1 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
3 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
4 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
6 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
7 Remove BSWAP handing for big-endian targets and use the form of
8 LD1RQ appropariate for the mode.
10 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
12 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
13 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
16 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
19 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
20 check for operands that need to go through aarch64_sve_reload_be.
22 2018-02-01 Jakub Jelinek <jakub@redhat.com>
24 PR tree-optimization/81661
25 PR tree-optimization/84117
26 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
27 * tree-eh.c: Include gimplify.h.
28 (find_trapping_overflow, replace_trapping_overflow,
29 rewrite_to_non_trapping_overflow): New functions.
30 * tree-vect-loop.c: Include tree-eh.h.
31 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
32 * tree-data-ref.c: Include tree-eh.h.
33 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
35 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
37 PR rtl-optimization/84123
38 * combine.c (change_zero_ext): Check if hard register satisfies
39 can_change_dest_mode before calling gen_lowpart_SUBREG.
41 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
44 * ira.c (ira_init_register_move_cost): Remove assert.
46 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
48 PR rtl-optimization/84071
49 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
50 * doc/tm.texi: Regenerate.
52 2018-01-31 Richard Biener <rguenther@suse.de>
54 PR tree-optimization/84132
55 * tree-data-ref.c (analyze_miv_subscript): Properly
56 check whether evolution_function_is_affine_multivariate_p
57 before calling gcd_of_steps_may_divide_p.
59 2018-01-31 Julia Koval <julia.koval@intel.com>
62 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
63 * config/i386/i386.md (rdpid_rex64) New.
64 (rdpid): Make 32bit only.
66 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
69 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
70 an IDENTIFIER_NODE for FUNCTION_TYPE's.
72 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
75 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
77 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
79 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
81 PR rtl-optimization/84071
82 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
83 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
85 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
87 * config/arc/arc.c (arc_handle_aux_attribute): New function.
88 (arc_attribute_table): Add 'aux' attribute.
89 (arc_in_small_data_p): Consider aux like variables.
90 (arc_is_aux_reg_p): New function.
91 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
92 (arc_get_aux_arg): New function.
93 (prepare_move_operands): Handle aux-register access.
94 (arc_handle_aux_attribute): New function.
95 * doc/extend.texi (ARC Variable attributes): Add subsection.
97 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
99 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
100 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
101 (arc_attribute_table): Add 'uncached' attribute.
102 (arc_print_operand): Print '.di' flag for uncached memory
104 (arc_in_small_data_p): Do not consider for small data the uncached
106 (arc_is_uncached_mem_p): New function.
107 * config/arc/predicates.md (compact_store_memory_operand): Check
108 for uncached memory accesses.
109 (nonvol_nonimm_operand): Likewise.
110 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
112 2018-01-31 Jakub Jelinek <jakub@redhat.com>
115 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
116 falign-loops=): Add Optimization flag.
118 2017-01-30 Jeff Law <law@redhat.com>
121 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
122 INT_REGISTERS_SAVED. Check it prior to calling
123 get_scratch_register_on_entry.
124 (ix86_adjust_stack_and_probe): Similarly.
125 (ix86_emit_probe_stack_range): Similarly.
126 (ix86_expand_prologue): Corresponding changes.
128 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
131 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
132 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
134 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
137 * lra-constraints.c (curr_insn_transform): Process AND in the
140 2018-01-30 Jakub Jelinek <jakub@redhat.com>
142 PR rtl-optimization/83986
143 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
144 dependence against last_pending_memory_flush in addition to
147 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
149 PR tree-optimization/81611
150 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
153 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
155 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
158 2018-01-30 Richard Biener <rguenther@suse.de>
159 Jakub Jelinek <jakub@redhat.com>
161 PR tree-optimization/84111
162 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
163 inner loops added during recursion, as they don't have up-to-date
166 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
169 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
170 (can_inline_edge_by_limits_p): ... here.
171 (can_early_inline_edge_p, check_callers,
172 update_caller_keys, update_callee_keys, recursive_inlining,
173 add_new_edges_to_heap, speculation_useful_p,
174 inline_small_functions,
175 inline_small_functions, flatten_function,
176 inline_to_all_callers_1): Update.
178 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
180 * profile-count.c (profile_count::combine_with_ipa_count): Handle
183 2018-01-30 Richard Biener <rguenther@suse.de>
185 PR tree-optimization/83008
186 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
187 invariant and constant vector uses in stmts when they need
190 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
193 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
194 * configure: Regenerate.
196 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
198 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
200 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
201 Use gen_rtx_REG rather than gen_lowpart.
203 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
205 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
206 rather than 0 when creating partial subregs.
208 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
210 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
213 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
216 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
217 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
218 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
219 flags. This restores the settings used before the 2017-07-24.
220 Turning off pre increment/decrement/modify allows IVOPTS to
221 optimize DF/SF loops where the index is an int.
223 2018-01-29 Richard Biener <rguenther@suse.de>
224 Kelvin Nilsen <kelvin@gcc.gnu.org>
227 * tree-vect-stmts.c (vectorizable_call): Don't call
228 targetm.vectorize_builtin_md_vectorized_function if callee is
231 2018-01-22 Carl Love <cel@us.ibm.com>
233 * doc/extend.tex: Fix typo in second arg in
234 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
236 2018-01-29 Richard Biener <rguenther@suse.de>
238 PR tree-optimization/84086
239 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
240 (flush_ssaname_freelist): When SSA names were released reset
243 2018-01-29 Richard Biener <rguenther@suse.de>
245 PR tree-optimization/84057
246 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
247 removed paths when removing edges.
249 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
251 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
252 -mfunction-return=@var{choice}.
254 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
257 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
259 (layout::print_source_line): Likewise.
260 (test_get_line_width_without_trailing_whitespace): Add test cases.
262 2018-01-27 Jakub Jelinek <jakub@redhat.com>
265 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
268 2018-01-26 Jim Wilson <jimw@sifive.com>
270 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
272 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
275 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
277 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
278 and CMP + SUB-immediate -> SUBS.
280 2018-01-26 Martin Sebor <msebor@redhat.com>
282 PR tree-optimization/83896
283 * tree-ssa-strlen.c (get_string_len): Rename...
284 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
285 Avoid assuming length is constant.
286 (handle_char_store): Use HOST_WIDE_INT for string length.
288 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
291 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
292 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
294 2018-01-26 Richard Biener <rguenther@suse.de>
296 PR rtl-optimization/84003
297 * dse.c (record_store): Only record redundant stores when
298 the earlier store aliases at least all accesses the later one does.
300 2018-01-26 Jakub Jelinek <jakub@redhat.com>
302 PR rtl-optimization/83985
303 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
304 REG_CFA_RESTORE insns.
305 (delete_unmarked_insns): Don't ignore separate shrink wrapping
306 REG_CFA_RESTORE insns here.
309 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
310 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
312 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
314 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
315 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
316 (arc_init): Likewise.
317 (arc_override_options): Likewise.
318 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
320 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
322 * config/arc/arc.h (TARGET_DBNZ): Define.
323 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
324 properly set the tune attribute.
325 (dbnz): Use TARGET_DBNZ guard.
326 * config/arc/arc.opt (mtune): Add core3 option.
328 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
330 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
331 recognize new pic like addresses.
332 (arc_delegitimize_address): Clean up.
334 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
336 * config/arc/arc-arches.def: Option mrf16 valid for all
338 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
339 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
340 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
341 * config/arc/arc-tables.opt: Regenerate.
342 * config/arc/arc.c (arc_conditional_register_usage): Handle
343 reduced register file case.
344 (arc_file_start): Set must have build attributes.
345 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
347 * config/arc/arc.opt (mrf16): Add new option.
348 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
349 * config/arc/genmultilib.awk: Handle new mrf16 option.
350 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
351 * config/arc/t-multilib: Regenerate.
352 * doc/invoke.texi (ARC Options): Document mrf16 option.
354 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
356 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
357 * config/arc/arc.c (arc_handle_secure_attribute): New function.
358 (arc_attribute_table): Add 'secure_call' attribute.
359 (arc_print_operand): Print secure call operand.
360 (arc_function_ok_for_sibcall): Don't optimize tail calls when
362 (arc_is_secure_call_p): New function. * config/arc/arc.md
363 (call_i): Add support for sjli instruction.
364 (call_value_i): Likewise.
365 * config/arc/constraints.md (Csc): New constraint.
367 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
368 John Eric Martin <John.Martin@emmicro-us.com>
370 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
371 * config/arc/arc.c (_arc_jli_section): New struct.
372 (arc_jli_section): New type.
373 (rc_jli_sections): New static variable.
374 (arc_handle_jli_attribute): New function.
375 (arc_attribute_table): Add jli_always and jli_fixed attribute.
376 (arc_file_end): New function.
377 (TARGET_ASM_FILE_END): Define.
378 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
379 (arc_add_jli_section): New function.
380 (jli_call_scan): Likewise.
381 (arc_reorg): Call jli_call_scan.
382 (arc_output_addsi): Remove 'S' from printing asm operand.
383 (arc_is_jli_call_p): New function.
384 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
386 (movhi_insn): Likewise.
387 (movsi_insn): Likewise.
388 (movsi_set_cc_insn): Likewise.
389 (loadqi_update): Likewise.
390 (load_zeroextendqisi_update): Likewise.
391 (load_signextendqisi_update): Likewise.
392 (loadhi_update): Likewise.
393 (load_zeroextendhisi_update): Likewise.
394 (load_signextendhisi_update): Likewise.
395 (loadsi_update): Likewise.
396 (loadsf_update): Likewise.
397 (movsicc_insn): Likewise.
398 (bset_insn): Likewise.
399 (bxor_insn): Likewise.
400 (bclr_insn): Likewise.
401 (bmsk_insn): Likewise.
402 (bicsi3_insn): Likewise.
403 (cmpsi_cc_c_insn): Likewise.
404 (movsi_ne): Likewise.
405 (movsi_cond_exec): Likewise.
406 (clrsbsi2): Likewise.
417 (call_i): Remove 'S' asm letter, add jli instruction.
418 (call_value_i): Likewise.
419 * config/arc/arc.op (mjli-always): New option.
420 * config/arc/constraints.md (Cji): New constraint.
421 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
423 (subsf3_fpx): Likewise.
424 (mulsf3_fpx): Likewise.
425 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
427 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
429 * doc/invoke.texi (ARC): Document mjli-always option.
431 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
433 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
434 and use incw and decw where possible
435 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
437 2018-01-26 Richard Biener <rguenther@suse.de>
439 PR tree-optimization/81082
440 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
441 association if it requires casting to unsigned.
442 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
443 from fold_plusminus_mult_expr to catch important cases late when
444 range info is available.
446 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
448 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
449 * configure.ac (hidden_linkonce): New test.
450 * configure: Regenerate.
451 * config.in: Regenerate.
453 2018-01-26 Julia Koval <julia.koval@intel.com>
455 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
456 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
457 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
458 _mm_mask_bitshuffle_epi64_mask): Fix type.
459 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
460 USI_FTYPE_V4DI_V4DI_USI): Remove.
461 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
462 __builtin_ia32_vpshufbitqmb256_mask,
463 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
464 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
465 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
467 2018-01-26 Alan Modra <amodra@gmail.com>
470 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
471 UNSPEC_VBPERMQ. Sort other unspecs.
473 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
475 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
477 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
480 * predict.c (drop_profile): Do not push/pop cfun; update also
482 (handle_missing_profiles): Fix logic looking for zero profiles.
484 2018-01-25 Jakub Jelinek <jakub@redhat.com>
487 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
488 on functions with #pragma omp declare simd or functions with simd
490 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
491 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
492 Remove trailing \n from warning_at calls.
494 2018-01-25 Tom de Vries <tom@codesourcery.com>
497 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
498 for neutered workers.
500 2018-01-24 Joseph Myers <joseph@codesourcery.com>
503 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
504 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
506 2017-01-24 Jeff Law <law@redhat.com>
509 * i386.c (get_probe_interval): Move to earlier point.
510 (ix86_compute_frame_layout): If -fstack-clash-protection and
511 the frame is larger than the probe interval, then use pushes
512 to save registers rather than reg->mem moves.
513 (ix86_expand_prologue): Remove conditional for int_registers_saved
516 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
519 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
520 min/max for never referenced object.
522 2018-01-24 Jakub Jelinek <jakub@redhat.com>
525 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
527 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
528 attributes from DECL_ATTRIBUTES (decl) without affecting
529 DECL_ATTRIBUTES (current_function_decl).
530 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
531 functions with non-NULL DECL_ABSTRACT_ORIGIN.
533 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
535 PR tree-optimization/83979
536 * fold-const.c (fold_comparison): Use constant_boolean_node
537 instead of boolean_{true,false}_node.
539 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
541 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
544 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
546 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
547 Simplify the clause that sets the length attribute.
548 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
549 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
550 clause that sets the length attribute.
551 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
553 2018-01-24 Tom de Vries <tom@codesourcery.com>
556 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
557 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
558 Add strict parameter.
559 (prevent_branch_around_nothing): Insert dummy insn between branch to
560 label and label with no ptx insn inbetween.
561 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
563 2018-01-24 Tom de Vries <tom@codesourcery.com>
566 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
567 for neutered threads in warp.
568 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
570 2018-01-24 Richard Biener <rguenther@suse.de>
572 PR tree-optimization/83176
573 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
576 2018-01-24 Richard Biener <rguenther@suse.de>
578 PR tree-optimization/82819
579 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
580 code generating pluses that are no-ops in the target precision.
582 2018-01-24 Richard Biener <rguenther@suse.de>
585 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
587 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
589 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
590 to merge probabilities.
591 * predict.c (probably_never_executed): Also mark as cold functions
592 with global 0 profile and guessed local profile.
593 * profile-count.c (profile_probability::combine_with_count): New
595 * profile-count.h (profile_probability::operator*,
596 profile_probability::operator*=, profile_probability::operator/,
597 profile_probability::operator/=): Reduce precision to adjusted
598 and set value to guessed on contradictory divisions.
599 (profile_probability::combine_with_freq): Remove.
600 (profile_probability::combine_wiht_count): Declare.
601 (profile_count::force_nonzero):: Set to adjusted.
602 (profile_count::probability_in):: Set quality to adjusted.
603 * tree-ssa-tail-merge.c (replace_block_by): Use
606 2018-01-23 Andrew Waterman <andrew@sifive.com>
607 Jim Wilson <jimw@sifive.com>
609 * config/riscv/riscv.c (riscv_stack_boundary): New.
610 (riscv_option_override): Set riscv_stack_boundary. Handle
611 riscv_preferred_stack_boundary_arg.
612 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
613 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
614 (STACK_BOUNDARY): Set to riscv_stack_boundary.
615 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
616 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
617 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
619 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
622 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
623 of struct ix86_frame.
624 (ix86_expand_epilogue): Likewise. Add a local variable for
625 the reg_save_offset field in struct ix86_frame.
627 2018-01-23 Bin Cheng <bin.cheng@arm.com>
629 PR tree-optimization/82604
630 * tree-loop-distribution.c (enum partition_kind): New enum item
631 PKIND_PARTIAL_MEMSET.
632 (partition_builtin_p): Support above new enum item.
633 (generate_code_for_partition): Ditto.
634 (compute_access_range): Differentiate cases that equality can be
635 proven at all loops, the innermost loops or no loops.
636 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
637 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
638 (finalize_partitions, distribute_loop): Don't fuse partition of
639 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
640 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
643 2018-01-23 Martin Liska <mliska@suse.cz>
645 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
646 order to ignore the predictor.
647 (PRED_POLYMORPHIC_CALL): Likewise.
648 (PRED_RECURSIVE_CALL): Likewise.
650 2018-01-23 Martin Liska <mliska@suse.cz>
652 * tree-profile.c (tree_profiling): Print function header to
653 aware reader which function we are working on.
654 * value-prof.c (gimple_find_values_to_profile): Do not print
655 not interesting value histograms.
657 2018-01-23 Martin Liska <mliska@suse.cz>
659 * profile-count.h (enum profile_quality): Add
660 profile_uninitialized as the first value. Do not number values
661 as they are zero based.
662 (profile_count::verify): Update sanity check.
663 (profile_probability::verify): Likewise.
665 2018-01-23 Nathan Sidwell <nathan@acm.org>
667 * doc/invoke.texi (ffor-scope): Deprecate.
669 2018-01-23 David Malcolm <dmalcolm@redhat.com>
671 PR tree-optimization/83510
672 * domwalk.c (set_all_edges_as_executable): New function.
673 (dom_walker::dom_walker): Convert bool param
674 "skip_unreachable_blocks" to enum reachability. Move setup of
675 edge flags to set_all_edges_as_executable and only do it when
676 reachability is REACHABLE_BLOCKS.
677 * domwalk.h (enum dom_walker::reachability): New enum.
678 (dom_walker::dom_walker): Convert bool param
679 "skip_unreachable_blocks" to enum reachability.
680 (set_all_edges_as_executable): New decl.
681 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
682 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
684 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
685 but converting true to REACHABLE_BLOCKS.
686 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
688 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
689 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
690 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
692 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
693 if check_all_array_refs will be called.
695 2018-01-23 David Malcolm <dmalcolm@redhat.com>
697 * tree.c (selftest::test_location_wrappers): Add more test
700 2018-01-23 David Malcolm <dmalcolm@redhat.com>
702 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
703 (selftest::test_bit_in_range): Likewise.
705 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
708 * doc/sourcebuild.texi (vect_float): Say that the selector
709 only describes the situation when -funsafe-math-optimizations is on.
710 (vect_float_strict): Document.
712 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
714 PR tree-optimization/83965
715 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
716 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
717 instead of checking only for a reduction.
718 (vect_recog_widen_sum_pattern): Likewise.
720 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
722 * predict.c (probably_never_executed): Only use precise profile info.
723 (compute_function_frequency): Skip after inlining hack since we now
724 have quality checking.
726 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
728 * profile-count.h (profile_probability::very_unlikely,
729 profile_probability::unlikely, profile_probability::even): Set
730 precision to guessed.
732 2018-01-23 Richard Biener <rguenther@suse.de>
734 PR tree-optimization/83963
735 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
736 Properly terminate dominator walk when crossing the exit edge not
737 when visiting its source block.
739 2018-01-23 Jakub Jelinek <jakub@redhat.com>
742 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
743 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
745 2018-01-22 Jakub Jelinek <jakub@redhat.com>
747 PR tree-optimization/83957
748 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
749 semicolon after for body surrounded by braces.
751 PR tree-optimization/83081
752 * profile-count.h (profile_probability::split): New method.
753 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
754 Use profile_probability::split.
755 (do_compare_rtx_and_jump): Fix adjustment of probabilities
756 when splitting a single conditional jump into 2.
758 2018-01-22 David Malcolm <dmalcolm@redhat.com>
760 PR tree-optimization/69452
761 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
764 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
766 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
767 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
768 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
770 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
772 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
773 * config/rl78/rl78.md: New define_expand "movdi"
774 * config/rl78/rl78.c: New function definition rl78_split_movdi
776 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
779 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
781 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
782 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
783 128-bit to produce an UNSPEC move to get the double word with the
784 signbit and then a shift directly to do signbit.
785 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
786 implementation with a new version that just does either a direct
787 move or a regular move. Move memory interface to separate insns.
788 Move insns so they are next to the expander.
789 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
790 with signbit move. Split big and little endian case.
791 (signbit<mode>2_dm_mem_le): Likewise.
792 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
793 (signbit<mode>2_dm2): Likewise.
795 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
797 * config/rl78/rl78.md: New define_expand "anddi3".
799 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
801 * config/rl78/rl78.md: New define_expand "umindi3".
803 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
805 * config/rl78/rl78.md: New define_expand "smindi3".
807 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
809 * config/rl78/rl78.md: New define_expand "smaxdi3".
811 2018-01-22 Carl Love <cel@us.ibm.com>
813 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
814 LVX_V1TI): Add macro expansion.
815 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
816 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
817 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
818 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
819 Change check to determine if the instruction is a byte reversing
820 entry. Fix typo in comment.
821 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
822 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
823 Add def_builtin calls for new builtins.
824 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
825 Add define_insn expansion.
827 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
829 * config/rl78/rl78.md: New define_expand "umaxdi3".
831 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
833 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
834 for non-QImode registers
836 2018-01-22 Richard Biener <rguenther@suse.de>
838 PR tree-optimization/83963
839 * graphite-scop-detection.c (scop_detection::get_sese): Delay
840 including the loop exit block.
841 (scop_detection::merge_sese): Likewise.
842 (scop_detection::add_scop): Do it here instead.
844 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
846 * doc/sourcebuild.texi (arm_softfloat): Document.
848 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
851 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
852 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
853 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
855 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
856 David Edelsohn <dje.gcc@gmail.com>
859 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
860 Change "crset eq" to "crset 2".
861 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
862 (*call_indirect_aix<mode>_nospec): Likewise.
863 (*call_value_indirect_aix<mode>_nospec): Likewise.
