1 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
3 PR rtl-optimization/84123
4 * combine.c (change_zero_ext): Check if hard register satisfies
5 can_change_dest_mode before calling gen_lowpart_SUBREG.
7 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
10 * ira.c (ira_init_register_move_cost): Remove assert.
12 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
14 PR rtl-optimization/84071
15 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
16 * doc/tm.texi: Regenerate.
18 2018-01-31 Richard Biener <rguenther@suse.de>
20 PR tree-optimization/84132
21 * tree-data-ref.c (analyze_miv_subscript): Properly
22 check whether evolution_function_is_affine_multivariate_p
23 before calling gcd_of_steps_may_divide_p.
25 2018-01-31 Julia Koval <julia.koval@intel.com>
28 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
29 * config/i386/i386.md (rdpid_rex64) New.
30 (rdpid): Make 32bit only.
32 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
35 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
36 an IDENTIFIER_NODE for FUNCTION_TYPE's.
38 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
41 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
43 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
45 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
47 PR rtl-optimization/84071
48 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
49 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
51 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
53 * config/arc/arc.c (arc_handle_aux_attribute): New function.
54 (arc_attribute_table): Add 'aux' attribute.
55 (arc_in_small_data_p): Consider aux like variables.
56 (arc_is_aux_reg_p): New function.
57 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
58 (arc_get_aux_arg): New function.
59 (prepare_move_operands): Handle aux-register access.
60 (arc_handle_aux_attribute): New function.
61 * doc/extend.texi (ARC Variable attributes): Add subsection.
63 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
65 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
66 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
67 (arc_attribute_table): Add 'uncached' attribute.
68 (arc_print_operand): Print '.di' flag for uncached memory
70 (arc_in_small_data_p): Do not consider for small data the uncached
72 (arc_is_uncached_mem_p): New function.
73 * config/arc/predicates.md (compact_store_memory_operand): Check
74 for uncached memory accesses.
75 (nonvol_nonimm_operand): Likewise.
76 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
78 2018-01-31 Jakub Jelinek <jakub@redhat.com>
81 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
82 falign-loops=): Add Optimization flag.
84 2017-01-30 Jeff Law <law@redhat.com>
87 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
88 INT_REGISTERS_SAVED. Check it prior to calling
89 get_scratch_register_on_entry.
90 (ix86_adjust_stack_and_probe): Similarly.
91 (ix86_emit_probe_stack_range): Similarly.
92 (ix86_expand_prologue): Corresponding changes.
94 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
97 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
98 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
100 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
103 * lra-constraints.c (curr_insn_transform): Process AND in the
106 2018-01-30 Jakub Jelinek <jakub@redhat.com>
108 PR rtl-optimization/83986
109 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
110 dependence against last_pending_memory_flush in addition to
113 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
115 PR tree-optimization/81611
116 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
119 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
121 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
124 2018-01-30 Richard Biener <rguenther@suse.de>
125 Jakub Jelinek <jakub@redhat.com>
127 PR tree-optimization/84111
128 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
129 inner loops added during recursion, as they don't have up-to-date
132 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
135 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
136 (can_inline_edge_by_limits_p): ... here.
137 (can_early_inline_edge_p, check_callers,
138 update_caller_keys, update_callee_keys, recursive_inlining,
139 add_new_edges_to_heap, speculation_useful_p,
140 inline_small_functions,
141 inline_small_functions, flatten_function,
142 inline_to_all_callers_1): Update.
144 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
146 * profile-count.c (profile_count::combine_with_ipa_count): Handle
149 2018-01-30 Richard Biener <rguenther@suse.de>
151 PR tree-optimization/83008
152 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
153 invariant and constant vector uses in stmts when they need
156 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
159 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
160 * configure: Regenerate.
162 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
164 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
166 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
167 Use gen_rtx_REG rather than gen_lowpart.
169 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
171 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
172 rather than 0 when creating partial subregs.
174 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
176 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
179 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
182 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
183 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
184 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
185 flags. This restores the settings used before the 2017-07-24.
186 Turning off pre increment/decrement/modify allows IVOPTS to
187 optimize DF/SF loops where the index is an int.
189 2018-01-29 Richard Biener <rguenther@suse.de>
190 Kelvin Nilsen <kelvin@gcc.gnu.org>
193 * tree-vect-stmts.c (vectorizable_call): Don't call
194 targetm.vectorize_builtin_md_vectorized_function if callee is
197 2018-01-22 Carl Love <cel@us.ibm.com>
199 * doc/extend.tex: Fix typo in second arg in
200 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
202 2018-01-29 Richard Biener <rguenther@suse.de>
204 PR tree-optimization/84086
205 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
206 (flush_ssaname_freelist): When SSA names were released reset
209 2018-01-29 Richard Biener <rguenther@suse.de>
211 PR tree-optimization/84057
212 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
213 removed paths when removing edges.
215 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
217 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
218 -mfunction-return=@var{choice}.
220 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
223 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
225 (layout::print_source_line): Likewise.
226 (test_get_line_width_without_trailing_whitespace): Add test cases.
228 2018-01-27 Jakub Jelinek <jakub@redhat.com>
231 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
234 2018-01-26 Jim Wilson <jimw@sifive.com>
236 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
238 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
241 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
243 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
244 and CMP + SUB-immediate -> SUBS.
246 2018-01-26 Martin Sebor <msebor@redhat.com>
248 PR tree-optimization/83896
249 * tree-ssa-strlen.c (get_string_len): Rename...
250 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
251 Avoid assuming length is constant.
252 (handle_char_store): Use HOST_WIDE_INT for string length.
254 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
257 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
258 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
260 2018-01-26 Richard Biener <rguenther@suse.de>
262 PR rtl-optimization/84003
263 * dse.c (record_store): Only record redundant stores when
264 the earlier store aliases at least all accesses the later one does.
266 2018-01-26 Jakub Jelinek <jakub@redhat.com>
268 PR rtl-optimization/83985
269 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
270 REG_CFA_RESTORE insns.
271 (delete_unmarked_insns): Don't ignore separate shrink wrapping
272 REG_CFA_RESTORE insns here.
275 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
276 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
278 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
280 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
281 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
282 (arc_init): Likewise.
283 (arc_override_options): Likewise.
284 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
286 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
288 * config/arc/arc.h (TARGET_DBNZ): Define.
289 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
290 properly set the tune attribute.
291 (dbnz): Use TARGET_DBNZ guard.
292 * config/arc/arc.opt (mtune): Add core3 option.
294 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
296 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
297 recognize new pic like addresses.
298 (arc_delegitimize_address): Clean up.
300 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
302 * config/arc/arc-arches.def: Option mrf16 valid for all
304 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
305 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
306 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
307 * config/arc/arc-tables.opt: Regenerate.
308 * config/arc/arc.c (arc_conditional_register_usage): Handle
309 reduced register file case.
310 (arc_file_start): Set must have build attributes.
311 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
313 * config/arc/arc.opt (mrf16): Add new option.
314 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
315 * config/arc/genmultilib.awk: Handle new mrf16 option.
316 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
317 * config/arc/t-multilib: Regenerate.
318 * doc/invoke.texi (ARC Options): Document mrf16 option.
320 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
322 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
323 * config/arc/arc.c (arc_handle_secure_attribute): New function.
324 (arc_attribute_table): Add 'secure_call' attribute.
325 (arc_print_operand): Print secure call operand.
326 (arc_function_ok_for_sibcall): Don't optimize tail calls when
328 (arc_is_secure_call_p): New function. * config/arc/arc.md
329 (call_i): Add support for sjli instruction.
330 (call_value_i): Likewise.
331 * config/arc/constraints.md (Csc): New constraint.
333 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
334 John Eric Martin <John.Martin@emmicro-us.com>
336 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
337 * config/arc/arc.c (_arc_jli_section): New struct.
338 (arc_jli_section): New type.
339 (rc_jli_sections): New static variable.
340 (arc_handle_jli_attribute): New function.
341 (arc_attribute_table): Add jli_always and jli_fixed attribute.
342 (arc_file_end): New function.
343 (TARGET_ASM_FILE_END): Define.
344 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
345 (arc_add_jli_section): New function.
346 (jli_call_scan): Likewise.
347 (arc_reorg): Call jli_call_scan.
348 (arc_output_addsi): Remove 'S' from printing asm operand.
349 (arc_is_jli_call_p): New function.
350 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
352 (movhi_insn): Likewise.
353 (movsi_insn): Likewise.
354 (movsi_set_cc_insn): Likewise.
355 (loadqi_update): Likewise.
356 (load_zeroextendqisi_update): Likewise.
357 (load_signextendqisi_update): Likewise.
358 (loadhi_update): Likewise.
359 (load_zeroextendhisi_update): Likewise.
360 (load_signextendhisi_update): Likewise.
361 (loadsi_update): Likewise.
362 (loadsf_update): Likewise.
363 (movsicc_insn): Likewise.
364 (bset_insn): Likewise.
365 (bxor_insn): Likewise.
366 (bclr_insn): Likewise.
367 (bmsk_insn): Likewise.
368 (bicsi3_insn): Likewise.
369 (cmpsi_cc_c_insn): Likewise.
370 (movsi_ne): Likewise.
371 (movsi_cond_exec): Likewise.
372 (clrsbsi2): Likewise.
383 (call_i): Remove 'S' asm letter, add jli instruction.
384 (call_value_i): Likewise.
385 * config/arc/arc.op (mjli-always): New option.
386 * config/arc/constraints.md (Cji): New constraint.
387 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
389 (subsf3_fpx): Likewise.
390 (mulsf3_fpx): Likewise.
391 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
393 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
395 * doc/invoke.texi (ARC): Document mjli-always option.
397 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
399 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
400 and use incw and decw where possible
401 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
403 2018-01-26 Richard Biener <rguenther@suse.de>
405 PR tree-optimization/81082
406 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
407 association if it requires casting to unsigned.
408 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
409 from fold_plusminus_mult_expr to catch important cases late when
410 range info is available.
412 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
414 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
415 * configure.ac (hidden_linkonce): New test.
416 * configure: Regenerate.
417 * config.in: Regenerate.
419 2018-01-26 Julia Koval <julia.koval@intel.com>
421 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
422 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
423 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
424 _mm_mask_bitshuffle_epi64_mask): Fix type.
425 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
426 USI_FTYPE_V4DI_V4DI_USI): Remove.
427 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
428 __builtin_ia32_vpshufbitqmb256_mask,
429 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
430 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
431 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
433 2018-01-26 Alan Modra <amodra@gmail.com>
436 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
437 UNSPEC_VBPERMQ. Sort other unspecs.
439 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
441 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
443 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
446 * predict.c (drop_profile): Do not push/pop cfun; update also
448 (handle_missing_profiles): Fix logic looking for zero profiles.
450 2018-01-25 Jakub Jelinek <jakub@redhat.com>
453 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
454 on functions with #pragma omp declare simd or functions with simd
456 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
457 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
458 Remove trailing \n from warning_at calls.
460 2018-01-25 Tom de Vries <tom@codesourcery.com>
463 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
464 for neutered workers.
466 2018-01-24 Joseph Myers <joseph@codesourcery.com>
469 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
470 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
472 2017-01-24 Jeff Law <law@redhat.com>
475 * i386.c (get_probe_interval): Move to earlier point.
476 (ix86_compute_frame_layout): If -fstack-clash-protection and
477 the frame is larger than the probe interval, then use pushes
478 to save registers rather than reg->mem moves.
479 (ix86_expand_prologue): Remove conditional for int_registers_saved
482 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
485 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
486 min/max for never referenced object.
488 2018-01-24 Jakub Jelinek <jakub@redhat.com>
491 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
493 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
494 attributes from DECL_ATTRIBUTES (decl) without affecting
495 DECL_ATTRIBUTES (current_function_decl).
496 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
497 functions with non-NULL DECL_ABSTRACT_ORIGIN.
499 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
501 PR tree-optimization/83979
502 * fold-const.c (fold_comparison): Use constant_boolean_node
503 instead of boolean_{true,false}_node.
505 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
507 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
510 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
512 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
513 Simplify the clause that sets the length attribute.
514 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
515 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
516 clause that sets the length attribute.
517 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
519 2018-01-24 Tom de Vries <tom@codesourcery.com>
522 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
523 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
524 Add strict parameter.
525 (prevent_branch_around_nothing): Insert dummy insn between branch to
526 label and label with no ptx insn inbetween.
527 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
529 2018-01-24 Tom de Vries <tom@codesourcery.com>
532 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
533 for neutered threads in warp.
534 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
536 2018-01-24 Richard Biener <rguenther@suse.de>
538 PR tree-optimization/83176
539 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
542 2018-01-24 Richard Biener <rguenther@suse.de>
544 PR tree-optimization/82819
545 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
546 code generating pluses that are no-ops in the target precision.
548 2018-01-24 Richard Biener <rguenther@suse.de>
551 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
553 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
555 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
556 to merge probabilities.
557 * predict.c (probably_never_executed): Also mark as cold functions
558 with global 0 profile and guessed local profile.
559 * profile-count.c (profile_probability::combine_with_count): New
561 * profile-count.h (profile_probability::operator*,
562 profile_probability::operator*=, profile_probability::operator/,
563 profile_probability::operator/=): Reduce precision to adjusted
564 and set value to guessed on contradictory divisions.
565 (profile_probability::combine_with_freq): Remove.
566 (profile_probability::combine_wiht_count): Declare.
567 (profile_count::force_nonzero):: Set to adjusted.
568 (profile_count::probability_in):: Set quality to adjusted.
569 * tree-ssa-tail-merge.c (replace_block_by): Use
572 2018-01-23 Andrew Waterman <andrew@sifive.com>
573 Jim Wilson <jimw@sifive.com>
575 * config/riscv/riscv.c (riscv_stack_boundary): New.
576 (riscv_option_override): Set riscv_stack_boundary. Handle
577 riscv_preferred_stack_boundary_arg.
578 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
579 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
580 (STACK_BOUNDARY): Set to riscv_stack_boundary.
581 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
582 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
583 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
585 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
588 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
589 of struct ix86_frame.
590 (ix86_expand_epilogue): Likewise. Add a local variable for
591 the reg_save_offset field in struct ix86_frame.
593 2018-01-23 Bin Cheng <bin.cheng@arm.com>
595 PR tree-optimization/82604
596 * tree-loop-distribution.c (enum partition_kind): New enum item
597 PKIND_PARTIAL_MEMSET.
598 (partition_builtin_p): Support above new enum item.
599 (generate_code_for_partition): Ditto.
600 (compute_access_range): Differentiate cases that equality can be
601 proven at all loops, the innermost loops or no loops.
602 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
603 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
604 (finalize_partitions, distribute_loop): Don't fuse partition of
605 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
606 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
609 2018-01-23 Martin Liska <mliska@suse.cz>
611 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
612 order to ignore the predictor.
613 (PRED_POLYMORPHIC_CALL): Likewise.
614 (PRED_RECURSIVE_CALL): Likewise.
616 2018-01-23 Martin Liska <mliska@suse.cz>
618 * tree-profile.c (tree_profiling): Print function header to
619 aware reader which function we are working on.
620 * value-prof.c (gimple_find_values_to_profile): Do not print
621 not interesting value histograms.
623 2018-01-23 Martin Liska <mliska@suse.cz>
625 * profile-count.h (enum profile_quality): Add
626 profile_uninitialized as the first value. Do not number values
627 as they are zero based.
628 (profile_count::verify): Update sanity check.
629 (profile_probability::verify): Likewise.
631 2018-01-23 Nathan Sidwell <nathan@acm.org>
633 * doc/invoke.texi (ffor-scope): Deprecate.
635 2018-01-23 David Malcolm <dmalcolm@redhat.com>
637 PR tree-optimization/83510
638 * domwalk.c (set_all_edges_as_executable): New function.
639 (dom_walker::dom_walker): Convert bool param
640 "skip_unreachable_blocks" to enum reachability. Move setup of
641 edge flags to set_all_edges_as_executable and only do it when
642 reachability is REACHABLE_BLOCKS.
643 * domwalk.h (enum dom_walker::reachability): New enum.
644 (dom_walker::dom_walker): Convert bool param
645 "skip_unreachable_blocks" to enum reachability.
646 (set_all_edges_as_executable): New decl.
647 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
648 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
650 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
651 but converting true to REACHABLE_BLOCKS.
652 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
654 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
655 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
656 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
658 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
659 if check_all_array_refs will be called.
661 2018-01-23 David Malcolm <dmalcolm@redhat.com>
663 * tree.c (selftest::test_location_wrappers): Add more test
666 2018-01-23 David Malcolm <dmalcolm@redhat.com>
668 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
669 (selftest::test_bit_in_range): Likewise.
671 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
674 * doc/sourcebuild.texi (vect_float): Say that the selector
675 only describes the situation when -funsafe-math-optimizations is on.
676 (vect_float_strict): Document.
678 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
680 PR tree-optimization/83965
681 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
682 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
683 instead of checking only for a reduction.
684 (vect_recog_widen_sum_pattern): Likewise.
686 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
688 * predict.c (probably_never_executed): Only use precise profile info.
689 (compute_function_frequency): Skip after inlining hack since we now
690 have quality checking.
692 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
694 * profile-count.h (profile_probability::very_unlikely,
695 profile_probability::unlikely, profile_probability::even): Set
696 precision to guessed.
698 2018-01-23 Richard Biener <rguenther@suse.de>
700 PR tree-optimization/83963
701 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
702 Properly terminate dominator walk when crossing the exit edge not
703 when visiting its source block.
705 2018-01-23 Jakub Jelinek <jakub@redhat.com>
708 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
709 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
711 2018-01-22 Jakub Jelinek <jakub@redhat.com>
713 PR tree-optimization/83957
714 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
715 semicolon after for body surrounded by braces.
717 PR tree-optimization/83081
718 * profile-count.h (profile_probability::split): New method.
719 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
720 Use profile_probability::split.
721 (do_compare_rtx_and_jump): Fix adjustment of probabilities
722 when splitting a single conditional jump into 2.
724 2018-01-22 David Malcolm <dmalcolm@redhat.com>
726 PR tree-optimization/69452
727 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
730 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
732 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
733 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
734 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
736 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
738 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
739 * config/rl78/rl78.md: New define_expand "movdi"
740 * config/rl78/rl78.c: New function definition rl78_split_movdi
742 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
745 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
747 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
748 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
749 128-bit to produce an UNSPEC move to get the double word with the
750 signbit and then a shift directly to do signbit.
751 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
752 implementation with a new version that just does either a direct
753 move or a regular move. Move memory interface to separate insns.
754 Move insns so they are next to the expander.
755 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
756 with signbit move. Split big and little endian case.
757 (signbit<mode>2_dm_mem_le): Likewise.
758 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
759 (signbit<mode>2_dm2): Likewise.
761 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
763 * config/rl78/rl78.md: New define_expand "anddi3".
765 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
767 * config/rl78/rl78.md: New define_expand "umindi3".
769 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
771 * config/rl78/rl78.md: New define_expand "smindi3".
