1 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
3 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
4 the TImode handling for big-endian targets.
6 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
8 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
9 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
11 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
12 Remove BSWAP handing for big-endian targets and use the form of
13 LD1RQ appropariate for the mode.
15 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
17 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
18 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
21 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
24 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
25 check for operands that need to go through aarch64_sve_reload_be.
27 2018-02-01 Jakub Jelinek <jakub@redhat.com>
29 PR tree-optimization/81661
30 PR tree-optimization/84117
31 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
32 * tree-eh.c: Include gimplify.h.
33 (find_trapping_overflow, replace_trapping_overflow,
34 rewrite_to_non_trapping_overflow): New functions.
35 * tree-vect-loop.c: Include tree-eh.h.
36 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
37 * tree-data-ref.c: Include tree-eh.h.
38 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
40 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
42 PR rtl-optimization/84123
43 * combine.c (change_zero_ext): Check if hard register satisfies
44 can_change_dest_mode before calling gen_lowpart_SUBREG.
46 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
49 * ira.c (ira_init_register_move_cost): Remove assert.
51 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
53 PR rtl-optimization/84071
54 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
55 * doc/tm.texi: Regenerate.
57 2018-01-31 Richard Biener <rguenther@suse.de>
59 PR tree-optimization/84132
60 * tree-data-ref.c (analyze_miv_subscript): Properly
61 check whether evolution_function_is_affine_multivariate_p
62 before calling gcd_of_steps_may_divide_p.
64 2018-01-31 Julia Koval <julia.koval@intel.com>
67 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
68 * config/i386/i386.md (rdpid_rex64) New.
69 (rdpid): Make 32bit only.
71 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
74 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
75 an IDENTIFIER_NODE for FUNCTION_TYPE's.
77 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
80 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
82 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
84 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
86 PR rtl-optimization/84071
87 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
88 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
90 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
92 * config/arc/arc.c (arc_handle_aux_attribute): New function.
93 (arc_attribute_table): Add 'aux' attribute.
94 (arc_in_small_data_p): Consider aux like variables.
95 (arc_is_aux_reg_p): New function.
96 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
97 (arc_get_aux_arg): New function.
98 (prepare_move_operands): Handle aux-register access.
99 (arc_handle_aux_attribute): New function.
100 * doc/extend.texi (ARC Variable attributes): Add subsection.
102 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
104 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
105 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
106 (arc_attribute_table): Add 'uncached' attribute.
107 (arc_print_operand): Print '.di' flag for uncached memory
109 (arc_in_small_data_p): Do not consider for small data the uncached
111 (arc_is_uncached_mem_p): New function.
112 * config/arc/predicates.md (compact_store_memory_operand): Check
113 for uncached memory accesses.
114 (nonvol_nonimm_operand): Likewise.
115 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
117 2018-01-31 Jakub Jelinek <jakub@redhat.com>
120 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
121 falign-loops=): Add Optimization flag.
123 2017-01-30 Jeff Law <law@redhat.com>
126 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
127 INT_REGISTERS_SAVED. Check it prior to calling
128 get_scratch_register_on_entry.
129 (ix86_adjust_stack_and_probe): Similarly.
130 (ix86_emit_probe_stack_range): Similarly.
131 (ix86_expand_prologue): Corresponding changes.
133 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
136 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
137 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
139 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
142 * lra-constraints.c (curr_insn_transform): Process AND in the
145 2018-01-30 Jakub Jelinek <jakub@redhat.com>
147 PR rtl-optimization/83986
148 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
149 dependence against last_pending_memory_flush in addition to
152 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
154 PR tree-optimization/81611
155 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
158 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
160 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
163 2018-01-30 Richard Biener <rguenther@suse.de>
164 Jakub Jelinek <jakub@redhat.com>
166 PR tree-optimization/84111
167 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
168 inner loops added during recursion, as they don't have up-to-date
171 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
174 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
175 (can_inline_edge_by_limits_p): ... here.
176 (can_early_inline_edge_p, check_callers,
177 update_caller_keys, update_callee_keys, recursive_inlining,
178 add_new_edges_to_heap, speculation_useful_p,
179 inline_small_functions,
180 inline_small_functions, flatten_function,
181 inline_to_all_callers_1): Update.
183 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
185 * profile-count.c (profile_count::combine_with_ipa_count): Handle
188 2018-01-30 Richard Biener <rguenther@suse.de>
190 PR tree-optimization/83008
191 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
192 invariant and constant vector uses in stmts when they need
195 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
198 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
199 * configure: Regenerate.
201 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
203 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
205 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
206 Use gen_rtx_REG rather than gen_lowpart.
208 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
210 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
211 rather than 0 when creating partial subregs.
213 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
215 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
218 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
221 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
222 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
223 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
224 flags. This restores the settings used before the 2017-07-24.
225 Turning off pre increment/decrement/modify allows IVOPTS to
226 optimize DF/SF loops where the index is an int.
228 2018-01-29 Richard Biener <rguenther@suse.de>
229 Kelvin Nilsen <kelvin@gcc.gnu.org>
232 * tree-vect-stmts.c (vectorizable_call): Don't call
233 targetm.vectorize_builtin_md_vectorized_function if callee is
236 2018-01-22 Carl Love <cel@us.ibm.com>
238 * doc/extend.tex: Fix typo in second arg in
239 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
241 2018-01-29 Richard Biener <rguenther@suse.de>
243 PR tree-optimization/84086
244 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
245 (flush_ssaname_freelist): When SSA names were released reset
248 2018-01-29 Richard Biener <rguenther@suse.de>
250 PR tree-optimization/84057
251 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
252 removed paths when removing edges.
254 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
256 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
257 -mfunction-return=@var{choice}.
259 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
262 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
264 (layout::print_source_line): Likewise.
265 (test_get_line_width_without_trailing_whitespace): Add test cases.
267 2018-01-27 Jakub Jelinek <jakub@redhat.com>
270 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
273 2018-01-26 Jim Wilson <jimw@sifive.com>
275 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
277 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
280 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
282 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
283 and CMP + SUB-immediate -> SUBS.
285 2018-01-26 Martin Sebor <msebor@redhat.com>
287 PR tree-optimization/83896
288 * tree-ssa-strlen.c (get_string_len): Rename...
289 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
290 Avoid assuming length is constant.
291 (handle_char_store): Use HOST_WIDE_INT for string length.
293 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
296 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
297 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
299 2018-01-26 Richard Biener <rguenther@suse.de>
301 PR rtl-optimization/84003
302 * dse.c (record_store): Only record redundant stores when
303 the earlier store aliases at least all accesses the later one does.
305 2018-01-26 Jakub Jelinek <jakub@redhat.com>
307 PR rtl-optimization/83985
308 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
309 REG_CFA_RESTORE insns.
310 (delete_unmarked_insns): Don't ignore separate shrink wrapping
311 REG_CFA_RESTORE insns here.
314 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
315 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
317 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
319 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
320 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
321 (arc_init): Likewise.
322 (arc_override_options): Likewise.
323 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
325 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
327 * config/arc/arc.h (TARGET_DBNZ): Define.
328 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
329 properly set the tune attribute.
330 (dbnz): Use TARGET_DBNZ guard.
331 * config/arc/arc.opt (mtune): Add core3 option.
333 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
335 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
336 recognize new pic like addresses.
337 (arc_delegitimize_address): Clean up.
339 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
341 * config/arc/arc-arches.def: Option mrf16 valid for all
343 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
344 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
345 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
346 * config/arc/arc-tables.opt: Regenerate.
347 * config/arc/arc.c (arc_conditional_register_usage): Handle
348 reduced register file case.
349 (arc_file_start): Set must have build attributes.
350 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
352 * config/arc/arc.opt (mrf16): Add new option.
353 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
354 * config/arc/genmultilib.awk: Handle new mrf16 option.
355 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
356 * config/arc/t-multilib: Regenerate.
357 * doc/invoke.texi (ARC Options): Document mrf16 option.
359 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
361 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
362 * config/arc/arc.c (arc_handle_secure_attribute): New function.
363 (arc_attribute_table): Add 'secure_call' attribute.
364 (arc_print_operand): Print secure call operand.
365 (arc_function_ok_for_sibcall): Don't optimize tail calls when
367 (arc_is_secure_call_p): New function. * config/arc/arc.md
368 (call_i): Add support for sjli instruction.
369 (call_value_i): Likewise.
370 * config/arc/constraints.md (Csc): New constraint.
372 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
373 John Eric Martin <John.Martin@emmicro-us.com>
375 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
376 * config/arc/arc.c (_arc_jli_section): New struct.
377 (arc_jli_section): New type.
378 (rc_jli_sections): New static variable.
379 (arc_handle_jli_attribute): New function.
380 (arc_attribute_table): Add jli_always and jli_fixed attribute.
381 (arc_file_end): New function.
382 (TARGET_ASM_FILE_END): Define.
383 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
384 (arc_add_jli_section): New function.
385 (jli_call_scan): Likewise.
386 (arc_reorg): Call jli_call_scan.
387 (arc_output_addsi): Remove 'S' from printing asm operand.
388 (arc_is_jli_call_p): New function.
389 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
391 (movhi_insn): Likewise.
392 (movsi_insn): Likewise.
393 (movsi_set_cc_insn): Likewise.
394 (loadqi_update): Likewise.
395 (load_zeroextendqisi_update): Likewise.
396 (load_signextendqisi_update): Likewise.
397 (loadhi_update): Likewise.
398 (load_zeroextendhisi_update): Likewise.
399 (load_signextendhisi_update): Likewise.
400 (loadsi_update): Likewise.
401 (loadsf_update): Likewise.
402 (movsicc_insn): Likewise.
403 (bset_insn): Likewise.
404 (bxor_insn): Likewise.
405 (bclr_insn): Likewise.
406 (bmsk_insn): Likewise.
407 (bicsi3_insn): Likewise.
408 (cmpsi_cc_c_insn): Likewise.
409 (movsi_ne): Likewise.
410 (movsi_cond_exec): Likewise.
411 (clrsbsi2): Likewise.
422 (call_i): Remove 'S' asm letter, add jli instruction.
423 (call_value_i): Likewise.
424 * config/arc/arc.op (mjli-always): New option.
425 * config/arc/constraints.md (Cji): New constraint.
426 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
428 (subsf3_fpx): Likewise.
429 (mulsf3_fpx): Likewise.
430 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
432 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
434 * doc/invoke.texi (ARC): Document mjli-always option.
436 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
438 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
439 and use incw and decw where possible
440 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
442 2018-01-26 Richard Biener <rguenther@suse.de>
444 PR tree-optimization/81082
445 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
446 association if it requires casting to unsigned.
447 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
448 from fold_plusminus_mult_expr to catch important cases late when
449 range info is available.
451 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
453 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
454 * configure.ac (hidden_linkonce): New test.
455 * configure: Regenerate.
456 * config.in: Regenerate.
458 2018-01-26 Julia Koval <julia.koval@intel.com>
460 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
461 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
462 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
463 _mm_mask_bitshuffle_epi64_mask): Fix type.
464 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
465 USI_FTYPE_V4DI_V4DI_USI): Remove.
466 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
467 __builtin_ia32_vpshufbitqmb256_mask,
468 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
469 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
470 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
472 2018-01-26 Alan Modra <amodra@gmail.com>
475 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
476 UNSPEC_VBPERMQ. Sort other unspecs.
478 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
480 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
482 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
485 * predict.c (drop_profile): Do not push/pop cfun; update also
487 (handle_missing_profiles): Fix logic looking for zero profiles.
489 2018-01-25 Jakub Jelinek <jakub@redhat.com>
492 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
493 on functions with #pragma omp declare simd or functions with simd
495 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
496 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
497 Remove trailing \n from warning_at calls.
499 2018-01-25 Tom de Vries <tom@codesourcery.com>
502 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
503 for neutered workers.
505 2018-01-24 Joseph Myers <joseph@codesourcery.com>
508 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
509 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
511 2017-01-24 Jeff Law <law@redhat.com>
514 * i386.c (get_probe_interval): Move to earlier point.
515 (ix86_compute_frame_layout): If -fstack-clash-protection and
516 the frame is larger than the probe interval, then use pushes
517 to save registers rather than reg->mem moves.
518 (ix86_expand_prologue): Remove conditional for int_registers_saved
521 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
524 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
525 min/max for never referenced object.
527 2018-01-24 Jakub Jelinek <jakub@redhat.com>
530 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
532 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
533 attributes from DECL_ATTRIBUTES (decl) without affecting
534 DECL_ATTRIBUTES (current_function_decl).
535 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
536 functions with non-NULL DECL_ABSTRACT_ORIGIN.
538 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
540 PR tree-optimization/83979
541 * fold-const.c (fold_comparison): Use constant_boolean_node
542 instead of boolean_{true,false}_node.
544 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
546 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
549 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
551 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
552 Simplify the clause that sets the length attribute.
553 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
554 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
555 clause that sets the length attribute.
556 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
558 2018-01-24 Tom de Vries <tom@codesourcery.com>
561 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
562 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
563 Add strict parameter.
564 (prevent_branch_around_nothing): Insert dummy insn between branch to
565 label and label with no ptx insn inbetween.
566 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
568 2018-01-24 Tom de Vries <tom@codesourcery.com>
571 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
572 for neutered threads in warp.
573 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
575 2018-01-24 Richard Biener <rguenther@suse.de>
577 PR tree-optimization/83176
578 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
581 2018-01-24 Richard Biener <rguenther@suse.de>
583 PR tree-optimization/82819
584 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
585 code generating pluses that are no-ops in the target precision.
587 2018-01-24 Richard Biener <rguenther@suse.de>
590 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
592 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
594 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
595 to merge probabilities.
596 * predict.c (probably_never_executed): Also mark as cold functions
597 with global 0 profile and guessed local profile.
598 * profile-count.c (profile_probability::combine_with_count): New
600 * profile-count.h (profile_probability::operator*,
601 profile_probability::operator*=, profile_probability::operator/,
602 profile_probability::operator/=): Reduce precision to adjusted
603 and set value to guessed on contradictory divisions.
604 (profile_probability::combine_with_freq): Remove.
605 (profile_probability::combine_wiht_count): Declare.
606 (profile_count::force_nonzero):: Set to adjusted.
607 (profile_count::probability_in):: Set quality to adjusted.
608 * tree-ssa-tail-merge.c (replace_block_by): Use
611 2018-01-23 Andrew Waterman <andrew@sifive.com>
612 Jim Wilson <jimw@sifive.com>
614 * config/riscv/riscv.c (riscv_stack_boundary): New.
615 (riscv_option_override): Set riscv_stack_boundary. Handle
616 riscv_preferred_stack_boundary_arg.
617 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
618 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
619 (STACK_BOUNDARY): Set to riscv_stack_boundary.
620 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
621 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
622 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
624 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
627 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
628 of struct ix86_frame.
629 (ix86_expand_epilogue): Likewise. Add a local variable for
630 the reg_save_offset field in struct ix86_frame.
632 2018-01-23 Bin Cheng <bin.cheng@arm.com>
634 PR tree-optimization/82604
635 * tree-loop-distribution.c (enum partition_kind): New enum item
636 PKIND_PARTIAL_MEMSET.
637 (partition_builtin_p): Support above new enum item.
638 (generate_code_for_partition): Ditto.
639 (compute_access_range): Differentiate cases that equality can be
640 proven at all loops, the innermost loops or no loops.
641 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
642 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
643 (finalize_partitions, distribute_loop): Don't fuse partition of
644 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
645 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
648 2018-01-23 Martin Liska <mliska@suse.cz>
650 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
651 order to ignore the predictor.
652 (PRED_POLYMORPHIC_CALL): Likewise.
653 (PRED_RECURSIVE_CALL): Likewise.
655 2018-01-23 Martin Liska <mliska@suse.cz>
657 * tree-profile.c (tree_profiling): Print function header to
658 aware reader which function we are working on.
659 * value-prof.c (gimple_find_values_to_profile): Do not print
660 not interesting value histograms.
662 2018-01-23 Martin Liska <mliska@suse.cz>
664 * profile-count.h (enum profile_quality): Add
665 profile_uninitialized as the first value. Do not number values
666 as they are zero based.
667 (profile_count::verify): Update sanity check.
668 (profile_probability::verify): Likewise.
670 2018-01-23 Nathan Sidwell <nathan@acm.org>
672 * doc/invoke.texi (ffor-scope): Deprecate.
674 2018-01-23 David Malcolm <dmalcolm@redhat.com>
676 PR tree-optimization/83510
677 * domwalk.c (set_all_edges_as_executable): New function.
678 (dom_walker::dom_walker): Convert bool param
679 "skip_unreachable_blocks" to enum reachability. Move setup of
680 edge flags to set_all_edges_as_executable and only do it when
681 reachability is REACHABLE_BLOCKS.
682 * domwalk.h (enum dom_walker::reachability): New enum.
683 (dom_walker::dom_walker): Convert bool param
684 "skip_unreachable_blocks" to enum reachability.
685 (set_all_edges_as_executable): New decl.
686 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
687 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
689 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
690 but converting true to REACHABLE_BLOCKS.
691 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
693 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
694 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
695 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
697 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
698 if check_all_array_refs will be called.
700 2018-01-23 David Malcolm <dmalcolm@redhat.com>
702 * tree.c (selftest::test_location_wrappers): Add more test
705 2018-01-23 David Malcolm <dmalcolm@redhat.com>
707 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
708 (selftest::test_bit_in_range): Likewise.
710 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
713 * doc/sourcebuild.texi (vect_float): Say that the selector
714 only describes the situation when -funsafe-math-optimizations is on.
715 (vect_float_strict): Document.
717 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
719 PR tree-optimization/83965
720 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
721 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
722 instead of checking only for a reduction.
723 (vect_recog_widen_sum_pattern): Likewise.
725 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
727 * predict.c (probably_never_executed): Only use precise profile info.
728 (compute_function_frequency): Skip after inlining hack since we now
729 have quality checking.
731 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
733 * profile-count.h (profile_probability::very_unlikely,
734 profile_probability::unlikely, profile_probability::even): Set
735 precision to guessed.
737 2018-01-23 Richard Biener <rguenther@suse.de>
739 PR tree-optimization/83963
740 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
741 Properly terminate dominator walk when crossing the exit edge not
742 when visiting its source block.
744 2018-01-23 Jakub Jelinek <jakub@redhat.com>
747 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
748 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
750 2018-01-22 Jakub Jelinek <jakub@redhat.com>
752 PR tree-optimization/83957
753 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
754 semicolon after for body surrounded by braces.
756 PR tree-optimization/83081
757 * profile-count.h (profile_probability::split): New method.
758 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
759 Use profile_probability::split.
760 (do_compare_rtx_and_jump): Fix adjustment of probabilities
761 when splitting a single conditional jump into 2.
763 2018-01-22 David Malcolm <dmalcolm@redhat.com>
765 PR tree-optimization/69452
766 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
769 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
771 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
772 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
773 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
775 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
777 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
778 * config/rl78/rl78.md: New define_expand "movdi"
779 * config/rl78/rl78.c: New function definition rl78_split_movdi
781 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
784 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
786 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
787 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
788 128-bit to produce an UNSPEC move to get the double word with the
789 signbit and then a shift directly to do signbit.
790 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
791 implementation with a new version that just does either a direct
792 move or a regular move. Move memory interface to separate insns.
793 Move insns so they are next to the expander.
794 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
795 with signbit move. Split big and little endian case.
796 (signbit<mode>2_dm_mem_le): Likewise.
797 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
798 (signbit<mode>2_dm2): Likewise.
800 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
802 * config/rl78/rl78.md: New define_expand "anddi3".
804 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
806 * config/rl78/rl78.md: New define_expand "umindi3".
808 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
810 * config/rl78/rl78.md: New define_expand "smindi3".
812 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
814 * config/rl78/rl78.md: New define_expand "smaxdi3".
816 2018-01-22 Carl Love <cel@us.ibm.com>
818 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
819 LVX_V1TI): Add macro expansion.
820 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
821 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
822 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
823 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
824 Change check to determine if the instruction is a byte reversing
825 entry. Fix typo in comment.
826 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
827 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
828 Add def_builtin calls for new builtins.
829 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
830 Add define_insn expansion.
832 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
834 * config/rl78/rl78.md: New define_expand "umaxdi3".
836 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
838 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
839 for non-QImode registers
841 2018-01-22 Richard Biener <rguenther@suse.de>
843 PR tree-optimization/83963
844 * graphite-scop-detection.c (scop_detection::get_sese): Delay
845 including the loop exit block.
846 (scop_detection::merge_sese): Likewise.
847 (scop_detection::add_scop): Do it here instead.
849 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
851 * doc/sourcebuild.texi (arm_softfloat): Document.
853 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
856 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
857 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
858 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
860 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
861 David Edelsohn <dje.gcc@gmail.com>
864 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
865 Change "crset eq" to "crset 2".
866 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
867 (*call_indirect_aix<mode>_nospec): Likewise.
868 (*call_value_indirect_aix<mode>_nospec): Likewise.
869 (*call_indirect_elfv2<mode>_nospec): Likewise.