864 (*call_indirect_elfv2<mode>_nospec): Likewise.
865 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
866 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
867 change assembly output from . to $.
868 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
869 (indirect_jump<mode>_nospec): Change assembly output from . to $.
870 (*tablejump<mode>_internal1_nospec): Likewise.
872 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
875 * config/sh/sh_optimize_sett_clrt.cc:
876 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
878 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
880 PR tree-optimization/83940
881 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
882 offset_dt to vect_constant_def rather than vect_unknown_def_type.
883 (vect_check_load_store_mask): Add a mask_dt_out parameter and
884 use it to pass back the definition type.
885 (vect_check_store_rhs): Likewise rhs_dt_out.
886 (vect_build_gather_load_calls): Add a mask_dt argument and use
887 it instead of a call to vect_is_simple_use.
888 (vectorizable_store): Update calls to vect_check_load_store_mask
889 and vect_check_store_rhs. Use the dt returned by the latter instead
890 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
891 instead of calls to vect_is_simple_use. Pass the scalar rather
892 than the vector operand to vect_is_simple_use when handling
893 second and subsequent copies of an rhs value.
894 (vectorizable_load): Update calls to vect_check_load_store_mask
895 and vect_build_gather_load_calls. Use the cached mask_dt and
896 gs_info.offset_dt instead of calls to vect_is_simple_use.
898 2018-01-20 Jakub Jelinek <jakub@redhat.com>
901 * tree-emutls.c: Include gimplify.h.
902 (lower_emutls_2): New function.
903 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
904 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
905 it before further processing.
908 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
909 UINTVAL (trueop1) instead of INTVAL (op1).
911 2018-01-19 Jakub Jelinek <jakub@redhat.com>
915 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
916 INCOMING_FRAME_SP_OFFSET if not defined.
917 (scan_trace): Add ENTRY argument. If true and
918 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
919 emit a note to adjust the CFA offset.
920 (create_cfi_notes): Adjust scan_trace callers.
921 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
922 INCOMING_FRAME_SP_OFFSET in the CIE.
923 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
924 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
926 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
927 * doc/tm.texi: Regenerated.
929 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
931 PR rtl-optimization/83147
932 * lra-constraints.c (remove_inheritance_pseudos): Use
933 lra_substitute_pseudo_within_insn.
935 2018-01-19 Tom de Vries <tom@codesourcery.com>
936 Cesar Philippidis <cesar@codesourcery.com>
939 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
941 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
944 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
945 spaces for function labels.
947 2018-01-19 Martin Liska <mliska@suse.cz>
949 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
950 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
951 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
952 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
953 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
954 (PRED_CONST_RETURN): Change from 69 to 65.
955 (PRED_NULL_RETURN): Change from 91 to 71.
956 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
957 (PRED_LOOP_GUARD): Change from 66 to 73.
959 2018-01-19 Martin Liska <mliska@suse.cz>
961 * predict.c (predict_insn_def): Add new assert.
962 (struct branch_predictor): Change type to signed integer.
963 (test_prediction_value_range): Amend test to cover
965 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
966 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
967 (PRED_LOOP_ITERATIONS_MAX): Likewise.
968 (PRED_LOOP_IV_COMPARE): Likewise.
969 * predict.h (PROB_UNINITIALIZED): Define new constant.
971 2018-01-19 Martin Liska <mliska@suse.cz>
973 * predict.c (dump_prediction): Add new format for
974 analyze_brprob.py script which is enabled with -details
976 * profile-count.h (precise_p): New function.
978 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
980 PR tree-optimization/83922
981 * tree-vect-loop.c (vect_verify_full_masking): Return false if
982 there are no statements that need masking.
983 (vect_active_double_reduction_p): New function.
984 (vect_analyze_loop_operations): Use it when handling phis that
985 are not in the loop header.
987 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
989 PR tree-optimization/83914
990 * tree-vect-loop.c (vectorizable_induction): Don't convert
991 init_expr or apply the peeling adjustment for inductions
992 that are nested within the vectorized loop.
994 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
996 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
999 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1003 * function.h (gimplify_parameters): Add gimple_seq * argument.
1004 * function.c: Include gimple.h and options.h.
1005 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1006 for the added local temporaries if needed.
1007 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1008 if there are any parameter cleanups, wrap whole body into a
1009 try/finally with the cleanups.
1011 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1014 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1015 Use GET_MODE_CLASS for scalar floating point.
1017 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1021 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1022 Fix call of call_cgraph_insertion_hooks.
1024 2018-01-18 Martin Sebor <msebor@redhat.com>
1026 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1028 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1031 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1034 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1037 * common.opt: (-ffile-prefix-map): New option.
1038 * opts.c (common_handle_option): Defer it.
1039 * opts-global.c (handle_common_deferred_options): Handle it.
1040 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1041 * file-prefix-map.h: New file.
1042 (remap_debug_filename, add_debug_prefix_map): ...here.
1043 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1044 * final.c (debug_prefix_map, add_debug_prefix_map
1045 remap_debug_filename): Move to...
1046 * file-prefix-map.c: New file.
1047 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1048 generalize, get rid of alloca(), use strrchr() instead of strchr().
1049 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1050 Implement in terms of add_prefix_map().
1051 (remap_macro_filename, remap_debug_filename): Implement in term of
1053 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1054 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1055 * dbxout.c: Include file-prefix-map.h.
1056 * varasm.c: Likewise.
1057 * vmsdbgout.c: Likewise.
1058 * xcoffout.c: Likewise.
1059 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1060 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1061 * doc/invoke.texi (-ffile-prefix-map): Document.
1062 (-fdebug-prefix-map): Update description.
1064 2018-01-18 Martin Liska <mliska@suse.cz>
1066 * config/i386/i386.c (indirect_thunk_name): Document that also
1068 (output_indirect_thunk): Document why both instructions
1069 (pause and lfence) are generated.
1071 2018-01-18 Richard Biener <rguenther@suse.de>
1073 PR tree-optimization/83887
1074 * graphite-scop-detection.c
1075 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1076 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1077 (scop_detection::merge_sese): Re-implement with a flood-fill
1078 algorithm that properly finds a SESE region if it exists.
1080 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1083 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1084 pointer_diff optimizations use view_convert instead of convert.
1086 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1088 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1089 Generate different code for -mno-speculate-indirect-jumps.
1090 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1091 (*call_indirect_aix<mode>): Disable for
1092 -mno-speculate-indirect-jumps.
1093 (*call_indirect_aix<mode>_nospec): New define_insn.
1094 (*call_value_indirect_aix<mode>): Disable for
1095 -mno-speculate-indirect-jumps.
1096 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1097 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1098 -mno-speculate-indirect-jumps.
1099 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1101 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1103 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1104 long double type, set the flags for noting the default long double
1105 type, even if we don't pass or return a long double type.
1107 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1110 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1113 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1115 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1116 support for merge[hl].
1117 (fold_mergehl_helper): New helper function.
1118 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1119 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1120 (altivec_vmrglw_direct): Add xxmrglw insn.
1122 2018-01-17 Andrew Waterman <andrew@sifive.com>
1124 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1125 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1127 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1130 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1131 call the lto_location_cache before reading the
1132 DECL_SOURCE_LOCATION of the types.
1134 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1135 Richard Sandiford <richard.sandiford@linaro.org>
1137 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1138 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1139 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1140 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1141 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1143 * config/aarch64/constraints.md (aarch64_movti_operand):
1145 * config/aarch64/predicates.md (Uti): Add new constraint.
1147 2018-01-17 Carl Love <cel@us.ibm.com>
1148 * config/rs6000/vsx.md (define_expand xl_len_r,
1149 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1150 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1152 (define_expand, define_insn): Move the shift left from the
1153 define_insn to the define_expand for lxvl and stxvl instructions.
1154 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1155 and XL_LEN_R definitions to PURE.
1157 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1159 * config/i386/i386.c (indirect_thunk_name): Declare regno
1160 as unsigned int. Compare regno with INVALID_REGNUM.
1161 (output_indirect_thunk): Ditto.
1162 (output_indirect_thunk_function): Ditto.
1163 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1164 in the call to output_indirect_thunk_function.
1166 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1169 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1170 rather than the size of inner_type to determine the stack slot size
1171 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1173 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1176 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1179 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1181 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1182 endian Linux systems to optionally enable multilibs for selecting
1183 the long double type if the user configured an explicit type.
1184 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1185 have no long double multilibs if not defined.
1186 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1187 warn if the user used -mabi={ieee,ibm}longdouble and we built
1188 multilibs for long double.
1189 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1190 appropriate multilib option.
1191 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1193 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1194 for building long double multilibs.
1195 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1197 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1199 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1202 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1204 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1207 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1210 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1213 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1215 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1216 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1218 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1220 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1221 different rtl trees depending on TARGET_64BIT.
1222 (rs6000_gen_lvx): Likewise.
1224 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1226 * config/visium/visium.md (nop): Tweak comment.
1227 (hazard_nop): Likewise.
1229 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1231 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1232 -mspeculate-indirect-jumps.
1233 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1234 for -mno-speculate-indirect-jumps.
1235 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1236 (*call_value_indirect_elfv2<mode>): Disable for
1237 -mno-speculate-indirect-jumps.
1238 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1239 (indirect_jump): Emit different RTL for
1240 -mno-speculate-indirect-jumps.
1241 (*indirect_jump<mode>): Disable for
1242 -mno-speculate-indirect-jumps.
1243 (*indirect_jump<mode>_nospec): New define_insn.
1244 (tablejump): Emit different RTL for
1245 -mno-speculate-indirect-jumps.
1246 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1247 (tablejumpsi_nospec): New define_expand.
1248 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1249 (tablejumpdi_nospec): New define_expand.
1250 (*tablejump<mode>_internal1): Disable for
1251 -mno-speculate-indirect-jumps.
1252 (*tablejump<mode>_internal1_nospec): New define_insn.
1253 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1256 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1258 * caller-save.c (insert_save): Drop unnecessary parameter. All
1261 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1262 Richard Biener <rguenth@suse.de>
1265 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1266 return early, inline manually is_gimple_sizepos. Make sure if we
1267 call gimplify_expr we don't end up with a gimple constant.
1268 * tree.c (variably_modified_type_p): Don't return true for
1269 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1270 * gimplify.h (is_gimple_sizepos): Remove.
1272 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1274 PR tree-optimization/83857
1275 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1276 vectorizable_live_operation for pure SLP statements.
1277 (vectorizable_live_operation): Handle PHIs.
1279 2018-01-16 Richard Biener <rguenther@suse.de>
1281 PR tree-optimization/83867
1282 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1283 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1285 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1288 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1289 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1290 If off is not INTEGER_CST, issue a may not be aligned warning
1291 rather than isn't aligned. Use isn%'t rather than isn't.
1292 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1294 <case MULT_EXPR>: Improve the case when bottom and one of the
1295 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1296 operand, in that case check if the other operand is multiple of
1297 bottom divided by the INTEGER_CST operand.
1299 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1302 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1303 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1304 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1305 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1306 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1307 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1308 (pa_function_arg_size): New function.
1310 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1312 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1313 in a separate statement.
1315 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1317 PR tree-optimization/83847
1318 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1319 group gathers and scatters.
1321 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1323 PR rtl-optimization/86620
1324 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1326 PR rtl-optimization/83213
1327 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1328 to last if both are JUMP_INSNs.
1330 PR tree-optimization/83843
1331 * gimple-ssa-store-merging.c
1332 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1333 store_immediate_info for bswap/nop orig_stores.
1335 2018-01-15 Andrew Waterman <andrew@sifive.com>
1337 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1339 <UDIV>: Increase cost if !TARGET_DIV.
1341 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1343 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1344 (define_attr "cr_logical_3op"): New.
1345 (cceq_ior_compare): Adjust.
1346 (cceq_ior_compare_complement): Adjust.
1347 (*cceq_rev_compare): Adjust.
1348 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1349 (is_cracked_insn): Adjust.
1350 (insn_must_be_first_in_group): Adjust.
1351 * config/rs6000/40x.md: Adjust.
1352 * config/rs6000/440.md: Adjust.
1353 * config/rs6000/476.md: Adjust.
1354 * config/rs6000/601.md: Adjust.
1355 * config/rs6000/603.md: Adjust.
1356 * config/rs6000/6xx.md: Adjust.
1357 * config/rs6000/7450.md: Adjust.
1358 * config/rs6000/7xx.md: Adjust.
1359 * config/rs6000/8540.md: Adjust.
1360 * config/rs6000/cell.md: Adjust.
1361 * config/rs6000/e300c2c3.md: Adjust.
1362 * config/rs6000/e500mc.md: Adjust.
1363 * config/rs6000/e500mc64.md: Adjust.
1364 * config/rs6000/e5500.md: Adjust.
1365 * config/rs6000/e6500.md: Adjust.
1366 * config/rs6000/mpc.md: Adjust.
1367 * config/rs6000/power4.md: Adjust.
1368 * config/rs6000/power5.md: Adjust.
1369 * config/rs6000/power6.md: Adjust.
1370 * config/rs6000/power7.md: Adjust.
1371 * config/rs6000/power8.md: Adjust.
1372 * config/rs6000/power9.md: Adjust.
1373 * config/rs6000/rs64.md: Adjust.
1374 * config/rs6000/titan.md: Adjust.
1376 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1378 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1379 ix86_indirect_branch_register logic.
1381 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1383 * config/i386/constraints.md (Bs): Update
1384 ix86_indirect_branch_register check. Don't check
1385 ix86_indirect_branch_register with GOT_memory_operand.
1387 * config/i386/predicates.md (GOT_memory_operand): Don't check
1388 ix86_indirect_branch_register here.
1389 (GOT32_symbol_operand): Likewise.
1391 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1393 * config/i386/predicates.md (constant_call_address_operand):
1394 Rewrite ix86_indirect_branch_register logic.
1395 (sibcall_insn_operand): Likewise.
1397 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1399 * config/i386/constraints.md (Bs): Replace
1400 ix86_indirect_branch_thunk_register with
1401 ix86_indirect_branch_register.
1403 * config/i386/i386.md (indirect_jump): Likewise.
1404 (tablejump): Likewise.
1405 (*sibcall_memory): Likewise.
1406 (*sibcall_value_memory): Likewise.
1407 Peepholes of indirect call and jump via memory: Likewise.
1408 * config/i386/i386.opt: Likewise.
1409 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1410 (GOT_memory_operand): Likewise.
1411 (call_insn_operand): Likewise.
1412 (sibcall_insn_operand): Likewise.
1413 (GOT32_symbol_operand): Likewise.
1415 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1418 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1419 type rather than type addr's type points to.
1420 (expand_omp_atomic_mutex): Likewise.
1421 (expand_omp_atomic): Likewise.
1423 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1426 * config/i386/i386.c (output_indirect_thunk_function): Use
1427 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1428 for __x86_return_thunk.
1430 2018-01-15 Richard Biener <rguenther@suse.de>
1433 * expmed.c (extract_bit_field_1): Fix typo.
1435 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1438 * config/arm/iterators.md (VF): New mode iterator.
1439 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1440 Remove integer-related logic from pattern.
1441 (neon_vabd<mode>_3): Likewise.
1443 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1446 * common.opt (fstrict-overflow): No longer an alias.
1447 (fwrapv-pointer): New option.
1448 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1449 also for pointer types based on flag_wrapv_pointer.
1450 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1451 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1452 opts->x_flag_wrapv got set.
1453 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1454 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1455 POINTER_TYPE_OVERFLOW_UNDEFINED.
1456 * match.pd: Likewise in address comparison pattern.
1457 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1459 2018-01-15 Richard Biener <rguenther@suse.de>
1462 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1463 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1464 Reset type names to their identifier if their TYPE_DECL doesn't
1465 have linkage (and thus is used for ODR and devirt).
1466 (save_debug_info_for_decl): Remove.
1467 (save_debug_info_for_type): Likewise.
1468 (add_tree_to_fld_list): Adjust.
1469 * tree-pretty-print.c (dump_generic_node): Make dumping of
1470 type names more robust.
1472 2018-01-15 Richard Biener <rguenther@suse.de>
1474 * BASE-VER: Bump to 8.0.1.
1476 2018-01-14 Martin Sebor <msebor@redhat.com>
1479 * builtins.c (check_access): Avoid warning when the no-warning bit
1482 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1484 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1485 * ira-color (allocno_hard_regs_compare): Likewise.
1487 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1490 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1491 Use .pushsection/.popsection.
1493 2018-01-14 Martin Sebor <msebor@redhat.com>
1496 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1498 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1500 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1501 entry from extra_headers.
1502 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1503 extra_headers, make the list bitwise identical to the i?86-*-* one.
1505 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1507 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1508 -mcmodel=large with -mindirect-branch=thunk,
1509 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1510 -mfunction-return=thunk-extern.
1511 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1512 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1513 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1515 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1517 * config/i386/i386.c (print_reg): Print the name of the full
1518 integer register without '%'.
1519 (ix86_print_operand): Handle 'V'.
1520 * doc/extend.texi: Document 'V' modifier.
1522 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1524 * config/i386/constraints.md (Bs): Disallow memory operand for
1525 -mindirect-branch-register.
1527 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1528 (GOT_memory_operand): Likewise.
1529 (call_insn_operand): Likewise.
1530 (sibcall_insn_operand): Likewise.
1531 (GOT32_symbol_operand): Likewise.
1532 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1533 for -mindirect-branch-register.
1534 (tablejump): Likewise.
1535 (*sibcall_memory): Likewise.
1536 (*sibcall_value_memory): Likewise.
1537 Disallow peepholes of indirect call and jump via memory for
1538 -mindirect-branch-register.
1539 (*call_pop): Replace m with Bw.
1540 (*call_value_pop): Likewise.
1541 (*sibcall_pop_memory): Replace m with Bs.
1542 * config/i386/i386.opt (mindirect-branch-register): New option.
1543 * doc/invoke.texi: Document -mindirect-branch-register option.
1545 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1547 * config/i386/i386-protos.h (ix86_output_function_return): New.
1548 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1549 set function_return_type.
1550 (indirect_thunk_name): Add ret_p to indicate thunk for function
1552 (output_indirect_thunk_function): Pass false to
1553 indirect_thunk_name.
1554 (ix86_output_indirect_branch_via_reg): Likewise.
1555 (ix86_output_indirect_branch_via_push): Likewise.
1556 (output_indirect_thunk_function): Create alias for function
1557 return thunk if regno < 0.
1558 (ix86_output_function_return): New function.
1559 (ix86_handle_fndecl_attribute): Handle function_return.
1560 (ix86_attribute_table): Add function_return.
1561 * config/i386/i386.h (machine_function): Add
1562 function_return_type.
1563 * config/i386/i386.md (simple_return_internal): Use
1564 ix86_output_function_return.
1565 (simple_return_internal_long): Likewise.
1566 * config/i386/i386.opt (mfunction-return=): New option.
1567 (indirect_branch): Mention -mfunction-return=.
1568 * doc/extend.texi: Document function_return function attribute.
1569 * doc/invoke.texi: Document -mfunction-return= option.
1571 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1573 * config/i386/i386-opts.h (indirect_branch): New.
1574 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1575 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1576 with local indirect jump when converting indirect call and jump.
1577 (ix86_set_indirect_branch_type): New.
1578 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1579 (indirectlabelno): New.
1580 (indirect_thunk_needed): Likewise.
1581 (indirect_thunk_bnd_needed): Likewise.
1582 (indirect_thunks_used): Likewise.
1583 (indirect_thunks_bnd_used): Likewise.
1584 (INDIRECT_LABEL): Likewise.
1585 (indirect_thunk_name): Likewise.
1586 (output_indirect_thunk): Likewise.
1587 (output_indirect_thunk_function): Likewise.
1588 (ix86_output_indirect_branch_via_reg): Likewise.
1589 (ix86_output_indirect_branch_via_push): Likewise.
1590 (ix86_output_indirect_branch): Likewise.
1591 (ix86_output_indirect_jmp): Likewise.
1592 (ix86_code_end): Call output_indirect_thunk_function if needed.
1593 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1595 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1596 (ix86_attribute_table): Add indirect_branch.
1597 * config/i386/i386.h (machine_function): Add indirect_branch_type
1598 and has_local_indirect_jump.
1599 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1601 (tablejump): Likewise.
1602 (*indirect_jump): Use ix86_output_indirect_jmp.
1603 (*tablejump_1): Likewise.
1604 (simple_return_indirect_internal): Likewise.
1605 * config/i386/i386.opt (mindirect-branch=): New option.
1606 (indirect_branch): New.
1609 (thunk-inline): Likewise.
1610 (thunk-extern): Likewise.
1611 * doc/extend.texi: Document indirect_branch function attribute.
1612 * doc/invoke.texi: Document -mindirect-branch= option.
1614 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1617 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1619 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1621 * ipa-inline.c (want_inline_small_function_p): Return false if
1622 inlining has already failed with CIF_FINAL_ERROR.