773 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
775 * config/rl78/rl78.md: New define_expand "smaxdi3".
777 2018-01-22 Carl Love <cel@us.ibm.com>
779 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
780 LVX_V1TI): Add macro expansion.
781 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
782 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
783 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
784 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
785 Change check to determine if the instruction is a byte reversing
786 entry. Fix typo in comment.
787 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
788 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
789 Add def_builtin calls for new builtins.
790 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
791 Add define_insn expansion.
793 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
795 * config/rl78/rl78.md: New define_expand "umaxdi3".
797 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
799 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
800 for non-QImode registers
802 2018-01-22 Richard Biener <rguenther@suse.de>
804 PR tree-optimization/83963
805 * graphite-scop-detection.c (scop_detection::get_sese): Delay
806 including the loop exit block.
807 (scop_detection::merge_sese): Likewise.
808 (scop_detection::add_scop): Do it here instead.
810 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
812 * doc/sourcebuild.texi (arm_softfloat): Document.
814 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
817 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
818 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
819 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
821 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
822 David Edelsohn <dje.gcc@gmail.com>
825 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
826 Change "crset eq" to "crset 2".
827 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
828 (*call_indirect_aix<mode>_nospec): Likewise.
829 (*call_value_indirect_aix<mode>_nospec): Likewise.
830 (*call_indirect_elfv2<mode>_nospec): Likewise.
831 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
832 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
833 change assembly output from . to $.
834 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
835 (indirect_jump<mode>_nospec): Change assembly output from . to $.
836 (*tablejump<mode>_internal1_nospec): Likewise.
838 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
841 * config/sh/sh_optimize_sett_clrt.cc:
842 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
844 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
846 PR tree-optimization/83940
847 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
848 offset_dt to vect_constant_def rather than vect_unknown_def_type.
849 (vect_check_load_store_mask): Add a mask_dt_out parameter and
850 use it to pass back the definition type.
851 (vect_check_store_rhs): Likewise rhs_dt_out.
852 (vect_build_gather_load_calls): Add a mask_dt argument and use
853 it instead of a call to vect_is_simple_use.
854 (vectorizable_store): Update calls to vect_check_load_store_mask
855 and vect_check_store_rhs. Use the dt returned by the latter instead
856 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
857 instead of calls to vect_is_simple_use. Pass the scalar rather
858 than the vector operand to vect_is_simple_use when handling
859 second and subsequent copies of an rhs value.
860 (vectorizable_load): Update calls to vect_check_load_store_mask
861 and vect_build_gather_load_calls. Use the cached mask_dt and
862 gs_info.offset_dt instead of calls to vect_is_simple_use.
864 2018-01-20 Jakub Jelinek <jakub@redhat.com>
867 * tree-emutls.c: Include gimplify.h.
868 (lower_emutls_2): New function.
869 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
870 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
871 it before further processing.
874 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
875 UINTVAL (trueop1) instead of INTVAL (op1).
877 2018-01-19 Jakub Jelinek <jakub@redhat.com>
881 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
882 INCOMING_FRAME_SP_OFFSET if not defined.
883 (scan_trace): Add ENTRY argument. If true and
884 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
885 emit a note to adjust the CFA offset.
886 (create_cfi_notes): Adjust scan_trace callers.
887 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
888 INCOMING_FRAME_SP_OFFSET in the CIE.
889 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
890 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
892 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
893 * doc/tm.texi: Regenerated.
895 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
897 PR rtl-optimization/83147
898 * lra-constraints.c (remove_inheritance_pseudos): Use
899 lra_substitute_pseudo_within_insn.
901 2018-01-19 Tom de Vries <tom@codesourcery.com>
902 Cesar Philippidis <cesar@codesourcery.com>
905 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
907 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
910 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
911 spaces for function labels.
913 2018-01-19 Martin Liska <mliska@suse.cz>
915 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
916 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
917 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
918 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
919 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
920 (PRED_CONST_RETURN): Change from 69 to 65.
921 (PRED_NULL_RETURN): Change from 91 to 71.
922 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
923 (PRED_LOOP_GUARD): Change from 66 to 73.
925 2018-01-19 Martin Liska <mliska@suse.cz>
927 * predict.c (predict_insn_def): Add new assert.
928 (struct branch_predictor): Change type to signed integer.
929 (test_prediction_value_range): Amend test to cover
931 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
932 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
933 (PRED_LOOP_ITERATIONS_MAX): Likewise.
934 (PRED_LOOP_IV_COMPARE): Likewise.
935 * predict.h (PROB_UNINITIALIZED): Define new constant.
937 2018-01-19 Martin Liska <mliska@suse.cz>
939 * predict.c (dump_prediction): Add new format for
940 analyze_brprob.py script which is enabled with -details
942 * profile-count.h (precise_p): New function.
944 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
946 PR tree-optimization/83922
947 * tree-vect-loop.c (vect_verify_full_masking): Return false if
948 there are no statements that need masking.
949 (vect_active_double_reduction_p): New function.
950 (vect_analyze_loop_operations): Use it when handling phis that
951 are not in the loop header.
953 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
955 PR tree-optimization/83914
956 * tree-vect-loop.c (vectorizable_induction): Don't convert
957 init_expr or apply the peeling adjustment for inductions
958 that are nested within the vectorized loop.
960 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
962 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
965 2018-01-18 Jakub Jelinek <jakub@redhat.com>
969 * function.h (gimplify_parameters): Add gimple_seq * argument.
970 * function.c: Include gimple.h and options.h.
971 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
972 for the added local temporaries if needed.
973 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
974 if there are any parameter cleanups, wrap whole body into a
975 try/finally with the cleanups.
977 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
980 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
981 Use GET_MODE_CLASS for scalar floating point.
983 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
987 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
988 Fix call of call_cgraph_insertion_hooks.
990 2018-01-18 Martin Sebor <msebor@redhat.com>
992 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
994 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
997 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1000 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1003 * common.opt: (-ffile-prefix-map): New option.
1004 * opts.c (common_handle_option): Defer it.
1005 * opts-global.c (handle_common_deferred_options): Handle it.
1006 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1007 * file-prefix-map.h: New file.
1008 (remap_debug_filename, add_debug_prefix_map): ...here.
1009 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1010 * final.c (debug_prefix_map, add_debug_prefix_map
1011 remap_debug_filename): Move to...
1012 * file-prefix-map.c: New file.
1013 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1014 generalize, get rid of alloca(), use strrchr() instead of strchr().
1015 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1016 Implement in terms of add_prefix_map().
1017 (remap_macro_filename, remap_debug_filename): Implement in term of
1019 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1020 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1021 * dbxout.c: Include file-prefix-map.h.
1022 * varasm.c: Likewise.
1023 * vmsdbgout.c: Likewise.
1024 * xcoffout.c: Likewise.
1025 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1026 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1027 * doc/invoke.texi (-ffile-prefix-map): Document.
1028 (-fdebug-prefix-map): Update description.
1030 2018-01-18 Martin Liska <mliska@suse.cz>
1032 * config/i386/i386.c (indirect_thunk_name): Document that also
1034 (output_indirect_thunk): Document why both instructions
1035 (pause and lfence) are generated.
1037 2018-01-18 Richard Biener <rguenther@suse.de>
1039 PR tree-optimization/83887
1040 * graphite-scop-detection.c
1041 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1042 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1043 (scop_detection::merge_sese): Re-implement with a flood-fill
1044 algorithm that properly finds a SESE region if it exists.
1046 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1049 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1050 pointer_diff optimizations use view_convert instead of convert.
1052 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1054 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1055 Generate different code for -mno-speculate-indirect-jumps.
1056 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1057 (*call_indirect_aix<mode>): Disable for
1058 -mno-speculate-indirect-jumps.
1059 (*call_indirect_aix<mode>_nospec): New define_insn.
1060 (*call_value_indirect_aix<mode>): Disable for
1061 -mno-speculate-indirect-jumps.
1062 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1063 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1064 -mno-speculate-indirect-jumps.
1065 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1067 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1069 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1070 long double type, set the flags for noting the default long double
1071 type, even if we don't pass or return a long double type.
1073 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1076 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1079 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1081 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1082 support for merge[hl].
1083 (fold_mergehl_helper): New helper function.
1084 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1085 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1086 (altivec_vmrglw_direct): Add xxmrglw insn.
1088 2018-01-17 Andrew Waterman <andrew@sifive.com>
1090 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1091 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1093 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1096 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1097 call the lto_location_cache before reading the
1098 DECL_SOURCE_LOCATION of the types.
1100 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1101 Richard Sandiford <richard.sandiford@linaro.org>
1103 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1104 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1105 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1106 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1107 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1109 * config/aarch64/constraints.md (aarch64_movti_operand):
1111 * config/aarch64/predicates.md (Uti): Add new constraint.
1113 2018-01-17 Carl Love <cel@us.ibm.com>
1114 * config/rs6000/vsx.md (define_expand xl_len_r,
1115 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1116 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1118 (define_expand, define_insn): Move the shift left from the
1119 define_insn to the define_expand for lxvl and stxvl instructions.
1120 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1121 and XL_LEN_R definitions to PURE.
1123 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1125 * config/i386/i386.c (indirect_thunk_name): Declare regno
1126 as unsigned int. Compare regno with INVALID_REGNUM.
1127 (output_indirect_thunk): Ditto.
1128 (output_indirect_thunk_function): Ditto.
1129 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1130 in the call to output_indirect_thunk_function.
1132 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1135 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1136 rather than the size of inner_type to determine the stack slot size
1137 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1139 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1142 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1145 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1147 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1148 endian Linux systems to optionally enable multilibs for selecting
1149 the long double type if the user configured an explicit type.
1150 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1151 have no long double multilibs if not defined.
1152 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1153 warn if the user used -mabi={ieee,ibm}longdouble and we built
1154 multilibs for long double.
1155 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1156 appropriate multilib option.
1157 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1159 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1160 for building long double multilibs.
1161 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1163 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1165 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1168 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1170 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1173 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1176 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1179 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1181 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1182 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1184 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1186 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1187 different rtl trees depending on TARGET_64BIT.
1188 (rs6000_gen_lvx): Likewise.
1190 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1192 * config/visium/visium.md (nop): Tweak comment.
1193 (hazard_nop): Likewise.
1195 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1197 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1198 -mspeculate-indirect-jumps.
1199 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1200 for -mno-speculate-indirect-jumps.
1201 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1202 (*call_value_indirect_elfv2<mode>): Disable for
1203 -mno-speculate-indirect-jumps.
1204 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1205 (indirect_jump): Emit different RTL for
1206 -mno-speculate-indirect-jumps.
1207 (*indirect_jump<mode>): Disable for
1208 -mno-speculate-indirect-jumps.
1209 (*indirect_jump<mode>_nospec): New define_insn.
1210 (tablejump): Emit different RTL for
1211 -mno-speculate-indirect-jumps.
1212 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1213 (tablejumpsi_nospec): New define_expand.
1214 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1215 (tablejumpdi_nospec): New define_expand.
1216 (*tablejump<mode>_internal1): Disable for
1217 -mno-speculate-indirect-jumps.
1218 (*tablejump<mode>_internal1_nospec): New define_insn.
1219 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1222 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1224 * caller-save.c (insert_save): Drop unnecessary parameter. All
1227 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1228 Richard Biener <rguenth@suse.de>
1231 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1232 return early, inline manually is_gimple_sizepos. Make sure if we
1233 call gimplify_expr we don't end up with a gimple constant.
1234 * tree.c (variably_modified_type_p): Don't return true for
1235 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1236 * gimplify.h (is_gimple_sizepos): Remove.
1238 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1240 PR tree-optimization/83857
1241 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1242 vectorizable_live_operation for pure SLP statements.
1243 (vectorizable_live_operation): Handle PHIs.
1245 2018-01-16 Richard Biener <rguenther@suse.de>
1247 PR tree-optimization/83867
1248 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1249 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1251 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1254 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1255 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1256 If off is not INTEGER_CST, issue a may not be aligned warning
1257 rather than isn't aligned. Use isn%'t rather than isn't.
1258 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1260 <case MULT_EXPR>: Improve the case when bottom and one of the
1261 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1262 operand, in that case check if the other operand is multiple of
1263 bottom divided by the INTEGER_CST operand.
1265 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1268 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1269 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1270 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1271 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1272 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1273 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1274 (pa_function_arg_size): New function.
1276 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1278 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1279 in a separate statement.
1281 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1283 PR tree-optimization/83847
1284 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1285 group gathers and scatters.
1287 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1289 PR rtl-optimization/86620
1290 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1292 PR rtl-optimization/83213
1293 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1294 to last if both are JUMP_INSNs.
1296 PR tree-optimization/83843
1297 * gimple-ssa-store-merging.c
1298 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1299 store_immediate_info for bswap/nop orig_stores.
1301 2018-01-15 Andrew Waterman <andrew@sifive.com>
1303 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1305 <UDIV>: Increase cost if !TARGET_DIV.
1307 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1309 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1310 (define_attr "cr_logical_3op"): New.
1311 (cceq_ior_compare): Adjust.
1312 (cceq_ior_compare_complement): Adjust.
1313 (*cceq_rev_compare): Adjust.
1314 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1315 (is_cracked_insn): Adjust.
1316 (insn_must_be_first_in_group): Adjust.
1317 * config/rs6000/40x.md: Adjust.
1318 * config/rs6000/440.md: Adjust.
1319 * config/rs6000/476.md: Adjust.
1320 * config/rs6000/601.md: Adjust.
1321 * config/rs6000/603.md: Adjust.
1322 * config/rs6000/6xx.md: Adjust.
1323 * config/rs6000/7450.md: Adjust.
1324 * config/rs6000/7xx.md: Adjust.
1325 * config/rs6000/8540.md: Adjust.
1326 * config/rs6000/cell.md: Adjust.
1327 * config/rs6000/e300c2c3.md: Adjust.
1328 * config/rs6000/e500mc.md: Adjust.
1329 * config/rs6000/e500mc64.md: Adjust.
1330 * config/rs6000/e5500.md: Adjust.
1331 * config/rs6000/e6500.md: Adjust.
1332 * config/rs6000/mpc.md: Adjust.
1333 * config/rs6000/power4.md: Adjust.
1334 * config/rs6000/power5.md: Adjust.
1335 * config/rs6000/power6.md: Adjust.
1336 * config/rs6000/power7.md: Adjust.
1337 * config/rs6000/power8.md: Adjust.
1338 * config/rs6000/power9.md: Adjust.
1339 * config/rs6000/rs64.md: Adjust.
1340 * config/rs6000/titan.md: Adjust.
1342 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1344 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1345 ix86_indirect_branch_register logic.
1347 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1349 * config/i386/constraints.md (Bs): Update
1350 ix86_indirect_branch_register check. Don't check
1351 ix86_indirect_branch_register with GOT_memory_operand.
1353 * config/i386/predicates.md (GOT_memory_operand): Don't check
1354 ix86_indirect_branch_register here.
1355 (GOT32_symbol_operand): Likewise.
1357 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1359 * config/i386/predicates.md (constant_call_address_operand):
1360 Rewrite ix86_indirect_branch_register logic.
1361 (sibcall_insn_operand): Likewise.
1363 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1365 * config/i386/constraints.md (Bs): Replace
1366 ix86_indirect_branch_thunk_register with
1367 ix86_indirect_branch_register.
1369 * config/i386/i386.md (indirect_jump): Likewise.
1370 (tablejump): Likewise.
1371 (*sibcall_memory): Likewise.
1372 (*sibcall_value_memory): Likewise.
1373 Peepholes of indirect call and jump via memory: Likewise.
1374 * config/i386/i386.opt: Likewise.
1375 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1376 (GOT_memory_operand): Likewise.
1377 (call_insn_operand): Likewise.
1378 (sibcall_insn_operand): Likewise.
1379 (GOT32_symbol_operand): Likewise.
1381 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1384 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1385 type rather than type addr's type points to.
1386 (expand_omp_atomic_mutex): Likewise.
1387 (expand_omp_atomic): Likewise.
1389 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1392 * config/i386/i386.c (output_indirect_thunk_function): Use
1393 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1394 for __x86_return_thunk.
1396 2018-01-15 Richard Biener <rguenther@suse.de>
1399 * expmed.c (extract_bit_field_1): Fix typo.
1401 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1404 * config/arm/iterators.md (VF): New mode iterator.
1405 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1406 Remove integer-related logic from pattern.
1407 (neon_vabd<mode>_3): Likewise.
1409 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1412 * common.opt (fstrict-overflow): No longer an alias.
1413 (fwrapv-pointer): New option.
1414 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1415 also for pointer types based on flag_wrapv_pointer.
1416 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1417 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1418 opts->x_flag_wrapv got set.
1419 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1420 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1421 POINTER_TYPE_OVERFLOW_UNDEFINED.
1422 * match.pd: Likewise in address comparison pattern.
1423 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1425 2018-01-15 Richard Biener <rguenther@suse.de>
1428 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1429 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1430 Reset type names to their identifier if their TYPE_DECL doesn't
1431 have linkage (and thus is used for ODR and devirt).
1432 (save_debug_info_for_decl): Remove.
1433 (save_debug_info_for_type): Likewise.
1434 (add_tree_to_fld_list): Adjust.
1435 * tree-pretty-print.c (dump_generic_node): Make dumping of
1436 type names more robust.
1438 2018-01-15 Richard Biener <rguenther@suse.de>
1440 * BASE-VER: Bump to 8.0.1.
1442 2018-01-14 Martin Sebor <msebor@redhat.com>
1445 * builtins.c (check_access): Avoid warning when the no-warning bit
1448 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1450 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1451 * ira-color (allocno_hard_regs_compare): Likewise.
1453 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1456 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1457 Use .pushsection/.popsection.
1459 2018-01-14 Martin Sebor <msebor@redhat.com>
1462 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1464 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1466 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1467 entry from extra_headers.
1468 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1469 extra_headers, make the list bitwise identical to the i?86-*-* one.
1471 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1473 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1474 -mcmodel=large with -mindirect-branch=thunk,
1475 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1476 -mfunction-return=thunk-extern.
1477 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1478 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1479 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1481 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1483 * config/i386/i386.c (print_reg): Print the name of the full
1484 integer register without '%'.
1485 (ix86_print_operand): Handle 'V'.
1486 * doc/extend.texi: Document 'V' modifier.
1488 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1490 * config/i386/constraints.md (Bs): Disallow memory operand for
1491 -mindirect-branch-register.
1493 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1494 (GOT_memory_operand): Likewise.
1495 (call_insn_operand): Likewise.
1496 (sibcall_insn_operand): Likewise.
1497 (GOT32_symbol_operand): Likewise.
1498 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1499 for -mindirect-branch-register.
1500 (tablejump): Likewise.
1501 (*sibcall_memory): Likewise.
1502 (*sibcall_value_memory): Likewise.
1503 Disallow peepholes of indirect call and jump via memory for
1504 -mindirect-branch-register.
1505 (*call_pop): Replace m with Bw.
1506 (*call_value_pop): Likewise.
1507 (*sibcall_pop_memory): Replace m with Bs.
