exists. Use targetm.vectorize.vec_perm_const instead of
5784 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5785 fit in the vector mode before using a variable permute.
5786 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5787 vec_perm_indices instead of an rtx.
5788 (expand_vec_perm): Replace with...
5789 (expand_vec_perm_const): ...this new function. Take the selector
5790 as a vec_perm_indices rather than an rtx. Also take the mode of
5791 the selector. Update call to shift_amt_for_vec_perm_mask.
5792 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5793 Use vec_perm_indices::new_expanded_vector to expand the original
5794 selector into bytes. Check whether the indices fit in the vector
5795 mode before using a variable permute.
5796 (expand_vec_perm_var): Make global.
5797 (expand_mult_highpart): Use expand_vec_perm_const.
5798 * fold-const.c: Includes vec-perm-indices.h.
5799 * tree-ssa-forwprop.c: Likewise.
5800 * tree-vect-data-refs.c: Likewise.
5801 * tree-vect-generic.c: Likewise.
5802 * tree-vect-loop.c: Likewise.
5803 * tree-vect-slp.c: Likewise.
5804 * tree-vect-stmts.c: Likewise.
5805 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5807 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5808 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5809 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5810 (aarch64_vectorize_vec_perm_const): ...this new function.
5811 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5812 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5813 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5814 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5815 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5816 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5817 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5819 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5820 check for NEON modes.
5821 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5822 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5823 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5824 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5826 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5827 the old VEC_PERM_CONST conditions.
5828 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5829 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5830 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5831 (ia64_vectorize_vec_perm_const_ok): Merge into...
5832 (ia64_vectorize_vec_perm_const): ...this new function.
5833 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5834 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5835 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5836 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5837 * config/mips/mips.c (mips_expand_vec_perm_const)
5838 (mips_vectorize_vec_perm_const_ok): Merge into...
5839 (mips_vectorize_vec_perm_const): ...this new function.
5840 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5841 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5842 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5843 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5844 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5845 (rs6000_expand_vec_perm_const): Delete.
5846 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5848 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5849 (altivec_expand_vec_perm_const_le): Take each operand individually.
5850 Operate on constant selectors rather than rtxes.
5851 (altivec_expand_vec_perm_const): Likewise. Update call to
5852 altivec_expand_vec_perm_const_le.
5853 (rs6000_expand_vec_perm_const): Delete.
5854 (rs6000_vectorize_vec_perm_const_ok): Delete.
5855 (rs6000_vectorize_vec_perm_const): New function.
5856 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5857 an element count and rtx array.
5858 (rs6000_expand_extract_even): Update call accordingly.
5859 (rs6000_expand_interleave): Likewise.
5860 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5861 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5862 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5863 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5864 (rs6000_expand_vec_perm_const): Delete.
5865 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5866 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5867 (altivec_expand_vec_perm_const_le): Take each operand individually.
5868 Operate on constant selectors rather than rtxes.
5869 (altivec_expand_vec_perm_const): Likewise. Update call to
5870 altivec_expand_vec_perm_const_le.
5871 (rs6000_expand_vec_perm_const): Delete.
5872 (rs6000_vectorize_vec_perm_const_ok): Delete.
5873 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5874 reference to the SPE evmerge intructions.
5875 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5876 an element count and rtx array.
5877 (rs6000_expand_extract_even): Update call accordingly.
5878 (rs6000_expand_interleave): Likewise.
5879 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5880 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5882 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5884 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5886 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5887 vector mode and that that mode matches the mode of the data
5889 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5890 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5891 directly using expand_vec_perm_1 when forcing selectors into
5893 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5895 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5897 * optabs-query.h (can_vec_perm_p): Delete.
5898 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5899 * optabs-query.c (can_vec_perm_p): Split into...
5900 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5901 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5902 particular selector is valid.
5903 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5904 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5905 (vect_grouped_load_supported): Likewise.
5906 (vect_shift_permute_load_chain): Likewise.
5907 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5908 (vect_transform_slp_perm_load): Likewise.
5909 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5910 (vectorizable_bswap): Likewise.
5911 (vect_gen_perm_mask_checked): Likewise.
5912 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5913 implementations of variable permutation vectors into account
5914 when deciding which selector to use.
5915 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5916 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5917 with a false third argument.
5918 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5919 to test whether the constant selector is valid and can_vec_perm_var_p
5920 to test whether a variable selector is valid.
5922 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5924 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5925 * optabs-query.c (can_vec_perm_p): Likewise.
5926 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5927 instead of vec_perm_indices.
5928 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5929 (vect_gen_perm_mask_checked): Likewise,
5930 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5931 (vect_gen_perm_mask_checked): Likewise,
5933 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5935 * optabs-query.h (qimode_for_vec_perm): Declare.
5936 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5937 (qimode_for_vec_perm): ...this new function.
5938 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5940 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5942 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5943 does not have a conditional at the top.
5945 2018-01-02 Richard Biener <rguenther@suse.de>
5947 * ipa-inline.c (big_speedup_p): Fix expression.
5949 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5952 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5955 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5959 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5960 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5961 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5962 cond_taken_branch_cost 3->4.
5964 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5966 PR tree-optimization/83581
5967 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5968 TODO_cleanup_cfg if any changes have been made.
5971 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5972 convert_modes if target mode has the right side, but different mode
5976 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5977 last argument when extracting from CONCAT. If either from_real or
5978 from_imag is NULL, use expansion through memory. If result is not
5979 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5980 the parts directly to inner mode, if even that fails, use expansion
5984 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5985 check for bswap in mode rather than HImode and use that in expand_unop
5988 Copyright (C) 2018 Free Software Foundation, Inc.
5990 Copying and distribution of this file, with or without modification,
5991 are permitted in any medium without royalty provided the copyright
5992 notice and this notice are preserved.