870 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
871 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
872 change assembly output from . to $.
873 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
874 (indirect_jump<mode>_nospec): Change assembly output from . to $.
875 (*tablejump<mode>_internal1_nospec): Likewise.
877 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
880 * config/sh/sh_optimize_sett_clrt.cc:
881 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
883 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
885 PR tree-optimization/83940
886 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
887 offset_dt to vect_constant_def rather than vect_unknown_def_type.
888 (vect_check_load_store_mask): Add a mask_dt_out parameter and
889 use it to pass back the definition type.
890 (vect_check_store_rhs): Likewise rhs_dt_out.
891 (vect_build_gather_load_calls): Add a mask_dt argument and use
892 it instead of a call to vect_is_simple_use.
893 (vectorizable_store): Update calls to vect_check_load_store_mask
894 and vect_check_store_rhs. Use the dt returned by the latter instead
895 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
896 instead of calls to vect_is_simple_use. Pass the scalar rather
897 than the vector operand to vect_is_simple_use when handling
898 second and subsequent copies of an rhs value.
899 (vectorizable_load): Update calls to vect_check_load_store_mask
900 and vect_build_gather_load_calls. Use the cached mask_dt and
901 gs_info.offset_dt instead of calls to vect_is_simple_use.
903 2018-01-20 Jakub Jelinek <jakub@redhat.com>
906 * tree-emutls.c: Include gimplify.h.
907 (lower_emutls_2): New function.
908 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
909 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
910 it before further processing.
913 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
914 UINTVAL (trueop1) instead of INTVAL (op1).
916 2018-01-19 Jakub Jelinek <jakub@redhat.com>
920 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
921 INCOMING_FRAME_SP_OFFSET if not defined.
922 (scan_trace): Add ENTRY argument. If true and
923 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
924 emit a note to adjust the CFA offset.
925 (create_cfi_notes): Adjust scan_trace callers.
926 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
927 INCOMING_FRAME_SP_OFFSET in the CIE.
928 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
929 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
931 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
932 * doc/tm.texi: Regenerated.
934 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
936 PR rtl-optimization/83147
937 * lra-constraints.c (remove_inheritance_pseudos): Use
938 lra_substitute_pseudo_within_insn.
940 2018-01-19 Tom de Vries <tom@codesourcery.com>
941 Cesar Philippidis <cesar@codesourcery.com>
944 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
946 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
949 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
950 spaces for function labels.
952 2018-01-19 Martin Liska <mliska@suse.cz>
954 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
955 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
956 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
957 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
958 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
959 (PRED_CONST_RETURN): Change from 69 to 65.
960 (PRED_NULL_RETURN): Change from 91 to 71.
961 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
962 (PRED_LOOP_GUARD): Change from 66 to 73.
964 2018-01-19 Martin Liska <mliska@suse.cz>
966 * predict.c (predict_insn_def): Add new assert.
967 (struct branch_predictor): Change type to signed integer.
968 (test_prediction_value_range): Amend test to cover
970 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
971 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
972 (PRED_LOOP_ITERATIONS_MAX): Likewise.
973 (PRED_LOOP_IV_COMPARE): Likewise.
974 * predict.h (PROB_UNINITIALIZED): Define new constant.
976 2018-01-19 Martin Liska <mliska@suse.cz>
978 * predict.c (dump_prediction): Add new format for
979 analyze_brprob.py script which is enabled with -details
981 * profile-count.h (precise_p): New function.
983 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
985 PR tree-optimization/83922
986 * tree-vect-loop.c (vect_verify_full_masking): Return false if
987 there are no statements that need masking.
988 (vect_active_double_reduction_p): New function.
989 (vect_analyze_loop_operations): Use it when handling phis that
990 are not in the loop header.
992 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
994 PR tree-optimization/83914
995 * tree-vect-loop.c (vectorizable_induction): Don't convert
996 init_expr or apply the peeling adjustment for inductions
997 that are nested within the vectorized loop.
999 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1001 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1004 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1008 * function.h (gimplify_parameters): Add gimple_seq * argument.
1009 * function.c: Include gimple.h and options.h.
1010 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1011 for the added local temporaries if needed.
1012 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1013 if there are any parameter cleanups, wrap whole body into a
1014 try/finally with the cleanups.
1016 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1019 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1020 Use GET_MODE_CLASS for scalar floating point.
1022 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1026 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1027 Fix call of call_cgraph_insertion_hooks.
1029 2018-01-18 Martin Sebor <msebor@redhat.com>
1031 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1033 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1036 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1039 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1042 * common.opt: (-ffile-prefix-map): New option.
1043 * opts.c (common_handle_option): Defer it.
1044 * opts-global.c (handle_common_deferred_options): Handle it.
1045 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1046 * file-prefix-map.h: New file.
1047 (remap_debug_filename, add_debug_prefix_map): ...here.
1048 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1049 * final.c (debug_prefix_map, add_debug_prefix_map
1050 remap_debug_filename): Move to...
1051 * file-prefix-map.c: New file.
1052 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1053 generalize, get rid of alloca(), use strrchr() instead of strchr().
1054 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1055 Implement in terms of add_prefix_map().
1056 (remap_macro_filename, remap_debug_filename): Implement in term of
1058 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1059 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1060 * dbxout.c: Include file-prefix-map.h.
1061 * varasm.c: Likewise.
1062 * vmsdbgout.c: Likewise.
1063 * xcoffout.c: Likewise.
1064 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1065 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1066 * doc/invoke.texi (-ffile-prefix-map): Document.
1067 (-fdebug-prefix-map): Update description.
1069 2018-01-18 Martin Liska <mliska@suse.cz>
1071 * config/i386/i386.c (indirect_thunk_name): Document that also
1073 (output_indirect_thunk): Document why both instructions
1074 (pause and lfence) are generated.
1076 2018-01-18 Richard Biener <rguenther@suse.de>
1078 PR tree-optimization/83887
1079 * graphite-scop-detection.c
1080 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1081 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1082 (scop_detection::merge_sese): Re-implement with a flood-fill
1083 algorithm that properly finds a SESE region if it exists.
1085 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1088 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1089 pointer_diff optimizations use view_convert instead of convert.
1091 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1093 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1094 Generate different code for -mno-speculate-indirect-jumps.
1095 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1096 (*call_indirect_aix<mode>): Disable for
1097 -mno-speculate-indirect-jumps.
1098 (*call_indirect_aix<mode>_nospec): New define_insn.
1099 (*call_value_indirect_aix<mode>): Disable for
1100 -mno-speculate-indirect-jumps.
1101 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1102 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1103 -mno-speculate-indirect-jumps.
1104 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1106 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1108 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1109 long double type, set the flags for noting the default long double
1110 type, even if we don't pass or return a long double type.
1112 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1115 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1118 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1120 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1121 support for merge[hl].
1122 (fold_mergehl_helper): New helper function.
1123 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1124 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1125 (altivec_vmrglw_direct): Add xxmrglw insn.
1127 2018-01-17 Andrew Waterman <andrew@sifive.com>
1129 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1130 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1132 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1135 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1136 call the lto_location_cache before reading the
1137 DECL_SOURCE_LOCATION of the types.
1139 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1140 Richard Sandiford <richard.sandiford@linaro.org>
1142 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1143 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1144 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1145 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1146 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1148 * config/aarch64/constraints.md (aarch64_movti_operand):
1150 * config/aarch64/predicates.md (Uti): Add new constraint.
1152 2018-01-17 Carl Love <cel@us.ibm.com>
1153 * config/rs6000/vsx.md (define_expand xl_len_r,
1154 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1155 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1157 (define_expand, define_insn): Move the shift left from the
1158 define_insn to the define_expand for lxvl and stxvl instructions.
1159 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1160 and XL_LEN_R definitions to PURE.
1162 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1164 * config/i386/i386.c (indirect_thunk_name): Declare regno
1165 as unsigned int. Compare regno with INVALID_REGNUM.
1166 (output_indirect_thunk): Ditto.
1167 (output_indirect_thunk_function): Ditto.
1168 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1169 in the call to output_indirect_thunk_function.
1171 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1174 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1175 rather than the size of inner_type to determine the stack slot size
1176 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1178 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1181 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1184 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1186 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1187 endian Linux systems to optionally enable multilibs for selecting
1188 the long double type if the user configured an explicit type.
1189 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1190 have no long double multilibs if not defined.
1191 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1192 warn if the user used -mabi={ieee,ibm}longdouble and we built
1193 multilibs for long double.
1194 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1195 appropriate multilib option.
1196 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1198 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1199 for building long double multilibs.
1200 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1202 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1204 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1207 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1209 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1212 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1215 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1218 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1220 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1221 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1223 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1225 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1226 different rtl trees depending on TARGET_64BIT.
1227 (rs6000_gen_lvx): Likewise.
1229 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1231 * config/visium/visium.md (nop): Tweak comment.
1232 (hazard_nop): Likewise.
1234 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1236 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1237 -mspeculate-indirect-jumps.
1238 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1239 for -mno-speculate-indirect-jumps.
1240 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1241 (*call_value_indirect_elfv2<mode>): Disable for
1242 -mno-speculate-indirect-jumps.
1243 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1244 (indirect_jump): Emit different RTL for
1245 -mno-speculate-indirect-jumps.
1246 (*indirect_jump<mode>): Disable for
1247 -mno-speculate-indirect-jumps.
1248 (*indirect_jump<mode>_nospec): New define_insn.
1249 (tablejump): Emit different RTL for
1250 -mno-speculate-indirect-jumps.
1251 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1252 (tablejumpsi_nospec): New define_expand.
1253 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1254 (tablejumpdi_nospec): New define_expand.
1255 (*tablejump<mode>_internal1): Disable for
1256 -mno-speculate-indirect-jumps.
1257 (*tablejump<mode>_internal1_nospec): New define_insn.
1258 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1261 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1263 * caller-save.c (insert_save): Drop unnecessary parameter. All
1266 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1267 Richard Biener <rguenth@suse.de>
1270 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1271 return early, inline manually is_gimple_sizepos. Make sure if we
1272 call gimplify_expr we don't end up with a gimple constant.
1273 * tree.c (variably_modified_type_p): Don't return true for
1274 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1275 * gimplify.h (is_gimple_sizepos): Remove.
1277 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1279 PR tree-optimization/83857
1280 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1281 vectorizable_live_operation for pure SLP statements.
1282 (vectorizable_live_operation): Handle PHIs.
1284 2018-01-16 Richard Biener <rguenther@suse.de>
1286 PR tree-optimization/83867
1287 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1288 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1290 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1293 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1294 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1295 If off is not INTEGER_CST, issue a may not be aligned warning
1296 rather than isn't aligned. Use isn%'t rather than isn't.
1297 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1299 <case MULT_EXPR>: Improve the case when bottom and one of the
1300 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1301 operand, in that case check if the other operand is multiple of
1302 bottom divided by the INTEGER_CST operand.
1304 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1307 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1308 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1309 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1310 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1311 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1312 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1313 (pa_function_arg_size): New function.
1315 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1317 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1318 in a separate statement.
1320 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1322 PR tree-optimization/83847
1323 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1324 group gathers and scatters.
1326 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1328 PR rtl-optimization/86620
1329 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1331 PR rtl-optimization/83213
1332 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1333 to last if both are JUMP_INSNs.
1335 PR tree-optimization/83843
1336 * gimple-ssa-store-merging.c
1337 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1338 store_immediate_info for bswap/nop orig_stores.
1340 2018-01-15 Andrew Waterman <andrew@sifive.com>
1342 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1344 <UDIV>: Increase cost if !TARGET_DIV.
1346 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1348 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1349 (define_attr "cr_logical_3op"): New.
1350 (cceq_ior_compare): Adjust.
1351 (cceq_ior_compare_complement): Adjust.
1352 (*cceq_rev_compare): Adjust.
1353 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1354 (is_cracked_insn): Adjust.
1355 (insn_must_be_first_in_group): Adjust.
1356 * config/rs6000/40x.md: Adjust.
1357 * config/rs6000/440.md: Adjust.
1358 * config/rs6000/476.md: Adjust.
1359 * config/rs6000/601.md: Adjust.
1360 * config/rs6000/603.md: Adjust.
1361 * config/rs6000/6xx.md: Adjust.
1362 * config/rs6000/7450.md: Adjust.
1363 * config/rs6000/7xx.md: Adjust.
1364 * config/rs6000/8540.md: Adjust.
1365 * config/rs6000/cell.md: Adjust.
1366 * config/rs6000/e300c2c3.md: Adjust.
1367 * config/rs6000/e500mc.md: Adjust.
1368 * config/rs6000/e500mc64.md: Adjust.
1369 * config/rs6000/e5500.md: Adjust.
1370 * config/rs6000/e6500.md: Adjust.
1371 * config/rs6000/mpc.md: Adjust.
1372 * config/rs6000/power4.md: Adjust.
1373 * config/rs6000/power5.md: Adjust.
1374 * config/rs6000/power6.md: Adjust.
1375 * config/rs6000/power7.md: Adjust.
1376 * config/rs6000/power8.md: Adjust.
1377 * config/rs6000/power9.md: Adjust.
1378 * config/rs6000/rs64.md: Adjust.
1379 * config/rs6000/titan.md: Adjust.
1381 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1383 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1384 ix86_indirect_branch_register logic.
1386 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1388 * config/i386/constraints.md (Bs): Update
1389 ix86_indirect_branch_register check. Don't check
1390 ix86_indirect_branch_register with GOT_memory_operand.
1392 * config/i386/predicates.md (GOT_memory_operand): Don't check
1393 ix86_indirect_branch_register here.
1394 (GOT32_symbol_operand): Likewise.
1396 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1398 * config/i386/predicates.md (constant_call_address_operand):
1399 Rewrite ix86_indirect_branch_register logic.
1400 (sibcall_insn_operand): Likewise.
1402 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1404 * config/i386/constraints.md (Bs): Replace
1405 ix86_indirect_branch_thunk_register with
1406 ix86_indirect_branch_register.
1408 * config/i386/i386.md (indirect_jump): Likewise.
1409 (tablejump): Likewise.
1410 (*sibcall_memory): Likewise.
1411 (*sibcall_value_memory): Likewise.
1412 Peepholes of indirect call and jump via memory: Likewise.
1413 * config/i386/i386.opt: Likewise.
1414 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1415 (GOT_memory_operand): Likewise.
1416 (call_insn_operand): Likewise.
1417 (sibcall_insn_operand): Likewise.
1418 (GOT32_symbol_operand): Likewise.
1420 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1423 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1424 type rather than type addr's type points to.
1425 (expand_omp_atomic_mutex): Likewise.
1426 (expand_omp_atomic): Likewise.
1428 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1431 * config/i386/i386.c (output_indirect_thunk_function): Use
1432 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1433 for __x86_return_thunk.
1435 2018-01-15 Richard Biener <rguenther@suse.de>
1438 * expmed.c (extract_bit_field_1): Fix typo.
1440 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1443 * config/arm/iterators.md (VF): New mode iterator.
1444 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1445 Remove integer-related logic from pattern.
1446 (neon_vabd<mode>_3): Likewise.
1448 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1451 * common.opt (fstrict-overflow): No longer an alias.
1452 (fwrapv-pointer): New option.
1453 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1454 also for pointer types based on flag_wrapv_pointer.
1455 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1456 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1457 opts->x_flag_wrapv got set.
1458 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1459 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1460 POINTER_TYPE_OVERFLOW_UNDEFINED.
1461 * match.pd: Likewise in address comparison pattern.
1462 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1464 2018-01-15 Richard Biener <rguenther@suse.de>
1467 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1468 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1469 Reset type names to their identifier if their TYPE_DECL doesn't
1470 have linkage (and thus is used for ODR and devirt).
1471 (save_debug_info_for_decl): Remove.
1472 (save_debug_info_for_type): Likewise.
1473 (add_tree_to_fld_list): Adjust.
1474 * tree-pretty-print.c (dump_generic_node): Make dumping of
1475 type names more robust.
1477 2018-01-15 Richard Biener <rguenther@suse.de>
1479 * BASE-VER: Bump to 8.0.1.
1481 2018-01-14 Martin Sebor <msebor@redhat.com>
1484 * builtins.c (check_access): Avoid warning when the no-warning bit
1487 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1489 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1490 * ira-color (allocno_hard_regs_compare): Likewise.
1492 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1495 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1496 Use .pushsection/.popsection.
1498 2018-01-14 Martin Sebor <msebor@redhat.com>
1501 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1503 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1505 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1506 entry from extra_headers.
1507 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1508 extra_headers, make the list bitwise identical to the i?86-*-* one.
1510 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1512 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1513 -mcmodel=large with -mindirect-branch=thunk,
1514 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1515 -mfunction-return=thunk-extern.
1516 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1517 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1518 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1520 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1522 * config/i386/i386.c (print_reg): Print the name of the full
1523 integer register without '%'.
1524 (ix86_print_operand): Handle 'V'.
1525 * doc/extend.texi: Document 'V' modifier.
1527 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1529 * config/i386/constraints.md (Bs): Disallow memory operand for
1530 -mindirect-branch-register.
1532 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1533 (GOT_memory_operand): Likewise.
1534 (call_insn_operand): Likewise.
1535 (sibcall_insn_operand): Likewise.
1536 (GOT32_symbol_operand): Likewise.
1537 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1538 for -mindirect-branch-register.
1539 (tablejump): Likewise.
1540 (*sibcall_memory): Likewise.
1541 (*sibcall_value_memory): Likewise.
1542 Disallow peepholes of indirect call and jump via memory for
1543 -mindirect-branch-register.
1544 (*call_pop): Replace m with Bw.
1545 (*call_value_pop): Likewise.
1546 (*sibcall_pop_memory): Replace m with Bs.
1547 * config/i386/i386.opt (mindirect-branch-register): New option.
1548 * doc/invoke.texi: Document -mindirect-branch-register option.
1550 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1552 * config/i386/i386-protos.h (ix86_output_function_return): New.
1553 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1554 set function_return_type.
1555 (indirect_thunk_name): Add ret_p to indicate thunk for function
1557 (output_indirect_thunk_function): Pass false to
1558 indirect_thunk_name.
1559 (ix86_output_indirect_branch_via_reg): Likewise.
1560 (ix86_output_indirect_branch_via_push): Likewise.
1561 (output_indirect_thunk_function): Create alias for function
1562 return thunk if regno < 0.
1563 (ix86_output_function_return): New function.
1564 (ix86_handle_fndecl_attribute): Handle function_return.
1565 (ix86_attribute_table): Add function_return.
1566 * config/i386/i386.h (machine_function): Add
1567 function_return_type.
1568 * config/i386/i386.md (simple_return_internal): Use
1569 ix86_output_function_return.
1570 (simple_return_internal_long): Likewise.
1571 * config/i386/i386.opt (mfunction-return=): New option.
1572 (indirect_branch): Mention -mfunction-return=.
1573 * doc/extend.texi: Document function_return function attribute.
1574 * doc/invoke.texi: Document -mfunction-return= option.
1576 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1578 * config/i386/i386-opts.h (indirect_branch): New.
1579 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1580 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1581 with local indirect jump when converting indirect call and jump.
1582 (ix86_set_indirect_branch_type): New.
1583 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1584 (indirectlabelno): New.
1585 (indirect_thunk_needed): Likewise.
1586 (indirect_thunk_bnd_needed): Likewise.
1587 (indirect_thunks_used): Likewise.
1588 (indirect_thunks_bnd_used): Likewise.
1589 (INDIRECT_LABEL): Likewise.
1590 (indirect_thunk_name): Likewise.
1591 (output_indirect_thunk): Likewise.
1592 (output_indirect_thunk_function): Likewise.
1593 (ix86_output_indirect_branch_via_reg): Likewise.
1594 (ix86_output_indirect_branch_via_push): Likewise.
1595 (ix86_output_indirect_branch): Likewise.
1596 (ix86_output_indirect_jmp): Likewise.
1597 (ix86_code_end): Call output_indirect_thunk_function if needed.
1598 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1600 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1601 (ix86_attribute_table): Add indirect_branch.
1602 * config/i386/i386.h (machine_function): Add indirect_branch_type
1603 and has_local_indirect_jump.
1604 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1606 (tablejump): Likewise.
1607 (*indirect_jump): Use ix86_output_indirect_jmp.
1608 (*tablejump_1): Likewise.
1609 (simple_return_indirect_internal): Likewise.
1610 * config/i386/i386.opt (mindirect-branch=): New option.
1611 (indirect_branch): New.
1614 (thunk-inline): Likewise.
1615 (thunk-extern): Likewise.
1616 * doc/extend.texi: Document indirect_branch function attribute.
1617 * doc/invoke.texi: Document -mindirect-branch= option.
1619 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1622 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1624 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1626 * ipa-inline.c (want_inline_small_function_p): Return false if
1627 inlining has already failed with CIF_FINAL_ERROR.
1628 (update_caller_keys): Call want_inline_small_function_p before
1630 (update_callee_keys): Likewise.
1632 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1634 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1636 (rs6000_quadword_masked_address_p): Likewise.
1637 (quad_aligned_load_p): Likewise.