1623 (update_caller_keys): Call want_inline_small_function_p before
1625 (update_callee_keys): Likewise.
1627 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1629 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1631 (rs6000_quadword_masked_address_p): Likewise.
1632 (quad_aligned_load_p): Likewise.
1633 (quad_aligned_store_p): Likewise.
1634 (const_load_sequence_p): Add comment to describe the outer-most loop.
1635 (mimic_memory_attributes_and_flags): New function.
1636 (rs6000_gen_stvx): Likewise.
1637 (replace_swapped_aligned_store): Likewise.
1638 (rs6000_gen_lvx): Likewise.
1639 (replace_swapped_aligned_load): Likewise.
1640 (replace_swapped_load_constant): Capitalize argument name in
1641 comment describing this function.
1642 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1643 and stores that access quad-word aligned addresses and replace
1644 with stvx or lvx instructions when appropriate.
1645 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1646 New function prototype.
1647 (rs6000_quadword_masked_address_p): Likewise.
1648 (rs6000_gen_lvx): Likewise.
1649 (rs6000_gen_stvx): Likewise.
1650 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1651 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1652 when memory address is aligned.
1653 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1654 this split to select lvx instruction when memory address is aligned.
1655 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1656 instruction when memory address is aligned.
1657 (*vsx_le_perm_load_v16qi): Likewise.
1658 (four unnamed splitters): Modify to select the stvx instruction
1659 when memory is aligned.
1661 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1663 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1664 which appears in the queue multiple times.
1666 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1667 Alan Hayward <alan.hayward@arm.com>
1668 David Sherwood <david.sherwood@arm.com>
1670 * tree-vectorizer.h (vec_lower_bound): New structure.
1671 (_loop_vec_info): Add check_nonzero and lower_bounds.
1672 (LOOP_VINFO_CHECK_NONZERO): New macro.
1673 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1674 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1675 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1676 fields. Make seg_len the distance travelled, not including the
1678 (dr_direction_indicator): Declare.
1679 (dr_zero_step_indicator): Likewise.
1680 (dr_known_forward_stride_p): Likewise.
1681 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1683 (runtime_alias_check_p): Allow runtime alias checks with
1685 (operator ==): Compare access_size and align.
1686 (prune_runtime_alias_test_list): Rework for new distinction between
1687 the access_size and seg_len.
1688 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1690 (get_segment_min_max): New function.
1691 (create_intersect_range_checks): Use it.
1692 (dr_step_indicator): New function.
1693 (dr_direction_indicator): Likewise.
1694 (dr_zero_step_indicator): Likewise.
1695 (dr_known_forward_stride_p): Likewise.
1696 * tree-loop-distribution.c (data_ref_segment_size): Return
1697 DR_STEP * (niters - 1).
1698 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1700 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1701 (vect_preserves_scalar_order_p): New function, split out from...
1702 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1703 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1704 (vect_vfa_access_size): New function.
1705 (vect_vfa_align): Likewise.
1706 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1707 (dump_lower_bound): New function.
1708 (vect_check_lower_bound): Likewise.
1709 (vect_small_gap_p): Likewise.
1710 (vectorizable_with_step_bound_p): Likewise.
1711 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1712 depencies if the vectorization factor is 1. Convert the checks
1713 for nonzero steps into checks on the bounds of DR_STEP. Try using
1714 a bunds check for variable steps if the minimum required step is
1715 relatively small. Update calls to the dr_with_seg_len
1716 constructor and to vect_compile_time_alias.
1717 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1719 (vect_loop_versioning): Call it.
1720 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1722 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1724 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1725 Alan Hayward <alan.hayward@arm.com>
1726 David Sherwood <david.sherwood@arm.com>
1728 * doc/sourcebuild.texi (vect_scatter_store): Document.
1729 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1731 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1733 * genopinit.c (main): Add supports_vec_scatter_store and
1734 supports_vec_scatter_store_cached to target_optabs.
1735 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1736 IFN_MASK_SCATTER_STORE.
1737 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1739 * internal-fn.h (internal_store_fn_p): Declare.
1740 (internal_fn_stored_value_index): Likewise.
1741 * internal-fn.c (scatter_store_direct): New macro.
1742 (expand_scatter_store_optab_fn): New function.
1743 (direct_scatter_store_optab_supported_p): New macro.
1744 (internal_store_fn_p): New function.
1745 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1746 IFN_MASK_SCATTER_STORE.
1747 (internal_fn_mask_index): Likewise.
1748 (internal_fn_stored_value_index): New function.
1749 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1751 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1752 * optabs-query.c (supports_vec_scatter_store_p): New function.
1753 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1754 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1755 true for scatter stores.
1756 (vect_gather_scatter_fn_p): Handle scatter stores too.
1757 (vect_check_gather_scatter): Consider using scatter stores if
1758 supports_vec_scatter_store_p.
1759 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1761 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1762 internal_fn_stored_value_index.
1763 (check_load_store_masking): Handle scatter stores too.
1764 (vect_get_store_rhs): Make public.
1765 (vectorizable_call): Use internal_store_fn_p.
1766 (vectorizable_store): Handle scatter store internal functions.
1767 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1768 when deciding whether the end of the group has been reached.
1769 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1770 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1771 (mask_scatter_store<mode>): New insns.
1773 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1774 Alan Hayward <alan.hayward@arm.com>
1775 David Sherwood <david.sherwood@arm.com>
1777 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1778 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1779 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1781 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1782 Use vect_truncate_gather_scatter_offset if we can't treat the
1783 operation as a normal gather load or scatter store.
1784 (get_group_load_store_type): Take the gather_scatter_info
1785 as argument. Try using a gather load or scatter store for
1786 single-element groups.
1787 (get_load_store_type): Update calls to get_group_load_store_type
1788 and vect_use_strided_gather_scatters_p.
1790 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1791 Alan Hayward <alan.hayward@arm.com>
1792 David Sherwood <david.sherwood@arm.com>
1794 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1795 optional tree argument.
1796 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1798 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1799 but continue to use the current value as a fallback.
1800 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1801 to compare the updates.
1802 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1803 (get_load_store_type): Use it when handling a strided access.
1804 (vect_get_strided_load_store_ops): New function.
1805 (vect_get_data_ptr_increment): Likewise.
1806 (vectorizable_load): Handle strided gather loads. Always pass
1807 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1809 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1810 Alan Hayward <alan.hayward@arm.com>
1811 David Sherwood <david.sherwood@arm.com>
1813 * doc/md.texi (gather_load@var{m}): Document.
1814 (mask_gather_load@var{m}): Likewise.
1815 * genopinit.c (main): Add supports_vec_gather_load and
1816 supports_vec_gather_load_cached to target_optabs.
1817 * optabs-tree.c (init_tree_optimization_optabs): Use
1818 ggc_cleared_alloc to allocate target_optabs.
1819 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1820 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1822 * internal-fn.h (internal_load_fn_p): Declare.
1823 (internal_gather_scatter_fn_p): Likewise.
1824 (internal_fn_mask_index): Likewise.
1825 (internal_gather_scatter_fn_supported_p): Likewise.
1826 * internal-fn.c (gather_load_direct): New macro.
1827 (expand_gather_load_optab_fn): New function.
1828 (direct_gather_load_optab_supported_p): New macro.
1829 (direct_internal_fn_optab): New function.
1830 (internal_load_fn_p): Likewise.
1831 (internal_gather_scatter_fn_p): Likewise.
1832 (internal_fn_mask_index): Likewise.
1833 (internal_gather_scatter_fn_supported_p): Likewise.
1834 * optabs-query.c (supports_at_least_one_mode_p): New function.
1835 (supports_vec_gather_load_p): Likewise.
1836 * optabs-query.h (supports_vec_gather_load_p): Declare.
1837 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1838 and memory_type field.
1839 (NUM_PATTERNS): Bump to 15.
1840 * tree-vect-data-refs.c: Include internal-fn.h.
1841 (vect_gather_scatter_fn_p): New function.
1842 (vect_describe_gather_scatter_call): Likewise.
1843 (vect_check_gather_scatter): Try using internal functions for
1844 gather loads. Recognize existing calls to a gather load function.
1845 (vect_analyze_data_refs): Consider using gather loads if
1846 supports_vec_gather_load_p.
1847 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1848 (vect_get_gather_scatter_offset_type): Likewise.
1849 (vect_convert_mask_for_vectype): Likewise.
1850 (vect_add_conversion_to_patterm): Likewise.
1851 (vect_try_gather_scatter_pattern): Likewise.
1852 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1853 (vect_vect_recog_func_ptrs): Add it.
1854 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1855 internal_fn_mask_index and internal_gather_scatter_fn_p.
1856 (check_load_store_masking): Take the gather_scatter_info as an
1857 argument and handle gather loads.
1858 (vect_get_gather_scatter_ops): New function.
1859 (vectorizable_call): Check internal_load_fn_p.
1860 (vectorizable_load): Likewise. Handle gather load internal
1862 (vectorizable_store): Update call to check_load_store_masking.
1863 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1864 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1865 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1866 (aarch64_gather_scale_operand_d): New predicates.
1867 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1868 (mask_gather_load<mode>): New insns.
1870 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1871 Alan Hayward <alan.hayward@arm.com>
1872 David Sherwood <david.sherwood@arm.com>
1874 * optabs.def (fold_left_plus_optab): New optab.
1875 * doc/md.texi (fold_left_plus_@var{m}): Document.
1876 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1877 * internal-fn.c (fold_left_direct): Define.
1878 (expand_fold_left_optab_fn): Likewise.
1879 (direct_fold_left_optab_supported_p): Likewise.
1880 * fold-const-call.c (fold_const_fold_left): New function.
1881 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1882 * tree-parloops.c (valid_reduction_p): New function.
1883 (gather_scalar_reductions): Use it.
1884 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1885 (vect_finish_replace_stmt): Declare.
1886 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1887 (needs_fold_left_reduction_p): New function, split out from...
1888 (vect_is_simple_reduction): ...here. Accept reductions that
1889 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1890 (vect_force_simple_reduction): Also store the reduction type in
1891 the assignment's STMT_VINFO_REDUC_TYPE.
1892 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1893 (merge_with_identity): New function.
1894 (vect_expand_fold_left): Likewise.
1895 (vectorize_fold_left_reduction): Likewise.
1896 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1897 scalar phi in place for it. Check for target support and reject
1898 cases that would reassociate the operation. Defer the transform
1899 phase to vectorize_fold_left_reduction.
1900 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1901 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1902 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1904 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1906 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1907 call to ifc_temp_var.
1909 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1910 Alan Hayward <alan.hayward@arm.com>
1911 David Sherwood <david.sherwood@arm.com>
1913 * target.def (legitimize_address_displacement): Take the original
1914 offset as a poly_int.
1915 * targhooks.h (default_legitimize_address_displacement): Update
1917 * targhooks.c (default_legitimize_address_displacement): Likewise.
1918 * doc/tm.texi: Regenerate.
1919 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1920 as an argument, moving assert of ad->disp == ad->disp_term to...
1921 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1922 Try calling targetm.legitimize_address_displacement before expanding
1923 the address rather than afterwards, and adjust for the new interface.
1924 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1925 Match the new hook interface. Handle SVE addresses.
1926 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1929 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1931 * Makefile.in (OBJS): Add early-remat.o.
1932 * target.def (select_early_remat_modes): New hook.
1933 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1934 * doc/tm.texi: Regenerate.
1935 * targhooks.h (default_select_early_remat_modes): Declare.
1936 * targhooks.c (default_select_early_remat_modes): New function.
1937 * timevar.def (TV_EARLY_REMAT): New timevar.
1938 * passes.def (pass_early_remat): New pass.
1939 * tree-pass.h (make_pass_early_remat): Declare.
1940 * early-remat.c: New file.
1941 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1943 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1945 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1946 Alan Hayward <alan.hayward@arm.com>
1947 David Sherwood <david.sherwood@arm.com>
1949 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1950 vfm1 with a bound_epilog parameter.
1951 (vect_do_peeling): Update calls accordingly, and move the prologue
1952 call earlier in the function. Treat the base bound_epilog as 0 for
1953 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1954 this base when peeling for gaps.
1955 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1956 with fully-masked loops.
1957 (vect_estimate_min_profitable_iters): Handle the single peeled
1958 iteration in that case.
1960 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1961 Alan Hayward <alan.hayward@arm.com>
1962 David Sherwood <david.sherwood@arm.com>
1964 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1965 single-element interleaving even if the size is not a power of 2.
1966 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1967 accesses for single-element interleaving if the group size is
1970 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1971 Alan Hayward <alan.hayward@arm.com>
1972 David Sherwood <david.sherwood@arm.com>
1974 * doc/md.texi (fold_extract_last_@var{m}): Document.
1975 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1976 * optabs.def (fold_extract_last_optab): New optab.
1977 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1978 * internal-fn.c (fold_extract_direct): New macro.
1979 (expand_fold_extract_optab_fn): Likewise.
1980 (direct_fold_extract_optab_supported_p): Likewise.
1981 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1982 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1983 EXTRACT_LAST_REDUCTION.
1984 (get_initial_def_for_reduction): Do not create an initial vector
1985 for EXTRACT_LAST_REDUCTION reductions.
1986 (vectorizable_reduction): Leave the scalar phi in place for
1987 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1988 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1989 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1990 transform phase to vectorizable_condition.
1991 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1993 (vect_finish_stmt_generation): ...here.
1994 (vect_finish_replace_stmt): New function.
1995 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1996 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1998 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2000 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2001 Alan Hayward <alan.hayward@arm.com>
2002 David Sherwood <david.sherwood@arm.com>
2004 * doc/md.texi (extract_last_@var{m}): Document.
2005 * optabs.def (extract_last_optab): New optab.
2006 * internal-fn.def (EXTRACT_LAST): New internal function.
2007 * internal-fn.c (cond_unary_direct): New macro.
2008 (expand_cond_unary_optab_fn): Likewise.
2009 (direct_cond_unary_optab_supported_p): Likewise.
2010 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2011 loops using EXTRACT_LAST.
2012 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2013 (extract_last_<mode>): ...this optab.
2014 (vec_extract<mode><Vel>): Update accordingly.
2016 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2017 Alan Hayward <alan.hayward@arm.com>
2018 David Sherwood <david.sherwood@arm.com>
2020 * target.def (empty_mask_is_expensive): New hook.
2021 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2022 * doc/tm.texi: Regenerate.
2023 * targhooks.h (default_empty_mask_is_expensive): Declare.
2024 * targhooks.c (default_empty_mask_is_expensive): New function.
2025 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2026 if the target says that empty masks are expensive.
2027 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2029 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2031 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2032 Alan Hayward <alan.hayward@arm.com>
2033 David Sherwood <david.sherwood@arm.com>
2035 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2036 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2037 (vect_use_loop_mask_for_alignment_p): New function.
2038 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2039 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2040 niters_skip argument. Make sure that the first niters_skip elements
2041 of the first iteration are inactive.
2042 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2043 Update call to vect_set_loop_masks_directly.
2044 (get_misalign_in_elems): New function, split out from...
2045 (vect_gen_prolog_loop_niters): ...here.
2046 (vect_update_init_of_dr): Take a code argument that specifies whether
2047 the adjustment should be added or subtracted.
2048 (vect_update_init_of_drs): Likewise.
2049 (vect_prepare_for_masked_peels): New function.
2050 (vect_do_peeling): Skip prologue peeling if we're using a mask
2051 instead. Update call to vect_update_inits_of_drs.
2052 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2054 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2055 alignment. Do not include the number of peeled iterations in
2056 the minimum threshold in that case.
2057 (vectorizable_induction): Adjust the start value down by
2058 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2059 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2060 Take the number of skipped iterations into account when calculating
2062 * tree-vect-stmts.c (vect_gen_while_not): New function.
2064 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2065 Alan Hayward <alan.hayward@arm.com>
2066 David Sherwood <david.sherwood@arm.com>
2068 * doc/sourcebuild.texi (vect_fully_masked): Document.
2069 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2071 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2073 (vect_analyze_loop_2): ...here. Don't check the vectorization
2074 factor against the number of loop iterations if the loop is
2077 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2078 Alan Hayward <alan.hayward@arm.com>
2079 David Sherwood <david.sherwood@arm.com>
2081 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2082 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2083 (dump_groups): Update accordingly.
2084 (iv_use::mem_type): New member variable.
2085 (address_p): New function.
2086 (record_use): Add a mem_type argument and initialize the new
2088 (record_group_use): Add a mem_type argument. Use address_p.
2089 Remove obsolete null checks of base_object. Update call to record_use.
2090 (find_interesting_uses_op): Update call to record_group_use.
2091 (find_interesting_uses_cond): Likewise.
2092 (find_interesting_uses_address): Likewise.
2093 (get_mem_type_for_internal_fn): New function.
2094 (find_address_like_use): Likewise.
2095 (find_interesting_uses_stmt): Try find_address_like_use before
2096 calling find_interesting_uses_op.
2097 (addr_offset_valid_p): Use the iv mem_type field as the type
2098 of the addressed memory.
2099 (add_autoinc_candidates): Likewise.
2100 (get_address_cost): Likewise.
2101 (split_small_address_groups_p): Use address_p.
2102 (split_address_groups): Likewise.
2103 (add_iv_candidate_for_use): Likewise.
2104 (autoinc_possible_for_pair): Likewise.
2105 (rewrite_groups): Likewise.
2106 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2107 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2108 (get_alias_ptr_type_for_ptr_address): New function.
2109 (rewrite_use_address): Rewrite address uses in calls that were
2110 identified by find_address_like_use.
2112 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2113 Alan Hayward <alan.hayward@arm.com>
2114 David Sherwood <david.sherwood@arm.com>
2116 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2118 * gimple-expr.h (is_gimple_addressable: Likewise.
2119 * gimple-expr.c (is_gimple_address): Likewise.
2120 * internal-fn.c (expand_call_mem_ref): New function.
2121 (expand_mask_load_optab_fn): Use it.
2122 (expand_mask_store_optab_fn): Likewise.
2124 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2125 Alan Hayward <alan.hayward@arm.com>
2126 David Sherwood <david.sherwood@arm.com>
2128 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2129 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2130 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2131 (cond_umax@var{mode}): Document.
2132 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2133 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2134 (cond_umin_optab, cond_umax_optab): New optabs.
2135 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2136 (COND_IOR, COND_XOR): New internal functions.
2137 * internal-fn.h (get_conditional_internal_fn): Declare.
2138 * internal-fn.c (cond_binary_direct): New macro.
2139 (expand_cond_binary_optab_fn): Likewise.
2140 (direct_cond_binary_optab_supported_p): Likewise.
2141 (get_conditional_internal_fn): New function.
2142 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2143 Cope with reduction statements that are vectorized as calls rather
2145 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2146 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2147 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2148 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2149 (UNSPEC_COND_EOR): New unspecs.
2150 (optab): Add mappings for them.
2151 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2152 (sve_int_op, sve_fp_op): New int attributes.
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 * optabs.def (while_ult_optab): New optab.
2159 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2160 * internal-fn.def (WHILE_ULT): New internal function.
2161 * internal-fn.h (direct_internal_fn_supported_p): New override
2162 that takes two types as argument.
2163 * internal-fn.c (while_direct): New macro.
2164 (expand_while_optab_fn): New function.
2165 (convert_optab_supported_p): Likewise.
2166 (direct_while_optab_supported_p): New macro.
2167 * wide-int.h (wi::udiv_ceil): New function.
2168 * tree-vectorizer.h (rgroup_masks): New structure.
2169 (vec_loop_masks): New typedef.
2170 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2172 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2173 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2174 (vect_max_vf): New function.
2175 (slpeel_make_loop_iterate_ntimes): Delete.
2176 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2177 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2178 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2179 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2180 internal-fn.h, stor-layout.h and optabs-query.h.
2181 (vect_set_loop_mask): New function.
2182 (add_preheader_seq): Likewise.
2183 (add_header_seq): Likewise.
2184 (interleave_supported_p): Likewise.
2185 (vect_maybe_permute_loop_masks): Likewise.
2186 (vect_set_loop_masks_directly): Likewise.
2187 (vect_set_loop_condition_masked): Likewise.
2188 (vect_set_loop_condition_unmasked): New function, split out from
2189 slpeel_make_loop_iterate_ntimes.
2190 (slpeel_make_loop_iterate_ntimes): Rename to..
2191 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2192 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2193 (vect_do_peeling): Update call accordingly.
2194 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2196 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2197 mask_compare_type, can_fully_mask_p and fully_masked_p.
2198 (release_vec_loop_masks): New function.
2199 (_loop_vec_info): Use it to free the loop masks.
2200 (can_produce_all_loop_masks_p): New function.
2201 (vect_get_max_nscalars_per_iter): Likewise.
2202 (vect_verify_full_masking): Likewise.
2203 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2204 retries, and free the mask rgroups before retrying. Check loop-wide
2205 reasons for disallowing fully-masked loops. Make the final decision
2206 about whether use a fully-masked loop or not.