1508 * config/i386/i386.opt (mindirect-branch-register): New option.
1509 * doc/invoke.texi: Document -mindirect-branch-register option.
1511 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1513 * config/i386/i386-protos.h (ix86_output_function_return): New.
1514 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1515 set function_return_type.
1516 (indirect_thunk_name): Add ret_p to indicate thunk for function
1518 (output_indirect_thunk_function): Pass false to
1519 indirect_thunk_name.
1520 (ix86_output_indirect_branch_via_reg): Likewise.
1521 (ix86_output_indirect_branch_via_push): Likewise.
1522 (output_indirect_thunk_function): Create alias for function
1523 return thunk if regno < 0.
1524 (ix86_output_function_return): New function.
1525 (ix86_handle_fndecl_attribute): Handle function_return.
1526 (ix86_attribute_table): Add function_return.
1527 * config/i386/i386.h (machine_function): Add
1528 function_return_type.
1529 * config/i386/i386.md (simple_return_internal): Use
1530 ix86_output_function_return.
1531 (simple_return_internal_long): Likewise.
1532 * config/i386/i386.opt (mfunction-return=): New option.
1533 (indirect_branch): Mention -mfunction-return=.
1534 * doc/extend.texi: Document function_return function attribute.
1535 * doc/invoke.texi: Document -mfunction-return= option.
1537 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1539 * config/i386/i386-opts.h (indirect_branch): New.
1540 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1541 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1542 with local indirect jump when converting indirect call and jump.
1543 (ix86_set_indirect_branch_type): New.
1544 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1545 (indirectlabelno): New.
1546 (indirect_thunk_needed): Likewise.
1547 (indirect_thunk_bnd_needed): Likewise.
1548 (indirect_thunks_used): Likewise.
1549 (indirect_thunks_bnd_used): Likewise.
1550 (INDIRECT_LABEL): Likewise.
1551 (indirect_thunk_name): Likewise.
1552 (output_indirect_thunk): Likewise.
1553 (output_indirect_thunk_function): Likewise.
1554 (ix86_output_indirect_branch_via_reg): Likewise.
1555 (ix86_output_indirect_branch_via_push): Likewise.
1556 (ix86_output_indirect_branch): Likewise.
1557 (ix86_output_indirect_jmp): Likewise.
1558 (ix86_code_end): Call output_indirect_thunk_function if needed.
1559 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1561 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1562 (ix86_attribute_table): Add indirect_branch.
1563 * config/i386/i386.h (machine_function): Add indirect_branch_type
1564 and has_local_indirect_jump.
1565 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1567 (tablejump): Likewise.
1568 (*indirect_jump): Use ix86_output_indirect_jmp.
1569 (*tablejump_1): Likewise.
1570 (simple_return_indirect_internal): Likewise.
1571 * config/i386/i386.opt (mindirect-branch=): New option.
1572 (indirect_branch): New.
1575 (thunk-inline): Likewise.
1576 (thunk-extern): Likewise.
1577 * doc/extend.texi: Document indirect_branch function attribute.
1578 * doc/invoke.texi: Document -mindirect-branch= option.
1580 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1583 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1585 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1587 * ipa-inline.c (want_inline_small_function_p): Return false if
1588 inlining has already failed with CIF_FINAL_ERROR.
1589 (update_caller_keys): Call want_inline_small_function_p before
1591 (update_callee_keys): Likewise.
1593 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1595 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1597 (rs6000_quadword_masked_address_p): Likewise.
1598 (quad_aligned_load_p): Likewise.
1599 (quad_aligned_store_p): Likewise.
1600 (const_load_sequence_p): Add comment to describe the outer-most loop.
1601 (mimic_memory_attributes_and_flags): New function.
1602 (rs6000_gen_stvx): Likewise.
1603 (replace_swapped_aligned_store): Likewise.
1604 (rs6000_gen_lvx): Likewise.
1605 (replace_swapped_aligned_load): Likewise.
1606 (replace_swapped_load_constant): Capitalize argument name in
1607 comment describing this function.
1608 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1609 and stores that access quad-word aligned addresses and replace
1610 with stvx or lvx instructions when appropriate.
1611 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1612 New function prototype.
1613 (rs6000_quadword_masked_address_p): Likewise.
1614 (rs6000_gen_lvx): Likewise.
1615 (rs6000_gen_stvx): Likewise.
1616 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1617 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1618 when memory address is aligned.
1619 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1620 this split to select lvx instruction when memory address is aligned.
1621 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1622 instruction when memory address is aligned.
1623 (*vsx_le_perm_load_v16qi): Likewise.
1624 (four unnamed splitters): Modify to select the stvx instruction
1625 when memory is aligned.
1627 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1629 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1630 which appears in the queue multiple times.
1632 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1633 Alan Hayward <alan.hayward@arm.com>
1634 David Sherwood <david.sherwood@arm.com>
1636 * tree-vectorizer.h (vec_lower_bound): New structure.
1637 (_loop_vec_info): Add check_nonzero and lower_bounds.
1638 (LOOP_VINFO_CHECK_NONZERO): New macro.
1639 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1640 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1641 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1642 fields. Make seg_len the distance travelled, not including the
1644 (dr_direction_indicator): Declare.
1645 (dr_zero_step_indicator): Likewise.
1646 (dr_known_forward_stride_p): Likewise.
1647 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1649 (runtime_alias_check_p): Allow runtime alias checks with
1651 (operator ==): Compare access_size and align.
1652 (prune_runtime_alias_test_list): Rework for new distinction between
1653 the access_size and seg_len.
1654 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1656 (get_segment_min_max): New function.
1657 (create_intersect_range_checks): Use it.
1658 (dr_step_indicator): New function.
1659 (dr_direction_indicator): Likewise.
1660 (dr_zero_step_indicator): Likewise.
1661 (dr_known_forward_stride_p): Likewise.
1662 * tree-loop-distribution.c (data_ref_segment_size): Return
1663 DR_STEP * (niters - 1).
1664 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1666 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1667 (vect_preserves_scalar_order_p): New function, split out from...
1668 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1669 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1670 (vect_vfa_access_size): New function.
1671 (vect_vfa_align): Likewise.
1672 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1673 (dump_lower_bound): New function.
1674 (vect_check_lower_bound): Likewise.
1675 (vect_small_gap_p): Likewise.
1676 (vectorizable_with_step_bound_p): Likewise.
1677 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1678 depencies if the vectorization factor is 1. Convert the checks
1679 for nonzero steps into checks on the bounds of DR_STEP. Try using
1680 a bunds check for variable steps if the minimum required step is
1681 relatively small. Update calls to the dr_with_seg_len
1682 constructor and to vect_compile_time_alias.
1683 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1685 (vect_loop_versioning): Call it.
1686 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1688 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1690 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1691 Alan Hayward <alan.hayward@arm.com>
1692 David Sherwood <david.sherwood@arm.com>
1694 * doc/sourcebuild.texi (vect_scatter_store): Document.
1695 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1697 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1699 * genopinit.c (main): Add supports_vec_scatter_store and
1700 supports_vec_scatter_store_cached to target_optabs.
1701 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1702 IFN_MASK_SCATTER_STORE.
1703 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1705 * internal-fn.h (internal_store_fn_p): Declare.
1706 (internal_fn_stored_value_index): Likewise.
1707 * internal-fn.c (scatter_store_direct): New macro.
1708 (expand_scatter_store_optab_fn): New function.
1709 (direct_scatter_store_optab_supported_p): New macro.
1710 (internal_store_fn_p): New function.
1711 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1712 IFN_MASK_SCATTER_STORE.
1713 (internal_fn_mask_index): Likewise.
1714 (internal_fn_stored_value_index): New function.
1715 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1717 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1718 * optabs-query.c (supports_vec_scatter_store_p): New function.
1719 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1720 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1721 true for scatter stores.
1722 (vect_gather_scatter_fn_p): Handle scatter stores too.
1723 (vect_check_gather_scatter): Consider using scatter stores if
1724 supports_vec_scatter_store_p.
1725 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1727 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1728 internal_fn_stored_value_index.
1729 (check_load_store_masking): Handle scatter stores too.
1730 (vect_get_store_rhs): Make public.
1731 (vectorizable_call): Use internal_store_fn_p.
1732 (vectorizable_store): Handle scatter store internal functions.
1733 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1734 when deciding whether the end of the group has been reached.
1735 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1736 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1737 (mask_scatter_store<mode>): New insns.
1739 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1740 Alan Hayward <alan.hayward@arm.com>
1741 David Sherwood <david.sherwood@arm.com>
1743 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1744 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1745 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1747 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1748 Use vect_truncate_gather_scatter_offset if we can't treat the
1749 operation as a normal gather load or scatter store.
1750 (get_group_load_store_type): Take the gather_scatter_info
1751 as argument. Try using a gather load or scatter store for
1752 single-element groups.
1753 (get_load_store_type): Update calls to get_group_load_store_type
1754 and vect_use_strided_gather_scatters_p.
1756 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1757 Alan Hayward <alan.hayward@arm.com>
1758 David Sherwood <david.sherwood@arm.com>
1760 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1761 optional tree argument.
1762 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1764 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1765 but continue to use the current value as a fallback.
1766 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1767 to compare the updates.
1768 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1769 (get_load_store_type): Use it when handling a strided access.
1770 (vect_get_strided_load_store_ops): New function.
1771 (vect_get_data_ptr_increment): Likewise.
1772 (vectorizable_load): Handle strided gather loads. Always pass
1773 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1775 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1776 Alan Hayward <alan.hayward@arm.com>
1777 David Sherwood <david.sherwood@arm.com>
1779 * doc/md.texi (gather_load@var{m}): Document.
1780 (mask_gather_load@var{m}): Likewise.
1781 * genopinit.c (main): Add supports_vec_gather_load and
1782 supports_vec_gather_load_cached to target_optabs.
1783 * optabs-tree.c (init_tree_optimization_optabs): Use
1784 ggc_cleared_alloc to allocate target_optabs.
1785 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1786 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1788 * internal-fn.h (internal_load_fn_p): Declare.
1789 (internal_gather_scatter_fn_p): Likewise.
1790 (internal_fn_mask_index): Likewise.
1791 (internal_gather_scatter_fn_supported_p): Likewise.
1792 * internal-fn.c (gather_load_direct): New macro.
1793 (expand_gather_load_optab_fn): New function.
1794 (direct_gather_load_optab_supported_p): New macro.
1795 (direct_internal_fn_optab): New function.
1796 (internal_load_fn_p): Likewise.
1797 (internal_gather_scatter_fn_p): Likewise.
1798 (internal_fn_mask_index): Likewise.
1799 (internal_gather_scatter_fn_supported_p): Likewise.
1800 * optabs-query.c (supports_at_least_one_mode_p): New function.
1801 (supports_vec_gather_load_p): Likewise.
1802 * optabs-query.h (supports_vec_gather_load_p): Declare.
1803 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1804 and memory_type field.
1805 (NUM_PATTERNS): Bump to 15.
1806 * tree-vect-data-refs.c: Include internal-fn.h.
1807 (vect_gather_scatter_fn_p): New function.
1808 (vect_describe_gather_scatter_call): Likewise.
1809 (vect_check_gather_scatter): Try using internal functions for
1810 gather loads. Recognize existing calls to a gather load function.
1811 (vect_analyze_data_refs): Consider using gather loads if
1812 supports_vec_gather_load_p.
1813 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1814 (vect_get_gather_scatter_offset_type): Likewise.
1815 (vect_convert_mask_for_vectype): Likewise.
1816 (vect_add_conversion_to_patterm): Likewise.
1817 (vect_try_gather_scatter_pattern): Likewise.
1818 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1819 (vect_vect_recog_func_ptrs): Add it.
1820 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1821 internal_fn_mask_index and internal_gather_scatter_fn_p.
1822 (check_load_store_masking): Take the gather_scatter_info as an
1823 argument and handle gather loads.
1824 (vect_get_gather_scatter_ops): New function.
1825 (vectorizable_call): Check internal_load_fn_p.
1826 (vectorizable_load): Likewise. Handle gather load internal
1828 (vectorizable_store): Update call to check_load_store_masking.
1829 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1830 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1831 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1832 (aarch64_gather_scale_operand_d): New predicates.
1833 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1834 (mask_gather_load<mode>): New insns.
1836 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1837 Alan Hayward <alan.hayward@arm.com>
1838 David Sherwood <david.sherwood@arm.com>
1840 * optabs.def (fold_left_plus_optab): New optab.
1841 * doc/md.texi (fold_left_plus_@var{m}): Document.
1842 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1843 * internal-fn.c (fold_left_direct): Define.
1844 (expand_fold_left_optab_fn): Likewise.
1845 (direct_fold_left_optab_supported_p): Likewise.
1846 * fold-const-call.c (fold_const_fold_left): New function.
1847 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1848 * tree-parloops.c (valid_reduction_p): New function.
1849 (gather_scalar_reductions): Use it.
1850 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1851 (vect_finish_replace_stmt): Declare.
1852 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1853 (needs_fold_left_reduction_p): New function, split out from...
1854 (vect_is_simple_reduction): ...here. Accept reductions that
1855 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1856 (vect_force_simple_reduction): Also store the reduction type in
1857 the assignment's STMT_VINFO_REDUC_TYPE.
1858 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1859 (merge_with_identity): New function.
1860 (vect_expand_fold_left): Likewise.
1861 (vectorize_fold_left_reduction): Likewise.
1862 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1863 scalar phi in place for it. Check for target support and reject
1864 cases that would reassociate the operation. Defer the transform
1865 phase to vectorize_fold_left_reduction.
1866 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1867 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1868 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1870 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1872 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1873 call to ifc_temp_var.
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 * target.def (legitimize_address_displacement): Take the original
1880 offset as a poly_int.
1881 * targhooks.h (default_legitimize_address_displacement): Update
1883 * targhooks.c (default_legitimize_address_displacement): Likewise.
1884 * doc/tm.texi: Regenerate.
1885 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1886 as an argument, moving assert of ad->disp == ad->disp_term to...
1887 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1888 Try calling targetm.legitimize_address_displacement before expanding
1889 the address rather than afterwards, and adjust for the new interface.
1890 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1891 Match the new hook interface. Handle SVE addresses.
1892 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1895 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1897 * Makefile.in (OBJS): Add early-remat.o.
1898 * target.def (select_early_remat_modes): New hook.
1899 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1900 * doc/tm.texi: Regenerate.
1901 * targhooks.h (default_select_early_remat_modes): Declare.
1902 * targhooks.c (default_select_early_remat_modes): New function.
1903 * timevar.def (TV_EARLY_REMAT): New timevar.
1904 * passes.def (pass_early_remat): New pass.
1905 * tree-pass.h (make_pass_early_remat): Declare.
1906 * early-remat.c: New file.
1907 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1909 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1911 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1912 Alan Hayward <alan.hayward@arm.com>
1913 David Sherwood <david.sherwood@arm.com>
1915 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1916 vfm1 with a bound_epilog parameter.
1917 (vect_do_peeling): Update calls accordingly, and move the prologue
1918 call earlier in the function. Treat the base bound_epilog as 0 for
1919 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1920 this base when peeling for gaps.
1921 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1922 with fully-masked loops.
1923 (vect_estimate_min_profitable_iters): Handle the single peeled
1924 iteration in that case.
1926 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1927 Alan Hayward <alan.hayward@arm.com>
1928 David Sherwood <david.sherwood@arm.com>
1930 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1931 single-element interleaving even if the size is not a power of 2.
1932 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1933 accesses for single-element interleaving if the group size is
1936 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1937 Alan Hayward <alan.hayward@arm.com>
1938 David Sherwood <david.sherwood@arm.com>
1940 * doc/md.texi (fold_extract_last_@var{m}): Document.
1941 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1942 * optabs.def (fold_extract_last_optab): New optab.
1943 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1944 * internal-fn.c (fold_extract_direct): New macro.
1945 (expand_fold_extract_optab_fn): Likewise.
1946 (direct_fold_extract_optab_supported_p): Likewise.
1947 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1948 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1949 EXTRACT_LAST_REDUCTION.
1950 (get_initial_def_for_reduction): Do not create an initial vector
1951 for EXTRACT_LAST_REDUCTION reductions.
1952 (vectorizable_reduction): Leave the scalar phi in place for
1953 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1954 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1955 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1956 transform phase to vectorizable_condition.
1957 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1959 (vect_finish_stmt_generation): ...here.
1960 (vect_finish_replace_stmt): New function.
1961 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1962 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1964 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1966 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1967 Alan Hayward <alan.hayward@arm.com>
1968 David Sherwood <david.sherwood@arm.com>
1970 * doc/md.texi (extract_last_@var{m}): Document.
1971 * optabs.def (extract_last_optab): New optab.
1972 * internal-fn.def (EXTRACT_LAST): New internal function.
1973 * internal-fn.c (cond_unary_direct): New macro.
1974 (expand_cond_unary_optab_fn): Likewise.
1975 (direct_cond_unary_optab_supported_p): Likewise.
1976 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1977 loops using EXTRACT_LAST.
1978 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1979 (extract_last_<mode>): ...this optab.
1980 (vec_extract<mode><Vel>): Update accordingly.
1982 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1983 Alan Hayward <alan.hayward@arm.com>
1984 David Sherwood <david.sherwood@arm.com>
1986 * target.def (empty_mask_is_expensive): New hook.
1987 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
1988 * doc/tm.texi: Regenerate.
1989 * targhooks.h (default_empty_mask_is_expensive): Declare.
1990 * targhooks.c (default_empty_mask_is_expensive): New function.
1991 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
1992 if the target says that empty masks are expensive.
1993 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
1995 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
1997 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1998 Alan Hayward <alan.hayward@arm.com>
1999 David Sherwood <david.sherwood@arm.com>
2001 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2002 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2003 (vect_use_loop_mask_for_alignment_p): New function.
2004 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2005 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2006 niters_skip argument. Make sure that the first niters_skip elements
2007 of the first iteration are inactive.
2008 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2009 Update call to vect_set_loop_masks_directly.
2010 (get_misalign_in_elems): New function, split out from...
2011 (vect_gen_prolog_loop_niters): ...here.
2012 (vect_update_init_of_dr): Take a code argument that specifies whether
2013 the adjustment should be added or subtracted.
2014 (vect_update_init_of_drs): Likewise.
2015 (vect_prepare_for_masked_peels): New function.
2016 (vect_do_peeling): Skip prologue peeling if we're using a mask
2017 instead. Update call to vect_update_inits_of_drs.
2018 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2020 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2021 alignment. Do not include the number of peeled iterations in
2022 the minimum threshold in that case.
2023 (vectorizable_induction): Adjust the start value down by
2024 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2025 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2026 Take the number of skipped iterations into account when calculating
2028 * tree-vect-stmts.c (vect_gen_while_not): New function.
2030 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2031 Alan Hayward <alan.hayward@arm.com>
2032 David Sherwood <david.sherwood@arm.com>
2034 * doc/sourcebuild.texi (vect_fully_masked): Document.