1638 (quad_aligned_store_p): Likewise.
1639 (const_load_sequence_p): Add comment to describe the outer-most loop.
1640 (mimic_memory_attributes_and_flags): New function.
1641 (rs6000_gen_stvx): Likewise.
1642 (replace_swapped_aligned_store): Likewise.
1643 (rs6000_gen_lvx): Likewise.
1644 (replace_swapped_aligned_load): Likewise.
1645 (replace_swapped_load_constant): Capitalize argument name in
1646 comment describing this function.
1647 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1648 and stores that access quad-word aligned addresses and replace
1649 with stvx or lvx instructions when appropriate.
1650 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1651 New function prototype.
1652 (rs6000_quadword_masked_address_p): Likewise.
1653 (rs6000_gen_lvx): Likewise.
1654 (rs6000_gen_stvx): Likewise.
1655 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1656 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1657 when memory address is aligned.
1658 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1659 this split to select lvx instruction when memory address is aligned.
1660 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1661 instruction when memory address is aligned.
1662 (*vsx_le_perm_load_v16qi): Likewise.
1663 (four unnamed splitters): Modify to select the stvx instruction
1664 when memory is aligned.
1666 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1668 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1669 which appears in the queue multiple times.
1671 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1672 Alan Hayward <alan.hayward@arm.com>
1673 David Sherwood <david.sherwood@arm.com>
1675 * tree-vectorizer.h (vec_lower_bound): New structure.
1676 (_loop_vec_info): Add check_nonzero and lower_bounds.
1677 (LOOP_VINFO_CHECK_NONZERO): New macro.
1678 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1679 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1680 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1681 fields. Make seg_len the distance travelled, not including the
1683 (dr_direction_indicator): Declare.
1684 (dr_zero_step_indicator): Likewise.
1685 (dr_known_forward_stride_p): Likewise.
1686 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1688 (runtime_alias_check_p): Allow runtime alias checks with
1690 (operator ==): Compare access_size and align.
1691 (prune_runtime_alias_test_list): Rework for new distinction between
1692 the access_size and seg_len.
1693 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1695 (get_segment_min_max): New function.
1696 (create_intersect_range_checks): Use it.
1697 (dr_step_indicator): New function.
1698 (dr_direction_indicator): Likewise.
1699 (dr_zero_step_indicator): Likewise.
1700 (dr_known_forward_stride_p): Likewise.
1701 * tree-loop-distribution.c (data_ref_segment_size): Return
1702 DR_STEP * (niters - 1).
1703 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1705 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1706 (vect_preserves_scalar_order_p): New function, split out from...
1707 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1708 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1709 (vect_vfa_access_size): New function.
1710 (vect_vfa_align): Likewise.
1711 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1712 (dump_lower_bound): New function.
1713 (vect_check_lower_bound): Likewise.
1714 (vect_small_gap_p): Likewise.
1715 (vectorizable_with_step_bound_p): Likewise.
1716 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1717 depencies if the vectorization factor is 1. Convert the checks
1718 for nonzero steps into checks on the bounds of DR_STEP. Try using
1719 a bunds check for variable steps if the minimum required step is
1720 relatively small. Update calls to the dr_with_seg_len
1721 constructor and to vect_compile_time_alias.
1722 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1724 (vect_loop_versioning): Call it.
1725 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1727 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1729 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1730 Alan Hayward <alan.hayward@arm.com>
1731 David Sherwood <david.sherwood@arm.com>
1733 * doc/sourcebuild.texi (vect_scatter_store): Document.
1734 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1736 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1738 * genopinit.c (main): Add supports_vec_scatter_store and
1739 supports_vec_scatter_store_cached to target_optabs.
1740 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1741 IFN_MASK_SCATTER_STORE.
1742 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1744 * internal-fn.h (internal_store_fn_p): Declare.
1745 (internal_fn_stored_value_index): Likewise.
1746 * internal-fn.c (scatter_store_direct): New macro.
1747 (expand_scatter_store_optab_fn): New function.
1748 (direct_scatter_store_optab_supported_p): New macro.
1749 (internal_store_fn_p): New function.
1750 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1751 IFN_MASK_SCATTER_STORE.
1752 (internal_fn_mask_index): Likewise.
1753 (internal_fn_stored_value_index): New function.
1754 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1756 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1757 * optabs-query.c (supports_vec_scatter_store_p): New function.
1758 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1759 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1760 true for scatter stores.
1761 (vect_gather_scatter_fn_p): Handle scatter stores too.
1762 (vect_check_gather_scatter): Consider using scatter stores if
1763 supports_vec_scatter_store_p.
1764 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1766 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1767 internal_fn_stored_value_index.
1768 (check_load_store_masking): Handle scatter stores too.
1769 (vect_get_store_rhs): Make public.
1770 (vectorizable_call): Use internal_store_fn_p.
1771 (vectorizable_store): Handle scatter store internal functions.
1772 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1773 when deciding whether the end of the group has been reached.
1774 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1775 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1776 (mask_scatter_store<mode>): New insns.
1778 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1779 Alan Hayward <alan.hayward@arm.com>
1780 David Sherwood <david.sherwood@arm.com>
1782 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1783 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1784 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1786 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1787 Use vect_truncate_gather_scatter_offset if we can't treat the
1788 operation as a normal gather load or scatter store.
1789 (get_group_load_store_type): Take the gather_scatter_info
1790 as argument. Try using a gather load or scatter store for
1791 single-element groups.
1792 (get_load_store_type): Update calls to get_group_load_store_type
1793 and vect_use_strided_gather_scatters_p.
1795 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1796 Alan Hayward <alan.hayward@arm.com>
1797 David Sherwood <david.sherwood@arm.com>
1799 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1800 optional tree argument.
1801 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1803 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1804 but continue to use the current value as a fallback.
1805 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1806 to compare the updates.
1807 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1808 (get_load_store_type): Use it when handling a strided access.
1809 (vect_get_strided_load_store_ops): New function.
1810 (vect_get_data_ptr_increment): Likewise.
1811 (vectorizable_load): Handle strided gather loads. Always pass
1812 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1814 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1815 Alan Hayward <alan.hayward@arm.com>
1816 David Sherwood <david.sherwood@arm.com>
1818 * doc/md.texi (gather_load@var{m}): Document.
1819 (mask_gather_load@var{m}): Likewise.
1820 * genopinit.c (main): Add supports_vec_gather_load and
1821 supports_vec_gather_load_cached to target_optabs.
1822 * optabs-tree.c (init_tree_optimization_optabs): Use
1823 ggc_cleared_alloc to allocate target_optabs.
1824 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1825 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1827 * internal-fn.h (internal_load_fn_p): Declare.
1828 (internal_gather_scatter_fn_p): Likewise.
1829 (internal_fn_mask_index): Likewise.
1830 (internal_gather_scatter_fn_supported_p): Likewise.
1831 * internal-fn.c (gather_load_direct): New macro.
1832 (expand_gather_load_optab_fn): New function.
1833 (direct_gather_load_optab_supported_p): New macro.
1834 (direct_internal_fn_optab): New function.
1835 (internal_load_fn_p): Likewise.
1836 (internal_gather_scatter_fn_p): Likewise.
1837 (internal_fn_mask_index): Likewise.
1838 (internal_gather_scatter_fn_supported_p): Likewise.
1839 * optabs-query.c (supports_at_least_one_mode_p): New function.
1840 (supports_vec_gather_load_p): Likewise.
1841 * optabs-query.h (supports_vec_gather_load_p): Declare.
1842 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1843 and memory_type field.
1844 (NUM_PATTERNS): Bump to 15.
1845 * tree-vect-data-refs.c: Include internal-fn.h.
1846 (vect_gather_scatter_fn_p): New function.
1847 (vect_describe_gather_scatter_call): Likewise.
1848 (vect_check_gather_scatter): Try using internal functions for
1849 gather loads. Recognize existing calls to a gather load function.
1850 (vect_analyze_data_refs): Consider using gather loads if
1851 supports_vec_gather_load_p.
1852 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1853 (vect_get_gather_scatter_offset_type): Likewise.
1854 (vect_convert_mask_for_vectype): Likewise.
1855 (vect_add_conversion_to_patterm): Likewise.
1856 (vect_try_gather_scatter_pattern): Likewise.
1857 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1858 (vect_vect_recog_func_ptrs): Add it.
1859 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1860 internal_fn_mask_index and internal_gather_scatter_fn_p.
1861 (check_load_store_masking): Take the gather_scatter_info as an
1862 argument and handle gather loads.
1863 (vect_get_gather_scatter_ops): New function.
1864 (vectorizable_call): Check internal_load_fn_p.
1865 (vectorizable_load): Likewise. Handle gather load internal
1867 (vectorizable_store): Update call to check_load_store_masking.
1868 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1869 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1870 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1871 (aarch64_gather_scale_operand_d): New predicates.
1872 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1873 (mask_gather_load<mode>): New insns.
1875 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1876 Alan Hayward <alan.hayward@arm.com>
1877 David Sherwood <david.sherwood@arm.com>
1879 * optabs.def (fold_left_plus_optab): New optab.
1880 * doc/md.texi (fold_left_plus_@var{m}): Document.
1881 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1882 * internal-fn.c (fold_left_direct): Define.
1883 (expand_fold_left_optab_fn): Likewise.
1884 (direct_fold_left_optab_supported_p): Likewise.
1885 * fold-const-call.c (fold_const_fold_left): New function.
1886 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1887 * tree-parloops.c (valid_reduction_p): New function.
1888 (gather_scalar_reductions): Use it.
1889 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1890 (vect_finish_replace_stmt): Declare.
1891 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1892 (needs_fold_left_reduction_p): New function, split out from...
1893 (vect_is_simple_reduction): ...here. Accept reductions that
1894 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1895 (vect_force_simple_reduction): Also store the reduction type in
1896 the assignment's STMT_VINFO_REDUC_TYPE.
1897 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1898 (merge_with_identity): New function.
1899 (vect_expand_fold_left): Likewise.
1900 (vectorize_fold_left_reduction): Likewise.
1901 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1902 scalar phi in place for it. Check for target support and reject
1903 cases that would reassociate the operation. Defer the transform
1904 phase to vectorize_fold_left_reduction.
1905 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1906 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1907 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1909 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1911 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1912 call to ifc_temp_var.
1914 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1915 Alan Hayward <alan.hayward@arm.com>
1916 David Sherwood <david.sherwood@arm.com>
1918 * target.def (legitimize_address_displacement): Take the original
1919 offset as a poly_int.
1920 * targhooks.h (default_legitimize_address_displacement): Update
1922 * targhooks.c (default_legitimize_address_displacement): Likewise.
1923 * doc/tm.texi: Regenerate.
1924 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1925 as an argument, moving assert of ad->disp == ad->disp_term to...
1926 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1927 Try calling targetm.legitimize_address_displacement before expanding
1928 the address rather than afterwards, and adjust for the new interface.
1929 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1930 Match the new hook interface. Handle SVE addresses.
1931 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1934 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1936 * Makefile.in (OBJS): Add early-remat.o.
1937 * target.def (select_early_remat_modes): New hook.
1938 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1939 * doc/tm.texi: Regenerate.
1940 * targhooks.h (default_select_early_remat_modes): Declare.
1941 * targhooks.c (default_select_early_remat_modes): New function.
1942 * timevar.def (TV_EARLY_REMAT): New timevar.
1943 * passes.def (pass_early_remat): New pass.
1944 * tree-pass.h (make_pass_early_remat): Declare.
1945 * early-remat.c: New file.
1946 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1948 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1950 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1951 Alan Hayward <alan.hayward@arm.com>
1952 David Sherwood <david.sherwood@arm.com>
1954 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1955 vfm1 with a bound_epilog parameter.
1956 (vect_do_peeling): Update calls accordingly, and move the prologue
1957 call earlier in the function. Treat the base bound_epilog as 0 for
1958 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1959 this base when peeling for gaps.
1960 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1961 with fully-masked loops.
1962 (vect_estimate_min_profitable_iters): Handle the single peeled
1963 iteration in that case.
1965 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1966 Alan Hayward <alan.hayward@arm.com>
1967 David Sherwood <david.sherwood@arm.com>
1969 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1970 single-element interleaving even if the size is not a power of 2.
1971 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1972 accesses for single-element interleaving if the group size is
1975 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1976 Alan Hayward <alan.hayward@arm.com>
1977 David Sherwood <david.sherwood@arm.com>
1979 * doc/md.texi (fold_extract_last_@var{m}): Document.
1980 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1981 * optabs.def (fold_extract_last_optab): New optab.
1982 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1983 * internal-fn.c (fold_extract_direct): New macro.
1984 (expand_fold_extract_optab_fn): Likewise.
1985 (direct_fold_extract_optab_supported_p): Likewise.
1986 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1987 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1988 EXTRACT_LAST_REDUCTION.
1989 (get_initial_def_for_reduction): Do not create an initial vector
1990 for EXTRACT_LAST_REDUCTION reductions.
1991 (vectorizable_reduction): Leave the scalar phi in place for
1992 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1993 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1994 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1995 transform phase to vectorizable_condition.
1996 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1998 (vect_finish_stmt_generation): ...here.
1999 (vect_finish_replace_stmt): New function.
2000 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2001 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2003 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2005 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2006 Alan Hayward <alan.hayward@arm.com>
2007 David Sherwood <david.sherwood@arm.com>
2009 * doc/md.texi (extract_last_@var{m}): Document.
2010 * optabs.def (extract_last_optab): New optab.
2011 * internal-fn.def (EXTRACT_LAST): New internal function.
2012 * internal-fn.c (cond_unary_direct): New macro.
2013 (expand_cond_unary_optab_fn): Likewise.
2014 (direct_cond_unary_optab_supported_p): Likewise.
2015 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2016 loops using EXTRACT_LAST.
2017 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2018 (extract_last_<mode>): ...this optab.
2019 (vec_extract<mode><Vel>): Update accordingly.
2021 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2022 Alan Hayward <alan.hayward@arm.com>
2023 David Sherwood <david.sherwood@arm.com>
2025 * target.def (empty_mask_is_expensive): New hook.
2026 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2027 * doc/tm.texi: Regenerate.
2028 * targhooks.h (default_empty_mask_is_expensive): Declare.
2029 * targhooks.c (default_empty_mask_is_expensive): New function.
2030 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2031 if the target says that empty masks are expensive.
2032 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2034 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2036 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2037 Alan Hayward <alan.hayward@arm.com>
2038 David Sherwood <david.sherwood@arm.com>
2040 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2041 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2042 (vect_use_loop_mask_for_alignment_p): New function.
2043 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2044 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2045 niters_skip argument. Make sure that the first niters_skip elements
2046 of the first iteration are inactive.
2047 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2048 Update call to vect_set_loop_masks_directly.
2049 (get_misalign_in_elems): New function, split out from...
2050 (vect_gen_prolog_loop_niters): ...here.
2051 (vect_update_init_of_dr): Take a code argument that specifies whether
2052 the adjustment should be added or subtracted.
2053 (vect_update_init_of_drs): Likewise.
2054 (vect_prepare_for_masked_peels): New function.
2055 (vect_do_peeling): Skip prologue peeling if we're using a mask
2056 instead. Update call to vect_update_inits_of_drs.
2057 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2059 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2060 alignment. Do not include the number of peeled iterations in
2061 the minimum threshold in that case.
2062 (vectorizable_induction): Adjust the start value down by
2063 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2064 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2065 Take the number of skipped iterations into account when calculating
2067 * tree-vect-stmts.c (vect_gen_while_not): New function.
2069 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2070 Alan Hayward <alan.hayward@arm.com>
2071 David Sherwood <david.sherwood@arm.com>
2073 * doc/sourcebuild.texi (vect_fully_masked): Document.
2074 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2076 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2078 (vect_analyze_loop_2): ...here. Don't check the vectorization
2079 factor against the number of loop iterations if the loop is
2082 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2083 Alan Hayward <alan.hayward@arm.com>
2084 David Sherwood <david.sherwood@arm.com>
2086 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2087 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2088 (dump_groups): Update accordingly.
2089 (iv_use::mem_type): New member variable.
2090 (address_p): New function.
2091 (record_use): Add a mem_type argument and initialize the new
2093 (record_group_use): Add a mem_type argument. Use address_p.
2094 Remove obsolete null checks of base_object. Update call to record_use.
2095 (find_interesting_uses_op): Update call to record_group_use.
2096 (find_interesting_uses_cond): Likewise.
2097 (find_interesting_uses_address): Likewise.
2098 (get_mem_type_for_internal_fn): New function.
2099 (find_address_like_use): Likewise.
2100 (find_interesting_uses_stmt): Try find_address_like_use before
2101 calling find_interesting_uses_op.
2102 (addr_offset_valid_p): Use the iv mem_type field as the type
2103 of the addressed memory.
2104 (add_autoinc_candidates): Likewise.
2105 (get_address_cost): Likewise.
2106 (split_small_address_groups_p): Use address_p.
2107 (split_address_groups): Likewise.
2108 (add_iv_candidate_for_use): Likewise.
2109 (autoinc_possible_for_pair): Likewise.
2110 (rewrite_groups): Likewise.
2111 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2112 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2113 (get_alias_ptr_type_for_ptr_address): New function.
2114 (rewrite_use_address): Rewrite address uses in calls that were
2115 identified by find_address_like_use.
2117 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2118 Alan Hayward <alan.hayward@arm.com>
2119 David Sherwood <david.sherwood@arm.com>
2121 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2123 * gimple-expr.h (is_gimple_addressable: Likewise.
2124 * gimple-expr.c (is_gimple_address): Likewise.
2125 * internal-fn.c (expand_call_mem_ref): New function.
2126 (expand_mask_load_optab_fn): Use it.
2127 (expand_mask_store_optab_fn): Likewise.
2129 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2130 Alan Hayward <alan.hayward@arm.com>
2131 David Sherwood <david.sherwood@arm.com>
2133 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2134 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2135 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2136 (cond_umax@var{mode}): Document.
2137 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2138 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2139 (cond_umin_optab, cond_umax_optab): New optabs.
2140 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2141 (COND_IOR, COND_XOR): New internal functions.
2142 * internal-fn.h (get_conditional_internal_fn): Declare.
2143 * internal-fn.c (cond_binary_direct): New macro.
2144 (expand_cond_binary_optab_fn): Likewise.
2145 (direct_cond_binary_optab_supported_p): Likewise.
2146 (get_conditional_internal_fn): New function.
2147 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2148 Cope with reduction statements that are vectorized as calls rather
2150 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2151 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2152 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2153 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2154 (UNSPEC_COND_EOR): New unspecs.
2155 (optab): Add mappings for them.
2156 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2157 (sve_int_op, sve_fp_op): New int attributes.
2159 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2160 Alan Hayward <alan.hayward@arm.com>
2161 David Sherwood <david.sherwood@arm.com>
2163 * optabs.def (while_ult_optab): New optab.
2164 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2165 * internal-fn.def (WHILE_ULT): New internal function.
2166 * internal-fn.h (direct_internal_fn_supported_p): New override
2167 that takes two types as argument.
2168 * internal-fn.c (while_direct): New macro.
2169 (expand_while_optab_fn): New function.
2170 (convert_optab_supported_p): Likewise.
2171 (direct_while_optab_supported_p): New macro.
2172 * wide-int.h (wi::udiv_ceil): New function.
2173 * tree-vectorizer.h (rgroup_masks): New structure.
2174 (vec_loop_masks): New typedef.
2175 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2177 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2178 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2179 (vect_max_vf): New function.
2180 (slpeel_make_loop_iterate_ntimes): Delete.
2181 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2182 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2183 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2184 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2185 internal-fn.h, stor-layout.h and optabs-query.h.
2186 (vect_set_loop_mask): New function.
2187 (add_preheader_seq): Likewise.
2188 (add_header_seq): Likewise.
2189 (interleave_supported_p): Likewise.
2190 (vect_maybe_permute_loop_masks): Likewise.
2191 (vect_set_loop_masks_directly): Likewise.
2192 (vect_set_loop_condition_masked): Likewise.
2193 (vect_set_loop_condition_unmasked): New function, split out from
2194 slpeel_make_loop_iterate_ntimes.
2195 (slpeel_make_loop_iterate_ntimes): Rename to..
2196 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2197 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2198 (vect_do_peeling): Update call accordingly.
2199 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2201 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2202 mask_compare_type, can_fully_mask_p and fully_masked_p.
2203 (release_vec_loop_masks): New function.
2204 (_loop_vec_info): Use it to free the loop masks.
2205 (can_produce_all_loop_masks_p): New function.
2206 (vect_get_max_nscalars_per_iter): Likewise.
2207 (vect_verify_full_masking): Likewise.
2208 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2209 retries, and free the mask rgroups before retrying. Check loop-wide
2210 reasons for disallowing fully-masked loops. Make the final decision
2211 about whether use a fully-masked loop or not.
2212 (vect_estimate_min_profitable_iters): Do not assume that peeling
2213 for the number of iterations will be needed for fully-masked loops.
2214 (vectorizable_reduction): Disable fully-masked loops.
2215 (vectorizable_live_operation): Likewise.