2207 (vect_estimate_min_profitable_iters): Do not assume that peeling
2208 for the number of iterations will be needed for fully-masked loops.
2209 (vectorizable_reduction): Disable fully-masked loops.
2210 (vectorizable_live_operation): Likewise.
2211 (vect_halve_mask_nunits): New function.
2212 (vect_double_mask_nunits): Likewise.
2213 (vect_record_loop_mask): Likewise.
2214 (vect_get_loop_mask): Likewise.
2215 (vect_transform_loop): Handle the case in which the final loop
2216 iteration might handle a partial vector. Call vect_set_loop_condition
2217 instead of slpeel_make_loop_iterate_ntimes.
2218 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2219 (check_load_store_masking): New function.
2220 (prepare_load_store_mask): Likewise.
2221 (vectorizable_store): Handle fully-masked loops.
2222 (vectorizable_load): Likewise.
2223 (supportable_widening_operation): Use vect_halve_mask_nunits for
2225 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2226 (vect_gen_while): New function.
2227 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2228 (aarch64_uqdec<mode>): New insn.
2230 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2231 Alan Hayward <alan.hayward@arm.com>
2232 David Sherwood <david.sherwood@arm.com>
2234 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2235 (reduc_xor_scal_optab): New optabs.
2236 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2237 (reduc_xor_scal_@var{m}): Document.
2238 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2239 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2241 * fold-const-call.c (fold_const_call): Handle them.
2242 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2243 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2244 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2245 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2246 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2247 (UNSPEC_XORV): New unspecs.
2248 (optab): Add entries for them.
2249 (BITWISEV): New int iterator.
2250 (bit_reduc_op): New int attributes.
2252 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2253 Alan Hayward <alan.hayward@arm.com>
2254 David Sherwood <david.sherwood@arm.com>
2256 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2257 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2258 * optabs.def (vec_shl_insert_optab): New optab.
2259 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2260 (duplicate_and_interleave): Likewise.
2261 * tree-vect-loop.c: Include internal-fn.h.
2262 (neutral_op_for_slp_reduction): New function, split out from
2263 get_initial_defs_for_reduction.
2264 (get_initial_def_for_reduction): Handle option 2 for variable-length
2265 vectors by loading the neutral value into a vector and then shifting
2266 the initial value into element 0.
2267 (get_initial_defs_for_reduction): Replace the code argument with
2268 the neutral value calculated by neutral_op_for_slp_reduction.
2269 Use gimple_build_vector for constant-length vectors.
2270 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2271 but the first group_size elements have a neutral value.
2272 Use duplicate_and_interleave otherwise.
2273 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2274 Update call to get_initial_defs_for_reduction. Handle SLP
2275 reductions for variable-length vectors by creating one vector
2276 result for each scalar result, with the elements associated
2277 with other scalar results stubbed out with the neutral value.
2278 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2279 Require IFN_VEC_SHL_INSERT for double reductions on
2280 variable-length vectors, or SLP reductions that have
2281 a neutral value. Require can_duplicate_and_interleave_p
2282 support for variable-length unchained SLP reductions if there
2283 is no neutral value, such as for MIN/MAX reductions. Also require
2284 the number of vector elements to be a multiple of the number of
2285 SLP statements when doing variable-length unchained SLP reductions.
2286 Update call to vect_create_epilog_for_reduction.
2287 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2288 and remove initial values.
2289 (duplicate_and_interleave): Make public.
2290 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2291 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2293 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2294 Alan Hayward <alan.hayward@arm.com>
2295 David Sherwood <david.sherwood@arm.com>
2297 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2298 (can_duplicate_and_interleave_p): New function.
2299 (vect_get_and_check_slp_defs): Take the vector of statements
2300 rather than just the current one. Remove excess parentheses.
2301 Restriction rejectinon of vect_constant_def and vect_external_def
2302 for variable-length vectors to boolean types, or types for which
2303 can_duplicate_and_interleave_p is false.
2304 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2305 (duplicate_and_interleave): New function.
2306 (vect_get_constant_vectors): Use gimple_build_vector for
2307 constant-length vectors and suitable variable-length constant
2308 vectors. Use duplicate_and_interleave for other variable-length
2309 vectors. Don't defer the update when inserting new statements.
2311 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2312 Alan Hayward <alan.hayward@arm.com>
2313 David Sherwood <david.sherwood@arm.com>
2315 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2316 min_profitable_iters doesn't go negative.
2318 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2319 Alan Hayward <alan.hayward@arm.com>
2320 David Sherwood <david.sherwood@arm.com>
2322 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2323 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2324 * optabs.def (vec_mask_load_lanes_optab): New optab.
2325 (vec_mask_store_lanes_optab): Likewise.
2326 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2327 (MASK_STORE_LANES): Likewise.
2328 * internal-fn.c (mask_load_lanes_direct): New macro.
2329 (mask_store_lanes_direct): Likewise.
2330 (expand_mask_load_optab_fn): Handle masked operations.
2331 (expand_mask_load_lanes_optab_fn): New macro.
2332 (expand_mask_store_optab_fn): Handle masked operations.
2333 (expand_mask_store_lanes_optab_fn): New macro.
2334 (direct_mask_load_lanes_optab_supported_p): Likewise.
2335 (direct_mask_store_lanes_optab_supported_p): Likewise.
2336 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2338 (vect_load_lanes_supported): Likewise.
2339 * tree-vect-data-refs.c (strip_conversion): New function.
2340 (can_group_stmts_p): Likewise.
2341 (vect_analyze_data_ref_accesses): Use it instead of checking
2342 for a pair of assignments.
2343 (vect_store_lanes_supported): Take a masked_p parameter.
2344 (vect_load_lanes_supported): Likewise.
2345 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2346 vect_store_lanes_supported and vect_load_lanes_supported.
2347 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2348 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2349 parameter. Don't allow gaps for masked accesses.
2350 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2351 and vect_load_lanes_supported.
2352 (get_load_store_type): Take a masked_p parameter and update
2353 call to get_group_load_store_type.
2354 (vectorizable_store): Update call to get_load_store_type.
2355 Handle IFN_MASK_STORE_LANES.
2356 (vectorizable_load): Update call to get_load_store_type.
2357 Handle IFN_MASK_LOAD_LANES.
2359 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2360 Alan Hayward <alan.hayward@arm.com>
2361 David Sherwood <david.sherwood@arm.com>
2363 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2365 * config/aarch64/aarch64-protos.h
2366 (aarch64_sve_struct_memory_operand_p): Declare.
2367 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2368 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2369 (VPRED, vpred): Handle SVE structure modes.
2370 * config/aarch64/constraints.md (Utx): New constraint.
2371 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2372 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2373 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2374 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2375 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2376 structure modes. Split into pieces after RA.
2377 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2378 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2380 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2381 SVE structure modes.
2382 (aarch64_classify_address): Likewise.
2383 (sizetochar): Move earlier in file.
2384 (aarch64_print_operand): Handle SVE register lists.
2385 (aarch64_array_mode): New function.
2386 (aarch64_sve_struct_memory_operand_p): Likewise.
2387 (TARGET_ARRAY_MODE): Redefine.
2389 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2390 Alan Hayward <alan.hayward@arm.com>
2391 David Sherwood <david.sherwood@arm.com>
2393 * target.def (array_mode): New target hook.
2394 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2395 * doc/tm.texi: Regenerate.
2396 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2397 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2398 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2400 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2403 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2404 Alan Hayward <alan.hayward@arm.com>
2405 David Sherwood <david.sherwood@arm.com>
2407 * fold-const.c (fold_binary_loc): Check the argument types
2408 rather than the result type when testing for a vector operation.
2410 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2412 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2413 * doc/tm.texi: Regenerate.
2415 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2416 Alan Hayward <alan.hayward@arm.com>
2417 David Sherwood <david.sherwood@arm.com>
2419 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2420 (sve): Document new AArch64 extension.
2421 * doc/md.texi (w): Extend the description of the AArch64
2422 constraint to include SVE vectors.
2423 (Upl, Upa): Document new AArch64 predicate constraints.
2424 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2426 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2427 (msve-vector-bits=): New option.
2428 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2429 SVE when these are disabled.
2430 (sve): New extension.
2431 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2432 modes. Adjust their number of units based on aarch64_sve_vg.
2433 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2434 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2435 aarch64_addr_query_type.
2436 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2437 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2438 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2439 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2440 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2441 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2442 (aarch64_simd_imm_zero_p): Delete.
2443 (aarch64_check_zero_based_sve_index_immediate): Declare.
2444 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2445 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2446 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2447 (aarch64_sve_float_mul_immediate_p): Likewise.
2448 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2450 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2451 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2452 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2453 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2454 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2455 (aarch64_regmode_natural_size): Likewise.
2456 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2457 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2459 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2460 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2461 for VG and the SVE predicate registers.
2462 (V_ALIASES): Add a "z"-prefixed alias.
2463 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2464 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2465 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2466 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2467 (REG_CLASS_NAMES): Add entries for them.
2468 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2469 and the predicate registers.
2470 (aarch64_sve_vg): Declare.
2471 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2472 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2473 (REGMODE_NATURAL_SIZE): Define.
2474 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2476 * config/aarch64/aarch64.c: Include cfgrtl.h.
2477 (simd_immediate_info): Add a constructor for series vectors,
2478 and an associated step field.
2479 (aarch64_sve_vg): New variable.
2480 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2481 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2482 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2483 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2484 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2485 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2486 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2487 (aarch64_get_mask_mode): New functions.
2488 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2489 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2490 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2491 predicate modes and predicate registers. Explicitly restrict
2492 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2493 to store a vector mode if it is recognized by
2494 aarch64_classify_vector_mode.
2495 (aarch64_regmode_natural_size): New function.
2496 (aarch64_hard_regno_caller_save_mode): Return the original mode
2498 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2499 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2500 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2501 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2503 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2504 does not overlap dest if the function is frame-related. Handle
2506 (aarch64_split_add_offset): New function.
2507 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2508 them aarch64_add_offset.
2509 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2510 and update call to aarch64_sub_sp.
2511 (aarch64_add_cfa_expression): New function.
2512 (aarch64_expand_prologue): Pass extra temporary registers to the
2513 functions above. Handle the case in which we need to emit new
2514 DW_CFA_expressions for registers that were originally saved
2515 relative to the stack pointer, but now have to be expressed
2516 relative to the frame pointer.
2517 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2519 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2520 IP0 and IP1 values for SVE frames.
2521 (aarch64_expand_vec_series): New function.
2522 (aarch64_expand_sve_widened_duplicate): Likewise.
2523 (aarch64_expand_sve_const_vector): Likewise.
2524 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2525 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2526 into the register, rather than emitting a SET directly.
2527 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2528 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2529 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2530 (offset_9bit_signed_scaled_p): New functions.
2531 (aarch64_replicate_bitmask_imm): New function.
2532 (aarch64_bitmask_imm): Use it.
2533 (aarch64_cannot_force_const_mem): Reject expressions involving
2534 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2535 (aarch64_classify_index): Handle SVE indices, by requiring
2536 a plain register index with a scale that matches the element size.
2537 (aarch64_classify_address): Handle SVE addresses. Assert that
2538 the mode of the address is VOIDmode or an integer mode.
2539 Update call to aarch64_classify_symbol.
2540 (aarch64_classify_symbolic_expression): Update call to
2541 aarch64_classify_symbol.
2542 (aarch64_const_vec_all_in_range_p): New function.
2543 (aarch64_print_vector_float_operand): Likewise.
2544 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2545 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2546 and the FP immediates 1.0 and 0.5.
2547 (aarch64_print_address_internal): Handle SVE addresses.
2548 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2549 (aarch64_regno_regclass): Handle predicate registers.
2550 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2552 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2553 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2554 (aarch64_convert_sve_vector_bits): New function.
2555 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2556 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2558 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2559 Handle SVE vector and predicate modes. Accept VL-based constants
2560 that need only one temporary register, and VL offsets that require
2561 no temporary registers.
2562 (aarch64_conditional_register_usage): Mark the predicate registers
2563 as fixed if SVE isn't available.
2564 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2565 Return true for SVE vector and predicate modes.
2566 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2567 rather than an unsigned int. Handle SVE modes.
2568 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2570 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2572 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2573 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2574 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2575 (aarch64_sve_float_mul_immediate_p): New functions.
2576 (aarch64_sve_valid_immediate): New function.
2577 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2578 Explicitly reject structure modes. Check for INDEX constants.
2579 Handle PTRUE and PFALSE constants.
2580 (aarch64_check_zero_based_sve_index_immediate): New function.
2581 (aarch64_simd_imm_zero_p): Delete.
2582 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2583 vector modes. Accept constants in the range of CNT[BHWD].
2584 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2585 ask for an Advanced SIMD mode.
2586 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2587 (aarch64_simd_vector_alignment): Handle SVE predicates.
2588 (aarch64_vectorize_preferred_vector_alignment): New function.
2589 (aarch64_simd_vector_alignment_reachable): Use it instead of
2591 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2592 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2594 (MAX_VECT_LEN): Delete.
2595 (expand_vec_perm_d): Add a vec_flags field.
2596 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2597 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2598 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2600 (aarch64_evpc_rev): Rename to...
2601 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2602 (aarch64_evpc_rev_global): New function.
2603 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2604 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2606 (aarch64_evpc_sve_tbl): New function.
2607 (aarch64_expand_vec_perm_const_1): Update after rename of
2608 aarch64_evpc_rev. Handle SVE permutes too, trying
2609 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2610 than aarch64_evpc_tbl.
2611 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2612 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2613 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2614 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2615 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2616 (aarch64_expand_sve_vcond): New functions.
2617 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2618 of aarch64_vector_mode_p.
2619 (aarch64_dwarf_poly_indeterminate_value): New function.
2620 (aarch64_compute_pressure_classes): Likewise.
2621 (aarch64_can_change_mode_class): Likewise.
2622 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2623 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2624 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2625 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2626 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2627 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2628 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2629 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2631 (Dn, Dl, Dr): Accept const as well as const_vector.
2632 (Dz): Likewise. Compare against CONST0_RTX.
2633 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2634 of "vector" where appropriate.
2635 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2636 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2637 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2638 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2639 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2640 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2641 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2642 (v_int_equiv): Extend to SVE modes.
2643 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2645 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2646 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2647 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2648 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2649 (SVE_COND_FP_CMP): New int iterators.
2650 (perm_hilo): Handle the new unpack unspecs.
2651 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2653 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2654 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2655 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2656 (aarch64_equality_operator, aarch64_constant_vector_operand)
2657 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2658 (aarch64_sve_nonimmediate_operand): Likewise.
2659 (aarch64_sve_general_operand): Likewise.
2660 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2661 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2662 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2663 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2664 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2665 (aarch64_sve_float_arith_immediate): Likewise.
2666 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2667 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2668 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2669 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2670 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2671 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2672 (aarch64_sve_float_arith_operand): Likewise.
2673 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2674 (aarch64_sve_float_mul_operand): Likewise.
2675 (aarch64_sve_vec_perm_operand): Likewise.
2676 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2677 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2678 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2679 as well as const_vector.
2680 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2681 in file. Use CONST0_RTX and CONSTM1_RTX.
2682 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2683 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2684 Use aarch64_simd_imm_zero.
2685 * config/aarch64/aarch64-sve.md: New file.
2686 * config/aarch64/aarch64.md: Include it.
2687 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2688 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2689 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2690 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2691 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2692 (sve): New attribute.
2693 (enabled): Disable instructions with the sve attribute unless
2695 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2696 aarch64_expand_mov_immediate.
2697 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2698 CNT[BHSD] immediates.
2699 (movti): Split CONST_POLY_INT moves into two halves.
2700 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2701 Split additions that need a temporary here if the destination
2702 is the stack pointer.
2703 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2704 (*add<mode>3_poly_1): New instruction.
2705 (set_clobber_cc): New expander.
2707 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2709 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2710 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2711 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2712 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2713 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2714 Change innermode from fixed_mode_size to machine_mode.
2715 (simplify_subreg): Update call accordingly. Handle a constant-sized
2716 subreg of a variable-length CONST_VECTOR.
2718 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2719 Alan Hayward <alan.hayward@arm.com>
2720 David Sherwood <david.sherwood@arm.com>
2722 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2723 (add_offset_to_base): New function, split out from...
2724 (create_mem_ref): ...here. When handling a scale other than 1,
2725 check first whether the address is valid without the offset.
2726 Add it into the base if so, leaving the index and scale as-is.
2728 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2731 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2732 fold_for_warn before checking if arg2 is INTEGER_CST.
2734 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2736 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2737 (store_multiple_operation): Delete.
2738 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2739 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2740 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2741 guarded by TARGET_STRING.
2742 (rs6000_output_load_multiple): Delete.
2743 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2744 OPTION_MASK_STRING / TARGET_STRING handling.
2745 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2746 (const rs6000_opt_masks) <"string">: Change mask to 0.
2747 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2748 (MASK_STRING): Delete.
2749 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2751 (load_multiple): Delete.
2758 (store_multiple): Delete.
2765 (movmemsi_8reg): Delete.
2766 (corresponding unnamed define_insn): Delete.
2767 (movmemsi_6reg): Delete.
2768 (corresponding unnamed define_insn): Delete.
2769 (movmemsi_4reg): Delete.
2770 (corresponding unnamed define_insn): Delete.
2771 (movmemsi_2reg): Delete.
2772 (corresponding unnamed define_insn): Delete.
2773 (movmemsi_1reg): Delete.
2774 (corresponding unnamed define_insn): Delete.
2775 * config/rs6000/rs6000.opt (mno-string): New.
2776 (mstring): Replace by deprecation warning stub.
2777 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2779 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2781 * regrename.c (regrename_do_replace): If replacing the same
2782 reg multiple times, try to reuse last created gen_raw_REG.
2785 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2786 main to workaround a bug in GDB.
2788 2018-01-12 Tom de Vries <tom@codesourcery.com>
2791 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2793 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2795 PR rtl-optimization/80481
2796 * ira-color.c (get_cap_member): New function.
2797 (allocnos_conflict_by_live_ranges_p): Use it.
2798 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2799 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2801 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2804 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2805 (*saddl_se_1): Ditto.
2807 (*ssubl_se_1): Ditto.
2809 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2811 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2812 rather than wi::to_widest for DR_INITs.
2813 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2814 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2815 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2817 (vect_analyze_group_access_1): Note that here.
2819 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2821 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2822 polynomial type sizes.
2824 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2826 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2827 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2828 (gimple_add_tmp_var): Likewise.
2830 2018-01-12 Martin Liska <mliska@suse.cz>
2832 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2833 (gimple_alloc_sizes): Likewise.
2834 (dump_gimple_statistics): Use PRIu64 in printf format.
2835 * gimple.h: Change uint64_t to int.
2837 2018-01-12 Martin Liska <mliska@suse.cz>
2839 * tree-core.h: Use uint64_t instead of int.
2840 * tree.c (tree_node_counts): Likewise.
2841 (tree_node_sizes): Likewise.
2842 (dump_tree_statistics): Use PRIu64 in printf format.
2844 2018-01-12 Martin Liska <mliska@suse.cz>
2846 * Makefile.in: As qsort_chk is implemented in vec.c, add
2847 vec.o to linkage of gencfn-macros.
2848 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2849 passing the info to record_node_allocation_statistics.
2850 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2852 * ggc-common.c (struct ggc_usage): Add operator== and use
2853 it in operator< and compare function.
2854 * mem-stats.h (struct mem_usage): Likewise.
2855 * vec.c (struct vec_usage): Remove operator< and compare
2856 function. Can be simply inherited.
2858 2018-01-12 Martin Jambor <mjambor@suse.cz>
2861 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2862 * tree-ssa-math-opts.c: Include domwalk.h.
2863 (convert_mult_to_fma_1): New function.
2864 (fma_transformation_info): New type.
2865 (fma_deferring_state): Likewise.
2866 (cancel_fma_deferring): New function.
2867 (result_of_phi): Likewise.
2868 (last_fma_candidate_feeds_initial_phi): Likewise.
2869 (convert_mult_to_fma): Added deferring logic, split actual
2870 transformation to convert_mult_to_fma_1.
2871 (math_opts_dom_walker): New type.
2872 (math_opts_dom_walker::after_dom_children): New method, body moved
2873 here from pass_optimize_widening_mul::execute, added deferring logic
2875 (pass_optimize_widening_mul::execute): Moved most of code to
2876 math_opts_dom_walker::after_dom_children.
2877 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2878 * config/i386/i386.c (ix86_option_override_internal): Added
2879 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2881 2018-01-12 Richard Biener <rguenther@suse.de>
2884 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2885 inline instance vars.
2887 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2890 * config/rx/rx.c (rx_is_restricted_memory_address):
2893 2018-01-12 Richard Biener <rguenther@suse.de>
2895 PR tree-optimization/80846
2896 * target.def (split_reduction): New target hook.
2897 * targhooks.c (default_split_reduction): New function.