2035 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2037 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2039 (vect_analyze_loop_2): ...here. Don't check the vectorization
2040 factor against the number of loop iterations if the loop is
2043 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2044 Alan Hayward <alan.hayward@arm.com>
2045 David Sherwood <david.sherwood@arm.com>
2047 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2048 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2049 (dump_groups): Update accordingly.
2050 (iv_use::mem_type): New member variable.
2051 (address_p): New function.
2052 (record_use): Add a mem_type argument and initialize the new
2054 (record_group_use): Add a mem_type argument. Use address_p.
2055 Remove obsolete null checks of base_object. Update call to record_use.
2056 (find_interesting_uses_op): Update call to record_group_use.
2057 (find_interesting_uses_cond): Likewise.
2058 (find_interesting_uses_address): Likewise.
2059 (get_mem_type_for_internal_fn): New function.
2060 (find_address_like_use): Likewise.
2061 (find_interesting_uses_stmt): Try find_address_like_use before
2062 calling find_interesting_uses_op.
2063 (addr_offset_valid_p): Use the iv mem_type field as the type
2064 of the addressed memory.
2065 (add_autoinc_candidates): Likewise.
2066 (get_address_cost): Likewise.
2067 (split_small_address_groups_p): Use address_p.
2068 (split_address_groups): Likewise.
2069 (add_iv_candidate_for_use): Likewise.
2070 (autoinc_possible_for_pair): Likewise.
2071 (rewrite_groups): Likewise.
2072 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2073 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2074 (get_alias_ptr_type_for_ptr_address): New function.
2075 (rewrite_use_address): Rewrite address uses in calls that were
2076 identified by find_address_like_use.
2078 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2079 Alan Hayward <alan.hayward@arm.com>
2080 David Sherwood <david.sherwood@arm.com>
2082 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2084 * gimple-expr.h (is_gimple_addressable: Likewise.
2085 * gimple-expr.c (is_gimple_address): Likewise.
2086 * internal-fn.c (expand_call_mem_ref): New function.
2087 (expand_mask_load_optab_fn): Use it.
2088 (expand_mask_store_optab_fn): Likewise.
2090 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2091 Alan Hayward <alan.hayward@arm.com>
2092 David Sherwood <david.sherwood@arm.com>
2094 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2095 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2096 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2097 (cond_umax@var{mode}): Document.
2098 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2099 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2100 (cond_umin_optab, cond_umax_optab): New optabs.
2101 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2102 (COND_IOR, COND_XOR): New internal functions.
2103 * internal-fn.h (get_conditional_internal_fn): Declare.
2104 * internal-fn.c (cond_binary_direct): New macro.
2105 (expand_cond_binary_optab_fn): Likewise.
2106 (direct_cond_binary_optab_supported_p): Likewise.
2107 (get_conditional_internal_fn): New function.
2108 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2109 Cope with reduction statements that are vectorized as calls rather
2111 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2112 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2113 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2114 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2115 (UNSPEC_COND_EOR): New unspecs.
2116 (optab): Add mappings for them.
2117 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2118 (sve_int_op, sve_fp_op): New int attributes.
2120 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2121 Alan Hayward <alan.hayward@arm.com>
2122 David Sherwood <david.sherwood@arm.com>
2124 * optabs.def (while_ult_optab): New optab.
2125 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2126 * internal-fn.def (WHILE_ULT): New internal function.
2127 * internal-fn.h (direct_internal_fn_supported_p): New override
2128 that takes two types as argument.
2129 * internal-fn.c (while_direct): New macro.
2130 (expand_while_optab_fn): New function.
2131 (convert_optab_supported_p): Likewise.
2132 (direct_while_optab_supported_p): New macro.
2133 * wide-int.h (wi::udiv_ceil): New function.
2134 * tree-vectorizer.h (rgroup_masks): New structure.
2135 (vec_loop_masks): New typedef.
2136 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2138 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2139 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2140 (vect_max_vf): New function.
2141 (slpeel_make_loop_iterate_ntimes): Delete.
2142 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2143 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2144 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2145 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2146 internal-fn.h, stor-layout.h and optabs-query.h.
2147 (vect_set_loop_mask): New function.
2148 (add_preheader_seq): Likewise.
2149 (add_header_seq): Likewise.
2150 (interleave_supported_p): Likewise.
2151 (vect_maybe_permute_loop_masks): Likewise.
2152 (vect_set_loop_masks_directly): Likewise.
2153 (vect_set_loop_condition_masked): Likewise.
2154 (vect_set_loop_condition_unmasked): New function, split out from
2155 slpeel_make_loop_iterate_ntimes.
2156 (slpeel_make_loop_iterate_ntimes): Rename to..
2157 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2158 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2159 (vect_do_peeling): Update call accordingly.
2160 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2162 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2163 mask_compare_type, can_fully_mask_p and fully_masked_p.
2164 (release_vec_loop_masks): New function.
2165 (_loop_vec_info): Use it to free the loop masks.
2166 (can_produce_all_loop_masks_p): New function.
2167 (vect_get_max_nscalars_per_iter): Likewise.
2168 (vect_verify_full_masking): Likewise.
2169 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2170 retries, and free the mask rgroups before retrying. Check loop-wide
2171 reasons for disallowing fully-masked loops. Make the final decision
2172 about whether use a fully-masked loop or not.
2173 (vect_estimate_min_profitable_iters): Do not assume that peeling
2174 for the number of iterations will be needed for fully-masked loops.
2175 (vectorizable_reduction): Disable fully-masked loops.
2176 (vectorizable_live_operation): Likewise.
2177 (vect_halve_mask_nunits): New function.
2178 (vect_double_mask_nunits): Likewise.
2179 (vect_record_loop_mask): Likewise.
2180 (vect_get_loop_mask): Likewise.
2181 (vect_transform_loop): Handle the case in which the final loop
2182 iteration might handle a partial vector. Call vect_set_loop_condition
2183 instead of slpeel_make_loop_iterate_ntimes.
2184 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2185 (check_load_store_masking): New function.
2186 (prepare_load_store_mask): Likewise.
2187 (vectorizable_store): Handle fully-masked loops.
2188 (vectorizable_load): Likewise.
2189 (supportable_widening_operation): Use vect_halve_mask_nunits for
2191 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2192 (vect_gen_while): New function.
2193 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2194 (aarch64_uqdec<mode>): New insn.
2196 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2197 Alan Hayward <alan.hayward@arm.com>
2198 David Sherwood <david.sherwood@arm.com>
2200 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2201 (reduc_xor_scal_optab): New optabs.
2202 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2203 (reduc_xor_scal_@var{m}): Document.
2204 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2205 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2207 * fold-const-call.c (fold_const_call): Handle them.
2208 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2209 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2210 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2211 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2212 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2213 (UNSPEC_XORV): New unspecs.
2214 (optab): Add entries for them.
2215 (BITWISEV): New int iterator.
2216 (bit_reduc_op): New int attributes.
2218 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2219 Alan Hayward <alan.hayward@arm.com>
2220 David Sherwood <david.sherwood@arm.com>
2222 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2223 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2224 * optabs.def (vec_shl_insert_optab): New optab.
2225 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2226 (duplicate_and_interleave): Likewise.
2227 * tree-vect-loop.c: Include internal-fn.h.
2228 (neutral_op_for_slp_reduction): New function, split out from
2229 get_initial_defs_for_reduction.
2230 (get_initial_def_for_reduction): Handle option 2 for variable-length
2231 vectors by loading the neutral value into a vector and then shifting
2232 the initial value into element 0.
2233 (get_initial_defs_for_reduction): Replace the code argument with
2234 the neutral value calculated by neutral_op_for_slp_reduction.
2235 Use gimple_build_vector for constant-length vectors.
2236 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2237 but the first group_size elements have a neutral value.
2238 Use duplicate_and_interleave otherwise.
2239 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2240 Update call to get_initial_defs_for_reduction. Handle SLP
2241 reductions for variable-length vectors by creating one vector
2242 result for each scalar result, with the elements associated
2243 with other scalar results stubbed out with the neutral value.
2244 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2245 Require IFN_VEC_SHL_INSERT for double reductions on
2246 variable-length vectors, or SLP reductions that have
2247 a neutral value. Require can_duplicate_and_interleave_p
2248 support for variable-length unchained SLP reductions if there
2249 is no neutral value, such as for MIN/MAX reductions. Also require
2250 the number of vector elements to be a multiple of the number of
2251 SLP statements when doing variable-length unchained SLP reductions.
2252 Update call to vect_create_epilog_for_reduction.
2253 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2254 and remove initial values.
2255 (duplicate_and_interleave): Make public.
2256 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2257 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2259 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2260 Alan Hayward <alan.hayward@arm.com>
2261 David Sherwood <david.sherwood@arm.com>
2263 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2264 (can_duplicate_and_interleave_p): New function.
2265 (vect_get_and_check_slp_defs): Take the vector of statements
2266 rather than just the current one. Remove excess parentheses.
2267 Restriction rejectinon of vect_constant_def and vect_external_def
2268 for variable-length vectors to boolean types, or types for which
2269 can_duplicate_and_interleave_p is false.
2270 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2271 (duplicate_and_interleave): New function.
2272 (vect_get_constant_vectors): Use gimple_build_vector for
2273 constant-length vectors and suitable variable-length constant
2274 vectors. Use duplicate_and_interleave for other variable-length
2275 vectors. Don't defer the update when inserting new statements.
2277 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2278 Alan Hayward <alan.hayward@arm.com>
2279 David Sherwood <david.sherwood@arm.com>
2281 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2282 min_profitable_iters doesn't go negative.
2284 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2285 Alan Hayward <alan.hayward@arm.com>
2286 David Sherwood <david.sherwood@arm.com>
2288 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2289 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2290 * optabs.def (vec_mask_load_lanes_optab): New optab.
2291 (vec_mask_store_lanes_optab): Likewise.
2292 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2293 (MASK_STORE_LANES): Likewise.
2294 * internal-fn.c (mask_load_lanes_direct): New macro.
2295 (mask_store_lanes_direct): Likewise.
2296 (expand_mask_load_optab_fn): Handle masked operations.
2297 (expand_mask_load_lanes_optab_fn): New macro.
2298 (expand_mask_store_optab_fn): Handle masked operations.
2299 (expand_mask_store_lanes_optab_fn): New macro.
2300 (direct_mask_load_lanes_optab_supported_p): Likewise.
2301 (direct_mask_store_lanes_optab_supported_p): Likewise.
2302 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2304 (vect_load_lanes_supported): Likewise.
2305 * tree-vect-data-refs.c (strip_conversion): New function.
2306 (can_group_stmts_p): Likewise.
2307 (vect_analyze_data_ref_accesses): Use it instead of checking
2308 for a pair of assignments.
2309 (vect_store_lanes_supported): Take a masked_p parameter.
2310 (vect_load_lanes_supported): Likewise.
2311 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2312 vect_store_lanes_supported and vect_load_lanes_supported.
2313 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2314 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2315 parameter. Don't allow gaps for masked accesses.
2316 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2317 and vect_load_lanes_supported.
2318 (get_load_store_type): Take a masked_p parameter and update
2319 call to get_group_load_store_type.
2320 (vectorizable_store): Update call to get_load_store_type.
2321 Handle IFN_MASK_STORE_LANES.
2322 (vectorizable_load): Update call to get_load_store_type.
2323 Handle IFN_MASK_LOAD_LANES.
2325 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2326 Alan Hayward <alan.hayward@arm.com>
2327 David Sherwood <david.sherwood@arm.com>
2329 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2331 * config/aarch64/aarch64-protos.h
2332 (aarch64_sve_struct_memory_operand_p): Declare.
2333 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2334 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2335 (VPRED, vpred): Handle SVE structure modes.
2336 * config/aarch64/constraints.md (Utx): New constraint.
2337 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2338 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2339 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2340 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2341 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2342 structure modes. Split into pieces after RA.
2343 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2344 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2346 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2347 SVE structure modes.
2348 (aarch64_classify_address): Likewise.
2349 (sizetochar): Move earlier in file.
2350 (aarch64_print_operand): Handle SVE register lists.
2351 (aarch64_array_mode): New function.
2352 (aarch64_sve_struct_memory_operand_p): Likewise.
2353 (TARGET_ARRAY_MODE): Redefine.
2355 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2356 Alan Hayward <alan.hayward@arm.com>
2357 David Sherwood <david.sherwood@arm.com>
2359 * target.def (array_mode): New target hook.
2360 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2361 * doc/tm.texi: Regenerate.
2362 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2363 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2364 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2366 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2369 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2370 Alan Hayward <alan.hayward@arm.com>
2371 David Sherwood <david.sherwood@arm.com>
2373 * fold-const.c (fold_binary_loc): Check the argument types
2374 rather than the result type when testing for a vector operation.
2376 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2378 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2379 * doc/tm.texi: Regenerate.
2381 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2382 Alan Hayward <alan.hayward@arm.com>
2383 David Sherwood <david.sherwood@arm.com>
2385 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2386 (sve): Document new AArch64 extension.
2387 * doc/md.texi (w): Extend the description of the AArch64
2388 constraint to include SVE vectors.
2389 (Upl, Upa): Document new AArch64 predicate constraints.
2390 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2392 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2393 (msve-vector-bits=): New option.
2394 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2395 SVE when these are disabled.
2396 (sve): New extension.
2397 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2398 modes. Adjust their number of units based on aarch64_sve_vg.
2399 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2400 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2401 aarch64_addr_query_type.
2402 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2403 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2404 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2405 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2406 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2407 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2408 (aarch64_simd_imm_zero_p): Delete.
2409 (aarch64_check_zero_based_sve_index_immediate): Declare.
2410 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2411 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2412 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2413 (aarch64_sve_float_mul_immediate_p): Likewise.
2414 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2416 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2417 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2418 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2419 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2420 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2421 (aarch64_regmode_natural_size): Likewise.
2422 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2423 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2425 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2426 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2427 for VG and the SVE predicate registers.
2428 (V_ALIASES): Add a "z"-prefixed alias.
2429 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2430 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2431 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2432 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2433 (REG_CLASS_NAMES): Add entries for them.
2434 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2435 and the predicate registers.
2436 (aarch64_sve_vg): Declare.
2437 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2438 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2439 (REGMODE_NATURAL_SIZE): Define.
2440 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2442 * config/aarch64/aarch64.c: Include cfgrtl.h.
2443 (simd_immediate_info): Add a constructor for series vectors,
2444 and an associated step field.
2445 (aarch64_sve_vg): New variable.
2446 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2447 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2448 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2449 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2450 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2451 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2452 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2453 (aarch64_get_mask_mode): New functions.
2454 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2455 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2456 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2457 predicate modes and predicate registers. Explicitly restrict
2458 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2459 to store a vector mode if it is recognized by
2460 aarch64_classify_vector_mode.
2461 (aarch64_regmode_natural_size): New function.
2462 (aarch64_hard_regno_caller_save_mode): Return the original mode
2464 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2465 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2466 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2467 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2469 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2470 does not overlap dest if the function is frame-related. Handle
2472 (aarch64_split_add_offset): New function.
2473 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2474 them aarch64_add_offset.
2475 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2476 and update call to aarch64_sub_sp.
2477 (aarch64_add_cfa_expression): New function.
2478 (aarch64_expand_prologue): Pass extra temporary registers to the
2479 functions above. Handle the case in which we need to emit new
2480 DW_CFA_expressions for registers that were originally saved
2481 relative to the stack pointer, but now have to be expressed
2482 relative to the frame pointer.
2483 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2485 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2486 IP0 and IP1 values for SVE frames.
2487 (aarch64_expand_vec_series): New function.
2488 (aarch64_expand_sve_widened_duplicate): Likewise.
2489 (aarch64_expand_sve_const_vector): Likewise.
2490 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2491 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2492 into the register, rather than emitting a SET directly.
2493 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2494 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2495 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2496 (offset_9bit_signed_scaled_p): New functions.
2497 (aarch64_replicate_bitmask_imm): New function.
2498 (aarch64_bitmask_imm): Use it.
2499 (aarch64_cannot_force_const_mem): Reject expressions involving
2500 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2501 (aarch64_classify_index): Handle SVE indices, by requiring
2502 a plain register index with a scale that matches the element size.
2503 (aarch64_classify_address): Handle SVE addresses. Assert that
2504 the mode of the address is VOIDmode or an integer mode.
2505 Update call to aarch64_classify_symbol.
2506 (aarch64_classify_symbolic_expression): Update call to
2507 aarch64_classify_symbol.
2508 (aarch64_const_vec_all_in_range_p): New function.
2509 (aarch64_print_vector_float_operand): Likewise.
2510 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2511 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2512 and the FP immediates 1.0 and 0.5.
2513 (aarch64_print_address_internal): Handle SVE addresses.
2514 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2515 (aarch64_regno_regclass): Handle predicate registers.
2516 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2518 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2519 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2520 (aarch64_convert_sve_vector_bits): New function.
2521 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2522 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2524 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2525 Handle SVE vector and predicate modes. Accept VL-based constants
2526 that need only one temporary register, and VL offsets that require
2527 no temporary registers.
2528 (aarch64_conditional_register_usage): Mark the predicate registers
2529 as fixed if SVE isn't available.
2530 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2531 Return true for SVE vector and predicate modes.
2532 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2533 rather than an unsigned int. Handle SVE modes.
2534 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2536 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2538 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2539 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2540 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2541 (aarch64_sve_float_mul_immediate_p): New functions.
2542 (aarch64_sve_valid_immediate): New function.
2543 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2544 Explicitly reject structure modes. Check for INDEX constants.
2545 Handle PTRUE and PFALSE constants.
2546 (aarch64_check_zero_based_sve_index_immediate): New function.
2547 (aarch64_simd_imm_zero_p): Delete.
2548 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2549 vector modes. Accept constants in the range of CNT[BHWD].
2550 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2551 ask for an Advanced SIMD mode.
2552 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2553 (aarch64_simd_vector_alignment): Handle SVE predicates.
2554 (aarch64_vectorize_preferred_vector_alignment): New function.
2555 (aarch64_simd_vector_alignment_reachable): Use it instead of
2557 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2558 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2560 (MAX_VECT_LEN): Delete.
2561 (expand_vec_perm_d): Add a vec_flags field.
2562 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2563 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2564 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2566 (aarch64_evpc_rev): Rename to...
2567 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2568 (aarch64_evpc_rev_global): New function.
2569 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2570 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2572 (aarch64_evpc_sve_tbl): New function.
2573 (aarch64_expand_vec_perm_const_1): Update after rename of
2574 aarch64_evpc_rev. Handle SVE permutes too, trying
2575 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2576 than aarch64_evpc_tbl.
2577 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2578 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2579 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2580 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2581 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2582 (aarch64_expand_sve_vcond): New functions.
2583 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2584 of aarch64_vector_mode_p.
2585 (aarch64_dwarf_poly_indeterminate_value): New function.
2586 (aarch64_compute_pressure_classes): Likewise.
2587 (aarch64_can_change_mode_class): Likewise.