2216 (vect_halve_mask_nunits): New function.
2217 (vect_double_mask_nunits): Likewise.
2218 (vect_record_loop_mask): Likewise.
2219 (vect_get_loop_mask): Likewise.
2220 (vect_transform_loop): Handle the case in which the final loop
2221 iteration might handle a partial vector. Call vect_set_loop_condition
2222 instead of slpeel_make_loop_iterate_ntimes.
2223 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2224 (check_load_store_masking): New function.
2225 (prepare_load_store_mask): Likewise.
2226 (vectorizable_store): Handle fully-masked loops.
2227 (vectorizable_load): Likewise.
2228 (supportable_widening_operation): Use vect_halve_mask_nunits for
2230 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2231 (vect_gen_while): New function.
2232 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2233 (aarch64_uqdec<mode>): New insn.
2235 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2236 Alan Hayward <alan.hayward@arm.com>
2237 David Sherwood <david.sherwood@arm.com>
2239 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2240 (reduc_xor_scal_optab): New optabs.
2241 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2242 (reduc_xor_scal_@var{m}): Document.
2243 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2244 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2246 * fold-const-call.c (fold_const_call): Handle them.
2247 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2248 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2249 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2250 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2251 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2252 (UNSPEC_XORV): New unspecs.
2253 (optab): Add entries for them.
2254 (BITWISEV): New int iterator.
2255 (bit_reduc_op): New int attributes.
2257 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2258 Alan Hayward <alan.hayward@arm.com>
2259 David Sherwood <david.sherwood@arm.com>
2261 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2262 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2263 * optabs.def (vec_shl_insert_optab): New optab.
2264 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2265 (duplicate_and_interleave): Likewise.
2266 * tree-vect-loop.c: Include internal-fn.h.
2267 (neutral_op_for_slp_reduction): New function, split out from
2268 get_initial_defs_for_reduction.
2269 (get_initial_def_for_reduction): Handle option 2 for variable-length
2270 vectors by loading the neutral value into a vector and then shifting
2271 the initial value into element 0.
2272 (get_initial_defs_for_reduction): Replace the code argument with
2273 the neutral value calculated by neutral_op_for_slp_reduction.
2274 Use gimple_build_vector for constant-length vectors.
2275 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2276 but the first group_size elements have a neutral value.
2277 Use duplicate_and_interleave otherwise.
2278 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2279 Update call to get_initial_defs_for_reduction. Handle SLP
2280 reductions for variable-length vectors by creating one vector
2281 result for each scalar result, with the elements associated
2282 with other scalar results stubbed out with the neutral value.
2283 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2284 Require IFN_VEC_SHL_INSERT for double reductions on
2285 variable-length vectors, or SLP reductions that have
2286 a neutral value. Require can_duplicate_and_interleave_p
2287 support for variable-length unchained SLP reductions if there
2288 is no neutral value, such as for MIN/MAX reductions. Also require
2289 the number of vector elements to be a multiple of the number of
2290 SLP statements when doing variable-length unchained SLP reductions.
2291 Update call to vect_create_epilog_for_reduction.
2292 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2293 and remove initial values.
2294 (duplicate_and_interleave): Make public.
2295 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2296 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2298 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2299 Alan Hayward <alan.hayward@arm.com>
2300 David Sherwood <david.sherwood@arm.com>
2302 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2303 (can_duplicate_and_interleave_p): New function.
2304 (vect_get_and_check_slp_defs): Take the vector of statements
2305 rather than just the current one. Remove excess parentheses.
2306 Restriction rejectinon of vect_constant_def and vect_external_def
2307 for variable-length vectors to boolean types, or types for which
2308 can_duplicate_and_interleave_p is false.
2309 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2310 (duplicate_and_interleave): New function.
2311 (vect_get_constant_vectors): Use gimple_build_vector for
2312 constant-length vectors and suitable variable-length constant
2313 vectors. Use duplicate_and_interleave for other variable-length
2314 vectors. Don't defer the update when inserting new statements.
2316 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2317 Alan Hayward <alan.hayward@arm.com>
2318 David Sherwood <david.sherwood@arm.com>
2320 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2321 min_profitable_iters doesn't go negative.
2323 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2324 Alan Hayward <alan.hayward@arm.com>
2325 David Sherwood <david.sherwood@arm.com>
2327 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2328 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2329 * optabs.def (vec_mask_load_lanes_optab): New optab.
2330 (vec_mask_store_lanes_optab): Likewise.
2331 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2332 (MASK_STORE_LANES): Likewise.
2333 * internal-fn.c (mask_load_lanes_direct): New macro.
2334 (mask_store_lanes_direct): Likewise.
2335 (expand_mask_load_optab_fn): Handle masked operations.
2336 (expand_mask_load_lanes_optab_fn): New macro.
2337 (expand_mask_store_optab_fn): Handle masked operations.
2338 (expand_mask_store_lanes_optab_fn): New macro.
2339 (direct_mask_load_lanes_optab_supported_p): Likewise.
2340 (direct_mask_store_lanes_optab_supported_p): Likewise.
2341 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2343 (vect_load_lanes_supported): Likewise.
2344 * tree-vect-data-refs.c (strip_conversion): New function.
2345 (can_group_stmts_p): Likewise.
2346 (vect_analyze_data_ref_accesses): Use it instead of checking
2347 for a pair of assignments.
2348 (vect_store_lanes_supported): Take a masked_p parameter.
2349 (vect_load_lanes_supported): Likewise.
2350 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2351 vect_store_lanes_supported and vect_load_lanes_supported.
2352 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2353 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2354 parameter. Don't allow gaps for masked accesses.
2355 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2356 and vect_load_lanes_supported.
2357 (get_load_store_type): Take a masked_p parameter and update
2358 call to get_group_load_store_type.
2359 (vectorizable_store): Update call to get_load_store_type.
2360 Handle IFN_MASK_STORE_LANES.
2361 (vectorizable_load): Update call to get_load_store_type.
2362 Handle IFN_MASK_LOAD_LANES.
2364 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2365 Alan Hayward <alan.hayward@arm.com>
2366 David Sherwood <david.sherwood@arm.com>
2368 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2370 * config/aarch64/aarch64-protos.h
2371 (aarch64_sve_struct_memory_operand_p): Declare.
2372 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2373 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2374 (VPRED, vpred): Handle SVE structure modes.
2375 * config/aarch64/constraints.md (Utx): New constraint.
2376 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2377 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2378 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2379 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2380 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2381 structure modes. Split into pieces after RA.
2382 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2383 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2385 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2386 SVE structure modes.
2387 (aarch64_classify_address): Likewise.
2388 (sizetochar): Move earlier in file.
2389 (aarch64_print_operand): Handle SVE register lists.
2390 (aarch64_array_mode): New function.
2391 (aarch64_sve_struct_memory_operand_p): Likewise.
2392 (TARGET_ARRAY_MODE): Redefine.
2394 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2395 Alan Hayward <alan.hayward@arm.com>
2396 David Sherwood <david.sherwood@arm.com>
2398 * target.def (array_mode): New target hook.
2399 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2400 * doc/tm.texi: Regenerate.
2401 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2402 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2403 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2405 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2408 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2409 Alan Hayward <alan.hayward@arm.com>
2410 David Sherwood <david.sherwood@arm.com>
2412 * fold-const.c (fold_binary_loc): Check the argument types
2413 rather than the result type when testing for a vector operation.
2415 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2417 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2418 * doc/tm.texi: Regenerate.
2420 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2421 Alan Hayward <alan.hayward@arm.com>
2422 David Sherwood <david.sherwood@arm.com>
2424 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2425 (sve): Document new AArch64 extension.
2426 * doc/md.texi (w): Extend the description of the AArch64
2427 constraint to include SVE vectors.
2428 (Upl, Upa): Document new AArch64 predicate constraints.
2429 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2431 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2432 (msve-vector-bits=): New option.
2433 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2434 SVE when these are disabled.
2435 (sve): New extension.
2436 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2437 modes. Adjust their number of units based on aarch64_sve_vg.
2438 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2439 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2440 aarch64_addr_query_type.
2441 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2442 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2443 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2444 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2445 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2446 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2447 (aarch64_simd_imm_zero_p): Delete.
2448 (aarch64_check_zero_based_sve_index_immediate): Declare.
2449 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2450 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2451 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2452 (aarch64_sve_float_mul_immediate_p): Likewise.
2453 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2455 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2456 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2457 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2458 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2459 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2460 (aarch64_regmode_natural_size): Likewise.
2461 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2462 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2464 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2465 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2466 for VG and the SVE predicate registers.
2467 (V_ALIASES): Add a "z"-prefixed alias.
2468 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2469 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2470 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2471 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2472 (REG_CLASS_NAMES): Add entries for them.
2473 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2474 and the predicate registers.
2475 (aarch64_sve_vg): Declare.
2476 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2477 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2478 (REGMODE_NATURAL_SIZE): Define.
2479 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2481 * config/aarch64/aarch64.c: Include cfgrtl.h.
2482 (simd_immediate_info): Add a constructor for series vectors,
2483 and an associated step field.
2484 (aarch64_sve_vg): New variable.
2485 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2486 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2487 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2488 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2489 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2490 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2491 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2492 (aarch64_get_mask_mode): New functions.
2493 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2494 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2495 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2496 predicate modes and predicate registers. Explicitly restrict
2497 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2498 to store a vector mode if it is recognized by
2499 aarch64_classify_vector_mode.
2500 (aarch64_regmode_natural_size): New function.
2501 (aarch64_hard_regno_caller_save_mode): Return the original mode
2503 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2504 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2505 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2506 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2508 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2509 does not overlap dest if the function is frame-related. Handle
2511 (aarch64_split_add_offset): New function.
2512 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2513 them aarch64_add_offset.
2514 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2515 and update call to aarch64_sub_sp.
2516 (aarch64_add_cfa_expression): New function.
2517 (aarch64_expand_prologue): Pass extra temporary registers to the
2518 functions above. Handle the case in which we need to emit new
2519 DW_CFA_expressions for registers that were originally saved
2520 relative to the stack pointer, but now have to be expressed
2521 relative to the frame pointer.
2522 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2524 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2525 IP0 and IP1 values for SVE frames.
2526 (aarch64_expand_vec_series): New function.
2527 (aarch64_expand_sve_widened_duplicate): Likewise.
2528 (aarch64_expand_sve_const_vector): Likewise.
2529 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2530 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2531 into the register, rather than emitting a SET directly.
2532 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2533 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2534 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2535 (offset_9bit_signed_scaled_p): New functions.
2536 (aarch64_replicate_bitmask_imm): New function.
2537 (aarch64_bitmask_imm): Use it.
2538 (aarch64_cannot_force_const_mem): Reject expressions involving
2539 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2540 (aarch64_classify_index): Handle SVE indices, by requiring
2541 a plain register index with a scale that matches the element size.
2542 (aarch64_classify_address): Handle SVE addresses. Assert that
2543 the mode of the address is VOIDmode or an integer mode.
2544 Update call to aarch64_classify_symbol.
2545 (aarch64_classify_symbolic_expression): Update call to
2546 aarch64_classify_symbol.
2547 (aarch64_const_vec_all_in_range_p): New function.
2548 (aarch64_print_vector_float_operand): Likewise.
2549 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2550 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2551 and the FP immediates 1.0 and 0.5.
2552 (aarch64_print_address_internal): Handle SVE addresses.
2553 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2554 (aarch64_regno_regclass): Handle predicate registers.
2555 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2557 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2558 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2559 (aarch64_convert_sve_vector_bits): New function.
2560 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2561 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2563 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2564 Handle SVE vector and predicate modes. Accept VL-based constants
2565 that need only one temporary register, and VL offsets that require
2566 no temporary registers.
2567 (aarch64_conditional_register_usage): Mark the predicate registers
2568 as fixed if SVE isn't available.
2569 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2570 Return true for SVE vector and predicate modes.
2571 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2572 rather than an unsigned int. Handle SVE modes.
2573 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2575 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2577 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2578 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2579 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2580 (aarch64_sve_float_mul_immediate_p): New functions.
2581 (aarch64_sve_valid_immediate): New function.
2582 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2583 Explicitly reject structure modes. Check for INDEX constants.
2584 Handle PTRUE and PFALSE constants.
2585 (aarch64_check_zero_based_sve_index_immediate): New function.
2586 (aarch64_simd_imm_zero_p): Delete.
2587 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2588 vector modes. Accept constants in the range of CNT[BHWD].
2589 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2590 ask for an Advanced SIMD mode.
2591 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2592 (aarch64_simd_vector_alignment): Handle SVE predicates.
2593 (aarch64_vectorize_preferred_vector_alignment): New function.
2594 (aarch64_simd_vector_alignment_reachable): Use it instead of
2596 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2597 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2599 (MAX_VECT_LEN): Delete.
2600 (expand_vec_perm_d): Add a vec_flags field.
2601 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2602 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2603 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2605 (aarch64_evpc_rev): Rename to...
2606 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2607 (aarch64_evpc_rev_global): New function.
2608 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2609 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2611 (aarch64_evpc_sve_tbl): New function.
2612 (aarch64_expand_vec_perm_const_1): Update after rename of
2613 aarch64_evpc_rev. Handle SVE permutes too, trying
2614 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2615 than aarch64_evpc_tbl.
2616 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2617 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2618 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2619 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2620 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2621 (aarch64_expand_sve_vcond): New functions.
2622 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2623 of aarch64_vector_mode_p.
2624 (aarch64_dwarf_poly_indeterminate_value): New function.
2625 (aarch64_compute_pressure_classes): Likewise.
2626 (aarch64_can_change_mode_class): Likewise.
2627 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2628 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2629 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2630 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2631 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2632 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2633 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2634 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2636 (Dn, Dl, Dr): Accept const as well as const_vector.
2637 (Dz): Likewise. Compare against CONST0_RTX.
2638 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2639 of "vector" where appropriate.
2640 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2641 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2642 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2643 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2644 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2645 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2646 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2647 (v_int_equiv): Extend to SVE modes.
2648 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2650 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2651 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2652 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2653 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2654 (SVE_COND_FP_CMP): New int iterators.
2655 (perm_hilo): Handle the new unpack unspecs.
2656 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2658 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2659 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2660 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2661 (aarch64_equality_operator, aarch64_constant_vector_operand)
2662 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2663 (aarch64_sve_nonimmediate_operand): Likewise.
2664 (aarch64_sve_general_operand): Likewise.
2665 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2666 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2667 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2668 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2669 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2670 (aarch64_sve_float_arith_immediate): Likewise.
2671 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2672 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2673 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2674 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2675 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2676 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2677 (aarch64_sve_float_arith_operand): Likewise.
2678 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2679 (aarch64_sve_float_mul_operand): Likewise.
2680 (aarch64_sve_vec_perm_operand): Likewise.
2681 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2682 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2683 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2684 as well as const_vector.
2685 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2686 in file. Use CONST0_RTX and CONSTM1_RTX.
2687 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2688 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2689 Use aarch64_simd_imm_zero.
2690 * config/aarch64/aarch64-sve.md: New file.
2691 * config/aarch64/aarch64.md: Include it.
2692 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2693 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2694 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2695 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2696 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2697 (sve): New attribute.
2698 (enabled): Disable instructions with the sve attribute unless
2700 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2701 aarch64_expand_mov_immediate.
2702 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2703 CNT[BHSD] immediates.
2704 (movti): Split CONST_POLY_INT moves into two halves.
2705 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2706 Split additions that need a temporary here if the destination
2707 is the stack pointer.
2708 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2709 (*add<mode>3_poly_1): New instruction.
2710 (set_clobber_cc): New expander.
2712 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2714 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2715 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2716 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2717 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2718 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2719 Change innermode from fixed_mode_size to machine_mode.
2720 (simplify_subreg): Update call accordingly. Handle a constant-sized
2721 subreg of a variable-length CONST_VECTOR.
2723 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2724 Alan Hayward <alan.hayward@arm.com>
2725 David Sherwood <david.sherwood@arm.com>
2727 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2728 (add_offset_to_base): New function, split out from...
2729 (create_mem_ref): ...here. When handling a scale other than 1,
2730 check first whether the address is valid without the offset.
2731 Add it into the base if so, leaving the index and scale as-is.
2733 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2736 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2737 fold_for_warn before checking if arg2 is INTEGER_CST.
2739 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2741 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2742 (store_multiple_operation): Delete.
2743 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2744 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2745 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2746 guarded by TARGET_STRING.
2747 (rs6000_output_load_multiple): Delete.
2748 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2749 OPTION_MASK_STRING / TARGET_STRING handling.
2750 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2751 (const rs6000_opt_masks) <"string">: Change mask to 0.
2752 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2753 (MASK_STRING): Delete.
2754 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2756 (load_multiple): Delete.
2763 (store_multiple): Delete.
2770 (movmemsi_8reg): Delete.
2771 (corresponding unnamed define_insn): Delete.
2772 (movmemsi_6reg): Delete.
2773 (corresponding unnamed define_insn): Delete.
2774 (movmemsi_4reg): Delete.
2775 (corresponding unnamed define_insn): Delete.
2776 (movmemsi_2reg): Delete.
2777 (corresponding unnamed define_insn): Delete.
2778 (movmemsi_1reg): Delete.
2779 (corresponding unnamed define_insn): Delete.
2780 * config/rs6000/rs6000.opt (mno-string): New.
2781 (mstring): Replace by deprecation warning stub.
2782 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2784 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2786 * regrename.c (regrename_do_replace): If replacing the same
2787 reg multiple times, try to reuse last created gen_raw_REG.
2790 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2791 main to workaround a bug in GDB.
2793 2018-01-12 Tom de Vries <tom@codesourcery.com>
2796 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2798 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2800 PR rtl-optimization/80481
2801 * ira-color.c (get_cap_member): New function.
2802 (allocnos_conflict_by_live_ranges_p): Use it.
2803 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2804 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2806 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2809 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2810 (*saddl_se_1): Ditto.
2812 (*ssubl_se_1): Ditto.
2814 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2816 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2817 rather than wi::to_widest for DR_INITs.
2818 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2819 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2820 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2822 (vect_analyze_group_access_1): Note that here.
2824 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2826 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2827 polynomial type sizes.
2829 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2831 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2832 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2833 (gimple_add_tmp_var): Likewise.
2835 2018-01-12 Martin Liska <mliska@suse.cz>
2837 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2838 (gimple_alloc_sizes): Likewise.
2839 (dump_gimple_statistics): Use PRIu64 in printf format.
2840 * gimple.h: Change uint64_t to int.
2842 2018-01-12 Martin Liska <mliska@suse.cz>
2844 * tree-core.h: Use uint64_t instead of int.
2845 * tree.c (tree_node_counts): Likewise.
2846 (tree_node_sizes): Likewise.
2847 (dump_tree_statistics): Use PRIu64 in printf format.
2849 2018-01-12 Martin Liska <mliska@suse.cz>
2851 * Makefile.in: As qsort_chk is implemented in vec.c, add
2852 vec.o to linkage of gencfn-macros.
2853 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2854 passing the info to record_node_allocation_statistics.
2855 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2857 * ggc-common.c (struct ggc_usage): Add operator== and use
2858 it in operator< and compare function.
2859 * mem-stats.h (struct mem_usage): Likewise.
2860 * vec.c (struct vec_usage): Remove operator< and compare
2861 function. Can be simply inherited.
2863 2018-01-12 Martin Jambor <mjambor@suse.cz>
2866 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2867 * tree-ssa-math-opts.c: Include domwalk.h.
2868 (convert_mult_to_fma_1): New function.
2869 (fma_transformation_info): New type.
2870 (fma_deferring_state): Likewise.
2871 (cancel_fma_deferring): New function.
2872 (result_of_phi): Likewise.
2873 (last_fma_candidate_feeds_initial_phi): Likewise.
2874 (convert_mult_to_fma): Added deferring logic, split actual
2875 transformation to convert_mult_to_fma_1.
2876 (math_opts_dom_walker): New type.
2877 (math_opts_dom_walker::after_dom_children): New method, body moved
2878 here from pass_optimize_widening_mul::execute, added deferring logic
2880 (pass_optimize_widening_mul::execute): Moved most of code to
2881 math_opts_dom_walker::after_dom_children.
2882 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2883 * config/i386/i386.c (ix86_option_override_internal): Added
2884 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2886 2018-01-12 Richard Biener <rguenther@suse.de>
2889 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2890 inline instance vars.
2892 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2895 * config/rx/rx.c (rx_is_restricted_memory_address):
2898 2018-01-12 Richard Biener <rguenther@suse.de>
2900 PR tree-optimization/80846
2901 * target.def (split_reduction): New target hook.
2902 * targhooks.c (default_split_reduction): New function.
2903 * targhooks.h (default_split_reduction): Declare.
2904 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2905 target requests first reduce vectors by combining low and high
2907 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2908 (get_vectype_for_scalar_type_and_size): Export.
2909 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2910 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2911 * doc/tm.texi: Regenerate.
2912 * config/i386/i386.c (ix86_split_reduction): Implement
2913 TARGET_VECTORIZE_SPLIT_REDUCTION.
2915 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2918 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2919 in PIC mode except for TARGET_VXWORKS_RTP.