2898 * targhooks.h (default_split_reduction): Declare.
2899 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2900 target requests first reduce vectors by combining low and high
2902 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2903 (get_vectype_for_scalar_type_and_size): Export.
2904 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2905 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2906 * doc/tm.texi: Regenerate.
2907 * config/i386/i386.c (ix86_split_reduction): Implement
2908 TARGET_VECTORIZE_SPLIT_REDUCTION.
2910 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2913 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2914 in PIC mode except for TARGET_VXWORKS_RTP.
2915 * config/sparc/sparc.c: Include cfgrtl.h.
2916 (TARGET_INIT_PIC_REG): Define.
2917 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2918 (sparc_pic_register_p): New predicate.
2919 (sparc_legitimate_address_p): Use it.
2920 (sparc_legitimize_pic_address): Likewise.
2921 (sparc_delegitimize_address): Likewise.
2922 (sparc_mode_dependent_address_p): Likewise.
2923 (gen_load_pcrel_sym): Remove 4th parameter.
2924 (load_got_register): Adjust call to above. Remove obsolete stuff.
2925 (sparc_expand_prologue): Do not call load_got_register here.
2926 (sparc_flat_expand_prologue): Likewise.
2927 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2928 (sparc_use_pseudo_pic_reg): New function.
2929 (sparc_init_pic_reg): Likewise.
2930 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2931 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2933 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2935 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2936 Add item for branch_cost.
2938 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2940 PR rtl-optimization/83565
2941 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2942 not extend the result to a larger mode for rotate operations.
2943 (num_sign_bit_copies1): Likewise.
2945 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2948 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2950 Use values-Xc.o for -pedantic.
2951 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2953 2018-01-12 Martin Liska <mliska@suse.cz>
2956 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2958 (possible_polymorphic_call_targets): Use it.
2959 (ipa_devirt): Likewise.
2961 2018-01-12 Martin Liska <mliska@suse.cz>
2963 * profile-count.h (enum profile_quality): Use 0 as invalid
2964 enum value of profile_quality.
2966 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2968 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2969 -mext-string options.
2971 2018-01-12 Richard Biener <rguenther@suse.de>
2973 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2974 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2975 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2977 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2979 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2981 * configure.ac (--with-long-double-format): Add support for the
2982 configuration option to change the default long double format on
2984 * config.gcc (powerpc*-linux*-*): Likewise.
2985 * configure: Regenerate.
2986 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2987 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2988 used without modification.
2990 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2992 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2993 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2994 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2995 MISC_BUILTIN_SPEC_BARRIER.
2996 (rs6000_init_builtins): Likewise.
2997 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2999 (speculation_barrier): New define_insn.
3000 * doc/extend.texi: Document __builtin_speculation_barrier.
3002 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3005 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3006 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3007 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3009 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3010 integral modes instead of "ss" and "sd".
3011 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3012 vectors with 32-bit and 64-bit elements.
3013 (vecdupssescalarmodesuffix): New mode attribute.
3014 (vec_dup<mode>): Use it.
3016 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3019 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3020 frame if argument is passed on stack.
3022 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3025 * ree.c (combine_reaching_defs): Optimize also
3026 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3027 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3029 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3032 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3034 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3037 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3038 after they are computed.
3040 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3042 PR tree-optimization/83695
3043 * gimple-loop-linterchange.cc
3044 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3045 reset cached scev information after interchange.
3046 (pass_linterchange::execute): Remove call to scev_reset_htab.
3048 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3050 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3051 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3052 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3053 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3054 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3055 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3056 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3057 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3058 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3059 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3060 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3061 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3062 (V_lane_reg): Likewise.
3063 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3065 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3066 (vfmal_lane_low<mode>_intrinsic,
3067 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3068 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3069 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3070 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3071 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3072 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3074 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3076 * config/arm/arm-cpus.in (fp16fml): New feature.
3077 (ALL_SIMD): Add fp16fml.
3078 (armv8.2-a): Add fp16fml as an option.
3079 (armv8.3-a): Likewise.
3080 (armv8.4-a): Add fp16fml as part of fp16.
3081 * config/arm/arm.h (TARGET_FP16FML): Define.
3082 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3084 * config/arm/arm-modes.def (V2HF): Define.
3085 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3086 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3087 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3088 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3089 vfmsl_low, vfmsl_high): New set of builtins.
3090 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3091 (vfml_op): New code attribute.
3092 (VFMLHALVES): New int iterator.
3093 (VFML, VFMLSEL): New mode attributes.
3094 (V_reg): Define mapping for V2HF.
3095 (V_hi, V_lo): New mode attributes.
3096 (VF_constraint): Likewise.
3097 (vfml_half, vfml_half_selector): New int attributes.
3098 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3100 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3101 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3103 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3104 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3105 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3106 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3108 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3109 Document new effective target and option set.
3111 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3113 * config/arm/arm-cpus.in (armv8_4): New feature.
3114 (ARMv8_4a): New fgroup.
3115 (armv8.4-a): New arch.
3116 * config/arm/arm-tables.opt: Regenerate.
3117 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3118 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3119 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3120 Add matching rules for -march=armv8.4-a and extensions.
3121 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3123 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3126 * config/rx/rx.md (BW): New mode attribute.
3127 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3129 2018-01-11 Richard Biener <rguenther@suse.de>
3131 PR tree-optimization/83435
3132 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3133 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3134 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3136 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3137 Alan Hayward <alan.hayward@arm.com>
3138 David Sherwood <david.sherwood@arm.com>
3140 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3142 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3143 (aarch64_print_address_internal): Use it to check for a zero offset.
3145 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3146 Alan Hayward <alan.hayward@arm.com>
3147 David Sherwood <david.sherwood@arm.com>
3149 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3150 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3151 Return a poly_int64 rather than a HOST_WIDE_INT.
3152 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3153 rather than a HOST_WIDE_INT.
3154 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3155 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3156 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3157 final_offset from HOST_WIDE_INT to poly_int64.
3158 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3159 to_constant when getting the number of units in an Advanced SIMD
3161 (aarch64_builtin_vectorized_function): Check for a constant number
3163 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3165 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3166 attribute instead of GET_MODE_NUNITS.
3167 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3168 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3169 GET_MODE_SIZE for fixed-size registers.
3170 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3171 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3172 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3173 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3174 (aarch64_print_operand, aarch64_print_address_internal)
3175 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3176 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3177 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3178 Handle polynomial GET_MODE_SIZE.
3179 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3180 wider than SImode without modification.
3181 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3182 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3183 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3184 passing and returning SVE modes.
3185 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3186 rather than GEN_INT.
3187 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3188 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3189 (aarch64_allocate_and_probe_stack_space): Likewise.
3190 (aarch64_layout_frame): Cope with polynomial offsets.
3191 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3192 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3194 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3195 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3196 poly_int64 rather than a HOST_WIDE_INT.
3197 (aarch64_get_separate_components, aarch64_process_components)
3198 (aarch64_expand_prologue, aarch64_expand_epilogue)
3199 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3200 (aarch64_anchor_offset): New function, split out from...
3201 (aarch64_legitimize_address): ...here.
3202 (aarch64_builtin_vectorization_cost): Handle polynomial
3203 TYPE_VECTOR_SUBPARTS.
3204 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3206 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3207 number of elements from the PARALLEL rather than the mode.
3208 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3209 rather than GET_MODE_BITSIZE.
3210 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3211 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3212 (aarch64_expand_vec_perm_const_1): Handle polynomial
3213 d->perm.length () and d->perm elements.
3214 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3215 Apply to_constant to d->perm elements.
3216 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3217 polynomial CONST_VECTOR_NUNITS.
3218 (aarch64_move_pointer): Take amount as a poly_int64 rather
3220 (aarch64_progress_pointer): Avoid temporary variable.
3221 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3222 the mode attribute instead of GET_MODE.
3224 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3225 Alan Hayward <alan.hayward@arm.com>
3226 David Sherwood <david.sherwood@arm.com>
3228 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3229 x exists before using it.
3230 (aarch64_add_constant_internal): Rename to...
3231 (aarch64_add_offset_1): ...this. Replace regnum with separate
3232 src and dest rtxes. Handle the case in which they're different,
3233 including when the offset is zero. Replace scratchreg with an rtx.
3234 Use 2 additions if there is no spare register into which we can
3235 move a 16-bit constant.
3236 (aarch64_add_constant): Delete.
3237 (aarch64_add_offset): Replace reg with separate src and dest
3238 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3239 Use aarch64_add_offset_1.
3240 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3241 an rtx rather than an int. Take the delta as a poly_int64
3242 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3243 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3244 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3245 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3246 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3248 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3249 aarch64_add_constant.
3251 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3253 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3254 Use scalar_float_mode.
3256 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3258 * config/aarch64/aarch64-simd.md
3259 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3260 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3261 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3262 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3263 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3264 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3265 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3266 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3267 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3268 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3270 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3273 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3274 targ_options->x_arm_arch_string is non NULL.
3276 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3278 * config/aarch64/aarch64.h
3279 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3281 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3284 * expmed.c (emit_store_flag_force): Swap if const op0
3285 and change VOIDmode to mode of op0.
3287 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3289 PR rtl-optimization/83761
3290 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3291 than bytes to mode_for_size.
3293 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3296 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3297 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3300 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3303 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3304 when in layout mode.
3305 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3306 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3309 2018-01-10 Michael Collison <michael.collison@arm.com>
3311 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3312 * config/aarch64/aarch64-option-extension.def: Add
3313 AARCH64_OPT_EXTENSION of 'fp16fml'.
3314 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3315 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3316 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3317 * config/aarch64/constraints.md (Ui7): New constraint.
3318 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3319 (VFMLA_SEL_W): Ditto.
3322 (VFMLA16_LOW): New int iterator.
3323 (VFMLA16_HIGH): Ditto.
3324 (UNSPEC_FMLAL): New unspec.
3325 (UNSPEC_FMLSL): Ditto.
3326 (UNSPEC_FMLAL2): Ditto.
3327 (UNSPEC_FMLSL2): Ditto.
3328 (f16mac): New code attribute.
3329 * config/aarch64/aarch64-simd-builtins.def
3330 (aarch64_fmlal_lowv2sf): Ditto.
3331 (aarch64_fmlsl_lowv2sf): Ditto.
3332 (aarch64_fmlalq_lowv4sf): Ditto.
3333 (aarch64_fmlslq_lowv4sf): Ditto.
3334 (aarch64_fmlal_highv2sf): Ditto.
3335 (aarch64_fmlsl_highv2sf): Ditto.
3336 (aarch64_fmlalq_highv4sf): Ditto.
3337 (aarch64_fmlslq_highv4sf): Ditto.
3338 (aarch64_fmlal_lane_lowv2sf): Ditto.
3339 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3340 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3341 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3342 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3343 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3344 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3345 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3346 (aarch64_fmlal_lane_highv2sf): Ditto.
3347 (aarch64_fmlsl_lane_highv2sf): Ditto.
3348 (aarch64_fmlal_laneq_highv2sf): Ditto.
3349 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3350 (aarch64_fmlalq_lane_highv4sf): Ditto.
3351 (aarch64_fmlsl_lane_highv4sf): Ditto.
3352 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3353 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3354 * config/aarch64/aarch64-simd.md:
3355 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3356 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3357 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3358 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3359 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3360 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3361 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3362 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3363 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3364 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3365 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3366 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3367 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3368 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3369 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3370 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3371 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3372 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3373 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3374 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3375 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3376 (vfmlsl_low_u32): Ditto.
3377 (vfmlalq_low_u32): Ditto.
3378 (vfmlslq_low_u32): Ditto.
3379 (vfmlal_high_u32): Ditto.
3380 (vfmlsl_high_u32): Ditto.
3381 (vfmlalq_high_u32): Ditto.
3382 (vfmlslq_high_u32): Ditto.
3383 (vfmlal_lane_low_u32): Ditto.
3384 (vfmlsl_lane_low_u32): Ditto.
3385 (vfmlal_laneq_low_u32): Ditto.
3386 (vfmlsl_laneq_low_u32): Ditto.
3387 (vfmlalq_lane_low_u32): Ditto.
3388 (vfmlslq_lane_low_u32): Ditto.
3389 (vfmlalq_laneq_low_u32): Ditto.
3390 (vfmlslq_laneq_low_u32): Ditto.
3391 (vfmlal_lane_high_u32): Ditto.
3392 (vfmlsl_lane_high_u32): Ditto.
3393 (vfmlal_laneq_high_u32): Ditto.
3394 (vfmlsl_laneq_high_u32): Ditto.
3395 (vfmlalq_lane_high_u32): Ditto.
3396 (vfmlslq_lane_high_u32): Ditto.
3397 (vfmlalq_laneq_high_u32): Ditto.
3398 (vfmlslq_laneq_high_u32): Ditto.
3399 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3400 (AARCH64_FL_FOR_ARCH8_4): New.
3401 (AARCH64_ISA_F16FML): New ISA flag.
3402 (TARGET_F16FML): New feature flag for fp16fml.
3403 (doc/invoke.texi): Document new fp16fml option.
3405 2018-01-10 Michael Collison <michael.collison@arm.com>
3407 * config/aarch64/aarch64-builtins.c:
3408 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3409 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3410 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3411 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3412 (AARCH64_ISA_SHA3): New ISA flag.
3413 (TARGET_SHA3): New feature flag for sha3.
3414 * config/aarch64/iterators.md (sha512_op): New int attribute.
3415 (CRYPTO_SHA512): New int iterator.
3416 (UNSPEC_SHA512H): New unspec.
3417 (UNSPEC_SHA512H2): Ditto.
3418 (UNSPEC_SHA512SU0): Ditto.
3419 (UNSPEC_SHA512SU1): Ditto.
3420 * config/aarch64/aarch64-simd-builtins.def
3421 (aarch64_crypto_sha512hqv2di): New builtin.
3422 (aarch64_crypto_sha512h2qv2di): Ditto.
3423 (aarch64_crypto_sha512su0qv2di): Ditto.
3424 (aarch64_crypto_sha512su1qv2di): Ditto.
3425 (aarch64_eor3qv8hi): Ditto.
3426 (aarch64_rax1qv2di): Ditto.
3427 (aarch64_xarqv2di): Ditto.
3428 (aarch64_bcaxqv8hi): Ditto.
3429 * config/aarch64/aarch64-simd.md:
3430 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3431 (aarch64_crypto_sha512su0qv2di): Ditto.
3432 (aarch64_crypto_sha512su1qv2di): Ditto.
3433 (aarch64_eor3qv8hi): Ditto.
3434 (aarch64_rax1qv2di): Ditto.
3435 (aarch64_xarqv2di): Ditto.
3436 (aarch64_bcaxqv8hi): Ditto.
3437 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3438 (vsha512h2q_u64): Ditto.
3439 (vsha512su0q_u64): Ditto.
3440 (vsha512su1q_u64): Ditto.
3441 (veor3q_u16): Ditto.
3442 (vrax1q_u64): Ditto.
3444 (vbcaxq_u16): Ditto.
3445 * config/arm/types.md (crypto_sha512): New type attribute.
3446 (crypto_sha3): Ditto.
3447 (doc/invoke.texi): Document new sha3 option.
3449 2018-01-10 Michael Collison <michael.collison@arm.com>
3451 * config/aarch64/aarch64-builtins.c:
3452 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3453 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3454 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3455 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3456 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3457 (AARCH64_ISA_SM4): New ISA flag.
3458 (TARGET_SM4): New feature flag for sm4.
3459 * config/aarch64/aarch64-simd-builtins.def
3460 (aarch64_sm3ss1qv4si): Ditto.
3461 (aarch64_sm3tt1aq4si): Ditto.
3462 (aarch64_sm3tt1bq4si): Ditto.
3463 (aarch64_sm3tt2aq4si): Ditto.
3464 (aarch64_sm3tt2bq4si): Ditto.
3465 (aarch64_sm3partw1qv4si): Ditto.
3466 (aarch64_sm3partw2qv4si): Ditto.
3467 (aarch64_sm4eqv4si): Ditto.
3468 (aarch64_sm4ekeyqv4si): Ditto.
3469 * config/aarch64/aarch64-simd.md:
3470 (aarch64_sm3ss1qv4si): Ditto.
3471 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3472 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3473 (aarch64_sm4eqv4si): Ditto.
3474 (aarch64_sm4ekeyqv4si): Ditto.
3475 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3476 (sm3part_op): Ditto.
3477 (CRYPTO_SM3TT): Ditto.
3478 (CRYPTO_SM3PART): Ditto.
3479 (UNSPEC_SM3SS1): New unspec.
3480 (UNSPEC_SM3TT1A): Ditto.
3481 (UNSPEC_SM3TT1B): Ditto.
3482 (UNSPEC_SM3TT2A): Ditto.
3483 (UNSPEC_SM3TT2B): Ditto.
3484 (UNSPEC_SM3PARTW1): Ditto.
3485 (UNSPEC_SM3PARTW2): Ditto.
3486 (UNSPEC_SM4E): Ditto.
3487 (UNSPEC_SM4EKEY): Ditto.
3488 * config/aarch64/constraints.md (Ui2): New constraint.
3489 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3490 * config/arm/types.md (crypto_sm3): New type attribute.
3491 (crypto_sm4): Ditto.
3492 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3493 (vsm3tt1aq_u32): Ditto.
3494 (vsm3tt1bq_u32): Ditto.
3495 (vsm3tt2aq_u32): Ditto.
3496 (vsm3tt2bq_u32): Ditto.
3497 (vsm3partw1q_u32): Ditto.
3498 (vsm3partw2q_u32): Ditto.
3499 (vsm4eq_u32): Ditto.
3500 (vsm4ekeyq_u32): Ditto.
3501 (doc/invoke.texi): Document new sm4 option.
3503 2018-01-10 Michael Collison <michael.collison@arm.com>
3505 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3506 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3507 (AARCH64_FL_FOR_ARCH8_4): New.
3508 (AARCH64_FL_V8_4): New flag.
3509 (doc/invoke.texi): Document new armv8.4-a option.
3511 2018-01-10 Michael Collison <michael.collison@arm.com>
3513 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3514 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3515 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3516 * config/aarch64/aarch64-option-extension.def: Add
3517 AARCH64_OPT_EXTENSION of 'sha2'.
3518 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3519 (crypto): Disable sha2 and aes if crypto disabled.
3520 (crypto): Enable aes and sha2 if enabled.
3521 (simd): Disable sha2 and aes if simd disabled.
3522 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3524 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3525 (TARGET_SHA2): New feature flag for sha2.
3526 (TARGET_AES): New feature flag for aes.
3527 * config/aarch64/aarch64-simd.md:
3528 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3529 conditional on TARGET_AES.
3530 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3531 (aarch64_crypto_sha1hsi): Make pattern conditional
3533 (aarch64_crypto_sha1hv4si): Ditto.
3534 (aarch64_be_crypto_sha1hv4si): Ditto.
3535 (aarch64_crypto_sha1su1v4si): Ditto.
3536 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3537 (aarch64_crypto_sha1su0v4si): Ditto.
3538 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3539 (aarch64_crypto_sha256su0v4si): Ditto.
3540 (aarch64_crypto_sha256su1v4si): Ditto.
3541 (doc/invoke.texi): Document new aes and sha2 options.
3543 2018-01-10 Martin Sebor <msebor@redhat.com>
3545 PR tree-optimization/83781
3546 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3549 2018-01-11 Martin Sebor <msebor@gmail.com>
3550 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3552 PR tree-optimization/83501
3553 PR tree-optimization/81703
3555 * tree-ssa-strlen.c (get_string_cst): Rename...
3556 (get_string_len): ...to this. Handle global constants.
3557 (handle_char_store): Adjust.
3559 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3560 Jim Wilson <jimw@sifive.com>
3562 * config/riscv/riscv-protos.h (riscv_output_return): New.
3563 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3564 (riscv_attribute_table, riscv_output_return),
3565 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3566 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3567 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3569 (riscv_expand_prologue): Add early return for naked function.
3570 (riscv_expand_epilogue): Likewise.
3571 (riscv_function_ok_for_sibcall): Return false for naked function.
3572 (riscv_set_current_function): New.
3573 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3574 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3575 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3576 * doc/extend.texi (RISC-V Function Attributes): New.
3578 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3580 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3581 check for 128-bit long double before checking TCmode.
3582 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3583 128-bit long doubles before checking TFmode or TCmode.
3584 (FLOAT128_IBM_P): Likewise.
3586 2018-01-10 Martin Sebor <msebor@redhat.com>
3588 PR tree-optimization/83671
3589 * builtins.c (c_strlen): Unconditionally return zero for the empty
3591 Use -Warray-bounds for warnings.
3592 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3593 for non-constant array indices with COMPONENT_REF, arrays of
3594 arrays, and pointers to arrays.
3595 (gimple_fold_builtin_strlen): Determine and set length range for
3596 non-constant character arrays.