2588 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2589 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2590 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2591 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2592 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2593 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2594 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2595 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2597 (Dn, Dl, Dr): Accept const as well as const_vector.
2598 (Dz): Likewise. Compare against CONST0_RTX.
2599 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2600 of "vector" where appropriate.
2601 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2602 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2603 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2604 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2605 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2606 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2607 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2608 (v_int_equiv): Extend to SVE modes.
2609 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2611 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2612 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2613 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2614 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2615 (SVE_COND_FP_CMP): New int iterators.
2616 (perm_hilo): Handle the new unpack unspecs.
2617 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2619 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2620 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2621 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2622 (aarch64_equality_operator, aarch64_constant_vector_operand)
2623 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2624 (aarch64_sve_nonimmediate_operand): Likewise.
2625 (aarch64_sve_general_operand): Likewise.
2626 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2627 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2628 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2629 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2630 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2631 (aarch64_sve_float_arith_immediate): Likewise.
2632 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2633 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2634 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2635 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2636 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2637 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2638 (aarch64_sve_float_arith_operand): Likewise.
2639 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2640 (aarch64_sve_float_mul_operand): Likewise.
2641 (aarch64_sve_vec_perm_operand): Likewise.
2642 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2643 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2644 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2645 as well as const_vector.
2646 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2647 in file. Use CONST0_RTX and CONSTM1_RTX.
2648 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2649 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2650 Use aarch64_simd_imm_zero.
2651 * config/aarch64/aarch64-sve.md: New file.
2652 * config/aarch64/aarch64.md: Include it.
2653 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2654 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2655 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2656 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2657 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2658 (sve): New attribute.
2659 (enabled): Disable instructions with the sve attribute unless
2661 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2662 aarch64_expand_mov_immediate.
2663 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2664 CNT[BHSD] immediates.
2665 (movti): Split CONST_POLY_INT moves into two halves.
2666 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2667 Split additions that need a temporary here if the destination
2668 is the stack pointer.
2669 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2670 (*add<mode>3_poly_1): New instruction.
2671 (set_clobber_cc): New expander.
2673 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2675 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2676 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2677 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2678 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2679 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2680 Change innermode from fixed_mode_size to machine_mode.
2681 (simplify_subreg): Update call accordingly. Handle a constant-sized
2682 subreg of a variable-length CONST_VECTOR.
2684 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2685 Alan Hayward <alan.hayward@arm.com>
2686 David Sherwood <david.sherwood@arm.com>
2688 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2689 (add_offset_to_base): New function, split out from...
2690 (create_mem_ref): ...here. When handling a scale other than 1,
2691 check first whether the address is valid without the offset.
2692 Add it into the base if so, leaving the index and scale as-is.
2694 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2697 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2698 fold_for_warn before checking if arg2 is INTEGER_CST.
2700 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2702 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2703 (store_multiple_operation): Delete.
2704 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2705 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2706 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2707 guarded by TARGET_STRING.
2708 (rs6000_output_load_multiple): Delete.
2709 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2710 OPTION_MASK_STRING / TARGET_STRING handling.
2711 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2712 (const rs6000_opt_masks) <"string">: Change mask to 0.
2713 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2714 (MASK_STRING): Delete.
2715 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2717 (load_multiple): Delete.
2724 (store_multiple): Delete.
2731 (movmemsi_8reg): Delete.
2732 (corresponding unnamed define_insn): Delete.
2733 (movmemsi_6reg): Delete.
2734 (corresponding unnamed define_insn): Delete.
2735 (movmemsi_4reg): Delete.
2736 (corresponding unnamed define_insn): Delete.
2737 (movmemsi_2reg): Delete.
2738 (corresponding unnamed define_insn): Delete.
2739 (movmemsi_1reg): Delete.
2740 (corresponding unnamed define_insn): Delete.
2741 * config/rs6000/rs6000.opt (mno-string): New.
2742 (mstring): Replace by deprecation warning stub.
2743 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2745 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2747 * regrename.c (regrename_do_replace): If replacing the same
2748 reg multiple times, try to reuse last created gen_raw_REG.
2751 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2752 main to workaround a bug in GDB.
2754 2018-01-12 Tom de Vries <tom@codesourcery.com>
2757 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2759 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2761 PR rtl-optimization/80481
2762 * ira-color.c (get_cap_member): New function.
2763 (allocnos_conflict_by_live_ranges_p): Use it.
2764 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2765 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2767 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2770 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2771 (*saddl_se_1): Ditto.
2773 (*ssubl_se_1): Ditto.
2775 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2777 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2778 rather than wi::to_widest for DR_INITs.
2779 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2780 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2781 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2783 (vect_analyze_group_access_1): Note that here.
2785 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2787 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2788 polynomial type sizes.
2790 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2792 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2793 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2794 (gimple_add_tmp_var): Likewise.
2796 2018-01-12 Martin Liska <mliska@suse.cz>
2798 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2799 (gimple_alloc_sizes): Likewise.
2800 (dump_gimple_statistics): Use PRIu64 in printf format.
2801 * gimple.h: Change uint64_t to int.
2803 2018-01-12 Martin Liska <mliska@suse.cz>
2805 * tree-core.h: Use uint64_t instead of int.
2806 * tree.c (tree_node_counts): Likewise.
2807 (tree_node_sizes): Likewise.
2808 (dump_tree_statistics): Use PRIu64 in printf format.
2810 2018-01-12 Martin Liska <mliska@suse.cz>
2812 * Makefile.in: As qsort_chk is implemented in vec.c, add
2813 vec.o to linkage of gencfn-macros.
2814 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2815 passing the info to record_node_allocation_statistics.
2816 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2818 * ggc-common.c (struct ggc_usage): Add operator== and use
2819 it in operator< and compare function.
2820 * mem-stats.h (struct mem_usage): Likewise.
2821 * vec.c (struct vec_usage): Remove operator< and compare
2822 function. Can be simply inherited.
2824 2018-01-12 Martin Jambor <mjambor@suse.cz>
2827 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2828 * tree-ssa-math-opts.c: Include domwalk.h.
2829 (convert_mult_to_fma_1): New function.
2830 (fma_transformation_info): New type.
2831 (fma_deferring_state): Likewise.
2832 (cancel_fma_deferring): New function.
2833 (result_of_phi): Likewise.
2834 (last_fma_candidate_feeds_initial_phi): Likewise.
2835 (convert_mult_to_fma): Added deferring logic, split actual
2836 transformation to convert_mult_to_fma_1.
2837 (math_opts_dom_walker): New type.
2838 (math_opts_dom_walker::after_dom_children): New method, body moved
2839 here from pass_optimize_widening_mul::execute, added deferring logic
2841 (pass_optimize_widening_mul::execute): Moved most of code to
2842 math_opts_dom_walker::after_dom_children.
2843 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2844 * config/i386/i386.c (ix86_option_override_internal): Added
2845 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2847 2018-01-12 Richard Biener <rguenther@suse.de>
2850 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2851 inline instance vars.
2853 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2856 * config/rx/rx.c (rx_is_restricted_memory_address):
2859 2018-01-12 Richard Biener <rguenther@suse.de>
2861 PR tree-optimization/80846
2862 * target.def (split_reduction): New target hook.
2863 * targhooks.c (default_split_reduction): New function.
2864 * targhooks.h (default_split_reduction): Declare.
2865 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2866 target requests first reduce vectors by combining low and high
2868 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2869 (get_vectype_for_scalar_type_and_size): Export.
2870 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2871 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2872 * doc/tm.texi: Regenerate.
2873 * config/i386/i386.c (ix86_split_reduction): Implement
2874 TARGET_VECTORIZE_SPLIT_REDUCTION.
2876 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2879 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2880 in PIC mode except for TARGET_VXWORKS_RTP.
2881 * config/sparc/sparc.c: Include cfgrtl.h.
2882 (TARGET_INIT_PIC_REG): Define.
2883 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2884 (sparc_pic_register_p): New predicate.
2885 (sparc_legitimate_address_p): Use it.
2886 (sparc_legitimize_pic_address): Likewise.
2887 (sparc_delegitimize_address): Likewise.
2888 (sparc_mode_dependent_address_p): Likewise.
2889 (gen_load_pcrel_sym): Remove 4th parameter.
2890 (load_got_register): Adjust call to above. Remove obsolete stuff.
2891 (sparc_expand_prologue): Do not call load_got_register here.
2892 (sparc_flat_expand_prologue): Likewise.
2893 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2894 (sparc_use_pseudo_pic_reg): New function.
2895 (sparc_init_pic_reg): Likewise.
2896 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2897 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2899 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2901 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2902 Add item for branch_cost.
2904 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2906 PR rtl-optimization/83565
2907 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2908 not extend the result to a larger mode for rotate operations.
2909 (num_sign_bit_copies1): Likewise.
2911 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2914 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2916 Use values-Xc.o for -pedantic.
2917 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2919 2018-01-12 Martin Liska <mliska@suse.cz>
2922 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2924 (possible_polymorphic_call_targets): Use it.
2925 (ipa_devirt): Likewise.
2927 2018-01-12 Martin Liska <mliska@suse.cz>
2929 * profile-count.h (enum profile_quality): Use 0 as invalid
2930 enum value of profile_quality.
2932 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2934 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2935 -mext-string options.
2937 2018-01-12 Richard Biener <rguenther@suse.de>
2939 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2940 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2941 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2943 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2945 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2947 * configure.ac (--with-long-double-format): Add support for the
2948 configuration option to change the default long double format on
2950 * config.gcc (powerpc*-linux*-*): Likewise.
2951 * configure: Regenerate.
2952 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2953 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2954 used without modification.
2956 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2958 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2959 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2960 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2961 MISC_BUILTIN_SPEC_BARRIER.
2962 (rs6000_init_builtins): Likewise.
2963 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2965 (speculation_barrier): New define_insn.
2966 * doc/extend.texi: Document __builtin_speculation_barrier.
2968 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2971 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2972 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2973 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2975 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2976 integral modes instead of "ss" and "sd".
2977 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2978 vectors with 32-bit and 64-bit elements.
2979 (vecdupssescalarmodesuffix): New mode attribute.
2980 (vec_dup<mode>): Use it.
2982 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2985 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2986 frame if argument is passed on stack.
2988 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2991 * ree.c (combine_reaching_defs): Optimize also
2992 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
2993 reg2=any_extend(exp); reg1=reg2;, formatting fix.
2995 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2998 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3000 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3003 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3004 after they are computed.
3006 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3008 PR tree-optimization/83695
3009 * gimple-loop-linterchange.cc
3010 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3011 reset cached scev information after interchange.
3012 (pass_linterchange::execute): Remove call to scev_reset_htab.
3014 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3016 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3017 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3018 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3019 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3020 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3021 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3022 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3023 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3024 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3025 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3026 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3027 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3028 (V_lane_reg): Likewise.
3029 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3031 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3032 (vfmal_lane_low<mode>_intrinsic,
3033 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3034 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3035 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3036 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3037 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3038 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3040 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3042 * config/arm/arm-cpus.in (fp16fml): New feature.
3043 (ALL_SIMD): Add fp16fml.
3044 (armv8.2-a): Add fp16fml as an option.
3045 (armv8.3-a): Likewise.
3046 (armv8.4-a): Add fp16fml as part of fp16.
3047 * config/arm/arm.h (TARGET_FP16FML): Define.
3048 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3050 * config/arm/arm-modes.def (V2HF): Define.
3051 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3052 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3053 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3054 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3055 vfmsl_low, vfmsl_high): New set of builtins.
3056 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3057 (vfml_op): New code attribute.
3058 (VFMLHALVES): New int iterator.
3059 (VFML, VFMLSEL): New mode attributes.
3060 (V_reg): Define mapping for V2HF.
3061 (V_hi, V_lo): New mode attributes.
3062 (VF_constraint): Likewise.
3063 (vfml_half, vfml_half_selector): New int attributes.
3064 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3066 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3067 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3069 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3070 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3071 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3072 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3074 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3075 Document new effective target and option set.
3077 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3079 * config/arm/arm-cpus.in (armv8_4): New feature.
3080 (ARMv8_4a): New fgroup.
3081 (armv8.4-a): New arch.
3082 * config/arm/arm-tables.opt: Regenerate.
3083 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3084 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3085 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3086 Add matching rules for -march=armv8.4-a and extensions.
3087 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3089 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3092 * config/rx/rx.md (BW): New mode attribute.
3093 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3095 2018-01-11 Richard Biener <rguenther@suse.de>
3097 PR tree-optimization/83435
3098 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3099 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3100 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3102 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3103 Alan Hayward <alan.hayward@arm.com>
3104 David Sherwood <david.sherwood@arm.com>
3106 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3108 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3109 (aarch64_print_address_internal): Use it to check for a zero offset.
3111 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3112 Alan Hayward <alan.hayward@arm.com>
3113 David Sherwood <david.sherwood@arm.com>
3115 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3116 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3117 Return a poly_int64 rather than a HOST_WIDE_INT.
3118 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3119 rather than a HOST_WIDE_INT.
3120 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3121 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3122 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3123 final_offset from HOST_WIDE_INT to poly_int64.
3124 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3125 to_constant when getting the number of units in an Advanced SIMD
3127 (aarch64_builtin_vectorized_function): Check for a constant number
3129 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3131 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3132 attribute instead of GET_MODE_NUNITS.
3133 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3134 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3135 GET_MODE_SIZE for fixed-size registers.
3136 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3137 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3138 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3139 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3140 (aarch64_print_operand, aarch64_print_address_internal)
3141 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3142 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3143 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3144 Handle polynomial GET_MODE_SIZE.
3145 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3146 wider than SImode without modification.
3147 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3148 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3149 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3150 passing and returning SVE modes.
3151 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3152 rather than GEN_INT.
3153 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3154 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3155 (aarch64_allocate_and_probe_stack_space): Likewise.
3156 (aarch64_layout_frame): Cope with polynomial offsets.
3157 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3158 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3160 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3161 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3162 poly_int64 rather than a HOST_WIDE_INT.
3163 (aarch64_get_separate_components, aarch64_process_components)
3164 (aarch64_expand_prologue, aarch64_expand_epilogue)
3165 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3166 (aarch64_anchor_offset): New function, split out from...
3167 (aarch64_legitimize_address): ...here.
3168 (aarch64_builtin_vectorization_cost): Handle polynomial
3169 TYPE_VECTOR_SUBPARTS.
3170 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3172 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3173 number of elements from the PARALLEL rather than the mode.
3174 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3175 rather than GET_MODE_BITSIZE.
3176 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3177 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3178 (aarch64_expand_vec_perm_const_1): Handle polynomial
3179 d->perm.length () and d->perm elements.
3180 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3181 Apply to_constant to d->perm elements.
3182 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3183 polynomial CONST_VECTOR_NUNITS.
3184 (aarch64_move_pointer): Take amount as a poly_int64 rather
3186 (aarch64_progress_pointer): Avoid temporary variable.
3187 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3188 the mode attribute instead of GET_MODE.
3190 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3191 Alan Hayward <alan.hayward@arm.com>
3192 David Sherwood <david.sherwood@arm.com>
3194 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3195 x exists before using it.
3196 (aarch64_add_constant_internal): Rename to...
3197 (aarch64_add_offset_1): ...this. Replace regnum with separate
3198 src and dest rtxes. Handle the case in which they're different,
3199 including when the offset is zero. Replace scratchreg with an rtx.
3200 Use 2 additions if there is no spare register into which we can
3201 move a 16-bit constant.
3202 (aarch64_add_constant): Delete.
3203 (aarch64_add_offset): Replace reg with separate src and dest
3204 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3205 Use aarch64_add_offset_1.
3206 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3207 an rtx rather than an int. Take the delta as a poly_int64
3208 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3209 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3210 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3211 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3212 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3214 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3215 aarch64_add_constant.
3217 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3219 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3220 Use scalar_float_mode.
3222 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3224 * config/aarch64/aarch64-simd.md
3225 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3226 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3227 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3228 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3229 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3230 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3231 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3232 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3233 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3234 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3236 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3239 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3240 targ_options->x_arm_arch_string is non NULL.
3242 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3244 * config/aarch64/aarch64.h
3245 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3247 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3250 * expmed.c (emit_store_flag_force): Swap if const op0
3251 and change VOIDmode to mode of op0.
3253 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3255 PR rtl-optimization/83761
3256 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3257 than bytes to mode_for_size.
3259 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3262 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3263 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3266 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3269 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3270 when in layout mode.
3271 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3272 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3275 2018-01-10 Michael Collison <michael.collison@arm.com>
3277 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3278 * config/aarch64/aarch64-option-extension.def: Add
3279 AARCH64_OPT_EXTENSION of 'fp16fml'.
3280 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3281 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3282 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3283 * config/aarch64/constraints.md (Ui7): New constraint.
3284 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3285 (VFMLA_SEL_W): Ditto.
3288 (VFMLA16_LOW): New int iterator.
3289 (VFMLA16_HIGH): Ditto.
3290 (UNSPEC_FMLAL): New unspec.
3291 (UNSPEC_FMLSL): Ditto.
3292 (UNSPEC_FMLAL2): Ditto.
3293 (UNSPEC_FMLSL2): Ditto.
3294 (f16mac): New code attribute.
3295 * config/aarch64/aarch64-simd-builtins.def
3296 (aarch64_fmlal_lowv2sf): Ditto.
3297 (aarch64_fmlsl_lowv2sf): Ditto.
3298 (aarch64_fmlalq_lowv4sf): Ditto.
3299 (aarch64_fmlslq_lowv4sf): Ditto.
3300 (aarch64_fmlal_highv2sf): Ditto.
3301 (aarch64_fmlsl_highv2sf): Ditto.
3302 (aarch64_fmlalq_highv4sf): Ditto.
3303 (aarch64_fmlslq_highv4sf): Ditto.
3304 (aarch64_fmlal_lane_lowv2sf): Ditto.
3305 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3306 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3307 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3308 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3309 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3310 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3311 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3312 (aarch64_fmlal_lane_highv2sf): Ditto.
3313 (aarch64_fmlsl_lane_highv2sf): Ditto.
3314 (aarch64_fmlal_laneq_highv2sf): Ditto.
3315 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3316 (aarch64_fmlalq_lane_highv4sf): Ditto.
3317 (aarch64_fmlsl_lane_highv4sf): Ditto.
3318 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3319 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3320 * config/aarch64/aarch64-simd.md:
3321 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3322 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3323 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3324 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3325 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3326 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3327 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3328 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3329 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3330 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3331 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3332 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3333 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3334 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3335 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3336 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3337 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3338 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3339 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3340 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3341 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3342 (vfmlsl_low_u32): Ditto.
3343 (vfmlalq_low_u32): Ditto.
3344 (vfmlslq_low_u32): Ditto.
3345 (vfmlal_high_u32): Ditto.
3346 (vfmlsl_high_u32): Ditto.
3347 (vfmlalq_high_u32): Ditto.
3348 (vfmlslq_high_u32): Ditto.
3349 (vfmlal_lane_low_u32): Ditto.
3350 (vfmlsl_lane_low_u32): Ditto.