2920 * config/sparc/sparc.c: Include cfgrtl.h.
2921 (TARGET_INIT_PIC_REG): Define.
2922 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2923 (sparc_pic_register_p): New predicate.
2924 (sparc_legitimate_address_p): Use it.
2925 (sparc_legitimize_pic_address): Likewise.
2926 (sparc_delegitimize_address): Likewise.
2927 (sparc_mode_dependent_address_p): Likewise.
2928 (gen_load_pcrel_sym): Remove 4th parameter.
2929 (load_got_register): Adjust call to above. Remove obsolete stuff.
2930 (sparc_expand_prologue): Do not call load_got_register here.
2931 (sparc_flat_expand_prologue): Likewise.
2932 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2933 (sparc_use_pseudo_pic_reg): New function.
2934 (sparc_init_pic_reg): Likewise.
2935 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2936 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2938 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2940 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2941 Add item for branch_cost.
2943 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2945 PR rtl-optimization/83565
2946 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2947 not extend the result to a larger mode for rotate operations.
2948 (num_sign_bit_copies1): Likewise.
2950 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2953 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2955 Use values-Xc.o for -pedantic.
2956 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2958 2018-01-12 Martin Liska <mliska@suse.cz>
2961 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2963 (possible_polymorphic_call_targets): Use it.
2964 (ipa_devirt): Likewise.
2966 2018-01-12 Martin Liska <mliska@suse.cz>
2968 * profile-count.h (enum profile_quality): Use 0 as invalid
2969 enum value of profile_quality.
2971 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2973 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2974 -mext-string options.
2976 2018-01-12 Richard Biener <rguenther@suse.de>
2978 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2979 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2980 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2982 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2984 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2986 * configure.ac (--with-long-double-format): Add support for the
2987 configuration option to change the default long double format on
2989 * config.gcc (powerpc*-linux*-*): Likewise.
2990 * configure: Regenerate.
2991 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2992 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2993 used without modification.
2995 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2997 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2998 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2999 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3000 MISC_BUILTIN_SPEC_BARRIER.
3001 (rs6000_init_builtins): Likewise.
3002 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3004 (speculation_barrier): New define_insn.
3005 * doc/extend.texi: Document __builtin_speculation_barrier.
3007 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3010 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3011 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3012 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3014 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3015 integral modes instead of "ss" and "sd".
3016 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3017 vectors with 32-bit and 64-bit elements.
3018 (vecdupssescalarmodesuffix): New mode attribute.
3019 (vec_dup<mode>): Use it.
3021 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3024 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3025 frame if argument is passed on stack.
3027 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3030 * ree.c (combine_reaching_defs): Optimize also
3031 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3032 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3034 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3037 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3039 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3042 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3043 after they are computed.
3045 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3047 PR tree-optimization/83695
3048 * gimple-loop-linterchange.cc
3049 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3050 reset cached scev information after interchange.
3051 (pass_linterchange::execute): Remove call to scev_reset_htab.
3053 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3055 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3056 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3057 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3058 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3059 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3060 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3061 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3062 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3063 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3064 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3065 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3066 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3067 (V_lane_reg): Likewise.
3068 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3070 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3071 (vfmal_lane_low<mode>_intrinsic,
3072 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3073 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3074 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3075 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3076 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3077 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3079 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3081 * config/arm/arm-cpus.in (fp16fml): New feature.
3082 (ALL_SIMD): Add fp16fml.
3083 (armv8.2-a): Add fp16fml as an option.
3084 (armv8.3-a): Likewise.
3085 (armv8.4-a): Add fp16fml as part of fp16.
3086 * config/arm/arm.h (TARGET_FP16FML): Define.
3087 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3089 * config/arm/arm-modes.def (V2HF): Define.
3090 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3091 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3092 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3093 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3094 vfmsl_low, vfmsl_high): New set of builtins.
3095 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3096 (vfml_op): New code attribute.
3097 (VFMLHALVES): New int iterator.
3098 (VFML, VFMLSEL): New mode attributes.
3099 (V_reg): Define mapping for V2HF.
3100 (V_hi, V_lo): New mode attributes.
3101 (VF_constraint): Likewise.
3102 (vfml_half, vfml_half_selector): New int attributes.
3103 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3105 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3106 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3108 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3109 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3110 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3111 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3113 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3114 Document new effective target and option set.
3116 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3118 * config/arm/arm-cpus.in (armv8_4): New feature.
3119 (ARMv8_4a): New fgroup.
3120 (armv8.4-a): New arch.
3121 * config/arm/arm-tables.opt: Regenerate.
3122 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3123 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3124 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3125 Add matching rules for -march=armv8.4-a and extensions.
3126 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3128 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3131 * config/rx/rx.md (BW): New mode attribute.
3132 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3134 2018-01-11 Richard Biener <rguenther@suse.de>
3136 PR tree-optimization/83435
3137 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3138 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3139 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3141 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3142 Alan Hayward <alan.hayward@arm.com>
3143 David Sherwood <david.sherwood@arm.com>
3145 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3147 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3148 (aarch64_print_address_internal): Use it to check for a zero offset.
3150 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3151 Alan Hayward <alan.hayward@arm.com>
3152 David Sherwood <david.sherwood@arm.com>
3154 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3155 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3156 Return a poly_int64 rather than a HOST_WIDE_INT.
3157 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3158 rather than a HOST_WIDE_INT.
3159 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3160 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3161 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3162 final_offset from HOST_WIDE_INT to poly_int64.
3163 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3164 to_constant when getting the number of units in an Advanced SIMD
3166 (aarch64_builtin_vectorized_function): Check for a constant number
3168 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3170 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3171 attribute instead of GET_MODE_NUNITS.
3172 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3173 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3174 GET_MODE_SIZE for fixed-size registers.
3175 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3176 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3177 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3178 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3179 (aarch64_print_operand, aarch64_print_address_internal)
3180 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3181 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3182 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3183 Handle polynomial GET_MODE_SIZE.
3184 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3185 wider than SImode without modification.
3186 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3187 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3188 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3189 passing and returning SVE modes.
3190 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3191 rather than GEN_INT.
3192 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3193 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3194 (aarch64_allocate_and_probe_stack_space): Likewise.
3195 (aarch64_layout_frame): Cope with polynomial offsets.
3196 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3197 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3199 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3200 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3201 poly_int64 rather than a HOST_WIDE_INT.
3202 (aarch64_get_separate_components, aarch64_process_components)
3203 (aarch64_expand_prologue, aarch64_expand_epilogue)
3204 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3205 (aarch64_anchor_offset): New function, split out from...
3206 (aarch64_legitimize_address): ...here.
3207 (aarch64_builtin_vectorization_cost): Handle polynomial
3208 TYPE_VECTOR_SUBPARTS.
3209 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3211 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3212 number of elements from the PARALLEL rather than the mode.
3213 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3214 rather than GET_MODE_BITSIZE.
3215 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3216 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3217 (aarch64_expand_vec_perm_const_1): Handle polynomial
3218 d->perm.length () and d->perm elements.
3219 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3220 Apply to_constant to d->perm elements.
3221 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3222 polynomial CONST_VECTOR_NUNITS.
3223 (aarch64_move_pointer): Take amount as a poly_int64 rather
3225 (aarch64_progress_pointer): Avoid temporary variable.
3226 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3227 the mode attribute instead of GET_MODE.
3229 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3230 Alan Hayward <alan.hayward@arm.com>
3231 David Sherwood <david.sherwood@arm.com>
3233 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3234 x exists before using it.
3235 (aarch64_add_constant_internal): Rename to...
3236 (aarch64_add_offset_1): ...this. Replace regnum with separate
3237 src and dest rtxes. Handle the case in which they're different,
3238 including when the offset is zero. Replace scratchreg with an rtx.
3239 Use 2 additions if there is no spare register into which we can
3240 move a 16-bit constant.
3241 (aarch64_add_constant): Delete.
3242 (aarch64_add_offset): Replace reg with separate src and dest
3243 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3244 Use aarch64_add_offset_1.
3245 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3246 an rtx rather than an int. Take the delta as a poly_int64
3247 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3248 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3249 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3250 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3251 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3253 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3254 aarch64_add_constant.
3256 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3258 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3259 Use scalar_float_mode.
3261 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3263 * config/aarch64/aarch64-simd.md
3264 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3265 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3266 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3267 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3268 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3269 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3270 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3271 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3272 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3273 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3275 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3278 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3279 targ_options->x_arm_arch_string is non NULL.
3281 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3283 * config/aarch64/aarch64.h
3284 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3286 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3289 * expmed.c (emit_store_flag_force): Swap if const op0
3290 and change VOIDmode to mode of op0.
3292 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3294 PR rtl-optimization/83761
3295 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3296 than bytes to mode_for_size.
3298 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3301 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3302 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3305 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3308 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3309 when in layout mode.
3310 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3311 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3314 2018-01-10 Michael Collison <michael.collison@arm.com>
3316 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3317 * config/aarch64/aarch64-option-extension.def: Add
3318 AARCH64_OPT_EXTENSION of 'fp16fml'.
3319 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3320 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3321 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3322 * config/aarch64/constraints.md (Ui7): New constraint.
3323 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3324 (VFMLA_SEL_W): Ditto.
3327 (VFMLA16_LOW): New int iterator.
3328 (VFMLA16_HIGH): Ditto.
3329 (UNSPEC_FMLAL): New unspec.
3330 (UNSPEC_FMLSL): Ditto.
3331 (UNSPEC_FMLAL2): Ditto.
3332 (UNSPEC_FMLSL2): Ditto.
3333 (f16mac): New code attribute.
3334 * config/aarch64/aarch64-simd-builtins.def
3335 (aarch64_fmlal_lowv2sf): Ditto.
3336 (aarch64_fmlsl_lowv2sf): Ditto.
3337 (aarch64_fmlalq_lowv4sf): Ditto.
3338 (aarch64_fmlslq_lowv4sf): Ditto.
3339 (aarch64_fmlal_highv2sf): Ditto.
3340 (aarch64_fmlsl_highv2sf): Ditto.
3341 (aarch64_fmlalq_highv4sf): Ditto.
3342 (aarch64_fmlslq_highv4sf): Ditto.
3343 (aarch64_fmlal_lane_lowv2sf): Ditto.
3344 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3345 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3346 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3347 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3348 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3349 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3350 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3351 (aarch64_fmlal_lane_highv2sf): Ditto.
3352 (aarch64_fmlsl_lane_highv2sf): Ditto.
3353 (aarch64_fmlal_laneq_highv2sf): Ditto.
3354 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3355 (aarch64_fmlalq_lane_highv4sf): Ditto.
3356 (aarch64_fmlsl_lane_highv4sf): Ditto.
3357 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3358 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3359 * config/aarch64/aarch64-simd.md:
3360 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3361 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3362 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3363 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3364 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3365 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3366 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3367 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3368 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3369 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3370 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3371 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3372 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3373 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3374 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3375 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3376 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3377 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3378 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3379 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3380 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3381 (vfmlsl_low_u32): Ditto.
3382 (vfmlalq_low_u32): Ditto.
3383 (vfmlslq_low_u32): Ditto.
3384 (vfmlal_high_u32): Ditto.
3385 (vfmlsl_high_u32): Ditto.
3386 (vfmlalq_high_u32): Ditto.
3387 (vfmlslq_high_u32): Ditto.
3388 (vfmlal_lane_low_u32): Ditto.
3389 (vfmlsl_lane_low_u32): Ditto.
3390 (vfmlal_laneq_low_u32): Ditto.
3391 (vfmlsl_laneq_low_u32): Ditto.
3392 (vfmlalq_lane_low_u32): Ditto.
3393 (vfmlslq_lane_low_u32): Ditto.
3394 (vfmlalq_laneq_low_u32): Ditto.
3395 (vfmlslq_laneq_low_u32): Ditto.
3396 (vfmlal_lane_high_u32): Ditto.
3397 (vfmlsl_lane_high_u32): Ditto.
3398 (vfmlal_laneq_high_u32): Ditto.
3399 (vfmlsl_laneq_high_u32): Ditto.
3400 (vfmlalq_lane_high_u32): Ditto.
3401 (vfmlslq_lane_high_u32): Ditto.
3402 (vfmlalq_laneq_high_u32): Ditto.
3403 (vfmlslq_laneq_high_u32): Ditto.
3404 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3405 (AARCH64_FL_FOR_ARCH8_4): New.
3406 (AARCH64_ISA_F16FML): New ISA flag.
3407 (TARGET_F16FML): New feature flag for fp16fml.
3408 (doc/invoke.texi): Document new fp16fml option.
3410 2018-01-10 Michael Collison <michael.collison@arm.com>
3412 * config/aarch64/aarch64-builtins.c:
3413 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3414 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3415 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3416 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3417 (AARCH64_ISA_SHA3): New ISA flag.
3418 (TARGET_SHA3): New feature flag for sha3.
3419 * config/aarch64/iterators.md (sha512_op): New int attribute.
3420 (CRYPTO_SHA512): New int iterator.
3421 (UNSPEC_SHA512H): New unspec.
3422 (UNSPEC_SHA512H2): Ditto.
3423 (UNSPEC_SHA512SU0): Ditto.
3424 (UNSPEC_SHA512SU1): Ditto.
3425 * config/aarch64/aarch64-simd-builtins.def
3426 (aarch64_crypto_sha512hqv2di): New builtin.
3427 (aarch64_crypto_sha512h2qv2di): Ditto.
3428 (aarch64_crypto_sha512su0qv2di): Ditto.
3429 (aarch64_crypto_sha512su1qv2di): Ditto.
3430 (aarch64_eor3qv8hi): Ditto.
3431 (aarch64_rax1qv2di): Ditto.
3432 (aarch64_xarqv2di): Ditto.
3433 (aarch64_bcaxqv8hi): Ditto.
3434 * config/aarch64/aarch64-simd.md:
3435 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3436 (aarch64_crypto_sha512su0qv2di): Ditto.
3437 (aarch64_crypto_sha512su1qv2di): Ditto.
3438 (aarch64_eor3qv8hi): Ditto.
3439 (aarch64_rax1qv2di): Ditto.
3440 (aarch64_xarqv2di): Ditto.
3441 (aarch64_bcaxqv8hi): Ditto.
3442 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3443 (vsha512h2q_u64): Ditto.
3444 (vsha512su0q_u64): Ditto.
3445 (vsha512su1q_u64): Ditto.
3446 (veor3q_u16): Ditto.
3447 (vrax1q_u64): Ditto.
3449 (vbcaxq_u16): Ditto.
3450 * config/arm/types.md (crypto_sha512): New type attribute.
3451 (crypto_sha3): Ditto.
3452 (doc/invoke.texi): Document new sha3 option.
3454 2018-01-10 Michael Collison <michael.collison@arm.com>
3456 * config/aarch64/aarch64-builtins.c:
3457 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3458 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3459 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3460 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3461 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3462 (AARCH64_ISA_SM4): New ISA flag.
3463 (TARGET_SM4): New feature flag for sm4.
3464 * config/aarch64/aarch64-simd-builtins.def
3465 (aarch64_sm3ss1qv4si): Ditto.
3466 (aarch64_sm3tt1aq4si): Ditto.
3467 (aarch64_sm3tt1bq4si): Ditto.
3468 (aarch64_sm3tt2aq4si): Ditto.
3469 (aarch64_sm3tt2bq4si): Ditto.
3470 (aarch64_sm3partw1qv4si): Ditto.
3471 (aarch64_sm3partw2qv4si): Ditto.
3472 (aarch64_sm4eqv4si): Ditto.
3473 (aarch64_sm4ekeyqv4si): Ditto.
3474 * config/aarch64/aarch64-simd.md:
3475 (aarch64_sm3ss1qv4si): Ditto.
3476 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3477 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3478 (aarch64_sm4eqv4si): Ditto.
3479 (aarch64_sm4ekeyqv4si): Ditto.
3480 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3481 (sm3part_op): Ditto.
3482 (CRYPTO_SM3TT): Ditto.
3483 (CRYPTO_SM3PART): Ditto.
3484 (UNSPEC_SM3SS1): New unspec.
3485 (UNSPEC_SM3TT1A): Ditto.
3486 (UNSPEC_SM3TT1B): Ditto.
3487 (UNSPEC_SM3TT2A): Ditto.
3488 (UNSPEC_SM3TT2B): Ditto.
3489 (UNSPEC_SM3PARTW1): Ditto.
3490 (UNSPEC_SM3PARTW2): Ditto.
3491 (UNSPEC_SM4E): Ditto.
3492 (UNSPEC_SM4EKEY): Ditto.
3493 * config/aarch64/constraints.md (Ui2): New constraint.
3494 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3495 * config/arm/types.md (crypto_sm3): New type attribute.
3496 (crypto_sm4): Ditto.
3497 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3498 (vsm3tt1aq_u32): Ditto.
3499 (vsm3tt1bq_u32): Ditto.
3500 (vsm3tt2aq_u32): Ditto.
3501 (vsm3tt2bq_u32): Ditto.
3502 (vsm3partw1q_u32): Ditto.
3503 (vsm3partw2q_u32): Ditto.
3504 (vsm4eq_u32): Ditto.
3505 (vsm4ekeyq_u32): Ditto.
3506 (doc/invoke.texi): Document new sm4 option.
3508 2018-01-10 Michael Collison <michael.collison@arm.com>
3510 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3511 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3512 (AARCH64_FL_FOR_ARCH8_4): New.
3513 (AARCH64_FL_V8_4): New flag.
3514 (doc/invoke.texi): Document new armv8.4-a option.
3516 2018-01-10 Michael Collison <michael.collison@arm.com>
3518 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3519 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3520 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3521 * config/aarch64/aarch64-option-extension.def: Add
3522 AARCH64_OPT_EXTENSION of 'sha2'.
3523 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3524 (crypto): Disable sha2 and aes if crypto disabled.
3525 (crypto): Enable aes and sha2 if enabled.
3526 (simd): Disable sha2 and aes if simd disabled.
3527 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3529 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3530 (TARGET_SHA2): New feature flag for sha2.
3531 (TARGET_AES): New feature flag for aes.
3532 * config/aarch64/aarch64-simd.md:
3533 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3534 conditional on TARGET_AES.
3535 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3536 (aarch64_crypto_sha1hsi): Make pattern conditional
3538 (aarch64_crypto_sha1hv4si): Ditto.
3539 (aarch64_be_crypto_sha1hv4si): Ditto.
3540 (aarch64_crypto_sha1su1v4si): Ditto.
3541 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3542 (aarch64_crypto_sha1su0v4si): Ditto.
3543 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3544 (aarch64_crypto_sha256su0v4si): Ditto.
3545 (aarch64_crypto_sha256su1v4si): Ditto.
3546 (doc/invoke.texi): Document new aes and sha2 options.
3548 2018-01-10 Martin Sebor <msebor@redhat.com>
3550 PR tree-optimization/83781
3551 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3554 2018-01-11 Martin Sebor <msebor@gmail.com>
3555 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3557 PR tree-optimization/83501
3558 PR tree-optimization/81703
3560 * tree-ssa-strlen.c (get_string_cst): Rename...
3561 (get_string_len): ...to this. Handle global constants.
3562 (handle_char_store): Adjust.
3564 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3565 Jim Wilson <jimw@sifive.com>
3567 * config/riscv/riscv-protos.h (riscv_output_return): New.
3568 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3569 (riscv_attribute_table, riscv_output_return),
3570 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3571 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3572 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3574 (riscv_expand_prologue): Add early return for naked function.
3575 (riscv_expand_epilogue): Likewise.
3576 (riscv_function_ok_for_sibcall): Return false for naked function.
3577 (riscv_set_current_function): New.
3578 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3579 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3580 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3581 * doc/extend.texi (RISC-V Function Attributes): New.
3583 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3585 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3586 check for 128-bit long double before checking TCmode.
3587 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3588 128-bit long doubles before checking TFmode or TCmode.
3589 (FLOAT128_IBM_P): Likewise.
3591 2018-01-10 Martin Sebor <msebor@redhat.com>
3593 PR tree-optimization/83671
3594 * builtins.c (c_strlen): Unconditionally return zero for the empty
3596 Use -Warray-bounds for warnings.
3597 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3598 for non-constant array indices with COMPONENT_REF, arrays of
3599 arrays, and pointers to arrays.
3600 (gimple_fold_builtin_strlen): Determine and set length range for
3601 non-constant character arrays.
3603 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3606 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3609 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3611 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3613 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3616 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3617 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3618 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3619 indexed_or_indirect_operand predicate.
3620 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3621 (*vsx_le_perm_load_v8hi): Likewise.
3622 (*vsx_le_perm_load_v16qi): Likewise.
3623 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3624 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3625 (*vsx_le_perm_store_v8hi): Likewise.
3626 (*vsx_le_perm_store_v16qi): Likewise.
3627 (eight unnamed splitters): Likewise.
3629 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3631 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3632 * config/rs6000/emmintrin.h: Likewise.
3633 * config/rs6000/mmintrin.h: Likewise.