3598 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3601 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3604 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3606 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3608 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3611 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3612 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3613 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3614 indexed_or_indirect_operand predicate.
3615 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3616 (*vsx_le_perm_load_v8hi): Likewise.
3617 (*vsx_le_perm_load_v16qi): Likewise.
3618 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3619 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3620 (*vsx_le_perm_store_v8hi): Likewise.
3621 (*vsx_le_perm_store_v16qi): Likewise.
3622 (eight unnamed splitters): Likewise.
3624 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3626 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3627 * config/rs6000/emmintrin.h: Likewise.
3628 * config/rs6000/mmintrin.h: Likewise.
3629 * config/rs6000/xmmintrin.h: Likewise.
3631 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3634 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3636 * tree.c (tree_nop_conversion): Return true for location wrapper
3638 (maybe_wrap_with_location): New function.
3639 (selftest::check_strip_nops): New function.
3640 (selftest::test_location_wrappers): New function.
3641 (selftest::tree_c_tests): Call it.
3642 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3643 (maybe_wrap_with_location): New decl.
3644 (EXPR_LOCATION_WRAPPER_P): New macro.
3645 (location_wrapper_p): New inline function.
3646 (tree_strip_any_location_wrapper): New inline function.
3648 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3651 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3652 stack_realign_offset for the largest alignment of stack slot
3654 (ix86_find_max_used_stack_alignment): New function.
3655 (ix86_finalize_stack_frame_flags): Use it. Set
3656 max_used_stack_alignment if we don't realign stack.
3657 * config/i386/i386.h (machine_function): Add
3658 max_used_stack_alignment.
3660 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3662 * config/arm/arm.opt (-mbranch-cost): New option.
3663 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3666 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3669 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3670 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3672 2018-01-10 Richard Biener <rguenther@suse.de>
3675 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3676 early out so it also covers the case where we have a non-NULL
3679 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3681 PR tree-optimization/83753
3682 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3683 for non-strided grouped accesses if the number of elements is 1.
3685 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3688 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3689 * i386.h (TARGET_USE_GATHER): Define.
3690 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3692 2018-01-10 Martin Liska <mliska@suse.cz>
3695 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3696 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3698 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3699 CLEANUP_NO_PARTITIONING is not set.
3701 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3703 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3704 for vectors, as a partial revert of r254296.
3705 * rtl.h (const_vec_p): Delete.
3706 (const_vec_duplicate_p): Don't test for vector CONSTs.
3707 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3708 * expmed.c (make_tree): Likewise.
3711 * common.md (E, F): Use CONSTANT_P instead of checking for
3713 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3714 checking for CONST_VECTOR.
3716 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3719 * predict.c (force_edge_cold): Handle in more sane way edges
3722 2018-01-09 Carl Love <cel@us.ibm.com>
3724 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3726 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3727 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3728 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3729 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3730 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3731 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3732 * config/rs6000/rs6000-protos.h: Add extern defition for
3733 rs6000_generate_float2_double_code.
3734 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3736 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3737 (float2_v2df): Add define_expand.
3739 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3742 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3743 op_mode in the force_to_mode call.
3745 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3747 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3748 instead of checking each element individually.
3749 (aarch64_evpc_uzp): Likewise.
3750 (aarch64_evpc_zip): Likewise.
3751 (aarch64_evpc_ext): Likewise.
3752 (aarch64_evpc_rev): Likewise.
3753 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3754 instead of checking each element individually. Return true without
3756 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3757 whether all selected elements come from the same input, instead of
3758 checking each element individually. Remove calls to gen_rtx_REG,
3759 start_sequence and end_sequence and instead assert that no rtl is
3762 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3764 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3765 order of HIGH and CONST checks.
3767 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3769 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3770 if the destination isn't an SSA_NAME.
3772 2018-01-09 Richard Biener <rguenther@suse.de>
3774 PR tree-optimization/83668
3775 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3777 (canonicalize_loop_form): ... here, renamed from ...
3778 (canonicalize_loop_closed_ssa_form): ... this and amended to
3779 swap successor edges for loop exit blocks to make us use
3780 the RPO order we need for initial schedule generation.
3782 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3784 PR tree-optimization/64811
3785 * match.pd: When optimizing comparisons with Inf, avoid
3786 introducing or losing exceptions from comparisons with NaN.
3788 2018-01-09 Martin Liska <mliska@suse.cz>
3791 * asan.c (shadow_mem_size): Add gcc_assert.
3793 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3795 Don't save registers in main().
3798 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3799 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3800 * config/avr/avr.c (avr_set_current_function): Don't error if
3801 naked, OS_task or OS_main are specified at the same time.
3802 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3804 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3806 * common/config/avr/avr-common.c (avr_option_optimization_table):
3807 Switch on -mmain-is-OS_task for optimizing compilations.
3809 2018-01-09 Richard Biener <rguenther@suse.de>
3811 PR tree-optimization/83572
3812 * graphite.c: Include cfganal.h.
3813 (graphite_transform_loops): Connect infinite loops to exit
3814 and remove fake edges at the end.
3816 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3818 * ipa-inline.c (edge_badness): Revert accidental checkin.
3820 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3823 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3824 symbols; not inline clones.
3826 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3829 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3830 hard registers. Formatting fixes.
3832 PR preprocessor/83722
3833 * gcc.c (try_generate_repro): Pass
3834 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3835 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3838 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3839 Kito Cheng <kito.cheng@gmail.com>
3841 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3842 (riscv_leaf_function_p): Delete.
3843 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3845 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3847 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3849 (do_ifelse): New function.
3850 (do_isel): New function.
3851 (do_sub3): New function.
3852 (do_add3): New function.
3853 (do_load_mask_compare): New function.
3854 (do_overlap_load_compare): New function.
3855 (expand_compare_loop): New function.
3856 (expand_block_compare): Call expand_compare_loop() when appropriate.
3857 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3859 (-mblock-compare-inline-loop-limit): New option.
3861 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3864 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3865 Reverse order of second and third operands in first alternative.
3866 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3867 of first and second elements in UNSPEC_VPERMR vector.
3868 (altivec_expand_vec_perm_le): Likewise.
3870 2017-01-08 Jeff Law <law@redhat.com>
3872 PR rtl-optimizatin/81308
3873 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3874 (process_switch): If group_case_labels makes a change, then set
3876 (pass_convert_switch::execute): If a switch is converted, then
3877 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3879 PR rtl-optimization/81308
3880 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3883 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3885 PR target/83663 - Revert r255946
3886 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3887 generation for cases where splatting a value is not useful.
3888 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3889 across a vec_duplicate and a paradoxical subreg forming a vector
3890 mode to a vec_concat.
3892 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3894 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3895 -march=armv8.3-a variants.
3896 * config/arm/t-multilib: Likewise.
3897 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3899 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3901 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3903 (cceq_ior_compare_complement): Give it a name so I can use it, and
3904 change boolean_or_operator predicate to boolean_operator so it can
3905 be used to generate a crand.
3906 (eqne): New code iterator.
3907 (bd/bd_neg): New code_attrs.
3908 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3909 a single define_insn.
3910 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3911 decrement (bdnzt/bdnzf/bdzt/bdzf).
3912 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3913 with the new names of the branch decrement patterns, and added the
3914 names of the branch decrement conditional patterns.
3916 2018-01-08 Richard Biener <rguenther@suse.de>
3918 PR tree-optimization/83563
3919 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3922 2018-01-08 Richard Biener <rguenther@suse.de>
3925 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3927 2018-01-08 Richard Biener <rguenther@suse.de>
3929 PR tree-optimization/83685
3930 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3931 references to abnormals.
3933 2018-01-08 Richard Biener <rguenther@suse.de>
3936 * dwarf2out.c (output_indirect_strings): Handle empty
3937 skeleton_debug_str_hash.
3938 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3940 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3942 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3943 (emit_store_direct): Likewise.
3944 (arc_trampoline_adjust_address): Likewise.
3945 (arc_asm_trampoline_template): New function.
3946 (arc_initialize_trampoline): Use asm_trampoline_template.
3947 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3948 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3949 * config/arc/arc.md (flush_icache): Delete pattern.
3951 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3953 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3954 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3957 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3960 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3961 by not USED_FOR_TARGET.
3962 (make_pass_resolve_sw_modes): Likewise.
3964 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3966 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3969 2018-01-08 Richard Biener <rguenther@suse.de>
3972 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3974 2018-01-08 Richard Biener <rguenther@suse.de>
3977 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3979 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3982 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3983 basic blocks with a small number of successors.
3984 (convert_control_dep_chain_into_preds): Improve handling of
3986 (dump_predicates): Split apart into...
3987 (dump_pred_chain): ...here...
3988 (dump_pred_info): ...and here.
3989 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3990 (can_chain_union_be_invalidated_p): Improve check for invalidation
3992 (uninit_uses_cannot_happen): Avoid unnecessary if
3993 convert_control_dep_chain_into_preds yielded nothing.
3995 2018-01-06 Martin Sebor <msebor@redhat.com>
3997 PR tree-optimization/83640
3998 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3999 subtracting negative offset from size.
4000 (builtin_access::overlap): Adjust offset bounds of the access to fall
4001 within the size of the object if possible.
4003 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4005 PR rtl-optimization/83699
4006 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4007 extract_bit_field_as_subreg to cases in which the extracted
4008 value is also a vector.
4010 * lra-constraints.c (process_alt_operands): Test for the equivalence
4011 substitutions when detecting a possible reload cycle.
4013 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4016 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4017 by default if flag_selective_schedling{,2}. Formatting fixes.
4019 PR rtl-optimization/83682
4020 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4021 if it has non-VECTOR_MODE element mode.
4022 (vec_duplicate_p): Likewise.
4025 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4026 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4028 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4031 * config/i386/i386-builtin.def
4032 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4033 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4034 Require also OPTION_MASK_ISA_AVX512F in addition to
4035 OPTION_MASK_ISA_GFNI.
4036 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4037 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4038 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4039 to OPTION_MASK_ISA_GFNI.
4040 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4041 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4042 OPTION_MASK_ISA_AVX512BW.
4043 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4044 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4045 addition to OPTION_MASK_ISA_GFNI.
4046 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4047 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4048 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4049 to OPTION_MASK_ISA_GFNI.
4050 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4051 a requirement for all ISAs rather than any of them with a few
4053 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4055 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4056 bitmasks to be enabled with 3 exceptions, instead of requiring any
4057 enabled ISA with lots of exceptions.
4058 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4059 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4060 Change avx512bw in isa attribute to avx512f.
4061 * config/i386/sgxintrin.h: Add license boilerplate.
4062 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4063 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4064 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4065 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4067 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4068 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4069 temporarily sse2 rather than sse if not enabled already.
4072 * config/i386/sse.md (VI248_VLBW): Rename to ...
4073 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4074 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4075 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4076 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4077 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4078 mode iterator instead of VI248_VLBW.
4080 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4082 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4083 (record_modified): Skip clobbers; add debug output.
4084 (param_change_prob): Use sreal frequencies.
4086 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4088 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4089 punt for user-aligned variables.
4091 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4093 * tree-chrec.c (chrec_contains_symbols): Return true for
4096 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4099 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4100 of (x|y) == x for BICS pattern.
4102 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4104 PR tree-optimization/83605
4105 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4106 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4109 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4111 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4112 * config/epiphany/rtems.h: New file.
4114 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4115 Uros Bizjak <ubizjak@gmail.com>
4118 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4119 QIreg_operand instead of register_operand predicate.
4120 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4121 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4122 comments instead of -fmitigate[-_]rop.
4124 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4127 * cgraphunit.c (symbol_table::compile): Switch to text_section
4128 before calling assembly_start debug hook.
4129 * run-rtl-passes.c (run_rtl_passes): Likewise.
4132 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4134 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4135 range_int_cst_p rather than !symbolic_range_p before calling
4136 extract_range_from_multiplicative_op_1.
4138 2017-01-04 Jeff Law <law@redhat.com>
4140 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4141 redundant test in assertion.
4143 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4145 * doc/rtl.texi: Document machine_mode wrapper classes.
4147 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4149 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4152 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4154 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4155 the VEC_PERM_EXPR fold to fail.
4157 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4160 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4161 to switched_sections.
4163 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4166 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4169 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4172 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4173 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4175 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4178 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4179 is BLKmode and bitpos not zero or mode change is needed.
4181 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4184 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4187 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4190 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4191 instead of MULT rtx. Update all corresponding splitters.
4193 (*ssub<modesuffix>): Ditto.
4195 (*cmp_sadd_di): Update split patterns.
4196 (*cmp_sadd_si): Ditto.
4197 (*cmp_sadd_sidi): Ditto.
4198 (*cmp_ssub_di): Ditto.
4199 (*cmp_ssub_si): Ditto.
4200 (*cmp_ssub_sidi): Ditto.
4201 * config/alpha/predicates.md (const23_operand): New predicate.
4202 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4203 Look for ASHIFT, not MULT inner operand.
4204 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4206 2018-01-04 Martin Liska <mliska@suse.cz>
4208 PR gcov-profile/83669
4209 * gcov.c (output_intermediate_file): Add version to intermediate
4211 * doc/gcov.texi: Document new field 'version' in intermediate
4212 file format. Fix location of '-k' option of gcov command.
4214 2018-01-04 Martin Liska <mliska@suse.cz>
4217 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4219 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4221 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4223 2018-01-03 Martin Sebor <msebor@redhat.com>
4225 PR tree-optimization/83655
4226 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4227 checking calls with invalid arguments.
4229 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4231 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4232 (vectorizable_mask_load_store): Delete.
4233 (vectorizable_call): Return false for masked loads and stores.
4234 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4235 instead of gimple_assign_rhs1.
4236 (vectorizable_load): Handle IFN_MASK_LOAD.
4237 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4239 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4241 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4243 (vectorizable_mask_load_store): ...here.
4244 (vectorizable_load): ...and here.
4246 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4248 * tree-vect-stmts.c (vect_build_all_ones_mask)
4249 (vect_build_zero_merge_argument): New functions, split out from...
4250 (vectorizable_load): ...here.
4252 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4254 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4256 (vectorizable_mask_load_store): ...here.
4257 (vectorizable_store): ...and here.
4259 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4261 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4263 (vectorizable_mask_load_store): ...here.
4265 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4267 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4268 (vect_model_store_cost): Take a vec_load_store_type instead of a
4270 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4271 (vect_model_store_cost): Take a vec_load_store_type instead of a
4273 (vectorizable_mask_load_store): Update accordingly.
4274 (vectorizable_store): Likewise.
4275 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4277 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4279 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4280 IFN_MASK_LOAD calls here rather than...
4281 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4283 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4284 Alan Hayward <alan.hayward@arm.com>
4285 David Sherwood <david.sherwood@arm.com>
4287 * expmed.c (extract_bit_field_1): For vector extracts,
4288 fall back to extract_bit_field_as_subreg if vec_extract
4291 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4292 Alan Hayward <alan.hayward@arm.com>
4293 David Sherwood <david.sherwood@arm.com>
4295 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4296 they are variable or constant sized.
4297 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4298 slots for constant-sized data.
4300 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4301 Alan Hayward <alan.hayward@arm.com>
4302 David Sherwood <david.sherwood@arm.com>
4304 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4305 handling COND_EXPRs with boolean comparisons, try to find a better
4306 basis for the mask type than the boolean itself.
4308 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4310 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4311 is calculated and how it can be overridden.
4312 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4313 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4315 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4318 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4319 Alan Hayward <alan.hayward@arm.com>
4320 David Sherwood <david.sherwood@arm.com>
4322 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4323 Remove the mode argument.
4324 (aarch64_simd_valid_immediate): Remove the mode and inverse
4326 * config/aarch64/iterators.md (bitsize): New iterator.
4327 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4328 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4329 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4330 aarch64_simd_valid_immediate.
4331 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4332 (aarch64_reg_or_bic_imm): Likewise.
4333 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4334 with an insn_type enum and msl with a modifier_type enum.
4335 Replace element_width with a scalar_mode. Change the shift
4336 to unsigned int. Add constructors for scalar_float_mode and
4337 scalar_int_mode elements.
4338 (aarch64_vect_float_const_representable_p): Delete.
4339 (aarch64_can_const_movi_rtx_p)
4340 (aarch64_simd_scalar_immediate_valid_for_move)
4341 (aarch64_simd_make_constant): Update call to
4342 aarch64_simd_valid_immediate.
4343 (aarch64_advsimd_valid_immediate_hs): New function.
4344 (aarch64_advsimd_valid_immediate): Likewise.
4345 (aarch64_simd_valid_immediate): Remove mode and inverse
4346 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4347 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4348 and aarch64_float_const_representable_p on the result.
4349 (aarch64_output_simd_mov_immediate): Remove mode argument.
4350 Update call to aarch64_simd_valid_immediate and use of
4351 simd_immediate_info.
4352 (aarch64_output_scalar_simd_mov_immediate): Update call
4355 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4356 Alan Hayward <alan.hayward@arm.com>
4357 David Sherwood <david.sherwood@arm.com>
4359 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4360 (mode_nunits): Likewise CONST_MODE_NUNITS.
4361 * machmode.def (ADJUST_NUNITS): Document.
4362 * genmodes.c (mode_data::need_nunits_adj): New field.
4363 (blank_mode): Update accordingly.
4364 (adj_nunits): New variable.
4365 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4367 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4368 listed in adj_nunits.
4369 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4370 listed in adj_nunits. Don't emit case statements for such modes.
4371 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4372 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4373 nothing if adj_nunits is nonnull.
4374 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4375 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4376 (emit_mode_fbit): Update use of print_maybe_const_decl.
4377 (emit_move_size): Likewise. Treat the array as non-const
4379 (emit_mode_adjustments): Handle adj_nunits.
4381 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4383 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4384 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4385 (VECTOR_MODES): Use it.
4386 (make_vector_modes): Take the prefix as an argument.
4388 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4389 Alan Hayward <alan.hayward@arm.com>
4390 David Sherwood <david.sherwood@arm.com>
4392 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4393 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4394 for MODE_VECTOR_BOOL.
4395 * machmode.def (VECTOR_BOOL_MODE): Document.
4396 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4397 (make_vector_bool_mode): New function.
4398 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4400 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4401 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4403 * stor-layout.c (int_mode_for_mode): Likewise.
4404 * tree.c (build_vector_type_for_mode): Likewise.
4405 * varasm.c (output_constant_pool_2): Likewise.
4406 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4407 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4408 for MODE_VECTOR_BOOL.
4409 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4410 of mode class checks.
4411 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4412 instead of a list of mode class checks.
4413 (expand_vector_scalar_condition): Likewise.
4414 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4416 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4417 Alan Hayward <alan.hayward@arm.com>
4418 David Sherwood <david.sherwood@arm.com>
4420 * machmode.h (mode_size): Change from unsigned short to
4422 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4423 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4424 or if measurement_type is not polynomial.
4425 (fixed_size_mode::includes_p): Check for constant-sized modes.
4426 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4427 return a poly_uint16 rather than an unsigned short.
4428 (emit_mode_size): Change the type of mode_size from unsigned short
4429 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4430 (emit_mode_adjustments): Cope with polynomial vector sizes.
4431 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4433 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4435 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4436 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4437 * caller-save.c (setup_save_areas): Likewise.
4438 (replace_reg_with_saved_mem): Likewise.
4439 * calls.c (emit_library_call_value_1): Likewise.
4440 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4441 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4442 (gen_lowpart_for_combine): Likewise.
4443 * convert.c (convert_to_integer_1): Likewise.
4444 * cse.c (equiv_constant, cse_insn): Likewise.
4445 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4446 (cselib_subst_to_values): Likewise.
4447 * dce.c (word_dce_process_block): Likewise.
4448 * df-problems.c (df_word_lr_mark_ref): Likewise.
4449 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4450 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4451 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4452 (rtl_for_decl_location): Likewise.
4453 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4454 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4455 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4456 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4457 (expand_expr_real_1): Likewise.
4458 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4459 (pad_below): Likewise.
4460 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4461 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4462 * ira.c (get_subreg_tracking_sizes): Likewise.
4463 * ira-build.c (ira_create_allocno_objects): Likewise.
4464 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4465 (ira_sort_regnos_for_alter_reg): Likewise.
4466 * ira-costs.c (record_operand_costs): Likewise.
4467 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4468 (resolve_simple_move): Likewise.
4469 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4470 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4471 (lra_constraints): Likewise.
4472 (CONST_POOL_OK_P): Reject variable-sized modes.
4473 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4474 (add_pseudo_to_slot, lra_spill): Likewise.
4475 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4476 * optabs-query.c (get_best_extraction_insn): Likewise.
4477 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4478 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4479 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4480 * recog.c (offsettable_address_addr_space_p): Likewise.
4481 * regcprop.c (maybe_mode_change): Likewise.
4482 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4483 * regrename.c (build_def_use): Likewise.
4484 * regstat.c (dump_reg_info): Likewise.
4485 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4486 (find_reloads, find_reloads_subreg_address): Likewise.