3351 (vfmlal_laneq_low_u32): Ditto.
3352 (vfmlsl_laneq_low_u32): Ditto.
3353 (vfmlalq_lane_low_u32): Ditto.
3354 (vfmlslq_lane_low_u32): Ditto.
3355 (vfmlalq_laneq_low_u32): Ditto.
3356 (vfmlslq_laneq_low_u32): Ditto.
3357 (vfmlal_lane_high_u32): Ditto.
3358 (vfmlsl_lane_high_u32): Ditto.
3359 (vfmlal_laneq_high_u32): Ditto.
3360 (vfmlsl_laneq_high_u32): Ditto.
3361 (vfmlalq_lane_high_u32): Ditto.
3362 (vfmlslq_lane_high_u32): Ditto.
3363 (vfmlalq_laneq_high_u32): Ditto.
3364 (vfmlslq_laneq_high_u32): Ditto.
3365 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3366 (AARCH64_FL_FOR_ARCH8_4): New.
3367 (AARCH64_ISA_F16FML): New ISA flag.
3368 (TARGET_F16FML): New feature flag for fp16fml.
3369 (doc/invoke.texi): Document new fp16fml option.
3371 2018-01-10 Michael Collison <michael.collison@arm.com>
3373 * config/aarch64/aarch64-builtins.c:
3374 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3375 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3376 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3377 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3378 (AARCH64_ISA_SHA3): New ISA flag.
3379 (TARGET_SHA3): New feature flag for sha3.
3380 * config/aarch64/iterators.md (sha512_op): New int attribute.
3381 (CRYPTO_SHA512): New int iterator.
3382 (UNSPEC_SHA512H): New unspec.
3383 (UNSPEC_SHA512H2): Ditto.
3384 (UNSPEC_SHA512SU0): Ditto.
3385 (UNSPEC_SHA512SU1): Ditto.
3386 * config/aarch64/aarch64-simd-builtins.def
3387 (aarch64_crypto_sha512hqv2di): New builtin.
3388 (aarch64_crypto_sha512h2qv2di): Ditto.
3389 (aarch64_crypto_sha512su0qv2di): Ditto.
3390 (aarch64_crypto_sha512su1qv2di): Ditto.
3391 (aarch64_eor3qv8hi): Ditto.
3392 (aarch64_rax1qv2di): Ditto.
3393 (aarch64_xarqv2di): Ditto.
3394 (aarch64_bcaxqv8hi): Ditto.
3395 * config/aarch64/aarch64-simd.md:
3396 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3397 (aarch64_crypto_sha512su0qv2di): Ditto.
3398 (aarch64_crypto_sha512su1qv2di): Ditto.
3399 (aarch64_eor3qv8hi): Ditto.
3400 (aarch64_rax1qv2di): Ditto.
3401 (aarch64_xarqv2di): Ditto.
3402 (aarch64_bcaxqv8hi): Ditto.
3403 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3404 (vsha512h2q_u64): Ditto.
3405 (vsha512su0q_u64): Ditto.
3406 (vsha512su1q_u64): Ditto.
3407 (veor3q_u16): Ditto.
3408 (vrax1q_u64): Ditto.
3410 (vbcaxq_u16): Ditto.
3411 * config/arm/types.md (crypto_sha512): New type attribute.
3412 (crypto_sha3): Ditto.
3413 (doc/invoke.texi): Document new sha3 option.
3415 2018-01-10 Michael Collison <michael.collison@arm.com>
3417 * config/aarch64/aarch64-builtins.c:
3418 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3419 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3420 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3421 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3422 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3423 (AARCH64_ISA_SM4): New ISA flag.
3424 (TARGET_SM4): New feature flag for sm4.
3425 * config/aarch64/aarch64-simd-builtins.def
3426 (aarch64_sm3ss1qv4si): Ditto.
3427 (aarch64_sm3tt1aq4si): Ditto.
3428 (aarch64_sm3tt1bq4si): Ditto.
3429 (aarch64_sm3tt2aq4si): Ditto.
3430 (aarch64_sm3tt2bq4si): Ditto.
3431 (aarch64_sm3partw1qv4si): Ditto.
3432 (aarch64_sm3partw2qv4si): Ditto.
3433 (aarch64_sm4eqv4si): Ditto.
3434 (aarch64_sm4ekeyqv4si): Ditto.
3435 * config/aarch64/aarch64-simd.md:
3436 (aarch64_sm3ss1qv4si): Ditto.
3437 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3438 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3439 (aarch64_sm4eqv4si): Ditto.
3440 (aarch64_sm4ekeyqv4si): Ditto.
3441 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3442 (sm3part_op): Ditto.
3443 (CRYPTO_SM3TT): Ditto.
3444 (CRYPTO_SM3PART): Ditto.
3445 (UNSPEC_SM3SS1): New unspec.
3446 (UNSPEC_SM3TT1A): Ditto.
3447 (UNSPEC_SM3TT1B): Ditto.
3448 (UNSPEC_SM3TT2A): Ditto.
3449 (UNSPEC_SM3TT2B): Ditto.
3450 (UNSPEC_SM3PARTW1): Ditto.
3451 (UNSPEC_SM3PARTW2): Ditto.
3452 (UNSPEC_SM4E): Ditto.
3453 (UNSPEC_SM4EKEY): Ditto.
3454 * config/aarch64/constraints.md (Ui2): New constraint.
3455 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3456 * config/arm/types.md (crypto_sm3): New type attribute.
3457 (crypto_sm4): Ditto.
3458 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3459 (vsm3tt1aq_u32): Ditto.
3460 (vsm3tt1bq_u32): Ditto.
3461 (vsm3tt2aq_u32): Ditto.
3462 (vsm3tt2bq_u32): Ditto.
3463 (vsm3partw1q_u32): Ditto.
3464 (vsm3partw2q_u32): Ditto.
3465 (vsm4eq_u32): Ditto.
3466 (vsm4ekeyq_u32): Ditto.
3467 (doc/invoke.texi): Document new sm4 option.
3469 2018-01-10 Michael Collison <michael.collison@arm.com>
3471 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3472 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3473 (AARCH64_FL_FOR_ARCH8_4): New.
3474 (AARCH64_FL_V8_4): New flag.
3475 (doc/invoke.texi): Document new armv8.4-a option.
3477 2018-01-10 Michael Collison <michael.collison@arm.com>
3479 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3480 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3481 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3482 * config/aarch64/aarch64-option-extension.def: Add
3483 AARCH64_OPT_EXTENSION of 'sha2'.
3484 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3485 (crypto): Disable sha2 and aes if crypto disabled.
3486 (crypto): Enable aes and sha2 if enabled.
3487 (simd): Disable sha2 and aes if simd disabled.
3488 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3490 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3491 (TARGET_SHA2): New feature flag for sha2.
3492 (TARGET_AES): New feature flag for aes.
3493 * config/aarch64/aarch64-simd.md:
3494 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3495 conditional on TARGET_AES.
3496 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3497 (aarch64_crypto_sha1hsi): Make pattern conditional
3499 (aarch64_crypto_sha1hv4si): Ditto.
3500 (aarch64_be_crypto_sha1hv4si): Ditto.
3501 (aarch64_crypto_sha1su1v4si): Ditto.
3502 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3503 (aarch64_crypto_sha1su0v4si): Ditto.
3504 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3505 (aarch64_crypto_sha256su0v4si): Ditto.
3506 (aarch64_crypto_sha256su1v4si): Ditto.
3507 (doc/invoke.texi): Document new aes and sha2 options.
3509 2018-01-10 Martin Sebor <msebor@redhat.com>
3511 PR tree-optimization/83781
3512 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3515 2018-01-11 Martin Sebor <msebor@gmail.com>
3516 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3518 PR tree-optimization/83501
3519 PR tree-optimization/81703
3521 * tree-ssa-strlen.c (get_string_cst): Rename...
3522 (get_string_len): ...to this. Handle global constants.
3523 (handle_char_store): Adjust.
3525 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3526 Jim Wilson <jimw@sifive.com>
3528 * config/riscv/riscv-protos.h (riscv_output_return): New.
3529 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3530 (riscv_attribute_table, riscv_output_return),
3531 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3532 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3533 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3535 (riscv_expand_prologue): Add early return for naked function.
3536 (riscv_expand_epilogue): Likewise.
3537 (riscv_function_ok_for_sibcall): Return false for naked function.
3538 (riscv_set_current_function): New.
3539 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3540 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3541 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3542 * doc/extend.texi (RISC-V Function Attributes): New.
3544 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3546 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3547 check for 128-bit long double before checking TCmode.
3548 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3549 128-bit long doubles before checking TFmode or TCmode.
3550 (FLOAT128_IBM_P): Likewise.
3552 2018-01-10 Martin Sebor <msebor@redhat.com>
3554 PR tree-optimization/83671
3555 * builtins.c (c_strlen): Unconditionally return zero for the empty
3557 Use -Warray-bounds for warnings.
3558 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3559 for non-constant array indices with COMPONENT_REF, arrays of
3560 arrays, and pointers to arrays.
3561 (gimple_fold_builtin_strlen): Determine and set length range for
3562 non-constant character arrays.
3564 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3567 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3570 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3572 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3574 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3577 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3578 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3579 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3580 indexed_or_indirect_operand predicate.
3581 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3582 (*vsx_le_perm_load_v8hi): Likewise.
3583 (*vsx_le_perm_load_v16qi): Likewise.
3584 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3585 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3586 (*vsx_le_perm_store_v8hi): Likewise.
3587 (*vsx_le_perm_store_v16qi): Likewise.
3588 (eight unnamed splitters): Likewise.
3590 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3592 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3593 * config/rs6000/emmintrin.h: Likewise.
3594 * config/rs6000/mmintrin.h: Likewise.
3595 * config/rs6000/xmmintrin.h: Likewise.
3597 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3600 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3602 * tree.c (tree_nop_conversion): Return true for location wrapper
3604 (maybe_wrap_with_location): New function.
3605 (selftest::check_strip_nops): New function.
3606 (selftest::test_location_wrappers): New function.
3607 (selftest::tree_c_tests): Call it.
3608 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3609 (maybe_wrap_with_location): New decl.
3610 (EXPR_LOCATION_WRAPPER_P): New macro.
3611 (location_wrapper_p): New inline function.
3612 (tree_strip_any_location_wrapper): New inline function.
3614 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3617 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3618 stack_realign_offset for the largest alignment of stack slot
3620 (ix86_find_max_used_stack_alignment): New function.
3621 (ix86_finalize_stack_frame_flags): Use it. Set
3622 max_used_stack_alignment if we don't realign stack.
3623 * config/i386/i386.h (machine_function): Add
3624 max_used_stack_alignment.
3626 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3628 * config/arm/arm.opt (-mbranch-cost): New option.
3629 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3632 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3635 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3636 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3638 2018-01-10 Richard Biener <rguenther@suse.de>
3641 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3642 early out so it also covers the case where we have a non-NULL
3645 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3647 PR tree-optimization/83753
3648 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3649 for non-strided grouped accesses if the number of elements is 1.
3651 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3654 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3655 * i386.h (TARGET_USE_GATHER): Define.
3656 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3658 2018-01-10 Martin Liska <mliska@suse.cz>
3661 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3662 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3664 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3665 CLEANUP_NO_PARTITIONING is not set.
3667 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3669 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3670 for vectors, as a partial revert of r254296.
3671 * rtl.h (const_vec_p): Delete.
3672 (const_vec_duplicate_p): Don't test for vector CONSTs.
3673 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3674 * expmed.c (make_tree): Likewise.
3677 * common.md (E, F): Use CONSTANT_P instead of checking for
3679 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3680 checking for CONST_VECTOR.
3682 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3685 * predict.c (force_edge_cold): Handle in more sane way edges
3688 2018-01-09 Carl Love <cel@us.ibm.com>
3690 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3692 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3693 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3694 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3695 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3696 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3697 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3698 * config/rs6000/rs6000-protos.h: Add extern defition for
3699 rs6000_generate_float2_double_code.
3700 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3702 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3703 (float2_v2df): Add define_expand.
3705 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3708 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3709 op_mode in the force_to_mode call.
3711 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3713 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3714 instead of checking each element individually.
3715 (aarch64_evpc_uzp): Likewise.
3716 (aarch64_evpc_zip): Likewise.
3717 (aarch64_evpc_ext): Likewise.
3718 (aarch64_evpc_rev): Likewise.
3719 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3720 instead of checking each element individually. Return true without
3722 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3723 whether all selected elements come from the same input, instead of
3724 checking each element individually. Remove calls to gen_rtx_REG,
3725 start_sequence and end_sequence and instead assert that no rtl is
3728 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3730 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3731 order of HIGH and CONST checks.
3733 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3735 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3736 if the destination isn't an SSA_NAME.
3738 2018-01-09 Richard Biener <rguenther@suse.de>
3740 PR tree-optimization/83668
3741 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3743 (canonicalize_loop_form): ... here, renamed from ...
3744 (canonicalize_loop_closed_ssa_form): ... this and amended to
3745 swap successor edges for loop exit blocks to make us use
3746 the RPO order we need for initial schedule generation.
3748 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3750 PR tree-optimization/64811
3751 * match.pd: When optimizing comparisons with Inf, avoid
3752 introducing or losing exceptions from comparisons with NaN.
3754 2018-01-09 Martin Liska <mliska@suse.cz>
3757 * asan.c (shadow_mem_size): Add gcc_assert.
3759 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3761 Don't save registers in main().
3764 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3765 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3766 * config/avr/avr.c (avr_set_current_function): Don't error if
3767 naked, OS_task or OS_main are specified at the same time.
3768 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3770 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3772 * common/config/avr/avr-common.c (avr_option_optimization_table):
3773 Switch on -mmain-is-OS_task for optimizing compilations.
3775 2018-01-09 Richard Biener <rguenther@suse.de>
3777 PR tree-optimization/83572
3778 * graphite.c: Include cfganal.h.
3779 (graphite_transform_loops): Connect infinite loops to exit
3780 and remove fake edges at the end.
3782 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3784 * ipa-inline.c (edge_badness): Revert accidental checkin.
3786 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3789 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3790 symbols; not inline clones.
3792 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3795 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3796 hard registers. Formatting fixes.
3798 PR preprocessor/83722
3799 * gcc.c (try_generate_repro): Pass
3800 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3801 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3804 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3805 Kito Cheng <kito.cheng@gmail.com>
3807 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3808 (riscv_leaf_function_p): Delete.
3809 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3811 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3813 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3815 (do_ifelse): New function.
3816 (do_isel): New function.
3817 (do_sub3): New function.
3818 (do_add3): New function.
3819 (do_load_mask_compare): New function.
3820 (do_overlap_load_compare): New function.
3821 (expand_compare_loop): New function.
3822 (expand_block_compare): Call expand_compare_loop() when appropriate.
3823 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3825 (-mblock-compare-inline-loop-limit): New option.
3827 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3830 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3831 Reverse order of second and third operands in first alternative.
3832 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3833 of first and second elements in UNSPEC_VPERMR vector.
3834 (altivec_expand_vec_perm_le): Likewise.
3836 2017-01-08 Jeff Law <law@redhat.com>
3838 PR rtl-optimizatin/81308
3839 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3840 (process_switch): If group_case_labels makes a change, then set
3842 (pass_convert_switch::execute): If a switch is converted, then
3843 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3845 PR rtl-optimization/81308
3846 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3849 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3851 PR target/83663 - Revert r255946
3852 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3853 generation for cases where splatting a value is not useful.
3854 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3855 across a vec_duplicate and a paradoxical subreg forming a vector
3856 mode to a vec_concat.
3858 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3860 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3861 -march=armv8.3-a variants.
3862 * config/arm/t-multilib: Likewise.
3863 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3865 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3867 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3869 (cceq_ior_compare_complement): Give it a name so I can use it, and
3870 change boolean_or_operator predicate to boolean_operator so it can
3871 be used to generate a crand.
3872 (eqne): New code iterator.
3873 (bd/bd_neg): New code_attrs.
3874 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3875 a single define_insn.
3876 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3877 decrement (bdnzt/bdnzf/bdzt/bdzf).
3878 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3879 with the new names of the branch decrement patterns, and added the
3880 names of the branch decrement conditional patterns.
3882 2018-01-08 Richard Biener <rguenther@suse.de>
3884 PR tree-optimization/83563
3885 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3888 2018-01-08 Richard Biener <rguenther@suse.de>
3891 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3893 2018-01-08 Richard Biener <rguenther@suse.de>
3895 PR tree-optimization/83685
3896 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3897 references to abnormals.
3899 2018-01-08 Richard Biener <rguenther@suse.de>
3902 * dwarf2out.c (output_indirect_strings): Handle empty
3903 skeleton_debug_str_hash.
3904 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3906 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3908 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3909 (emit_store_direct): Likewise.
3910 (arc_trampoline_adjust_address): Likewise.
3911 (arc_asm_trampoline_template): New function.
3912 (arc_initialize_trampoline): Use asm_trampoline_template.
3913 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3914 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3915 * config/arc/arc.md (flush_icache): Delete pattern.
3917 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3919 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3920 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3923 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3926 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3927 by not USED_FOR_TARGET.
3928 (make_pass_resolve_sw_modes): Likewise.
3930 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3932 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3935 2018-01-08 Richard Biener <rguenther@suse.de>
3938 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3940 2018-01-08 Richard Biener <rguenther@suse.de>
3943 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3945 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3948 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3949 basic blocks with a small number of successors.
3950 (convert_control_dep_chain_into_preds): Improve handling of
3952 (dump_predicates): Split apart into...
3953 (dump_pred_chain): ...here...
3954 (dump_pred_info): ...and here.
3955 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3956 (can_chain_union_be_invalidated_p): Improve check for invalidation
3958 (uninit_uses_cannot_happen): Avoid unnecessary if
3959 convert_control_dep_chain_into_preds yielded nothing.
3961 2018-01-06 Martin Sebor <msebor@redhat.com>
3963 PR tree-optimization/83640
3964 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3965 subtracting negative offset from size.
3966 (builtin_access::overlap): Adjust offset bounds of the access to fall
3967 within the size of the object if possible.
3969 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3971 PR rtl-optimization/83699
3972 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3973 extract_bit_field_as_subreg to cases in which the extracted
3974 value is also a vector.
3976 * lra-constraints.c (process_alt_operands): Test for the equivalence
3977 substitutions when detecting a possible reload cycle.
3979 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3982 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3983 by default if flag_selective_schedling{,2}. Formatting fixes.
3985 PR rtl-optimization/83682
3986 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
3987 if it has non-VECTOR_MODE element mode.
3988 (vec_duplicate_p): Likewise.
3991 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
3992 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
3994 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3997 * config/i386/i386-builtin.def
3998 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
3999 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4000 Require also OPTION_MASK_ISA_AVX512F in addition to
4001 OPTION_MASK_ISA_GFNI.
4002 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4003 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4004 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4005 to OPTION_MASK_ISA_GFNI.
4006 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4007 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4008 OPTION_MASK_ISA_AVX512BW.