3634 * config/rs6000/xmmintrin.h: Likewise.
3636 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3639 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3641 * tree.c (tree_nop_conversion): Return true for location wrapper
3643 (maybe_wrap_with_location): New function.
3644 (selftest::check_strip_nops): New function.
3645 (selftest::test_location_wrappers): New function.
3646 (selftest::tree_c_tests): Call it.
3647 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3648 (maybe_wrap_with_location): New decl.
3649 (EXPR_LOCATION_WRAPPER_P): New macro.
3650 (location_wrapper_p): New inline function.
3651 (tree_strip_any_location_wrapper): New inline function.
3653 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3656 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3657 stack_realign_offset for the largest alignment of stack slot
3659 (ix86_find_max_used_stack_alignment): New function.
3660 (ix86_finalize_stack_frame_flags): Use it. Set
3661 max_used_stack_alignment if we don't realign stack.
3662 * config/i386/i386.h (machine_function): Add
3663 max_used_stack_alignment.
3665 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3667 * config/arm/arm.opt (-mbranch-cost): New option.
3668 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3671 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3674 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3675 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3677 2018-01-10 Richard Biener <rguenther@suse.de>
3680 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3681 early out so it also covers the case where we have a non-NULL
3684 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3686 PR tree-optimization/83753
3687 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3688 for non-strided grouped accesses if the number of elements is 1.
3690 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3693 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3694 * i386.h (TARGET_USE_GATHER): Define.
3695 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3697 2018-01-10 Martin Liska <mliska@suse.cz>
3700 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3701 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3703 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3704 CLEANUP_NO_PARTITIONING is not set.
3706 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3708 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3709 for vectors, as a partial revert of r254296.
3710 * rtl.h (const_vec_p): Delete.
3711 (const_vec_duplicate_p): Don't test for vector CONSTs.
3712 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3713 * expmed.c (make_tree): Likewise.
3716 * common.md (E, F): Use CONSTANT_P instead of checking for
3718 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3719 checking for CONST_VECTOR.
3721 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3724 * predict.c (force_edge_cold): Handle in more sane way edges
3727 2018-01-09 Carl Love <cel@us.ibm.com>
3729 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3731 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3732 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3733 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3734 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3735 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3736 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3737 * config/rs6000/rs6000-protos.h: Add extern defition for
3738 rs6000_generate_float2_double_code.
3739 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3741 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3742 (float2_v2df): Add define_expand.
3744 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3747 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3748 op_mode in the force_to_mode call.
3750 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3752 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3753 instead of checking each element individually.
3754 (aarch64_evpc_uzp): Likewise.
3755 (aarch64_evpc_zip): Likewise.
3756 (aarch64_evpc_ext): Likewise.
3757 (aarch64_evpc_rev): Likewise.
3758 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3759 instead of checking each element individually. Return true without
3761 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3762 whether all selected elements come from the same input, instead of
3763 checking each element individually. Remove calls to gen_rtx_REG,
3764 start_sequence and end_sequence and instead assert that no rtl is
3767 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3769 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3770 order of HIGH and CONST checks.
3772 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3774 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3775 if the destination isn't an SSA_NAME.
3777 2018-01-09 Richard Biener <rguenther@suse.de>
3779 PR tree-optimization/83668
3780 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3782 (canonicalize_loop_form): ... here, renamed from ...
3783 (canonicalize_loop_closed_ssa_form): ... this and amended to
3784 swap successor edges for loop exit blocks to make us use
3785 the RPO order we need for initial schedule generation.
3787 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3789 PR tree-optimization/64811
3790 * match.pd: When optimizing comparisons with Inf, avoid
3791 introducing or losing exceptions from comparisons with NaN.
3793 2018-01-09 Martin Liska <mliska@suse.cz>
3796 * asan.c (shadow_mem_size): Add gcc_assert.
3798 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3800 Don't save registers in main().
3803 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3804 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3805 * config/avr/avr.c (avr_set_current_function): Don't error if
3806 naked, OS_task or OS_main are specified at the same time.
3807 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3809 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3811 * common/config/avr/avr-common.c (avr_option_optimization_table):
3812 Switch on -mmain-is-OS_task for optimizing compilations.
3814 2018-01-09 Richard Biener <rguenther@suse.de>
3816 PR tree-optimization/83572
3817 * graphite.c: Include cfganal.h.
3818 (graphite_transform_loops): Connect infinite loops to exit
3819 and remove fake edges at the end.
3821 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3823 * ipa-inline.c (edge_badness): Revert accidental checkin.
3825 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3828 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3829 symbols; not inline clones.
3831 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3834 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3835 hard registers. Formatting fixes.
3837 PR preprocessor/83722
3838 * gcc.c (try_generate_repro): Pass
3839 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3840 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3843 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3844 Kito Cheng <kito.cheng@gmail.com>
3846 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3847 (riscv_leaf_function_p): Delete.
3848 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3850 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3852 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3854 (do_ifelse): New function.
3855 (do_isel): New function.
3856 (do_sub3): New function.
3857 (do_add3): New function.
3858 (do_load_mask_compare): New function.
3859 (do_overlap_load_compare): New function.
3860 (expand_compare_loop): New function.
3861 (expand_block_compare): Call expand_compare_loop() when appropriate.
3862 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3864 (-mblock-compare-inline-loop-limit): New option.
3866 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3869 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3870 Reverse order of second and third operands in first alternative.
3871 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3872 of first and second elements in UNSPEC_VPERMR vector.
3873 (altivec_expand_vec_perm_le): Likewise.
3875 2017-01-08 Jeff Law <law@redhat.com>
3877 PR rtl-optimizatin/81308
3878 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3879 (process_switch): If group_case_labels makes a change, then set
3881 (pass_convert_switch::execute): If a switch is converted, then
3882 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3884 PR rtl-optimization/81308
3885 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3888 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3890 PR target/83663 - Revert r255946
3891 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3892 generation for cases where splatting a value is not useful.
3893 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3894 across a vec_duplicate and a paradoxical subreg forming a vector
3895 mode to a vec_concat.
3897 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3899 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3900 -march=armv8.3-a variants.
3901 * config/arm/t-multilib: Likewise.
3902 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3904 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3906 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3908 (cceq_ior_compare_complement): Give it a name so I can use it, and
3909 change boolean_or_operator predicate to boolean_operator so it can
3910 be used to generate a crand.
3911 (eqne): New code iterator.
3912 (bd/bd_neg): New code_attrs.
3913 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3914 a single define_insn.
3915 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3916 decrement (bdnzt/bdnzf/bdzt/bdzf).
3917 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3918 with the new names of the branch decrement patterns, and added the
3919 names of the branch decrement conditional patterns.
3921 2018-01-08 Richard Biener <rguenther@suse.de>
3923 PR tree-optimization/83563
3924 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3927 2018-01-08 Richard Biener <rguenther@suse.de>
3930 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3932 2018-01-08 Richard Biener <rguenther@suse.de>
3934 PR tree-optimization/83685
3935 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3936 references to abnormals.
3938 2018-01-08 Richard Biener <rguenther@suse.de>
3941 * dwarf2out.c (output_indirect_strings): Handle empty
3942 skeleton_debug_str_hash.
3943 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3945 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3947 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3948 (emit_store_direct): Likewise.
3949 (arc_trampoline_adjust_address): Likewise.
3950 (arc_asm_trampoline_template): New function.
3951 (arc_initialize_trampoline): Use asm_trampoline_template.
3952 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3953 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3954 * config/arc/arc.md (flush_icache): Delete pattern.
3956 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3958 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3959 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3962 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3965 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3966 by not USED_FOR_TARGET.
3967 (make_pass_resolve_sw_modes): Likewise.
3969 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3971 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3974 2018-01-08 Richard Biener <rguenther@suse.de>
3977 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3979 2018-01-08 Richard Biener <rguenther@suse.de>
3982 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3984 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3987 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3988 basic blocks with a small number of successors.
3989 (convert_control_dep_chain_into_preds): Improve handling of
3991 (dump_predicates): Split apart into...
3992 (dump_pred_chain): ...here...
3993 (dump_pred_info): ...and here.
3994 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3995 (can_chain_union_be_invalidated_p): Improve check for invalidation
3997 (uninit_uses_cannot_happen): Avoid unnecessary if
3998 convert_control_dep_chain_into_preds yielded nothing.
4000 2018-01-06 Martin Sebor <msebor@redhat.com>
4002 PR tree-optimization/83640
4003 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4004 subtracting negative offset from size.
4005 (builtin_access::overlap): Adjust offset bounds of the access to fall
4006 within the size of the object if possible.
4008 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4010 PR rtl-optimization/83699
4011 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4012 extract_bit_field_as_subreg to cases in which the extracted
4013 value is also a vector.
4015 * lra-constraints.c (process_alt_operands): Test for the equivalence
4016 substitutions when detecting a possible reload cycle.
4018 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4021 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4022 by default if flag_selective_schedling{,2}. Formatting fixes.
4024 PR rtl-optimization/83682
4025 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4026 if it has non-VECTOR_MODE element mode.
4027 (vec_duplicate_p): Likewise.
4030 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4031 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4033 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4036 * config/i386/i386-builtin.def
4037 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4038 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4039 Require also OPTION_MASK_ISA_AVX512F in addition to
4040 OPTION_MASK_ISA_GFNI.
4041 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4042 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4043 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4044 to OPTION_MASK_ISA_GFNI.
4045 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4046 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4047 OPTION_MASK_ISA_AVX512BW.
4048 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4049 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4050 addition to OPTION_MASK_ISA_GFNI.
4051 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4052 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4053 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4054 to OPTION_MASK_ISA_GFNI.
4055 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4056 a requirement for all ISAs rather than any of them with a few
4058 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4060 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4061 bitmasks to be enabled with 3 exceptions, instead of requiring any
4062 enabled ISA with lots of exceptions.
4063 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4064 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4065 Change avx512bw in isa attribute to avx512f.
4066 * config/i386/sgxintrin.h: Add license boilerplate.
4067 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4068 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4069 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4070 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4072 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4073 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4074 temporarily sse2 rather than sse if not enabled already.
4077 * config/i386/sse.md (VI248_VLBW): Rename to ...
4078 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4079 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4080 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4081 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4082 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4083 mode iterator instead of VI248_VLBW.
4085 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4087 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4088 (record_modified): Skip clobbers; add debug output.
4089 (param_change_prob): Use sreal frequencies.
4091 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4093 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4094 punt for user-aligned variables.
4096 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4098 * tree-chrec.c (chrec_contains_symbols): Return true for
4101 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4104 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4105 of (x|y) == x for BICS pattern.
4107 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4109 PR tree-optimization/83605
4110 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4111 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4114 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4116 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4117 * config/epiphany/rtems.h: New file.
4119 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4120 Uros Bizjak <ubizjak@gmail.com>
4123 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4124 QIreg_operand instead of register_operand predicate.
4125 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4126 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4127 comments instead of -fmitigate[-_]rop.
4129 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4132 * cgraphunit.c (symbol_table::compile): Switch to text_section
4133 before calling assembly_start debug hook.
4134 * run-rtl-passes.c (run_rtl_passes): Likewise.
4137 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4139 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4140 range_int_cst_p rather than !symbolic_range_p before calling
4141 extract_range_from_multiplicative_op_1.
4143 2017-01-04 Jeff Law <law@redhat.com>
4145 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4146 redundant test in assertion.
4148 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4150 * doc/rtl.texi: Document machine_mode wrapper classes.
4152 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4154 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4157 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4159 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4160 the VEC_PERM_EXPR fold to fail.
4162 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4165 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4166 to switched_sections.
4168 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4171 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4174 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4177 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4178 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4180 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4183 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4184 is BLKmode and bitpos not zero or mode change is needed.
4186 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4189 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4192 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4195 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4196 instead of MULT rtx. Update all corresponding splitters.
4198 (*ssub<modesuffix>): Ditto.
4200 (*cmp_sadd_di): Update split patterns.
4201 (*cmp_sadd_si): Ditto.
4202 (*cmp_sadd_sidi): Ditto.
4203 (*cmp_ssub_di): Ditto.
4204 (*cmp_ssub_si): Ditto.
4205 (*cmp_ssub_sidi): Ditto.
4206 * config/alpha/predicates.md (const23_operand): New predicate.
4207 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4208 Look for ASHIFT, not MULT inner operand.
4209 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4211 2018-01-04 Martin Liska <mliska@suse.cz>
4213 PR gcov-profile/83669
4214 * gcov.c (output_intermediate_file): Add version to intermediate
4216 * doc/gcov.texi: Document new field 'version' in intermediate
4217 file format. Fix location of '-k' option of gcov command.
4219 2018-01-04 Martin Liska <mliska@suse.cz>
4222 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4224 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4226 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4228 2018-01-03 Martin Sebor <msebor@redhat.com>
4230 PR tree-optimization/83655
4231 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4232 checking calls with invalid arguments.
4234 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4236 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4237 (vectorizable_mask_load_store): Delete.
4238 (vectorizable_call): Return false for masked loads and stores.
4239 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4240 instead of gimple_assign_rhs1.
4241 (vectorizable_load): Handle IFN_MASK_LOAD.
4242 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4244 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4246 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4248 (vectorizable_mask_load_store): ...here.
4249 (vectorizable_load): ...and here.
4251 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4253 * tree-vect-stmts.c (vect_build_all_ones_mask)
4254 (vect_build_zero_merge_argument): New functions, split out from...
4255 (vectorizable_load): ...here.
4257 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4259 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4261 (vectorizable_mask_load_store): ...here.
4262 (vectorizable_store): ...and here.
4264 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4266 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4268 (vectorizable_mask_load_store): ...here.
4270 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4272 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4273 (vect_model_store_cost): Take a vec_load_store_type instead of a
4275 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4276 (vect_model_store_cost): Take a vec_load_store_type instead of a
4278 (vectorizable_mask_load_store): Update accordingly.
4279 (vectorizable_store): Likewise.
4280 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4282 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4284 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4285 IFN_MASK_LOAD calls here rather than...
4286 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4288 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4289 Alan Hayward <alan.hayward@arm.com>
4290 David Sherwood <david.sherwood@arm.com>
4292 * expmed.c (extract_bit_field_1): For vector extracts,
4293 fall back to extract_bit_field_as_subreg if vec_extract
4296 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4297 Alan Hayward <alan.hayward@arm.com>
4298 David Sherwood <david.sherwood@arm.com>
4300 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4301 they are variable or constant sized.
4302 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4303 slots for constant-sized data.
4305 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4306 Alan Hayward <alan.hayward@arm.com>
4307 David Sherwood <david.sherwood@arm.com>
4309 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4310 handling COND_EXPRs with boolean comparisons, try to find a better
4311 basis for the mask type than the boolean itself.
4313 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4315 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4316 is calculated and how it can be overridden.
4317 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4318 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4320 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4323 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4324 Alan Hayward <alan.hayward@arm.com>
4325 David Sherwood <david.sherwood@arm.com>
4327 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4328 Remove the mode argument.
4329 (aarch64_simd_valid_immediate): Remove the mode and inverse
4331 * config/aarch64/iterators.md (bitsize): New iterator.
4332 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4333 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4334 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4335 aarch64_simd_valid_immediate.
4336 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4337 (aarch64_reg_or_bic_imm): Likewise.
4338 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4339 with an insn_type enum and msl with a modifier_type enum.
4340 Replace element_width with a scalar_mode. Change the shift
4341 to unsigned int. Add constructors for scalar_float_mode and
4342 scalar_int_mode elements.
4343 (aarch64_vect_float_const_representable_p): Delete.
4344 (aarch64_can_const_movi_rtx_p)
4345 (aarch64_simd_scalar_immediate_valid_for_move)
4346 (aarch64_simd_make_constant): Update call to
4347 aarch64_simd_valid_immediate.
4348 (aarch64_advsimd_valid_immediate_hs): New function.
4349 (aarch64_advsimd_valid_immediate): Likewise.
4350 (aarch64_simd_valid_immediate): Remove mode and inverse
4351 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4352 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4353 and aarch64_float_const_representable_p on the result.
4354 (aarch64_output_simd_mov_immediate): Remove mode argument.
4355 Update call to aarch64_simd_valid_immediate and use of
4356 simd_immediate_info.
4357 (aarch64_output_scalar_simd_mov_immediate): Update call
4360 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4361 Alan Hayward <alan.hayward@arm.com>
4362 David Sherwood <david.sherwood@arm.com>
4364 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4365 (mode_nunits): Likewise CONST_MODE_NUNITS.
4366 * machmode.def (ADJUST_NUNITS): Document.
4367 * genmodes.c (mode_data::need_nunits_adj): New field.
4368 (blank_mode): Update accordingly.
4369 (adj_nunits): New variable.
4370 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4372 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4373 listed in adj_nunits.
4374 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4375 listed in adj_nunits. Don't emit case statements for such modes.
4376 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4377 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4378 nothing if adj_nunits is nonnull.
4379 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4380 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4381 (emit_mode_fbit): Update use of print_maybe_const_decl.
4382 (emit_move_size): Likewise. Treat the array as non-const
4384 (emit_mode_adjustments): Handle adj_nunits.
4386 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4388 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4389 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4390 (VECTOR_MODES): Use it.
4391 (make_vector_modes): Take the prefix as an argument.
4393 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4394 Alan Hayward <alan.hayward@arm.com>
4395 David Sherwood <david.sherwood@arm.com>
4397 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4398 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4399 for MODE_VECTOR_BOOL.
4400 * machmode.def (VECTOR_BOOL_MODE): Document.
4401 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4402 (make_vector_bool_mode): New function.
4403 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4405 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4406 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4408 * stor-layout.c (int_mode_for_mode): Likewise.
4409 * tree.c (build_vector_type_for_mode): Likewise.
4410 * varasm.c (output_constant_pool_2): Likewise.
4411 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4412 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4413 for MODE_VECTOR_BOOL.
4414 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4415 of mode class checks.
4416 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4417 instead of a list of mode class checks.
4418 (expand_vector_scalar_condition): Likewise.
4419 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4421 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4422 Alan Hayward <alan.hayward@arm.com>
4423 David Sherwood <david.sherwood@arm.com>
4425 * machmode.h (mode_size): Change from unsigned short to
4427 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4428 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4429 or if measurement_type is not polynomial.
4430 (fixed_size_mode::includes_p): Check for constant-sized modes.
4431 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4432 return a poly_uint16 rather than an unsigned short.
4433 (emit_mode_size): Change the type of mode_size from unsigned short
4434 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4435 (emit_mode_adjustments): Cope with polynomial vector sizes.
4436 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4438 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4440 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4441 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4442 * caller-save.c (setup_save_areas): Likewise.
4443 (replace_reg_with_saved_mem): Likewise.
4444 * calls.c (emit_library_call_value_1): Likewise.
4445 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4446 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4447 (gen_lowpart_for_combine): Likewise.
4448 * convert.c (convert_to_integer_1): Likewise.
4449 * cse.c (equiv_constant, cse_insn): Likewise.
4450 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4451 (cselib_subst_to_values): Likewise.
4452 * dce.c (word_dce_process_block): Likewise.
4453 * df-problems.c (df_word_lr_mark_ref): Likewise.
4454 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4455 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4456 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4457 (rtl_for_decl_location): Likewise.
4458 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4459 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4460 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4461 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4462 (expand_expr_real_1): Likewise.
4463 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4464 (pad_below): Likewise.
4465 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4466 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4467 * ira.c (get_subreg_tracking_sizes): Likewise.
4468 * ira-build.c (ira_create_allocno_objects): Likewise.
4469 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4470 (ira_sort_regnos_for_alter_reg): Likewise.
4471 * ira-costs.c (record_operand_costs): Likewise.
4472 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4473 (resolve_simple_move): Likewise.
4474 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4475 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4476 (lra_constraints): Likewise.
4477 (CONST_POOL_OK_P): Reject variable-sized modes.
4478 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4479 (add_pseudo_to_slot, lra_spill): Likewise.
4480 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4481 * optabs-query.c (get_best_extraction_insn): Likewise.
4482 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4483 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4484 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4485 * recog.c (offsettable_address_addr_space_p): Likewise.
4486 * regcprop.c (maybe_mode_change): Likewise.
4487 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4488 * regrename.c (build_def_use): Likewise.
4489 * regstat.c (dump_reg_info): Likewise.
4490 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4491 (find_reloads, find_reloads_subreg_address): Likewise.
4492 * reload1.c (eliminate_regs_1): Likewise.
4493 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4494 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4495 (simplify_binary_operation_1, simplify_subreg): Likewise.
4496 * targhooks.c (default_function_arg_padding): Likewise.
4497 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4498 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4499 (verify_gimple_assign_ternary): Likewise.
4500 * tree-inline.c (estimate_move_cost): Likewise.
4501 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4502 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4503 (get_address_cost_ainc): Likewise.
4504 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4505 (vect_supportable_dr_alignment): Likewise.
4506 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4507 (vectorizable_reduction): Likewise.
4508 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4509 (vectorizable_operation, vectorizable_load): Likewise.
4510 * tree.c (build_same_sized_truth_vector_type): Likewise.