4487 * reload1.c (eliminate_regs_1): Likewise.
4488 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4489 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4490 (simplify_binary_operation_1, simplify_subreg): Likewise.
4491 * targhooks.c (default_function_arg_padding): Likewise.
4492 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4493 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4494 (verify_gimple_assign_ternary): Likewise.
4495 * tree-inline.c (estimate_move_cost): Likewise.
4496 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4497 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4498 (get_address_cost_ainc): Likewise.
4499 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4500 (vect_supportable_dr_alignment): Likewise.
4501 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4502 (vectorizable_reduction): Likewise.
4503 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4504 (vectorizable_operation, vectorizable_load): Likewise.
4505 * tree.c (build_same_sized_truth_vector_type): Likewise.
4506 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4507 * var-tracking.c (emit_note_insn_var_location): Likewise.
4508 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4509 (ADDR_VEC_ALIGN): Likewise.
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 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4517 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4518 or if measurement_type is polynomial.
4519 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4520 * combine.c (make_extraction): Likewise.
4521 * dse.c (find_shift_sequence): Likewise.
4522 * dwarf2out.c (mem_loc_descriptor): Likewise.
4523 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4524 (extract_bit_field, extract_low_bits): Likewise.
4525 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4526 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4527 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4528 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4529 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4530 * reload.c (find_reloads): Likewise.
4531 * reload1.c (alter_reg): Likewise.
4532 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4533 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4534 * tree-if-conv.c (predicate_mem_writes): Likewise.
4535 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4536 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4537 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4538 * valtrack.c (dead_debug_insert_temp): Likewise.
4539 * varasm.c (mergeable_constant_section): Likewise.
4540 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4542 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4543 Alan Hayward <alan.hayward@arm.com>
4544 David Sherwood <david.sherwood@arm.com>
4546 * expr.c (expand_assignment): Cope with polynomial mode sizes
4547 when assigning to a CONCAT.
4549 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4550 Alan Hayward <alan.hayward@arm.com>
4551 David Sherwood <david.sherwood@arm.com>
4553 * machmode.h (mode_precision): Change from unsigned short to
4555 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4557 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4558 or if measurement_type is not polynomial.
4559 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4560 in which the mode is already known to be a scalar_int_mode.
4561 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4562 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4564 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4565 for GET_MODE_PRECISION.
4566 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4567 for GET_MODE_PRECISION.
4568 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4570 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4571 (expand_field_assignment, make_extraction): Likewise.
4572 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4573 (get_last_value): Likewise.
4574 * convert.c (convert_to_integer_1): Likewise.
4575 * cse.c (cse_insn): Likewise.
4576 * expr.c (expand_expr_real_1): Likewise.
4577 * lra-constraints.c (simplify_operand_subreg): Likewise.
4578 * optabs-query.c (can_atomic_load_p): Likewise.
4579 * optabs.c (expand_atomic_load): Likewise.
4580 (expand_atomic_store): Likewise.
4581 * ree.c (combine_reaching_defs): Likewise.
4582 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4583 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4584 * tree.h (type_has_mode_precision_p): Likewise.
4585 * ubsan.c (instrument_si_overflow): Likewise.
4587 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4588 Alan Hayward <alan.hayward@arm.com>
4589 David Sherwood <david.sherwood@arm.com>
4591 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4592 polynomial numbers of units.
4593 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4594 (valid_vector_subparts_p): New function.
4595 (build_vector_type): Remove temporary shim and take the number
4596 of units as a poly_uint64 rather than an int.
4597 (build_opaque_vector_type): Take the number of units as a
4598 poly_uint64 rather than an int.
4599 * tree.c (build_vector_from_ctor): Handle polynomial
4600 TYPE_VECTOR_SUBPARTS.
4601 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4602 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4603 (build_vector_from_val): If the number of units is variable,
4604 use build_vec_duplicate_cst for constant operands and
4605 VEC_DUPLICATE_EXPR otherwise.
4606 (make_vector_type): Remove temporary is_constant ().
4607 (build_vector_type, build_opaque_vector_type): Take the number of
4608 units as a poly_uint64 rather than an int.
4609 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4611 * cfgexpand.c (expand_debug_expr): Likewise.
4612 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4613 (store_constructor, expand_expr_real_1): Likewise.
4614 (const_scalar_mask_from_tree): Likewise.
4615 * fold-const-call.c (fold_const_reduction): Likewise.
4616 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4617 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4618 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4619 (fold_relational_const): Likewise.
4620 (native_interpret_vector): Likewise. Change the size from an
4621 int to an unsigned int.
4622 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4623 TYPE_VECTOR_SUBPARTS.
4624 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4625 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4626 duplicating a non-constant operand into a variable-length vector.
4627 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4628 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4629 * ipa-icf.c (sem_variable::equals): Likewise.
4630 * match.pd: Likewise.
4631 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4632 * print-tree.c (print_node): Likewise.
4633 * stor-layout.c (layout_type): Likewise.
4634 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4635 * tree-cfg.c (verify_gimple_comparison): Likewise.
4636 (verify_gimple_assign_binary): Likewise.
4637 (verify_gimple_assign_ternary): Likewise.
4638 (verify_gimple_assign_single): Likewise.
4639 * tree-pretty-print.c (dump_generic_node): Likewise.
4640 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4641 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4642 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4643 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4644 (vect_shift_permute_load_chain): Likewise.
4645 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4646 (expand_vector_condition, optimize_vector_constructor): Likewise.
4647 (lower_vec_perm, get_compute_type): Likewise.
4648 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4649 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4650 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4651 (vect_recog_mask_conversion_pattern): Likewise.
4652 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4653 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4654 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4655 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4656 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4657 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4658 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4659 (supportable_widening_operation): Likewise.
4660 (supportable_narrowing_operation): Likewise.
4661 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4663 * varasm.c (output_constant): Likewise.
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 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4670 so that both the length == 3 and length != 3 cases set up their
4671 own permute vectors. Add comments explaining why we know the
4672 number of elements is constant.
4673 (vect_permute_load_chain): Likewise.
4675 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4676 Alan Hayward <alan.hayward@arm.com>
4677 David Sherwood <david.sherwood@arm.com>
4679 * machmode.h (mode_nunits): Change from unsigned char to
4681 (ONLY_FIXED_SIZE_MODES): New macro.
4682 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4683 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4684 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4686 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4687 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4688 or if measurement_type is not polynomial.
4689 * genmodes.c (ZERO_COEFFS): New macro.
4690 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4692 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4693 Use ZERO_COEFFS when emitting initializers.
4694 * data-streamer.h (bp_pack_poly_value): New function.
4695 (bp_unpack_poly_value): Likewise.
4696 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4697 for GET_MODE_NUNITS.
4698 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4699 for GET_MODE_NUNITS.
4700 * tree.c (make_vector_type): Remove temporary shim and make
4701 the real function take the number of units as a poly_uint64
4703 (build_vector_type_for_mode): Handle polynomial nunits.
4704 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4705 * emit-rtl.c (const_vec_series_p_1): Likewise.
4706 (gen_rtx_CONST_VECTOR): Likewise.
4707 * fold-const.c (test_vec_duplicate_folding): Likewise.
4708 * genrecog.c (validate_pattern): Likewise.
4709 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4710 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4711 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4712 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4713 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4714 * rtlanal.c (subreg_get_info): Likewise.
4715 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4716 (vect_grouped_load_supported): Likewise.
4717 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4718 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4719 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4720 (simplify_const_unary_operation, simplify_binary_operation_1)
4721 (simplify_const_binary_operation, simplify_ternary_operation)
4722 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4723 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4724 instead of CONST_VECTOR_NUNITS.
4725 * varasm.c (output_constant_pool_2): Likewise.
4726 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4727 explicit-encoded elements in the XVEC for variable-length vectors.
4729 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4731 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4733 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4734 Alan Hayward <alan.hayward@arm.com>
4735 David Sherwood <david.sherwood@arm.com>
4737 * coretypes.h (fixed_size_mode): Declare.
4738 (fixed_size_mode_pod): New typedef.
4739 * builtins.h (target_builtins::x_apply_args_mode)
4740 (target_builtins::x_apply_result_mode): Change type to
4741 fixed_size_mode_pod.
4742 * builtins.c (apply_args_size, apply_result_size, result_vector)
4743 (expand_builtin_apply_args_1, expand_builtin_apply)
4744 (expand_builtin_return): Update accordingly.
4746 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4748 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4749 * cselib.c (cselib_hash_rtx): Likewise.
4750 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4751 CONST_VECTOR encoding.
4753 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4754 Jeff Law <law@redhat.com>
4757 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4758 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4759 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4760 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4763 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4764 explicitly probe *sp in a noreturn function if there were any callee
4765 register saves or frame pointer is needed.
4767 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4770 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4771 BLKmode for ternary, binary or unary expressions.
4774 * var-tracking.c (delete_vta_debug_insn): New inline function.
4775 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4776 insns from get_insns () to NULL instead of each bb separately.
4777 Use delete_vta_debug_insn. No longer static.
4778 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4779 delete_vta_debug_insns callers.
4780 * rtl.h (delete_vta_debug_insns): Declare.
4781 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4782 instead of variable_tracking_main.
4784 2018-01-03 Martin Sebor <msebor@redhat.com>
4786 PR tree-optimization/83603
4787 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4788 arguments past the endof the argument list in functions declared
4789 without a prototype.
4790 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4791 Avoid checking when arguments are null.
4793 2018-01-03 Martin Sebor <msebor@redhat.com>
4796 * doc/extend.texi (attribute const): Fix a typo.
4797 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4798 issuing -Wsuggest-attribute for void functions.
4800 2018-01-03 Martin Sebor <msebor@redhat.com>
4802 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4803 offset_int::from instead of wide_int::to_shwi.
4804 (maybe_diag_overlap): Remove assertion.
4805 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4806 * gimple-ssa-sprintf.c (format_directive): Same.
4807 (parse_directive): Same.
4808 (sprintf_dom_walker::compute_format_length): Same.
4809 (try_substitute_return_value): Same.
4811 2017-01-03 Jeff Law <law@redhat.com>
4814 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4815 non-constant residual for zero at runtime and avoid probing in
4816 that case. Reorganize code for trailing problem to mirror handling
4819 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4821 PR tree-optimization/83501
4822 * tree-ssa-strlen.c (get_string_cst): New.
4823 (handle_char_store): Call get_string_cst.
4825 2018-01-03 Martin Liska <mliska@suse.cz>
4827 PR tree-optimization/83593
4828 * tree-ssa-strlen.c: Include tree-cfg.h.
4829 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4830 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4831 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4833 (strlen_dom_walker::before_dom_children): Call
4834 gimple_purge_dead_eh_edges. Dump tranformation with details
4836 (strlen_dom_walker::before_dom_children): Update call by adding
4837 new argument cleanup_eh.
4838 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4840 2018-01-03 Martin Liska <mliska@suse.cz>
4843 * cif-code.def (VARIADIC_THUNK): New enum value.
4844 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4847 2018-01-03 Jan Beulich <jbeulich@suse.com>
4849 * sse.md (mov<mode>_internal): Tighten condition for when to use
4850 vmovdqu<ssescalarsize> for TI and OI modes.
4852 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4854 Update copyright years.
4856 2018-01-03 Martin Liska <mliska@suse.cz>
4859 * ipa-visibility.c (function_and_variable_visibility): Skip
4860 functions with noipa attribure.
4862 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4864 * gcc.c (process_command): Update copyright notice dates.
4865 * gcov-dump.c (print_version): Ditto.
4866 * gcov.c (print_version): Ditto.
4867 * gcov-tool.c (print_version): Ditto.
4868 * gengtype.c (create_file): Ditto.
4869 * doc/cpp.texi: Bump @copying's copyright year.
4870 * doc/cppinternals.texi: Ditto.
4871 * doc/gcc.texi: Ditto.
4872 * doc/gccint.texi: Ditto.
4873 * doc/gcov.texi: Ditto.
4874 * doc/install.texi: Ditto.
4875 * doc/invoke.texi: Ditto.
4877 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4879 * vector-builder.h (vector_builder::m_full_nelts): Change from
4880 unsigned int to poly_uint64.
4881 (vector_builder::full_nelts): Update prototype accordingly.
4882 (vector_builder::new_vector): Likewise.
4883 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4884 (vector_builder::operator ==): Likewise.
4885 (vector_builder::finalize): Likewise.
4886 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4887 Take the number of elements as a poly_uint64 rather than an
4889 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4890 from unsigned int to poly_uint64.
4891 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4892 (vec_perm_indices::new_vector): Likewise.
4893 (vec_perm_indices::length): Likewise.
4894 (vec_perm_indices::nelts_per_input): Likewise.
4895 (vec_perm_indices::input_nelts): Likewise.
4896 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4897 number of elements per input as a poly_uint64 rather than an
4898 unsigned int. Use the original encoding for variable-length
4899 vectors, rather than clamping each individual element.
4900 For the second and subsequent elements in each pattern,
4901 clamp the step and base before clamping their sum.
4902 (vec_perm_indices::series_p): Handle polynomial element counts.
4903 (vec_perm_indices::all_in_range_p): Likewise.
4904 (vec_perm_indices_to_tree): Likewise.
4905 (vec_perm_indices_to_rtx): Likewise.
4906 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4907 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4908 (tree_vector_builder::new_binary_operation): Handle polynomial
4909 element counts. Return false if we need to know the number
4910 of elements at compile time.
4911 * fold-const.c (fold_vec_perm): Punt if the number of elements
4912 isn't known at compile time.
4914 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4916 * vec-perm-indices.h (vec_perm_builder): Change element type
4917 from HOST_WIDE_INT to poly_int64.
4918 (vec_perm_indices::element_type): Update accordingly.
4919 (vec_perm_indices::clamp): Handle polynomial element_types.
4920 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4921 (vec_perm_indices::all_in_range_p): Likewise.
4922 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4924 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4925 polynomial vec_perm_indices element types.
4926 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4927 * fold-const.c (fold_vec_perm): Likewise.
4928 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4929 * tree-vect-generic.c (lower_vec_perm): Likewise.
4930 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4931 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4932 element type to HOST_WIDE_INT.
4934 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4935 Alan Hayward <alan.hayward@arm.com>
4936 David Sherwood <david.sherwood@arm.com>
4938 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4939 rather than an int. Use plus_constant.
4940 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4941 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4943 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4944 Alan Hayward <alan.hayward@arm.com>
4945 David Sherwood <david.sherwood@arm.com>
4947 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4948 a HOST_WIDE_INT to a poly_int64.
4950 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4951 Alan Hayward <alan.hayward@arm.com>
4952 David Sherwood <david.sherwood@arm.com>
4954 * calls.c (load_register_parameters): Cope with polynomial
4955 mode sizes. Require a constant size for BLKmode parameters
4956 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4957 forces a parameter to be padded at the lsb end in order to
4958 fill a complete number of words, require the parameter size
4959 to be ordered wrt UNITS_PER_WORD.
4961 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4962 Alan Hayward <alan.hayward@arm.com>
4963 David Sherwood <david.sherwood@arm.com>
4965 * reload1.c (spill_stack_slot_width): Change element type
4966 from unsigned int to poly_uint64_pod.
4967 (alter_reg): Treat mode sizes as polynomial.
4969 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4970 Alan Hayward <alan.hayward@arm.com>
4971 David Sherwood <david.sherwood@arm.com>
4973 * reload.c (complex_word_subreg_p): New function.
4974 (reload_inner_reg_of_subreg, push_reload): Use it.
4976 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4977 Alan Hayward <alan.hayward@arm.com>
4978 David Sherwood <david.sherwood@arm.com>
4980 * lra-constraints.c (process_alt_operands): Reject matched
4981 operands whose sizes aren't ordered.
4982 (match_reload): Refer to this check here.
4984 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4985 Alan Hayward <alan.hayward@arm.com>
4986 David Sherwood <david.sherwood@arm.com>
4988 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4989 that the mode size is in the set {1, 2, 4, 8, 16}.
4991 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4992 Alan Hayward <alan.hayward@arm.com>
4993 David Sherwood <david.sherwood@arm.com>
4995 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4996 Use plus_constant instead of gen_rtx_PLUS.
4998 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4999 Alan Hayward <alan.hayward@arm.com>
5000 David Sherwood <david.sherwood@arm.com>
5002 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5003 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5004 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5005 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5006 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5007 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5008 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5009 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5010 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5011 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5013 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5014 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5015 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5016 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5017 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5018 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5019 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5020 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5021 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5022 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5024 * expr.c (emit_move_resolve_push): Treat the input and result
5025 of PUSH_ROUNDING as a poly_int64.
5026 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5027 (emit_push_insn): Likewise.
5028 * lra-eliminations.c (mark_not_eliminable): Likewise.
5029 * recog.c (push_operand): Likewise.
5030 * reload1.c (elimination_effects): Likewise.
5031 * rtlanal.c (nonzero_bits1): Likewise.
5032 * calls.c (store_one_arg): Likewise. Require the padding to be
5033 known at compile time.
5035 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5036 Alan Hayward <alan.hayward@arm.com>
5037 David Sherwood <david.sherwood@arm.com>
5039 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5040 Use plus_constant instead of gen_rtx_PLUS.
5042 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5043 Alan Hayward <alan.hayward@arm.com>
5044 David Sherwood <david.sherwood@arm.com>
5046 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5049 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5050 Alan Hayward <alan.hayward@arm.com>
5051 David Sherwood <david.sherwood@arm.com>
5053 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5054 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5055 via stack temporaries. Treat the mode size as polynomial too.
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 * expr.c (expand_expr_real_2): When handling conversions involving
5062 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5063 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5064 as a poly_uint64 too.
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 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5072 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5073 Alan Hayward <alan.hayward@arm.com>
5074 David Sherwood <david.sherwood@arm.com>
5076 * combine.c (can_change_dest_mode): Handle polynomial
5077 REGMODE_NATURAL_SIZE.
5078 * expmed.c (store_bit_field_1): Likewise.
5079 * expr.c (store_constructor): Likewise.
5080 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5081 and polynomial REGMODE_NATURAL_SIZE.
5082 (gen_lowpart_common): Likewise.
5083 * reginfo.c (record_subregs_of_mode): Likewise.
5084 * rtlanal.c (read_modify_subreg_p): Likewise.
5086 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5087 Alan Hayward <alan.hayward@arm.com>
5088 David Sherwood <david.sherwood@arm.com>
5090 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5091 numbers of elements.
5093 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5094 Alan Hayward <alan.hayward@arm.com>
5095 David Sherwood <david.sherwood@arm.com>
5097 * match.pd: Cope with polynomial numbers of vector elements.
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 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5104 in a POINTER_PLUS_EXPR.
5106 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5107 Alan Hayward <alan.hayward@arm.com>
5108 David Sherwood <david.sherwood@arm.com>
5110 * omp-simd-clone.c (simd_clone_subparts): New function.
5111 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5112 (ipa_simd_modify_function_body): Likewise.
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 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5119 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5120 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5121 (expand_vector_condition, vector_element): Likewise.
5122 (subparts_gt): New function.
5123 (get_compute_type): Use subparts_gt.
5124 (count_type_subparts): Delete.
5125 (expand_vector_operations_1): Use subparts_gt instead of
5126 count_type_subparts.
5128 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5129 Alan Hayward <alan.hayward@arm.com>
5130 David Sherwood <david.sherwood@arm.com>
5132 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5133 (vect_compile_time_alias): ...this new function. Do the calculation
5134 on poly_ints rather than trees.
5135 (vect_prune_runtime_alias_test_list): Update call accordingly.
5137 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5138 Alan Hayward <alan.hayward@arm.com>
5139 David Sherwood <david.sherwood@arm.com>
5141 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5143 (vect_schedule_slp_instance): Likewise.
5145 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5146 Alan Hayward <alan.hayward@arm.com>
5147 David Sherwood <david.sherwood@arm.com>
5149 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5150 constant and extern definitions for variable-length vectors.
5151 (vect_get_constant_vectors): Note that the number of units
5152 is known to be constant.
5154 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5155 Alan Hayward <alan.hayward@arm.com>
5156 David Sherwood <david.sherwood@arm.com>
5158 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5159 of units as polynomial. Choose between WIDE and NARROW based
5162 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5163 Alan Hayward <alan.hayward@arm.com>
5164 David Sherwood <david.sherwood@arm.com>
5166 * tree-vect-stmts.c (simd_clone_subparts): New function.
5167 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5169 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5170 Alan Hayward <alan.hayward@arm.com>
5171 David Sherwood <david.sherwood@arm.com>
5173 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5174 vectors as polynomial. Use build_index_vector for
5177 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5178 Alan Hayward <alan.hayward@arm.com>
5179 David Sherwood <david.sherwood@arm.com>
5181 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5182 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5183 for variable-length vectors.
5184 (vectorizable_mask_load_store): Treat the number of units as
5185 polynomial, asserting that it is constant if the condition has
5186 already been enforced.
5187 (vectorizable_store, vectorizable_load): Likewise.