4009 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4010 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4011 addition to OPTION_MASK_ISA_GFNI.
4012 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4013 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4014 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4015 to OPTION_MASK_ISA_GFNI.
4016 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4017 a requirement for all ISAs rather than any of them with a few
4019 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4021 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4022 bitmasks to be enabled with 3 exceptions, instead of requiring any
4023 enabled ISA with lots of exceptions.
4024 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4025 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4026 Change avx512bw in isa attribute to avx512f.
4027 * config/i386/sgxintrin.h: Add license boilerplate.
4028 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4029 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4030 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4031 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4033 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4034 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4035 temporarily sse2 rather than sse if not enabled already.
4038 * config/i386/sse.md (VI248_VLBW): Rename to ...
4039 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4040 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4041 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4042 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4043 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4044 mode iterator instead of VI248_VLBW.
4046 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4048 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4049 (record_modified): Skip clobbers; add debug output.
4050 (param_change_prob): Use sreal frequencies.
4052 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4054 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4055 punt for user-aligned variables.
4057 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4059 * tree-chrec.c (chrec_contains_symbols): Return true for
4062 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4065 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4066 of (x|y) == x for BICS pattern.
4068 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4070 PR tree-optimization/83605
4071 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4072 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4075 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4077 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4078 * config/epiphany/rtems.h: New file.
4080 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4081 Uros Bizjak <ubizjak@gmail.com>
4084 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4085 QIreg_operand instead of register_operand predicate.
4086 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4087 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4088 comments instead of -fmitigate[-_]rop.
4090 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4093 * cgraphunit.c (symbol_table::compile): Switch to text_section
4094 before calling assembly_start debug hook.
4095 * run-rtl-passes.c (run_rtl_passes): Likewise.
4098 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4100 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4101 range_int_cst_p rather than !symbolic_range_p before calling
4102 extract_range_from_multiplicative_op_1.
4104 2017-01-04 Jeff Law <law@redhat.com>
4106 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4107 redundant test in assertion.
4109 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4111 * doc/rtl.texi: Document machine_mode wrapper classes.
4113 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4115 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4118 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4120 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4121 the VEC_PERM_EXPR fold to fail.
4123 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4126 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4127 to switched_sections.
4129 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4132 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4135 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4138 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4139 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4141 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4144 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4145 is BLKmode and bitpos not zero or mode change is needed.
4147 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4150 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4153 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4156 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4157 instead of MULT rtx. Update all corresponding splitters.
4159 (*ssub<modesuffix>): Ditto.
4161 (*cmp_sadd_di): Update split patterns.
4162 (*cmp_sadd_si): Ditto.
4163 (*cmp_sadd_sidi): Ditto.
4164 (*cmp_ssub_di): Ditto.
4165 (*cmp_ssub_si): Ditto.
4166 (*cmp_ssub_sidi): Ditto.
4167 * config/alpha/predicates.md (const23_operand): New predicate.
4168 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4169 Look for ASHIFT, not MULT inner operand.
4170 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4172 2018-01-04 Martin Liska <mliska@suse.cz>
4174 PR gcov-profile/83669
4175 * gcov.c (output_intermediate_file): Add version to intermediate
4177 * doc/gcov.texi: Document new field 'version' in intermediate
4178 file format. Fix location of '-k' option of gcov command.
4180 2018-01-04 Martin Liska <mliska@suse.cz>
4183 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4185 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4187 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4189 2018-01-03 Martin Sebor <msebor@redhat.com>
4191 PR tree-optimization/83655
4192 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4193 checking calls with invalid arguments.
4195 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4197 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4198 (vectorizable_mask_load_store): Delete.
4199 (vectorizable_call): Return false for masked loads and stores.
4200 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4201 instead of gimple_assign_rhs1.
4202 (vectorizable_load): Handle IFN_MASK_LOAD.
4203 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4205 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4207 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4209 (vectorizable_mask_load_store): ...here.
4210 (vectorizable_load): ...and here.
4212 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4214 * tree-vect-stmts.c (vect_build_all_ones_mask)
4215 (vect_build_zero_merge_argument): New functions, split out from...
4216 (vectorizable_load): ...here.
4218 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4220 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4222 (vectorizable_mask_load_store): ...here.
4223 (vectorizable_store): ...and here.
4225 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4227 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4229 (vectorizable_mask_load_store): ...here.
4231 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4233 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4234 (vect_model_store_cost): Take a vec_load_store_type instead of a
4236 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4237 (vect_model_store_cost): Take a vec_load_store_type instead of a
4239 (vectorizable_mask_load_store): Update accordingly.
4240 (vectorizable_store): Likewise.
4241 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4243 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4245 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4246 IFN_MASK_LOAD calls here rather than...
4247 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4249 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4250 Alan Hayward <alan.hayward@arm.com>
4251 David Sherwood <david.sherwood@arm.com>
4253 * expmed.c (extract_bit_field_1): For vector extracts,
4254 fall back to extract_bit_field_as_subreg if vec_extract
4257 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4258 Alan Hayward <alan.hayward@arm.com>
4259 David Sherwood <david.sherwood@arm.com>
4261 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4262 they are variable or constant sized.
4263 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4264 slots for constant-sized data.
4266 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4267 Alan Hayward <alan.hayward@arm.com>
4268 David Sherwood <david.sherwood@arm.com>
4270 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4271 handling COND_EXPRs with boolean comparisons, try to find a better
4272 basis for the mask type than the boolean itself.
4274 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4276 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4277 is calculated and how it can be overridden.
4278 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4279 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4281 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4284 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4285 Alan Hayward <alan.hayward@arm.com>
4286 David Sherwood <david.sherwood@arm.com>
4288 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4289 Remove the mode argument.
4290 (aarch64_simd_valid_immediate): Remove the mode and inverse
4292 * config/aarch64/iterators.md (bitsize): New iterator.
4293 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4294 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4295 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4296 aarch64_simd_valid_immediate.
4297 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4298 (aarch64_reg_or_bic_imm): Likewise.
4299 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4300 with an insn_type enum and msl with a modifier_type enum.
4301 Replace element_width with a scalar_mode. Change the shift
4302 to unsigned int. Add constructors for scalar_float_mode and
4303 scalar_int_mode elements.
4304 (aarch64_vect_float_const_representable_p): Delete.
4305 (aarch64_can_const_movi_rtx_p)
4306 (aarch64_simd_scalar_immediate_valid_for_move)
4307 (aarch64_simd_make_constant): Update call to
4308 aarch64_simd_valid_immediate.
4309 (aarch64_advsimd_valid_immediate_hs): New function.
4310 (aarch64_advsimd_valid_immediate): Likewise.
4311 (aarch64_simd_valid_immediate): Remove mode and inverse
4312 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4313 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4314 and aarch64_float_const_representable_p on the result.
4315 (aarch64_output_simd_mov_immediate): Remove mode argument.
4316 Update call to aarch64_simd_valid_immediate and use of
4317 simd_immediate_info.
4318 (aarch64_output_scalar_simd_mov_immediate): Update call
4321 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4322 Alan Hayward <alan.hayward@arm.com>
4323 David Sherwood <david.sherwood@arm.com>
4325 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4326 (mode_nunits): Likewise CONST_MODE_NUNITS.
4327 * machmode.def (ADJUST_NUNITS): Document.
4328 * genmodes.c (mode_data::need_nunits_adj): New field.
4329 (blank_mode): Update accordingly.
4330 (adj_nunits): New variable.
4331 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4333 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4334 listed in adj_nunits.
4335 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4336 listed in adj_nunits. Don't emit case statements for such modes.
4337 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4338 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4339 nothing if adj_nunits is nonnull.
4340 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4341 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4342 (emit_mode_fbit): Update use of print_maybe_const_decl.
4343 (emit_move_size): Likewise. Treat the array as non-const
4345 (emit_mode_adjustments): Handle adj_nunits.
4347 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4349 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4350 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4351 (VECTOR_MODES): Use it.
4352 (make_vector_modes): Take the prefix as an argument.
4354 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4355 Alan Hayward <alan.hayward@arm.com>
4356 David Sherwood <david.sherwood@arm.com>
4358 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4359 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4360 for MODE_VECTOR_BOOL.
4361 * machmode.def (VECTOR_BOOL_MODE): Document.
4362 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4363 (make_vector_bool_mode): New function.
4364 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4366 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4367 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4369 * stor-layout.c (int_mode_for_mode): Likewise.
4370 * tree.c (build_vector_type_for_mode): Likewise.
4371 * varasm.c (output_constant_pool_2): Likewise.
4372 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4373 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4374 for MODE_VECTOR_BOOL.
4375 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4376 of mode class checks.
4377 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4378 instead of a list of mode class checks.
4379 (expand_vector_scalar_condition): Likewise.
4380 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4382 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4383 Alan Hayward <alan.hayward@arm.com>
4384 David Sherwood <david.sherwood@arm.com>
4386 * machmode.h (mode_size): Change from unsigned short to
4388 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4389 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4390 or if measurement_type is not polynomial.
4391 (fixed_size_mode::includes_p): Check for constant-sized modes.
4392 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4393 return a poly_uint16 rather than an unsigned short.
4394 (emit_mode_size): Change the type of mode_size from unsigned short
4395 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4396 (emit_mode_adjustments): Cope with polynomial vector sizes.
4397 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4399 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4401 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4402 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4403 * caller-save.c (setup_save_areas): Likewise.
4404 (replace_reg_with_saved_mem): Likewise.
4405 * calls.c (emit_library_call_value_1): Likewise.
4406 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4407 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4408 (gen_lowpart_for_combine): Likewise.
4409 * convert.c (convert_to_integer_1): Likewise.
4410 * cse.c (equiv_constant, cse_insn): Likewise.
4411 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4412 (cselib_subst_to_values): Likewise.
4413 * dce.c (word_dce_process_block): Likewise.
4414 * df-problems.c (df_word_lr_mark_ref): Likewise.
4415 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4416 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4417 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4418 (rtl_for_decl_location): Likewise.
4419 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4420 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4421 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4422 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4423 (expand_expr_real_1): Likewise.
4424 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4425 (pad_below): Likewise.
4426 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4427 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4428 * ira.c (get_subreg_tracking_sizes): Likewise.
4429 * ira-build.c (ira_create_allocno_objects): Likewise.
4430 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4431 (ira_sort_regnos_for_alter_reg): Likewise.
4432 * ira-costs.c (record_operand_costs): Likewise.
4433 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4434 (resolve_simple_move): Likewise.
4435 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4436 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4437 (lra_constraints): Likewise.
4438 (CONST_POOL_OK_P): Reject variable-sized modes.
4439 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4440 (add_pseudo_to_slot, lra_spill): Likewise.
4441 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4442 * optabs-query.c (get_best_extraction_insn): Likewise.
4443 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4444 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4445 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4446 * recog.c (offsettable_address_addr_space_p): Likewise.
4447 * regcprop.c (maybe_mode_change): Likewise.
4448 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4449 * regrename.c (build_def_use): Likewise.
4450 * regstat.c (dump_reg_info): Likewise.
4451 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4452 (find_reloads, find_reloads_subreg_address): Likewise.
4453 * reload1.c (eliminate_regs_1): Likewise.
4454 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4455 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4456 (simplify_binary_operation_1, simplify_subreg): Likewise.
4457 * targhooks.c (default_function_arg_padding): Likewise.
4458 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4459 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4460 (verify_gimple_assign_ternary): Likewise.
4461 * tree-inline.c (estimate_move_cost): Likewise.
4462 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4463 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4464 (get_address_cost_ainc): Likewise.
4465 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4466 (vect_supportable_dr_alignment): Likewise.
4467 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4468 (vectorizable_reduction): Likewise.
4469 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4470 (vectorizable_operation, vectorizable_load): Likewise.
4471 * tree.c (build_same_sized_truth_vector_type): Likewise.
4472 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4473 * var-tracking.c (emit_note_insn_var_location): Likewise.
4474 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4475 (ADDR_VEC_ALIGN): Likewise.
4477 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4478 Alan Hayward <alan.hayward@arm.com>
4479 David Sherwood <david.sherwood@arm.com>
4481 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4483 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4484 or if measurement_type is polynomial.
4485 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4486 * combine.c (make_extraction): Likewise.
4487 * dse.c (find_shift_sequence): Likewise.
4488 * dwarf2out.c (mem_loc_descriptor): Likewise.
4489 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4490 (extract_bit_field, extract_low_bits): Likewise.
4491 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4492 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4493 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4494 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4495 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4496 * reload.c (find_reloads): Likewise.
4497 * reload1.c (alter_reg): Likewise.
4498 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4499 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4500 * tree-if-conv.c (predicate_mem_writes): Likewise.
4501 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4502 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4503 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4504 * valtrack.c (dead_debug_insert_temp): Likewise.
4505 * varasm.c (mergeable_constant_section): Likewise.
4506 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4508 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4509 Alan Hayward <alan.hayward@arm.com>
4510 David Sherwood <david.sherwood@arm.com>
4512 * expr.c (expand_assignment): Cope with polynomial mode sizes
4513 when assigning to a CONCAT.
4515 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4516 Alan Hayward <alan.hayward@arm.com>
4517 David Sherwood <david.sherwood@arm.com>
4519 * machmode.h (mode_precision): Change from unsigned short to
4521 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4523 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4524 or if measurement_type is not polynomial.
4525 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4526 in which the mode is already known to be a scalar_int_mode.
4527 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4528 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4530 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4531 for GET_MODE_PRECISION.
4532 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4533 for GET_MODE_PRECISION.
4534 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4536 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4537 (expand_field_assignment, make_extraction): Likewise.
4538 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4539 (get_last_value): Likewise.
4540 * convert.c (convert_to_integer_1): Likewise.
4541 * cse.c (cse_insn): Likewise.
4542 * expr.c (expand_expr_real_1): Likewise.
4543 * lra-constraints.c (simplify_operand_subreg): Likewise.
4544 * optabs-query.c (can_atomic_load_p): Likewise.
4545 * optabs.c (expand_atomic_load): Likewise.
4546 (expand_atomic_store): Likewise.
4547 * ree.c (combine_reaching_defs): Likewise.
4548 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4549 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4550 * tree.h (type_has_mode_precision_p): Likewise.
4551 * ubsan.c (instrument_si_overflow): Likewise.
4553 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4554 Alan Hayward <alan.hayward@arm.com>
4555 David Sherwood <david.sherwood@arm.com>
4557 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4558 polynomial numbers of units.
4559 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4560 (valid_vector_subparts_p): New function.
4561 (build_vector_type): Remove temporary shim and take the number
4562 of units as a poly_uint64 rather than an int.
4563 (build_opaque_vector_type): Take the number of units as a
4564 poly_uint64 rather than an int.
4565 * tree.c (build_vector_from_ctor): Handle polynomial
4566 TYPE_VECTOR_SUBPARTS.
4567 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4568 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4569 (build_vector_from_val): If the number of units is variable,
4570 use build_vec_duplicate_cst for constant operands and
4571 VEC_DUPLICATE_EXPR otherwise.
4572 (make_vector_type): Remove temporary is_constant ().
4573 (build_vector_type, build_opaque_vector_type): Take the number of
4574 units as a poly_uint64 rather than an int.
4575 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4577 * cfgexpand.c (expand_debug_expr): Likewise.
4578 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4579 (store_constructor, expand_expr_real_1): Likewise.
4580 (const_scalar_mask_from_tree): Likewise.
4581 * fold-const-call.c (fold_const_reduction): Likewise.
4582 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4583 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4584 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4585 (fold_relational_const): Likewise.
4586 (native_interpret_vector): Likewise. Change the size from an
4587 int to an unsigned int.
4588 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4589 TYPE_VECTOR_SUBPARTS.
4590 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4591 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4592 duplicating a non-constant operand into a variable-length vector.
4593 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4594 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4595 * ipa-icf.c (sem_variable::equals): Likewise.
4596 * match.pd: Likewise.
4597 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4598 * print-tree.c (print_node): Likewise.
4599 * stor-layout.c (layout_type): Likewise.
4600 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4601 * tree-cfg.c (verify_gimple_comparison): Likewise.
4602 (verify_gimple_assign_binary): Likewise.
4603 (verify_gimple_assign_ternary): Likewise.
4604 (verify_gimple_assign_single): Likewise.
4605 * tree-pretty-print.c (dump_generic_node): Likewise.
4606 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4607 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4608 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4609 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4610 (vect_shift_permute_load_chain): Likewise.
4611 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4612 (expand_vector_condition, optimize_vector_constructor): Likewise.
4613 (lower_vec_perm, get_compute_type): Likewise.
4614 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4615 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4616 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4617 (vect_recog_mask_conversion_pattern): Likewise.
4618 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4619 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4620 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4621 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4622 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4623 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4624 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4625 (supportable_widening_operation): Likewise.
4626 (supportable_narrowing_operation): Likewise.
4627 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4629 * varasm.c (output_constant): Likewise.
4631 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4632 Alan Hayward <alan.hayward@arm.com>
4633 David Sherwood <david.sherwood@arm.com>
4635 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4636 so that both the length == 3 and length != 3 cases set up their
4637 own permute vectors. Add comments explaining why we know the
4638 number of elements is constant.
4639 (vect_permute_load_chain): Likewise.
4641 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4642 Alan Hayward <alan.hayward@arm.com>
4643 David Sherwood <david.sherwood@arm.com>
4645 * machmode.h (mode_nunits): Change from unsigned char to
4647 (ONLY_FIXED_SIZE_MODES): New macro.
4648 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4649 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4650 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4652 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4653 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4654 or if measurement_type is not polynomial.
4655 * genmodes.c (ZERO_COEFFS): New macro.
4656 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4658 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4659 Use ZERO_COEFFS when emitting initializers.
4660 * data-streamer.h (bp_pack_poly_value): New function.
4661 (bp_unpack_poly_value): Likewise.
4662 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4663 for GET_MODE_NUNITS.
4664 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4665 for GET_MODE_NUNITS.
4666 * tree.c (make_vector_type): Remove temporary shim and make
4667 the real function take the number of units as a poly_uint64
4669 (build_vector_type_for_mode): Handle polynomial nunits.
4670 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4671 * emit-rtl.c (const_vec_series_p_1): Likewise.
4672 (gen_rtx_CONST_VECTOR): Likewise.
4673 * fold-const.c (test_vec_duplicate_folding): Likewise.
4674 * genrecog.c (validate_pattern): Likewise.
4675 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4676 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4677 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4678 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4679 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4680 * rtlanal.c (subreg_get_info): Likewise.
4681 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4682 (vect_grouped_load_supported): Likewise.
4683 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4684 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4685 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4686 (simplify_const_unary_operation, simplify_binary_operation_1)
4687 (simplify_const_binary_operation, simplify_ternary_operation)
4688 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4689 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4690 instead of CONST_VECTOR_NUNITS.
4691 * varasm.c (output_constant_pool_2): Likewise.
4692 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4693 explicit-encoded elements in the XVEC for variable-length vectors.
4695 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4697 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4699 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4700 Alan Hayward <alan.hayward@arm.com>
4701 David Sherwood <david.sherwood@arm.com>
4703 * coretypes.h (fixed_size_mode): Declare.