4511 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4512 * var-tracking.c (emit_note_insn_var_location): Likewise.
4513 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4514 (ADDR_VEC_ALIGN): Likewise.
4516 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4517 Alan Hayward <alan.hayward@arm.com>
4518 David Sherwood <david.sherwood@arm.com>
4520 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4522 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4523 or if measurement_type is polynomial.
4524 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4525 * combine.c (make_extraction): Likewise.
4526 * dse.c (find_shift_sequence): Likewise.
4527 * dwarf2out.c (mem_loc_descriptor): Likewise.
4528 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4529 (extract_bit_field, extract_low_bits): Likewise.
4530 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4531 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4532 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4533 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4534 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4535 * reload.c (find_reloads): Likewise.
4536 * reload1.c (alter_reg): Likewise.
4537 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4538 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4539 * tree-if-conv.c (predicate_mem_writes): Likewise.
4540 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4541 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4542 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4543 * valtrack.c (dead_debug_insert_temp): Likewise.
4544 * varasm.c (mergeable_constant_section): Likewise.
4545 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4547 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4548 Alan Hayward <alan.hayward@arm.com>
4549 David Sherwood <david.sherwood@arm.com>
4551 * expr.c (expand_assignment): Cope with polynomial mode sizes
4552 when assigning to a CONCAT.
4554 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4555 Alan Hayward <alan.hayward@arm.com>
4556 David Sherwood <david.sherwood@arm.com>
4558 * machmode.h (mode_precision): Change from unsigned short to
4560 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4562 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4563 or if measurement_type is not polynomial.
4564 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4565 in which the mode is already known to be a scalar_int_mode.
4566 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4567 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4569 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4570 for GET_MODE_PRECISION.
4571 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4572 for GET_MODE_PRECISION.
4573 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4575 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4576 (expand_field_assignment, make_extraction): Likewise.
4577 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4578 (get_last_value): Likewise.
4579 * convert.c (convert_to_integer_1): Likewise.
4580 * cse.c (cse_insn): Likewise.
4581 * expr.c (expand_expr_real_1): Likewise.
4582 * lra-constraints.c (simplify_operand_subreg): Likewise.
4583 * optabs-query.c (can_atomic_load_p): Likewise.
4584 * optabs.c (expand_atomic_load): Likewise.
4585 (expand_atomic_store): Likewise.
4586 * ree.c (combine_reaching_defs): Likewise.
4587 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4588 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4589 * tree.h (type_has_mode_precision_p): Likewise.
4590 * ubsan.c (instrument_si_overflow): Likewise.
4592 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4593 Alan Hayward <alan.hayward@arm.com>
4594 David Sherwood <david.sherwood@arm.com>
4596 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4597 polynomial numbers of units.
4598 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4599 (valid_vector_subparts_p): New function.
4600 (build_vector_type): Remove temporary shim and take the number
4601 of units as a poly_uint64 rather than an int.
4602 (build_opaque_vector_type): Take the number of units as a
4603 poly_uint64 rather than an int.
4604 * tree.c (build_vector_from_ctor): Handle polynomial
4605 TYPE_VECTOR_SUBPARTS.
4606 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4607 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4608 (build_vector_from_val): If the number of units is variable,
4609 use build_vec_duplicate_cst for constant operands and
4610 VEC_DUPLICATE_EXPR otherwise.
4611 (make_vector_type): Remove temporary is_constant ().
4612 (build_vector_type, build_opaque_vector_type): Take the number of
4613 units as a poly_uint64 rather than an int.
4614 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4616 * cfgexpand.c (expand_debug_expr): Likewise.
4617 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4618 (store_constructor, expand_expr_real_1): Likewise.
4619 (const_scalar_mask_from_tree): Likewise.
4620 * fold-const-call.c (fold_const_reduction): Likewise.
4621 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4622 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4623 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4624 (fold_relational_const): Likewise.
4625 (native_interpret_vector): Likewise. Change the size from an
4626 int to an unsigned int.
4627 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4628 TYPE_VECTOR_SUBPARTS.
4629 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4630 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4631 duplicating a non-constant operand into a variable-length vector.
4632 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4633 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4634 * ipa-icf.c (sem_variable::equals): Likewise.
4635 * match.pd: Likewise.
4636 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4637 * print-tree.c (print_node): Likewise.
4638 * stor-layout.c (layout_type): Likewise.
4639 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4640 * tree-cfg.c (verify_gimple_comparison): Likewise.
4641 (verify_gimple_assign_binary): Likewise.
4642 (verify_gimple_assign_ternary): Likewise.
4643 (verify_gimple_assign_single): Likewise.
4644 * tree-pretty-print.c (dump_generic_node): Likewise.
4645 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4646 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4647 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4648 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4649 (vect_shift_permute_load_chain): Likewise.
4650 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4651 (expand_vector_condition, optimize_vector_constructor): Likewise.
4652 (lower_vec_perm, get_compute_type): Likewise.
4653 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4654 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4655 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4656 (vect_recog_mask_conversion_pattern): Likewise.
4657 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4658 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4659 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4660 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4661 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4662 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4663 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4664 (supportable_widening_operation): Likewise.
4665 (supportable_narrowing_operation): Likewise.
4666 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4668 * varasm.c (output_constant): Likewise.
4670 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4671 Alan Hayward <alan.hayward@arm.com>
4672 David Sherwood <david.sherwood@arm.com>
4674 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4675 so that both the length == 3 and length != 3 cases set up their
4676 own permute vectors. Add comments explaining why we know the
4677 number of elements is constant.
4678 (vect_permute_load_chain): Likewise.
4680 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4681 Alan Hayward <alan.hayward@arm.com>
4682 David Sherwood <david.sherwood@arm.com>
4684 * machmode.h (mode_nunits): Change from unsigned char to
4686 (ONLY_FIXED_SIZE_MODES): New macro.
4687 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4688 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4689 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4691 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4692 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4693 or if measurement_type is not polynomial.
4694 * genmodes.c (ZERO_COEFFS): New macro.
4695 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4697 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4698 Use ZERO_COEFFS when emitting initializers.
4699 * data-streamer.h (bp_pack_poly_value): New function.
4700 (bp_unpack_poly_value): Likewise.
4701 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4702 for GET_MODE_NUNITS.
4703 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4704 for GET_MODE_NUNITS.
4705 * tree.c (make_vector_type): Remove temporary shim and make
4706 the real function take the number of units as a poly_uint64
4708 (build_vector_type_for_mode): Handle polynomial nunits.
4709 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4710 * emit-rtl.c (const_vec_series_p_1): Likewise.
4711 (gen_rtx_CONST_VECTOR): Likewise.
4712 * fold-const.c (test_vec_duplicate_folding): Likewise.
4713 * genrecog.c (validate_pattern): Likewise.
4714 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4715 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4716 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4717 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4718 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4719 * rtlanal.c (subreg_get_info): Likewise.
4720 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4721 (vect_grouped_load_supported): Likewise.
4722 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4723 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4724 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4725 (simplify_const_unary_operation, simplify_binary_operation_1)
4726 (simplify_const_binary_operation, simplify_ternary_operation)
4727 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4728 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4729 instead of CONST_VECTOR_NUNITS.
4730 * varasm.c (output_constant_pool_2): Likewise.
4731 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4732 explicit-encoded elements in the XVEC for variable-length vectors.
4734 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4736 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4738 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4739 Alan Hayward <alan.hayward@arm.com>
4740 David Sherwood <david.sherwood@arm.com>
4742 * coretypes.h (fixed_size_mode): Declare.
4743 (fixed_size_mode_pod): New typedef.
4744 * builtins.h (target_builtins::x_apply_args_mode)
4745 (target_builtins::x_apply_result_mode): Change type to
4746 fixed_size_mode_pod.
4747 * builtins.c (apply_args_size, apply_result_size, result_vector)
4748 (expand_builtin_apply_args_1, expand_builtin_apply)
4749 (expand_builtin_return): Update accordingly.
4751 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4753 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4754 * cselib.c (cselib_hash_rtx): Likewise.
4755 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4756 CONST_VECTOR encoding.
4758 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4759 Jeff Law <law@redhat.com>
4762 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4763 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4764 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4765 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4768 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4769 explicitly probe *sp in a noreturn function if there were any callee
4770 register saves or frame pointer is needed.
4772 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4775 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4776 BLKmode for ternary, binary or unary expressions.
4779 * var-tracking.c (delete_vta_debug_insn): New inline function.
4780 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4781 insns from get_insns () to NULL instead of each bb separately.
4782 Use delete_vta_debug_insn. No longer static.
4783 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4784 delete_vta_debug_insns callers.
4785 * rtl.h (delete_vta_debug_insns): Declare.
4786 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4787 instead of variable_tracking_main.
4789 2018-01-03 Martin Sebor <msebor@redhat.com>
4791 PR tree-optimization/83603
4792 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4793 arguments past the endof the argument list in functions declared
4794 without a prototype.
4795 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4796 Avoid checking when arguments are null.
4798 2018-01-03 Martin Sebor <msebor@redhat.com>
4801 * doc/extend.texi (attribute const): Fix a typo.
4802 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4803 issuing -Wsuggest-attribute for void functions.
4805 2018-01-03 Martin Sebor <msebor@redhat.com>
4807 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4808 offset_int::from instead of wide_int::to_shwi.
4809 (maybe_diag_overlap): Remove assertion.
4810 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4811 * gimple-ssa-sprintf.c (format_directive): Same.
4812 (parse_directive): Same.
4813 (sprintf_dom_walker::compute_format_length): Same.
4814 (try_substitute_return_value): Same.
4816 2017-01-03 Jeff Law <law@redhat.com>
4819 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4820 non-constant residual for zero at runtime and avoid probing in
4821 that case. Reorganize code for trailing problem to mirror handling
4824 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4826 PR tree-optimization/83501
4827 * tree-ssa-strlen.c (get_string_cst): New.
4828 (handle_char_store): Call get_string_cst.
4830 2018-01-03 Martin Liska <mliska@suse.cz>
4832 PR tree-optimization/83593
4833 * tree-ssa-strlen.c: Include tree-cfg.h.
4834 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4835 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4836 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4838 (strlen_dom_walker::before_dom_children): Call
4839 gimple_purge_dead_eh_edges. Dump tranformation with details
4841 (strlen_dom_walker::before_dom_children): Update call by adding
4842 new argument cleanup_eh.
4843 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4845 2018-01-03 Martin Liska <mliska@suse.cz>
4848 * cif-code.def (VARIADIC_THUNK): New enum value.
4849 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4852 2018-01-03 Jan Beulich <jbeulich@suse.com>
4854 * sse.md (mov<mode>_internal): Tighten condition for when to use
4855 vmovdqu<ssescalarsize> for TI and OI modes.
4857 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4859 Update copyright years.
4861 2018-01-03 Martin Liska <mliska@suse.cz>
4864 * ipa-visibility.c (function_and_variable_visibility): Skip
4865 functions with noipa attribure.
4867 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4869 * gcc.c (process_command): Update copyright notice dates.
4870 * gcov-dump.c (print_version): Ditto.
4871 * gcov.c (print_version): Ditto.
4872 * gcov-tool.c (print_version): Ditto.
4873 * gengtype.c (create_file): Ditto.
4874 * doc/cpp.texi: Bump @copying's copyright year.
4875 * doc/cppinternals.texi: Ditto.
4876 * doc/gcc.texi: Ditto.
4877 * doc/gccint.texi: Ditto.
4878 * doc/gcov.texi: Ditto.
4879 * doc/install.texi: Ditto.
4880 * doc/invoke.texi: Ditto.
4882 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4884 * vector-builder.h (vector_builder::m_full_nelts): Change from
4885 unsigned int to poly_uint64.
4886 (vector_builder::full_nelts): Update prototype accordingly.
4887 (vector_builder::new_vector): Likewise.
4888 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4889 (vector_builder::operator ==): Likewise.
4890 (vector_builder::finalize): Likewise.
4891 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4892 Take the number of elements as a poly_uint64 rather than an
4894 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4895 from unsigned int to poly_uint64.
4896 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4897 (vec_perm_indices::new_vector): Likewise.
4898 (vec_perm_indices::length): Likewise.
4899 (vec_perm_indices::nelts_per_input): Likewise.
4900 (vec_perm_indices::input_nelts): Likewise.
4901 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4902 number of elements per input as a poly_uint64 rather than an
4903 unsigned int. Use the original encoding for variable-length
4904 vectors, rather than clamping each individual element.
4905 For the second and subsequent elements in each pattern,
4906 clamp the step and base before clamping their sum.
4907 (vec_perm_indices::series_p): Handle polynomial element counts.
4908 (vec_perm_indices::all_in_range_p): Likewise.
4909 (vec_perm_indices_to_tree): Likewise.
4910 (vec_perm_indices_to_rtx): Likewise.
4911 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4912 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4913 (tree_vector_builder::new_binary_operation): Handle polynomial
4914 element counts. Return false if we need to know the number
4915 of elements at compile time.
4916 * fold-const.c (fold_vec_perm): Punt if the number of elements
4917 isn't known at compile time.
4919 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4921 * vec-perm-indices.h (vec_perm_builder): Change element type
4922 from HOST_WIDE_INT to poly_int64.
4923 (vec_perm_indices::element_type): Update accordingly.
4924 (vec_perm_indices::clamp): Handle polynomial element_types.
4925 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4926 (vec_perm_indices::all_in_range_p): Likewise.
4927 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4929 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4930 polynomial vec_perm_indices element types.
4931 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4932 * fold-const.c (fold_vec_perm): Likewise.
4933 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4934 * tree-vect-generic.c (lower_vec_perm): Likewise.
4935 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4936 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4937 element type to HOST_WIDE_INT.
4939 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4940 Alan Hayward <alan.hayward@arm.com>
4941 David Sherwood <david.sherwood@arm.com>
4943 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4944 rather than an int. Use plus_constant.
4945 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4946 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4948 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4949 Alan Hayward <alan.hayward@arm.com>
4950 David Sherwood <david.sherwood@arm.com>
4952 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4953 a HOST_WIDE_INT to a poly_int64.
4955 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4956 Alan Hayward <alan.hayward@arm.com>
4957 David Sherwood <david.sherwood@arm.com>
4959 * calls.c (load_register_parameters): Cope with polynomial
4960 mode sizes. Require a constant size for BLKmode parameters
4961 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4962 forces a parameter to be padded at the lsb end in order to
4963 fill a complete number of words, require the parameter size
4964 to be ordered wrt UNITS_PER_WORD.
4966 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4967 Alan Hayward <alan.hayward@arm.com>
4968 David Sherwood <david.sherwood@arm.com>
4970 * reload1.c (spill_stack_slot_width): Change element type
4971 from unsigned int to poly_uint64_pod.
4972 (alter_reg): Treat mode sizes as polynomial.
4974 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4975 Alan Hayward <alan.hayward@arm.com>
4976 David Sherwood <david.sherwood@arm.com>
4978 * reload.c (complex_word_subreg_p): New function.
4979 (reload_inner_reg_of_subreg, push_reload): Use it.
4981 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4982 Alan Hayward <alan.hayward@arm.com>
4983 David Sherwood <david.sherwood@arm.com>
4985 * lra-constraints.c (process_alt_operands): Reject matched
4986 operands whose sizes aren't ordered.
4987 (match_reload): Refer to this check here.
4989 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4990 Alan Hayward <alan.hayward@arm.com>
4991 David Sherwood <david.sherwood@arm.com>
4993 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4994 that the mode size is in the set {1, 2, 4, 8, 16}.
4996 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4997 Alan Hayward <alan.hayward@arm.com>
4998 David Sherwood <david.sherwood@arm.com>
5000 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5001 Use plus_constant instead of gen_rtx_PLUS.
5003 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5004 Alan Hayward <alan.hayward@arm.com>
5005 David Sherwood <david.sherwood@arm.com>
5007 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5008 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5009 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5010 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5011 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5012 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5013 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5014 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5015 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5016 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5018 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5019 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5020 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5021 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5022 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5023 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5024 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5025 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5026 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5027 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5029 * expr.c (emit_move_resolve_push): Treat the input and result
5030 of PUSH_ROUNDING as a poly_int64.
5031 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5032 (emit_push_insn): Likewise.
5033 * lra-eliminations.c (mark_not_eliminable): Likewise.
5034 * recog.c (push_operand): Likewise.
5035 * reload1.c (elimination_effects): Likewise.
5036 * rtlanal.c (nonzero_bits1): Likewise.
5037 * calls.c (store_one_arg): Likewise. Require the padding to be
5038 known at compile time.
5040 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5041 Alan Hayward <alan.hayward@arm.com>
5042 David Sherwood <david.sherwood@arm.com>
5044 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5045 Use plus_constant instead of gen_rtx_PLUS.
5047 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5048 Alan Hayward <alan.hayward@arm.com>
5049 David Sherwood <david.sherwood@arm.com>
5051 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5054 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5055 Alan Hayward <alan.hayward@arm.com>
5056 David Sherwood <david.sherwood@arm.com>
5058 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5059 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5060 via stack temporaries. Treat the mode size as polynomial too.
5062 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5063 Alan Hayward <alan.hayward@arm.com>
5064 David Sherwood <david.sherwood@arm.com>
5066 * expr.c (expand_expr_real_2): When handling conversions involving
5067 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5068 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5069 as a poly_uint64 too.
5071 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5072 Alan Hayward <alan.hayward@arm.com>
5073 David Sherwood <david.sherwood@arm.com>
5075 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5077 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5078 Alan Hayward <alan.hayward@arm.com>
5079 David Sherwood <david.sherwood@arm.com>
5081 * combine.c (can_change_dest_mode): Handle polynomial
5082 REGMODE_NATURAL_SIZE.
5083 * expmed.c (store_bit_field_1): Likewise.
5084 * expr.c (store_constructor): Likewise.
5085 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5086 and polynomial REGMODE_NATURAL_SIZE.
5087 (gen_lowpart_common): Likewise.
5088 * reginfo.c (record_subregs_of_mode): Likewise.
5089 * rtlanal.c (read_modify_subreg_p): Likewise.
5091 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5092 Alan Hayward <alan.hayward@arm.com>
5093 David Sherwood <david.sherwood@arm.com>
5095 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5096 numbers of elements.
5098 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5099 Alan Hayward <alan.hayward@arm.com>
5100 David Sherwood <david.sherwood@arm.com>
5102 * match.pd: Cope with polynomial numbers of vector elements.
5104 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5105 Alan Hayward <alan.hayward@arm.com>
5106 David Sherwood <david.sherwood@arm.com>
5108 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5109 in a POINTER_PLUS_EXPR.
5111 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5112 Alan Hayward <alan.hayward@arm.com>
5113 David Sherwood <david.sherwood@arm.com>
5115 * omp-simd-clone.c (simd_clone_subparts): New function.
5116 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5117 (ipa_simd_modify_function_body): Likewise.
5119 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5120 Alan Hayward <alan.hayward@arm.com>
5121 David Sherwood <david.sherwood@arm.com>
5123 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5124 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5125 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5126 (expand_vector_condition, vector_element): Likewise.
5127 (subparts_gt): New function.
5128 (get_compute_type): Use subparts_gt.
5129 (count_type_subparts): Delete.
5130 (expand_vector_operations_1): Use subparts_gt instead of
5131 count_type_subparts.
5133 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5134 Alan Hayward <alan.hayward@arm.com>
5135 David Sherwood <david.sherwood@arm.com>
5137 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5138 (vect_compile_time_alias): ...this new function. Do the calculation
5139 on poly_ints rather than trees.
5140 (vect_prune_runtime_alias_test_list): Update call accordingly.
5142 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5143 Alan Hayward <alan.hayward@arm.com>
5144 David Sherwood <david.sherwood@arm.com>
5146 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5148 (vect_schedule_slp_instance): Likewise.
5150 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5151 Alan Hayward <alan.hayward@arm.com>
5152 David Sherwood <david.sherwood@arm.com>
5154 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5155 constant and extern definitions for variable-length vectors.
5156 (vect_get_constant_vectors): Note that the number of units
5157 is known to be constant.
5159 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5160 Alan Hayward <alan.hayward@arm.com>
5161 David Sherwood <david.sherwood@arm.com>
5163 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5164 of units as polynomial. Choose between WIDE and NARROW based
5167 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5168 Alan Hayward <alan.hayward@arm.com>
5169 David Sherwood <david.sherwood@arm.com>
5171 * tree-vect-stmts.c (simd_clone_subparts): New function.
5172 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5174 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5175 Alan Hayward <alan.hayward@arm.com>
5176 David Sherwood <david.sherwood@arm.com>
5178 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5179 vectors as polynomial. Use build_index_vector for
5182 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5183 Alan Hayward <alan.hayward@arm.com>
5184 David Sherwood <david.sherwood@arm.com>
5186 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5187 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5188 for variable-length vectors.
5189 (vectorizable_mask_load_store): Treat the number of units as
5190 polynomial, asserting that it is constant if the condition has
5191 already been enforced.
5192 (vectorizable_store, vectorizable_load): Likewise.