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 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5194 of units as polynomial. Punt if we can't tell at compile time
5195 which vector contains the final result.
5197 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5198 Alan Hayward <alan.hayward@arm.com>
5199 David Sherwood <david.sherwood@arm.com>
5201 * tree-vect-loop.c (vectorizable_induction): Treat the number
5202 of units as polynomial. Punt on SLP inductions. Use an integer
5203 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5204 cast of such a series for variable-length floating-point
5207 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5208 Alan Hayward <alan.hayward@arm.com>
5209 David Sherwood <david.sherwood@arm.com>
5211 * tree.h (build_index_vector): Declare.
5212 * tree.c (build_index_vector): New function.
5213 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5214 of units as polynomial, forcibly converting it to a constant if
5215 vectorizable_reduction has already enforced the condition.
5216 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5217 to create a {1,2,3,...} vector.
5218 (vectorizable_reduction): Treat the number of units as polynomial.
5219 Choose vectype_in based on the largest scalar element size rather
5220 than the smallest number of units. Enforce the restrictions
5223 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5224 Alan Hayward <alan.hayward@arm.com>
5225 David Sherwood <david.sherwood@arm.com>
5227 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5228 number of units as polynomial.
5230 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5231 Alan Hayward <alan.hayward@arm.com>
5232 David Sherwood <david.sherwood@arm.com>
5234 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5235 * target.def (autovectorize_vector_sizes): Return the vector sizes
5236 by pointer, using vector_sizes rather than a bitmask.
5237 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5238 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5239 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5241 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5242 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5243 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5244 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5245 * omp-general.c (omp_max_vf): Likewise.
5246 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5247 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5248 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5249 * tree-vect-slp.c (vect_slp_bb): Likewise.
5250 * doc/tm.texi: Regenerate.
5251 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5253 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5254 the vector size as a poly_uint64 rather than an unsigned int.
5255 (current_vector_size): Change from an unsigned int to a poly_uint64.
5256 (get_vectype_for_scalar_type): Update accordingly.
5257 * tree.h (build_truth_vector_type): Take the size and number of
5258 units as a poly_uint64 rather than an unsigned int.
5259 (build_vector_type): Add a temporary overload that takes
5260 the number of units as a poly_uint64 rather than an unsigned int.
5261 * tree.c (make_vector_type): Likewise.
5262 (build_truth_vector_type): Take the number of units as a poly_uint64
5263 rather than an unsigned int.
5265 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5266 Alan Hayward <alan.hayward@arm.com>
5267 David Sherwood <david.sherwood@arm.com>
5269 * target.def (get_mask_mode): Take the number of units and length
5270 as poly_uint64s rather than unsigned ints.
5271 * targhooks.h (default_get_mask_mode): Update accordingly.
5272 * targhooks.c (default_get_mask_mode): Likewise.
5273 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5274 * doc/tm.texi: Regenerate.
5276 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5277 Alan Hayward <alan.hayward@arm.com>
5278 David Sherwood <david.sherwood@arm.com>
5280 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5281 * omp-general.c (omp_max_vf): Likewise.
5282 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5283 (expand_omp_simd): Handle polynomial safelen.
5284 * omp-low.c (omplow_simd_context): Add a default constructor.
5285 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5286 (lower_rec_simd_input_clauses): Update accordingly.
5287 (lower_rec_input_clauses): Likewise.
5289 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5290 Alan Hayward <alan.hayward@arm.com>
5291 David Sherwood <david.sherwood@arm.com>
5293 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5294 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5295 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5296 (vect_analyze_slp_cost): Likewise.
5297 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5298 (vect_model_load_cost): Likewise.
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-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5305 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5306 from an unsigned int * to a poly_uint64_pod *.
5307 (calculate_unrolling_factor): New function.
5308 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5310 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5311 Alan Hayward <alan.hayward@arm.com>
5312 David Sherwood <david.sherwood@arm.com>
5314 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5315 from an unsigned int to a poly_uint64.
5316 (_loop_vec_info::slp_unrolling_factor): Likewise.
5317 (_loop_vec_info::vectorization_factor): Change from an int
5319 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5320 (vect_get_num_vectors): New function.
5321 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5322 (vect_get_num_copies): Use vect_get_num_vectors.
5323 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5324 to an unsigned int *.
5325 (vect_analyze_data_refs): Change min_vf from an int * to a
5327 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5328 than an unsigned HOST_WIDE_INT.
5329 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5330 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5331 to an unsigned int *.
5332 (vect_analyze_data_ref_dependences): Likewise.
5333 (vect_compute_data_ref_alignment): Handle polynomial vf.
5334 (vect_enhance_data_refs_alignment): Likewise.
5335 (vect_prune_runtime_alias_test_list): Likewise.
5336 (vect_shift_permute_load_chain): Likewise.
5337 (vect_supportable_dr_alignment): Likewise.
5338 (dependence_distance_ge_vf): Take the vectorization factor as a
5339 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5340 (vect_analyze_data_refs): Change min_vf from an int * to a
5342 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5343 vfm1 as a poly_uint64 rather than an int. Make the same change
5344 for the returned bound_scalar.
5345 (vect_gen_vector_loop_niters): Handle polynomial vf.
5346 (vect_do_peeling): Likewise. Update call to
5347 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5348 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5350 * tree-vect-loop.c (vect_determine_vectorization_factor)
5351 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5352 (vect_get_known_peeling_cost): Likewise.
5353 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5354 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5355 (vect_transform_loop): Likewise. Use the lowest possible VF when
5356 updating the upper bounds of the loop.
5357 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5359 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5360 polynomial unroll factors.
5361 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5362 (vect_make_slp_decision): Likewise.
5363 (vect_supported_load_permutation_p): Likewise, and polynomial
5365 (vect_analyze_slp_cost): Handle polynomial vf.
5366 (vect_slp_analyze_node_operations): Likewise.
5367 (vect_slp_analyze_bb_1): Likewise.
5368 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5369 than an unsigned HOST_WIDE_INT.
5370 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5371 (vectorizable_load): Handle polynomial vf.
5372 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5374 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5376 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5377 Alan Hayward <alan.hayward@arm.com>
5378 David Sherwood <david.sherwood@arm.com>
5380 * match.pd: Handle bit operations involving three constants
5381 and try to fold one pair.
5383 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5385 * tree-vect-loop-manip.c: Include gimple-fold.h.
5386 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5387 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5388 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5389 Add a path that uses a step of VF instead of 1, but disable it
5391 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5392 and niters_no_overflow parameters. Update calls to
5393 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5394 Create a new SSA name if the latter choses to use a ste other
5395 than zero, and return it via niters_vector_mult_vf_var.
5396 * tree-vect-loop.c (vect_transform_loop): Update calls to
5397 vect_do_peeling, vect_gen_vector_loop_niters and
5398 slpeel_make_loop_iterate_ntimes.
5399 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5400 (vect_gen_vector_loop_niters): Update declarations after above changes.
5402 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5404 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5405 128-bit round to integer instructions.
5406 (ceil<mode>2): Likewise.
5407 (btrunc<mode>2): Likewise.
5408 (round<mode>2): Likewise.
5410 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5412 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5413 unaligned VSX load/store on P8/P9.
5414 (expand_block_clear): Allow the use of unaligned VSX
5415 load/store on P8/P9.
5417 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5419 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5421 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5422 swap associated with both a load and a store.
5424 2018-01-02 Andrew Waterman <andrew@sifive.com>
5426 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5427 * config/riscv/riscv.md (clear_cache): Use it.
5429 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5431 * web.c: Remove out-of-date comment.
5433 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5435 * expr.c (fixup_args_size_notes): Check that any existing
5436 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5437 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5438 (emit_single_push_insn): ...here.
5440 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5442 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5443 (const_vector_encoded_nelts): New function.
5444 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5445 (const_vector_int_elt, const_vector_elt): Declare.
5446 * emit-rtl.c (const_vector_int_elt_1): New function.
5447 (const_vector_elt): Likewise.
5448 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5449 of CONST_VECTOR_ELT.
5451 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5453 * expr.c: Include rtx-vector-builder.h.
5454 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5455 directly on the tree encoding.
5456 (const_vector_from_tree): Likewise.
5457 * optabs.c: Include rtx-vector-builder.h.
5458 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5459 sequence of "u" values.
5460 * vec-perm-indices.c: Include rtx-vector-builder.h.
5461 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5462 directly on the vec_perm_indices encoding.
5464 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5466 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5467 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5468 * rtx-vector-builder.h: New file.
5469 * rtx-vector-builder.c: Likewise.
5470 * rtl.h (rtx_def::u2): Add a const_vector field.
5471 (CONST_VECTOR_NPATTERNS): New macro.
5472 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5473 (CONST_VECTOR_DUPLICATE_P): Likewise.
5474 (CONST_VECTOR_STEPPED_P): Likewise.
5475 (CONST_VECTOR_ENCODED_ELT): Likewise.
5476 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5477 (unwrap_const_vec_duplicate): Likewise.
5478 (const_vec_series_p): Check for a non-duplicated vector encoding.
5479 Say that the function only returns true for integer vectors.
5480 * emit-rtl.c: Include rtx-vector-builder.h.
5481 (gen_const_vec_duplicate_1): Delete.
5482 (gen_const_vector): Call gen_const_vec_duplicate instead of
5483 gen_const_vec_duplicate_1.
5484 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5485 (gen_const_vec_duplicate): Use rtx_vector_builder.
5486 (gen_const_vec_series): Likewise.
5487 (gen_rtx_CONST_VECTOR): Likewise.
5488 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5489 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5490 Build a new vector rather than modifying a CONST_VECTOR in-place.
5491 (handle_special_swappables): Update call accordingly.
5492 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5493 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5494 Build a new vector rather than modifying a CONST_VECTOR in-place.
5495 (handle_special_swappables): Update call accordingly.
5497 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5499 * simplify-rtx.c (simplify_const_binary_operation): Use
5500 CONST_VECTOR_ELT instead of XVECEXP.
5502 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5504 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5505 the selector elements to be different from the data elements
5506 if the selector is a VECTOR_CST.
5507 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5508 ssizetype for the selector.
5510 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5512 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5513 before testing each element individually.
5514 * tree-vect-generic.c (lower_vec_perm): Likewise.
5516 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5518 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5519 * selftest-run-tests.c (selftest::run_tests): Call it.
5520 * vector-builder.h (vector_builder::operator ==): New function.
5521 (vector_builder::operator !=): Likewise.
5522 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5523 (vec_perm_indices::all_from_input_p): New function.
5524 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5525 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5526 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5527 instead of reading the VECTOR_CST directly. Detect whether both
5528 vector inputs are the same before constructing the vec_perm_indices,
5529 and update the number of inputs argument accordingly. Use the
5530 utility functions added above. Only construct sel2 if we need to.
5532 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5534 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5535 the broadcast of the low byte.
5536 (expand_mult_highpart): Use an explicit encoding for the permutes.
5537 * optabs-query.c (can_mult_highpart_p): Likewise.
5538 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5539 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5540 (vectorizable_bswap): Likewise.
5541 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5542 explicit encoding for the power-of-2 permutes.
5543 (vect_permute_store_chain): Likewise.
5544 (vect_grouped_load_supported): Likewise.
5545 (vect_permute_load_chain): Likewise.
5547 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5549 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5550 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5551 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5552 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5553 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5554 (vect_gen_perm_mask_any): Likewise.
5556 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5558 * int-vector-builder.h: New file.
5559 * vec-perm-indices.h: Include int-vector-builder.h.
5560 (vec_perm_indices): Redefine as an int_vector_builder.
5561 (auto_vec_perm_indices): Delete.
5562 (vec_perm_builder): Redefine as a stand-alone class.
5563 (vec_perm_indices::vec_perm_indices): New function.
5564 (vec_perm_indices::clamp): Likewise.
5565 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5566 (vec_perm_indices::new_vector): New function.
5567 (vec_perm_indices::new_expanded_vector): Update for new
5568 vec_perm_indices class.
5569 (vec_perm_indices::rotate_inputs): New function.
5570 (vec_perm_indices::all_in_range_p): Operate directly on the
5571 encoded form, without computing elided elements.
5572 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5573 encoding. Update for new vec_perm_indices class.
5574 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5575 the given vec_perm_builder.
5576 (expand_vec_perm_var): Update vec_perm_builder constructor.
5577 (expand_mult_highpart): Use vec_perm_builder instead of
5578 auto_vec_perm_indices.
5579 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5580 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5581 or double series encoding as appropriate.
5582 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5583 vec_perm_indices instead of auto_vec_perm_indices.
5584 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5585 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5586 (vect_permute_store_chain): Likewise.
5587 (vect_grouped_load_supported): Likewise.
5588 (vect_permute_load_chain): Likewise.
5589 (vect_shift_permute_load_chain): Likewise.
5590 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5591 (vect_transform_slp_perm_load): Likewise.
5592 (vect_schedule_slp_instance): Likewise.
5593 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5594 (vectorizable_mask_load_store): Likewise.
5595 (vectorizable_bswap): Likewise.
5596 (vectorizable_store): Likewise.
5597 (vectorizable_load): Likewise.
5598 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5599 vec_perm_indices instead of auto_vec_perm_indices. Use
5600 tree_to_vec_perm_builder to read the vector from a tree.
5601 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5602 vec_perm_builder instead of a vec_perm_indices.
5603 (have_whole_vector_shift): Use vec_perm_builder and
5604 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5605 truncation to calc_vec_perm_mask_for_shift.
5606 (vect_create_epilog_for_reduction): Likewise.
5607 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5608 from auto_vec_perm_indices to vec_perm_indices.
5609 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5610 instead of changing individual elements.
5611 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5612 the vector in d.perm.
5613 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5614 from auto_vec_perm_indices to vec_perm_indices.
5615 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5616 instead of changing individual elements.
5617 (arm_vectorize_vec_perm_const): Use new_vector to install
5618 the vector in d.perm.
5619 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5620 Update vec_perm_builder constructor.
5621 (rs6000_expand_interleave): Likewise.
5622 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5623 (rs6000_expand_interleave): Likewise.
5625 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5627 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5628 to qimode could truncate the indices.
5629 * optabs.c (expand_vec_perm_var): Likewise.
5631 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5633 * Makefile.in (OBJS): Add vec-perm-indices.o.
5634 * vec-perm-indices.h: New file.
5635 * vec-perm-indices.c: Likewise.
5636 * target.h (vec_perm_indices): Replace with a forward class
5638 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5639 * optabs.h: Include vec-perm-indices.h.
5640 (expand_vec_perm): Delete.
5641 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5642 (expand_vec_perm_const): Declare.
5643 * target.def (vec_perm_const_ok): Replace with...
5644 (vec_perm_const): ...this new hook.
5645 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5646 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5647 * doc/tm.texi: Regenerate.
5648 * optabs.def (vec_perm_const): Delete.
5649 * doc/md.texi (vec_perm_const): Likewise.
5650 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5651 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5652 expand_vec_perm for constant permutation vectors. Assert that
5653 the mode of variable permutation vectors is the integer equivalent
5654 of the mode that is being permuted.
5655 * optabs-query.h (selector_fits_mode_p): Declare.
5656 * optabs-query.c: Include vec-perm-indices.h.
5657 (selector_fits_mode_p): New function.
5658 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5659 is defined, instead of checking whether the vec_perm_const_optab
5660 exists. Use targetm.vectorize.vec_perm_const instead of
5661 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5662 fit in the vector mode before using a variable permute.
5663 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5664 vec_perm_indices instead of an rtx.
5665 (expand_vec_perm): Replace with...
5666 (expand_vec_perm_const): ...this new function. Take the selector
5667 as a vec_perm_indices rather than an rtx. Also take the mode of
5668 the selector. Update call to shift_amt_for_vec_perm_mask.
5669 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5670 Use vec_perm_indices::new_expanded_vector to expand the original
5671 selector into bytes. Check whether the indices fit in the vector
5672 mode before using a variable permute.
5673 (expand_vec_perm_var): Make global.
5674 (expand_mult_highpart): Use expand_vec_perm_const.
5675 * fold-const.c: Includes vec-perm-indices.h.
5676 * tree-ssa-forwprop.c: Likewise.
5677 * tree-vect-data-refs.c: Likewise.
5678 * tree-vect-generic.c: Likewise.
5679 * tree-vect-loop.c: Likewise.
5680 * tree-vect-slp.c: Likewise.
5681 * tree-vect-stmts.c: Likewise.
5682 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5684 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5685 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5686 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5687 (aarch64_vectorize_vec_perm_const): ...this new function.
5688 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5689 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5690 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5691 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5692 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5693 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5694 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5696 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5697 check for NEON modes.
5698 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5699 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5700 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5701 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5703 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5704 the old VEC_PERM_CONST conditions.
5705 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5706 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5707 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5708 (ia64_vectorize_vec_perm_const_ok): Merge into...
5709 (ia64_vectorize_vec_perm_const): ...this new function.
5710 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5711 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5712 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5713 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5714 * config/mips/mips.c (mips_expand_vec_perm_const)
5715 (mips_vectorize_vec_perm_const_ok): Merge into...
5716 (mips_vectorize_vec_perm_const): ...this new function.
5717 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5718 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5719 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5720 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5721 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5722 (rs6000_expand_vec_perm_const): Delete.
5723 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5725 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5726 (altivec_expand_vec_perm_const_le): Take each operand individually.
5727 Operate on constant selectors rather than rtxes.
5728 (altivec_expand_vec_perm_const): Likewise. Update call to
5729 altivec_expand_vec_perm_const_le.
5730 (rs6000_expand_vec_perm_const): Delete.
5731 (rs6000_vectorize_vec_perm_const_ok): Delete.
5732 (rs6000_vectorize_vec_perm_const): New function.
5733 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5734 an element count and rtx array.
5735 (rs6000_expand_extract_even): Update call accordingly.
5736 (rs6000_expand_interleave): Likewise.
5737 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5738 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5739 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5740 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5741 (rs6000_expand_vec_perm_const): Delete.
5742 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5743 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5744 (altivec_expand_vec_perm_const_le): Take each operand individually.
5745 Operate on constant selectors rather than rtxes.
5746 (altivec_expand_vec_perm_const): Likewise. Update call to
5747 altivec_expand_vec_perm_const_le.
5748 (rs6000_expand_vec_perm_const): Delete.
5749 (rs6000_vectorize_vec_perm_const_ok): Delete.
5750 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5751 reference to the SPE evmerge intructions.
5752 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5753 an element count and rtx array.
5754 (rs6000_expand_extract_even): Update call accordingly.
5755 (rs6000_expand_interleave): Likewise.
5756 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5757 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5759 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5761 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5763 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5764 vector mode and that that mode matches the mode of the data
5766 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5767 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5768 directly using expand_vec_perm_1 when forcing selectors into
5770 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5772 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5774 * optabs-query.h (can_vec_perm_p): Delete.
5775 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5776 * optabs-query.c (can_vec_perm_p): Split into...
5777 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5778 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5779 particular selector is valid.
5780 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5781 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5782 (vect_grouped_load_supported): Likewise.
5783 (vect_shift_permute_load_chain): Likewise.
5784 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5785 (vect_transform_slp_perm_load): Likewise.
5786 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5787 (vectorizable_bswap): Likewise.
5788 (vect_gen_perm_mask_checked): Likewise.
5789 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5790 implementations of variable permutation vectors into account
5791 when deciding which selector to use.
5792 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5793 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5794 with a false third argument.
5795 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5796 to test whether the constant selector is valid and can_vec_perm_var_p
5797 to test whether a variable selector is valid.
5799 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5801 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5802 * optabs-query.c (can_vec_perm_p): Likewise.
5803 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5804 instead of vec_perm_indices.
5805 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5806 (vect_gen_perm_mask_checked): Likewise,
5807 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5808 (vect_gen_perm_mask_checked): Likewise,
5810 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5812 * optabs-query.h (qimode_for_vec_perm): Declare.
5813 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5814 (qimode_for_vec_perm): ...this new function.
5815 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5817 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5819 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5820 does not have a conditional at the top.
5822 2018-01-02 Richard Biener <rguenther@suse.de>
5824 * ipa-inline.c (big_speedup_p): Fix expression.
5826 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5829 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5832 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5836 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5837 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5838 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5839 cond_taken_branch_cost 3->4.
5841 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5843 PR tree-optimization/83581
5844 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5845 TODO_cleanup_cfg if any changes have been made.
5848 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5849 convert_modes if target mode has the right side, but different mode
5853 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5854 last argument when extracting from CONCAT. If either from_real or
5855 from_imag is NULL, use expansion through memory. If result is not
5856 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5857 the parts directly to inner mode, if even that fails, use expansion
5861 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5862 check for bswap in mode rather than HImode and use that in expand_unop
5865 Copyright (C) 2018 Free Software Foundation, Inc.
5867 Copying and distribution of this file, with or without modification,
5868 are permitted in any medium without royalty provided the copyright
5869 notice and this notice are preserved.