5194 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5195 Alan Hayward <alan.hayward@arm.com>
5196 David Sherwood <david.sherwood@arm.com>
5198 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5199 of units as polynomial. Punt if we can't tell at compile time
5200 which vector contains the final result.
5202 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5203 Alan Hayward <alan.hayward@arm.com>
5204 David Sherwood <david.sherwood@arm.com>
5206 * tree-vect-loop.c (vectorizable_induction): Treat the number
5207 of units as polynomial. Punt on SLP inductions. Use an integer
5208 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5209 cast of such a series for variable-length floating-point
5212 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5213 Alan Hayward <alan.hayward@arm.com>
5214 David Sherwood <david.sherwood@arm.com>
5216 * tree.h (build_index_vector): Declare.
5217 * tree.c (build_index_vector): New function.
5218 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5219 of units as polynomial, forcibly converting it to a constant if
5220 vectorizable_reduction has already enforced the condition.
5221 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5222 to create a {1,2,3,...} vector.
5223 (vectorizable_reduction): Treat the number of units as polynomial.
5224 Choose vectype_in based on the largest scalar element size rather
5225 than the smallest number of units. Enforce the restrictions
5228 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5229 Alan Hayward <alan.hayward@arm.com>
5230 David Sherwood <david.sherwood@arm.com>
5232 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5233 number of units as polynomial.
5235 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5236 Alan Hayward <alan.hayward@arm.com>
5237 David Sherwood <david.sherwood@arm.com>
5239 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5240 * target.def (autovectorize_vector_sizes): Return the vector sizes
5241 by pointer, using vector_sizes rather than a bitmask.
5242 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5243 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5244 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5246 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5247 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5248 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5249 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5250 * omp-general.c (omp_max_vf): Likewise.
5251 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5252 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5253 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5254 * tree-vect-slp.c (vect_slp_bb): Likewise.
5255 * doc/tm.texi: Regenerate.
5256 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5258 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5259 the vector size as a poly_uint64 rather than an unsigned int.
5260 (current_vector_size): Change from an unsigned int to a poly_uint64.
5261 (get_vectype_for_scalar_type): Update accordingly.
5262 * tree.h (build_truth_vector_type): Take the size and number of
5263 units as a poly_uint64 rather than an unsigned int.
5264 (build_vector_type): Add a temporary overload that takes
5265 the number of units as a poly_uint64 rather than an unsigned int.
5266 * tree.c (make_vector_type): Likewise.
5267 (build_truth_vector_type): Take the number of units as a poly_uint64
5268 rather than an unsigned int.
5270 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5271 Alan Hayward <alan.hayward@arm.com>
5272 David Sherwood <david.sherwood@arm.com>
5274 * target.def (get_mask_mode): Take the number of units and length
5275 as poly_uint64s rather than unsigned ints.
5276 * targhooks.h (default_get_mask_mode): Update accordingly.
5277 * targhooks.c (default_get_mask_mode): Likewise.
5278 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5279 * doc/tm.texi: Regenerate.
5281 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5282 Alan Hayward <alan.hayward@arm.com>
5283 David Sherwood <david.sherwood@arm.com>
5285 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5286 * omp-general.c (omp_max_vf): Likewise.
5287 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5288 (expand_omp_simd): Handle polynomial safelen.
5289 * omp-low.c (omplow_simd_context): Add a default constructor.
5290 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5291 (lower_rec_simd_input_clauses): Update accordingly.
5292 (lower_rec_input_clauses): Likewise.
5294 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5295 Alan Hayward <alan.hayward@arm.com>
5296 David Sherwood <david.sherwood@arm.com>
5298 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5299 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5300 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5301 (vect_analyze_slp_cost): Likewise.
5302 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5303 (vect_model_load_cost): Likewise.
5305 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5306 Alan Hayward <alan.hayward@arm.com>
5307 David Sherwood <david.sherwood@arm.com>
5309 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5310 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5311 from an unsigned int * to a poly_uint64_pod *.
5312 (calculate_unrolling_factor): New function.
5313 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5315 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5316 Alan Hayward <alan.hayward@arm.com>
5317 David Sherwood <david.sherwood@arm.com>
5319 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5320 from an unsigned int to a poly_uint64.
5321 (_loop_vec_info::slp_unrolling_factor): Likewise.
5322 (_loop_vec_info::vectorization_factor): Change from an int
5324 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5325 (vect_get_num_vectors): New function.
5326 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5327 (vect_get_num_copies): Use vect_get_num_vectors.
5328 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5329 to an unsigned int *.
5330 (vect_analyze_data_refs): Change min_vf from an int * to a
5332 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5333 than an unsigned HOST_WIDE_INT.
5334 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5335 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5336 to an unsigned int *.
5337 (vect_analyze_data_ref_dependences): Likewise.
5338 (vect_compute_data_ref_alignment): Handle polynomial vf.
5339 (vect_enhance_data_refs_alignment): Likewise.
5340 (vect_prune_runtime_alias_test_list): Likewise.
5341 (vect_shift_permute_load_chain): Likewise.
5342 (vect_supportable_dr_alignment): Likewise.
5343 (dependence_distance_ge_vf): Take the vectorization factor as a
5344 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5345 (vect_analyze_data_refs): Change min_vf from an int * to a
5347 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5348 vfm1 as a poly_uint64 rather than an int. Make the same change
5349 for the returned bound_scalar.
5350 (vect_gen_vector_loop_niters): Handle polynomial vf.
5351 (vect_do_peeling): Likewise. Update call to
5352 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5353 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5355 * tree-vect-loop.c (vect_determine_vectorization_factor)
5356 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5357 (vect_get_known_peeling_cost): Likewise.
5358 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5359 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5360 (vect_transform_loop): Likewise. Use the lowest possible VF when
5361 updating the upper bounds of the loop.
5362 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5364 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5365 polynomial unroll factors.
5366 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5367 (vect_make_slp_decision): Likewise.
5368 (vect_supported_load_permutation_p): Likewise, and polynomial
5370 (vect_analyze_slp_cost): Handle polynomial vf.
5371 (vect_slp_analyze_node_operations): Likewise.
5372 (vect_slp_analyze_bb_1): Likewise.
5373 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5374 than an unsigned HOST_WIDE_INT.
5375 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5376 (vectorizable_load): Handle polynomial vf.
5377 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5379 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5381 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5382 Alan Hayward <alan.hayward@arm.com>
5383 David Sherwood <david.sherwood@arm.com>
5385 * match.pd: Handle bit operations involving three constants
5386 and try to fold one pair.
5388 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5390 * tree-vect-loop-manip.c: Include gimple-fold.h.
5391 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5392 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5393 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5394 Add a path that uses a step of VF instead of 1, but disable it
5396 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5397 and niters_no_overflow parameters. Update calls to
5398 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5399 Create a new SSA name if the latter choses to use a ste other
5400 than zero, and return it via niters_vector_mult_vf_var.
5401 * tree-vect-loop.c (vect_transform_loop): Update calls to
5402 vect_do_peeling, vect_gen_vector_loop_niters and
5403 slpeel_make_loop_iterate_ntimes.
5404 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5405 (vect_gen_vector_loop_niters): Update declarations after above changes.
5407 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5409 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5410 128-bit round to integer instructions.
5411 (ceil<mode>2): Likewise.
5412 (btrunc<mode>2): Likewise.
5413 (round<mode>2): Likewise.
5415 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5417 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5418 unaligned VSX load/store on P8/P9.
5419 (expand_block_clear): Allow the use of unaligned VSX
5420 load/store on P8/P9.
5422 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5424 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5426 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5427 swap associated with both a load and a store.
5429 2018-01-02 Andrew Waterman <andrew@sifive.com>
5431 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5432 * config/riscv/riscv.md (clear_cache): Use it.
5434 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5436 * web.c: Remove out-of-date comment.
5438 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5440 * expr.c (fixup_args_size_notes): Check that any existing
5441 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5442 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5443 (emit_single_push_insn): ...here.
5445 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5447 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5448 (const_vector_encoded_nelts): New function.
5449 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5450 (const_vector_int_elt, const_vector_elt): Declare.
5451 * emit-rtl.c (const_vector_int_elt_1): New function.
5452 (const_vector_elt): Likewise.
5453 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5454 of CONST_VECTOR_ELT.
5456 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5458 * expr.c: Include rtx-vector-builder.h.
5459 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5460 directly on the tree encoding.
5461 (const_vector_from_tree): Likewise.
5462 * optabs.c: Include rtx-vector-builder.h.
5463 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5464 sequence of "u" values.
5465 * vec-perm-indices.c: Include rtx-vector-builder.h.
5466 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5467 directly on the vec_perm_indices encoding.
5469 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5471 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5472 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5473 * rtx-vector-builder.h: New file.
5474 * rtx-vector-builder.c: Likewise.
5475 * rtl.h (rtx_def::u2): Add a const_vector field.
5476 (CONST_VECTOR_NPATTERNS): New macro.
5477 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5478 (CONST_VECTOR_DUPLICATE_P): Likewise.
5479 (CONST_VECTOR_STEPPED_P): Likewise.
5480 (CONST_VECTOR_ENCODED_ELT): Likewise.
5481 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5482 (unwrap_const_vec_duplicate): Likewise.
5483 (const_vec_series_p): Check for a non-duplicated vector encoding.
5484 Say that the function only returns true for integer vectors.
5485 * emit-rtl.c: Include rtx-vector-builder.h.
5486 (gen_const_vec_duplicate_1): Delete.
5487 (gen_const_vector): Call gen_const_vec_duplicate instead of
5488 gen_const_vec_duplicate_1.
5489 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5490 (gen_const_vec_duplicate): Use rtx_vector_builder.
5491 (gen_const_vec_series): Likewise.
5492 (gen_rtx_CONST_VECTOR): Likewise.
5493 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5494 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5495 Build a new vector rather than modifying a CONST_VECTOR in-place.
5496 (handle_special_swappables): Update call accordingly.
5497 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5498 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5499 Build a new vector rather than modifying a CONST_VECTOR in-place.
5500 (handle_special_swappables): Update call accordingly.
5502 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5504 * simplify-rtx.c (simplify_const_binary_operation): Use
5505 CONST_VECTOR_ELT instead of XVECEXP.
5507 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5509 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5510 the selector elements to be different from the data elements
5511 if the selector is a VECTOR_CST.
5512 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5513 ssizetype for the selector.
5515 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5517 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5518 before testing each element individually.
5519 * tree-vect-generic.c (lower_vec_perm): Likewise.
5521 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5523 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5524 * selftest-run-tests.c (selftest::run_tests): Call it.
5525 * vector-builder.h (vector_builder::operator ==): New function.
5526 (vector_builder::operator !=): Likewise.
5527 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5528 (vec_perm_indices::all_from_input_p): New function.
5529 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5530 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5531 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5532 instead of reading the VECTOR_CST directly. Detect whether both
5533 vector inputs are the same before constructing the vec_perm_indices,
5534 and update the number of inputs argument accordingly. Use the
5535 utility functions added above. Only construct sel2 if we need to.
5537 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5539 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5540 the broadcast of the low byte.
5541 (expand_mult_highpart): Use an explicit encoding for the permutes.
5542 * optabs-query.c (can_mult_highpart_p): Likewise.
5543 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5544 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5545 (vectorizable_bswap): Likewise.
5546 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5547 explicit encoding for the power-of-2 permutes.
5548 (vect_permute_store_chain): Likewise.
5549 (vect_grouped_load_supported): Likewise.
5550 (vect_permute_load_chain): Likewise.
5552 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5554 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5555 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5556 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5557 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5558 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5559 (vect_gen_perm_mask_any): Likewise.
5561 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5563 * int-vector-builder.h: New file.
5564 * vec-perm-indices.h: Include int-vector-builder.h.
5565 (vec_perm_indices): Redefine as an int_vector_builder.
5566 (auto_vec_perm_indices): Delete.
5567 (vec_perm_builder): Redefine as a stand-alone class.
5568 (vec_perm_indices::vec_perm_indices): New function.
5569 (vec_perm_indices::clamp): Likewise.
5570 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5571 (vec_perm_indices::new_vector): New function.
5572 (vec_perm_indices::new_expanded_vector): Update for new
5573 vec_perm_indices class.
5574 (vec_perm_indices::rotate_inputs): New function.
5575 (vec_perm_indices::all_in_range_p): Operate directly on the
5576 encoded form, without computing elided elements.
5577 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5578 encoding. Update for new vec_perm_indices class.
5579 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5580 the given vec_perm_builder.
5581 (expand_vec_perm_var): Update vec_perm_builder constructor.
5582 (expand_mult_highpart): Use vec_perm_builder instead of
5583 auto_vec_perm_indices.
5584 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5585 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5586 or double series encoding as appropriate.
5587 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5588 vec_perm_indices instead of auto_vec_perm_indices.
5589 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5590 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5591 (vect_permute_store_chain): Likewise.
5592 (vect_grouped_load_supported): Likewise.
5593 (vect_permute_load_chain): Likewise.
5594 (vect_shift_permute_load_chain): Likewise.
5595 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5596 (vect_transform_slp_perm_load): Likewise.
5597 (vect_schedule_slp_instance): Likewise.
5598 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5599 (vectorizable_mask_load_store): Likewise.
5600 (vectorizable_bswap): Likewise.
5601 (vectorizable_store): Likewise.
5602 (vectorizable_load): Likewise.
5603 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5604 vec_perm_indices instead of auto_vec_perm_indices. Use
5605 tree_to_vec_perm_builder to read the vector from a tree.
5606 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5607 vec_perm_builder instead of a vec_perm_indices.
5608 (have_whole_vector_shift): Use vec_perm_builder and
5609 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5610 truncation to calc_vec_perm_mask_for_shift.
5611 (vect_create_epilog_for_reduction): Likewise.
5612 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5613 from auto_vec_perm_indices to vec_perm_indices.
5614 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5615 instead of changing individual elements.
5616 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5617 the vector in d.perm.
5618 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5619 from auto_vec_perm_indices to vec_perm_indices.
5620 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5621 instead of changing individual elements.
5622 (arm_vectorize_vec_perm_const): Use new_vector to install
5623 the vector in d.perm.
5624 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5625 Update vec_perm_builder constructor.
5626 (rs6000_expand_interleave): Likewise.
5627 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5628 (rs6000_expand_interleave): Likewise.
5630 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5632 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5633 to qimode could truncate the indices.
5634 * optabs.c (expand_vec_perm_var): Likewise.
5636 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5638 * Makefile.in (OBJS): Add vec-perm-indices.o.
5639 * vec-perm-indices.h: New file.
5640 * vec-perm-indices.c: Likewise.
5641 * target.h (vec_perm_indices): Replace with a forward class
5643 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5644 * optabs.h: Include vec-perm-indices.h.
5645 (expand_vec_perm): Delete.
5646 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5647 (expand_vec_perm_const): Declare.
5648 * target.def (vec_perm_const_ok): Replace with...
5649 (vec_perm_const): ...this new hook.
5650 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5651 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5652 * doc/tm.texi: Regenerate.
5653 * optabs.def (vec_perm_const): Delete.
5654 * doc/md.texi (vec_perm_const): Likewise.
5655 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5656 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5657 expand_vec_perm for constant permutation vectors. Assert that
5658 the mode of variable permutation vectors is the integer equivalent
5659 of the mode that is being permuted.
5660 * optabs-query.h (selector_fits_mode_p): Declare.
5661 * optabs-query.c: Include vec-perm-indices.h.
5662 (selector_fits_mode_p): New function.
5663 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5664 is defined, instead of checking whether the vec_perm_const_optab
5665 exists. Use targetm.vectorize.vec_perm_const instead of
5666 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5667 fit in the vector mode before using a variable permute.
5668 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5669 vec_perm_indices instead of an rtx.
5670 (expand_vec_perm): Replace with...
5671 (expand_vec_perm_const): ...this new function. Take the selector
5672 as a vec_perm_indices rather than an rtx. Also take the mode of
5673 the selector. Update call to shift_amt_for_vec_perm_mask.
5674 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5675 Use vec_perm_indices::new_expanded_vector to expand the original
5676 selector into bytes. Check whether the indices fit in the vector
5677 mode before using a variable permute.
5678 (expand_vec_perm_var): Make global.
5679 (expand_mult_highpart): Use expand_vec_perm_const.
5680 * fold-const.c: Includes vec-perm-indices.h.
5681 * tree-ssa-forwprop.c: Likewise.
5682 * tree-vect-data-refs.c: Likewise.
5683 * tree-vect-generic.c: Likewise.
5684 * tree-vect-loop.c: Likewise.
5685 * tree-vect-slp.c: Likewise.
5686 * tree-vect-stmts.c: Likewise.
5687 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5689 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5690 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5691 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5692 (aarch64_vectorize_vec_perm_const): ...this new function.
5693 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5694 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5695 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5696 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5697 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5698 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5699 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5701 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5702 check for NEON modes.
5703 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5704 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5705 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5706 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5708 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5709 the old VEC_PERM_CONST conditions.
5710 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5711 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5712 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5713 (ia64_vectorize_vec_perm_const_ok): Merge into...
5714 (ia64_vectorize_vec_perm_const): ...this new function.
5715 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5716 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5717 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5718 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5719 * config/mips/mips.c (mips_expand_vec_perm_const)
5720 (mips_vectorize_vec_perm_const_ok): Merge into...
5721 (mips_vectorize_vec_perm_const): ...this new function.
5722 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5723 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5724 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5725 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5726 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5727 (rs6000_expand_vec_perm_const): Delete.
5728 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5730 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5731 (altivec_expand_vec_perm_const_le): Take each operand individually.
5732 Operate on constant selectors rather than rtxes.
5733 (altivec_expand_vec_perm_const): Likewise. Update call to
5734 altivec_expand_vec_perm_const_le.
5735 (rs6000_expand_vec_perm_const): Delete.
5736 (rs6000_vectorize_vec_perm_const_ok): Delete.
5737 (rs6000_vectorize_vec_perm_const): New function.
5738 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5739 an element count and rtx array.
5740 (rs6000_expand_extract_even): Update call accordingly.
5741 (rs6000_expand_interleave): Likewise.
5742 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5743 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5744 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5745 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5746 (rs6000_expand_vec_perm_const): Delete.
5747 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5748 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5749 (altivec_expand_vec_perm_const_le): Take each operand individually.
5750 Operate on constant selectors rather than rtxes.
5751 (altivec_expand_vec_perm_const): Likewise. Update call to
5752 altivec_expand_vec_perm_const_le.
5753 (rs6000_expand_vec_perm_const): Delete.
5754 (rs6000_vectorize_vec_perm_const_ok): Delete.
5755 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5756 reference to the SPE evmerge intructions.
5757 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5758 an element count and rtx array.
5759 (rs6000_expand_extract_even): Update call accordingly.
5760 (rs6000_expand_interleave): Likewise.
5761 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5762 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5764 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5766 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5768 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5769 vector mode and that that mode matches the mode of the data
5771 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5772 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5773 directly using expand_vec_perm_1 when forcing selectors into
5775 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5777 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5779 * optabs-query.h (can_vec_perm_p): Delete.
5780 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5781 * optabs-query.c (can_vec_perm_p): Split into...
5782 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5783 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5784 particular selector is valid.
5785 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5786 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5787 (vect_grouped_load_supported): Likewise.
5788 (vect_shift_permute_load_chain): Likewise.
5789 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5790 (vect_transform_slp_perm_load): Likewise.
5791 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5792 (vectorizable_bswap): Likewise.
5793 (vect_gen_perm_mask_checked): Likewise.
5794 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5795 implementations of variable permutation vectors into account
5796 when deciding which selector to use.
5797 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5798 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5799 with a false third argument.
5800 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5801 to test whether the constant selector is valid and can_vec_perm_var_p
5802 to test whether a variable selector is valid.
5804 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5806 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5807 * optabs-query.c (can_vec_perm_p): Likewise.
5808 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5809 instead of vec_perm_indices.
5810 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5811 (vect_gen_perm_mask_checked): Likewise,
5812 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5813 (vect_gen_perm_mask_checked): Likewise,
5815 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5817 * optabs-query.h (qimode_for_vec_perm): Declare.
5818 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5819 (qimode_for_vec_perm): ...this new function.
5820 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5822 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5824 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5825 does not have a conditional at the top.
5827 2018-01-02 Richard Biener <rguenther@suse.de>
5829 * ipa-inline.c (big_speedup_p): Fix expression.
5831 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5834 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5837 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5841 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5842 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5843 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5844 cond_taken_branch_cost 3->4.
5846 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5848 PR tree-optimization/83581
5849 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5850 TODO_cleanup_cfg if any changes have been made.
5853 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5854 convert_modes if target mode has the right side, but different mode
5858 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5859 last argument when extracting from CONCAT. If either from_real or
5860 from_imag is NULL, use expansion through memory. If result is not
5861 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5862 the parts directly to inner mode, if even that fails, use expansion
5866 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5867 check for bswap in mode rather than HImode and use that in expand_unop
5870 Copyright (C) 2018 Free Software Foundation, Inc.
5872 Copying and distribution of this file, with or without modification,
5873 are permitted in any medium without royalty provided the copyright
5874 notice and this notice are preserved.