4704 (fixed_size_mode_pod): New typedef.
4705 * builtins.h (target_builtins::x_apply_args_mode)
4706 (target_builtins::x_apply_result_mode): Change type to
4707 fixed_size_mode_pod.
4708 * builtins.c (apply_args_size, apply_result_size, result_vector)
4709 (expand_builtin_apply_args_1, expand_builtin_apply)
4710 (expand_builtin_return): Update accordingly.
4712 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4714 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4715 * cselib.c (cselib_hash_rtx): Likewise.
4716 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4717 CONST_VECTOR encoding.
4719 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4720 Jeff Law <law@redhat.com>
4723 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4724 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4725 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4726 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4729 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4730 explicitly probe *sp in a noreturn function if there were any callee
4731 register saves or frame pointer is needed.
4733 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4736 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4737 BLKmode for ternary, binary or unary expressions.
4740 * var-tracking.c (delete_vta_debug_insn): New inline function.
4741 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4742 insns from get_insns () to NULL instead of each bb separately.
4743 Use delete_vta_debug_insn. No longer static.
4744 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4745 delete_vta_debug_insns callers.
4746 * rtl.h (delete_vta_debug_insns): Declare.
4747 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4748 instead of variable_tracking_main.
4750 2018-01-03 Martin Sebor <msebor@redhat.com>
4752 PR tree-optimization/83603
4753 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4754 arguments past the endof the argument list in functions declared
4755 without a prototype.
4756 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4757 Avoid checking when arguments are null.
4759 2018-01-03 Martin Sebor <msebor@redhat.com>
4762 * doc/extend.texi (attribute const): Fix a typo.
4763 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4764 issuing -Wsuggest-attribute for void functions.
4766 2018-01-03 Martin Sebor <msebor@redhat.com>
4768 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4769 offset_int::from instead of wide_int::to_shwi.
4770 (maybe_diag_overlap): Remove assertion.
4771 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4772 * gimple-ssa-sprintf.c (format_directive): Same.
4773 (parse_directive): Same.
4774 (sprintf_dom_walker::compute_format_length): Same.
4775 (try_substitute_return_value): Same.
4777 2017-01-03 Jeff Law <law@redhat.com>
4780 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4781 non-constant residual for zero at runtime and avoid probing in
4782 that case. Reorganize code for trailing problem to mirror handling
4785 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4787 PR tree-optimization/83501
4788 * tree-ssa-strlen.c (get_string_cst): New.
4789 (handle_char_store): Call get_string_cst.
4791 2018-01-03 Martin Liska <mliska@suse.cz>
4793 PR tree-optimization/83593
4794 * tree-ssa-strlen.c: Include tree-cfg.h.
4795 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4796 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4797 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4799 (strlen_dom_walker::before_dom_children): Call
4800 gimple_purge_dead_eh_edges. Dump tranformation with details
4802 (strlen_dom_walker::before_dom_children): Update call by adding
4803 new argument cleanup_eh.
4804 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4806 2018-01-03 Martin Liska <mliska@suse.cz>
4809 * cif-code.def (VARIADIC_THUNK): New enum value.
4810 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4813 2018-01-03 Jan Beulich <jbeulich@suse.com>
4815 * sse.md (mov<mode>_internal): Tighten condition for when to use
4816 vmovdqu<ssescalarsize> for TI and OI modes.
4818 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4820 Update copyright years.
4822 2018-01-03 Martin Liska <mliska@suse.cz>
4825 * ipa-visibility.c (function_and_variable_visibility): Skip
4826 functions with noipa attribure.
4828 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4830 * gcc.c (process_command): Update copyright notice dates.
4831 * gcov-dump.c (print_version): Ditto.
4832 * gcov.c (print_version): Ditto.
4833 * gcov-tool.c (print_version): Ditto.
4834 * gengtype.c (create_file): Ditto.
4835 * doc/cpp.texi: Bump @copying's copyright year.
4836 * doc/cppinternals.texi: Ditto.
4837 * doc/gcc.texi: Ditto.
4838 * doc/gccint.texi: Ditto.
4839 * doc/gcov.texi: Ditto.
4840 * doc/install.texi: Ditto.
4841 * doc/invoke.texi: Ditto.
4843 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4845 * vector-builder.h (vector_builder::m_full_nelts): Change from
4846 unsigned int to poly_uint64.
4847 (vector_builder::full_nelts): Update prototype accordingly.
4848 (vector_builder::new_vector): Likewise.
4849 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4850 (vector_builder::operator ==): Likewise.
4851 (vector_builder::finalize): Likewise.
4852 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4853 Take the number of elements as a poly_uint64 rather than an
4855 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4856 from unsigned int to poly_uint64.
4857 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4858 (vec_perm_indices::new_vector): Likewise.
4859 (vec_perm_indices::length): Likewise.
4860 (vec_perm_indices::nelts_per_input): Likewise.
4861 (vec_perm_indices::input_nelts): Likewise.
4862 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4863 number of elements per input as a poly_uint64 rather than an
4864 unsigned int. Use the original encoding for variable-length
4865 vectors, rather than clamping each individual element.
4866 For the second and subsequent elements in each pattern,
4867 clamp the step and base before clamping their sum.
4868 (vec_perm_indices::series_p): Handle polynomial element counts.
4869 (vec_perm_indices::all_in_range_p): Likewise.
4870 (vec_perm_indices_to_tree): Likewise.
4871 (vec_perm_indices_to_rtx): Likewise.
4872 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4873 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4874 (tree_vector_builder::new_binary_operation): Handle polynomial
4875 element counts. Return false if we need to know the number
4876 of elements at compile time.
4877 * fold-const.c (fold_vec_perm): Punt if the number of elements
4878 isn't known at compile time.
4880 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4882 * vec-perm-indices.h (vec_perm_builder): Change element type
4883 from HOST_WIDE_INT to poly_int64.
4884 (vec_perm_indices::element_type): Update accordingly.
4885 (vec_perm_indices::clamp): Handle polynomial element_types.
4886 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4887 (vec_perm_indices::all_in_range_p): Likewise.
4888 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4890 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4891 polynomial vec_perm_indices element types.
4892 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4893 * fold-const.c (fold_vec_perm): Likewise.
4894 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4895 * tree-vect-generic.c (lower_vec_perm): Likewise.
4896 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4897 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4898 element type to HOST_WIDE_INT.
4900 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4901 Alan Hayward <alan.hayward@arm.com>
4902 David Sherwood <david.sherwood@arm.com>
4904 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4905 rather than an int. Use plus_constant.
4906 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4907 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4909 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4910 Alan Hayward <alan.hayward@arm.com>
4911 David Sherwood <david.sherwood@arm.com>
4913 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4914 a HOST_WIDE_INT to a poly_int64.
4916 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4917 Alan Hayward <alan.hayward@arm.com>
4918 David Sherwood <david.sherwood@arm.com>
4920 * calls.c (load_register_parameters): Cope with polynomial
4921 mode sizes. Require a constant size for BLKmode parameters
4922 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4923 forces a parameter to be padded at the lsb end in order to
4924 fill a complete number of words, require the parameter size
4925 to be ordered wrt UNITS_PER_WORD.
4927 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4928 Alan Hayward <alan.hayward@arm.com>
4929 David Sherwood <david.sherwood@arm.com>
4931 * reload1.c (spill_stack_slot_width): Change element type
4932 from unsigned int to poly_uint64_pod.
4933 (alter_reg): Treat mode sizes as polynomial.
4935 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4936 Alan Hayward <alan.hayward@arm.com>
4937 David Sherwood <david.sherwood@arm.com>
4939 * reload.c (complex_word_subreg_p): New function.
4940 (reload_inner_reg_of_subreg, push_reload): Use it.
4942 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4943 Alan Hayward <alan.hayward@arm.com>
4944 David Sherwood <david.sherwood@arm.com>
4946 * lra-constraints.c (process_alt_operands): Reject matched
4947 operands whose sizes aren't ordered.
4948 (match_reload): Refer to this check here.
4950 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4951 Alan Hayward <alan.hayward@arm.com>
4952 David Sherwood <david.sherwood@arm.com>
4954 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4955 that the mode size is in the set {1, 2, 4, 8, 16}.
4957 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4958 Alan Hayward <alan.hayward@arm.com>
4959 David Sherwood <david.sherwood@arm.com>
4961 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4962 Use plus_constant instead of gen_rtx_PLUS.
4964 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4965 Alan Hayward <alan.hayward@arm.com>
4966 David Sherwood <david.sherwood@arm.com>
4968 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4969 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4970 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4971 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4972 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4973 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4974 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4975 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4976 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4977 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4979 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4980 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4981 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4982 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4983 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4984 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4985 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4986 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
4987 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
4988 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
4990 * expr.c (emit_move_resolve_push): Treat the input and result
4991 of PUSH_ROUNDING as a poly_int64.
4992 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
4993 (emit_push_insn): Likewise.
4994 * lra-eliminations.c (mark_not_eliminable): Likewise.
4995 * recog.c (push_operand): Likewise.
4996 * reload1.c (elimination_effects): Likewise.
4997 * rtlanal.c (nonzero_bits1): Likewise.
4998 * calls.c (store_one_arg): Likewise. Require the padding to be
4999 known at compile time.
5001 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5002 Alan Hayward <alan.hayward@arm.com>
5003 David Sherwood <david.sherwood@arm.com>
5005 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5006 Use plus_constant instead of gen_rtx_PLUS.
5008 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5009 Alan Hayward <alan.hayward@arm.com>
5010 David Sherwood <david.sherwood@arm.com>
5012 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5015 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5016 Alan Hayward <alan.hayward@arm.com>
5017 David Sherwood <david.sherwood@arm.com>
5019 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5020 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5021 via stack temporaries. Treat the mode size as polynomial too.
5023 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5024 Alan Hayward <alan.hayward@arm.com>
5025 David Sherwood <david.sherwood@arm.com>
5027 * expr.c (expand_expr_real_2): When handling conversions involving
5028 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5029 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5030 as a poly_uint64 too.
5032 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5033 Alan Hayward <alan.hayward@arm.com>
5034 David Sherwood <david.sherwood@arm.com>
5036 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5038 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5039 Alan Hayward <alan.hayward@arm.com>
5040 David Sherwood <david.sherwood@arm.com>
5042 * combine.c (can_change_dest_mode): Handle polynomial
5043 REGMODE_NATURAL_SIZE.
5044 * expmed.c (store_bit_field_1): Likewise.
5045 * expr.c (store_constructor): Likewise.
5046 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5047 and polynomial REGMODE_NATURAL_SIZE.
5048 (gen_lowpart_common): Likewise.
5049 * reginfo.c (record_subregs_of_mode): Likewise.
5050 * rtlanal.c (read_modify_subreg_p): Likewise.
5052 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5053 Alan Hayward <alan.hayward@arm.com>
5054 David Sherwood <david.sherwood@arm.com>
5056 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5057 numbers of elements.
5059 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5060 Alan Hayward <alan.hayward@arm.com>
5061 David Sherwood <david.sherwood@arm.com>
5063 * match.pd: Cope with polynomial numbers of vector elements.
5065 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5066 Alan Hayward <alan.hayward@arm.com>
5067 David Sherwood <david.sherwood@arm.com>
5069 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5070 in a POINTER_PLUS_EXPR.
5072 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5073 Alan Hayward <alan.hayward@arm.com>
5074 David Sherwood <david.sherwood@arm.com>
5076 * omp-simd-clone.c (simd_clone_subparts): New function.
5077 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5078 (ipa_simd_modify_function_body): Likewise.
5080 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5081 Alan Hayward <alan.hayward@arm.com>
5082 David Sherwood <david.sherwood@arm.com>
5084 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5085 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5086 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5087 (expand_vector_condition, vector_element): Likewise.
5088 (subparts_gt): New function.
5089 (get_compute_type): Use subparts_gt.
5090 (count_type_subparts): Delete.
5091 (expand_vector_operations_1): Use subparts_gt instead of
5092 count_type_subparts.
5094 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5095 Alan Hayward <alan.hayward@arm.com>
5096 David Sherwood <david.sherwood@arm.com>
5098 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5099 (vect_compile_time_alias): ...this new function. Do the calculation
5100 on poly_ints rather than trees.
5101 (vect_prune_runtime_alias_test_list): Update call accordingly.
5103 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5104 Alan Hayward <alan.hayward@arm.com>
5105 David Sherwood <david.sherwood@arm.com>
5107 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5109 (vect_schedule_slp_instance): Likewise.
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 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5116 constant and extern definitions for variable-length vectors.
5117 (vect_get_constant_vectors): Note that the number of units
5118 is known to be constant.
5120 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5121 Alan Hayward <alan.hayward@arm.com>
5122 David Sherwood <david.sherwood@arm.com>
5124 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5125 of units as polynomial. Choose between WIDE and NARROW based
5128 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5129 Alan Hayward <alan.hayward@arm.com>
5130 David Sherwood <david.sherwood@arm.com>
5132 * tree-vect-stmts.c (simd_clone_subparts): New function.
5133 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5135 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5136 Alan Hayward <alan.hayward@arm.com>
5137 David Sherwood <david.sherwood@arm.com>
5139 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5140 vectors as polynomial. Use build_index_vector for
5143 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5144 Alan Hayward <alan.hayward@arm.com>
5145 David Sherwood <david.sherwood@arm.com>
5147 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5148 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5149 for variable-length vectors.
5150 (vectorizable_mask_load_store): Treat the number of units as
5151 polynomial, asserting that it is constant if the condition has
5152 already been enforced.
5153 (vectorizable_store, vectorizable_load): Likewise.
5155 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5156 Alan Hayward <alan.hayward@arm.com>
5157 David Sherwood <david.sherwood@arm.com>
5159 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5160 of units as polynomial. Punt if we can't tell at compile time
5161 which vector contains the final result.
5163 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5164 Alan Hayward <alan.hayward@arm.com>
5165 David Sherwood <david.sherwood@arm.com>
5167 * tree-vect-loop.c (vectorizable_induction): Treat the number
5168 of units as polynomial. Punt on SLP inductions. Use an integer
5169 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5170 cast of such a series for variable-length floating-point
5173 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5174 Alan Hayward <alan.hayward@arm.com>
5175 David Sherwood <david.sherwood@arm.com>
5177 * tree.h (build_index_vector): Declare.
5178 * tree.c (build_index_vector): New function.
5179 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5180 of units as polynomial, forcibly converting it to a constant if
5181 vectorizable_reduction has already enforced the condition.
5182 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5183 to create a {1,2,3,...} vector.
5184 (vectorizable_reduction): Treat the number of units as polynomial.
5185 Choose vectype_in based on the largest scalar element size rather
5186 than the smallest number of units. Enforce the restrictions
5189 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5190 Alan Hayward <alan.hayward@arm.com>
5191 David Sherwood <david.sherwood@arm.com>
5193 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5194 number of units as polynomial.
5196 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5197 Alan Hayward <alan.hayward@arm.com>
5198 David Sherwood <david.sherwood@arm.com>
5200 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5201 * target.def (autovectorize_vector_sizes): Return the vector sizes
5202 by pointer, using vector_sizes rather than a bitmask.
5203 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5204 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5205 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5207 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5208 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5209 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5210 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5211 * omp-general.c (omp_max_vf): Likewise.
5212 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5213 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5214 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5215 * tree-vect-slp.c (vect_slp_bb): Likewise.
5216 * doc/tm.texi: Regenerate.
5217 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5219 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5220 the vector size as a poly_uint64 rather than an unsigned int.
5221 (current_vector_size): Change from an unsigned int to a poly_uint64.
5222 (get_vectype_for_scalar_type): Update accordingly.
5223 * tree.h (build_truth_vector_type): Take the size and number of
5224 units as a poly_uint64 rather than an unsigned int.
5225 (build_vector_type): Add a temporary overload that takes
5226 the number of units as a poly_uint64 rather than an unsigned int.
5227 * tree.c (make_vector_type): Likewise.
5228 (build_truth_vector_type): Take the number of units as a poly_uint64
5229 rather than an unsigned int.
5231 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5232 Alan Hayward <alan.hayward@arm.com>
5233 David Sherwood <david.sherwood@arm.com>
5235 * target.def (get_mask_mode): Take the number of units and length
5236 as poly_uint64s rather than unsigned ints.
5237 * targhooks.h (default_get_mask_mode): Update accordingly.
5238 * targhooks.c (default_get_mask_mode): Likewise.
5239 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5240 * doc/tm.texi: Regenerate.
5242 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5243 Alan Hayward <alan.hayward@arm.com>
5244 David Sherwood <david.sherwood@arm.com>
5246 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5247 * omp-general.c (omp_max_vf): Likewise.
5248 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5249 (expand_omp_simd): Handle polynomial safelen.
5250 * omp-low.c (omplow_simd_context): Add a default constructor.
5251 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5252 (lower_rec_simd_input_clauses): Update accordingly.
5253 (lower_rec_input_clauses): Likewise.
5255 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5256 Alan Hayward <alan.hayward@arm.com>
5257 David Sherwood <david.sherwood@arm.com>
5259 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5260 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5261 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5262 (vect_analyze_slp_cost): Likewise.
5263 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5264 (vect_model_load_cost): Likewise.
5266 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5267 Alan Hayward <alan.hayward@arm.com>
5268 David Sherwood <david.sherwood@arm.com>
5270 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5271 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5272 from an unsigned int * to a poly_uint64_pod *.
5273 (calculate_unrolling_factor): New function.
5274 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5276 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5277 Alan Hayward <alan.hayward@arm.com>
5278 David Sherwood <david.sherwood@arm.com>
5280 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5281 from an unsigned int to a poly_uint64.
5282 (_loop_vec_info::slp_unrolling_factor): Likewise.
5283 (_loop_vec_info::vectorization_factor): Change from an int
5285 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5286 (vect_get_num_vectors): New function.
5287 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5288 (vect_get_num_copies): Use vect_get_num_vectors.
5289 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5290 to an unsigned int *.
5291 (vect_analyze_data_refs): Change min_vf from an int * to a
5293 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5294 than an unsigned HOST_WIDE_INT.
5295 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5296 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5297 to an unsigned int *.
5298 (vect_analyze_data_ref_dependences): Likewise.
5299 (vect_compute_data_ref_alignment): Handle polynomial vf.
5300 (vect_enhance_data_refs_alignment): Likewise.
5301 (vect_prune_runtime_alias_test_list): Likewise.
5302 (vect_shift_permute_load_chain): Likewise.
5303 (vect_supportable_dr_alignment): Likewise.
5304 (dependence_distance_ge_vf): Take the vectorization factor as a
5305 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5306 (vect_analyze_data_refs): Change min_vf from an int * to a
5308 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5309 vfm1 as a poly_uint64 rather than an int. Make the same change
5310 for the returned bound_scalar.
5311 (vect_gen_vector_loop_niters): Handle polynomial vf.
5312 (vect_do_peeling): Likewise. Update call to
5313 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5314 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5316 * tree-vect-loop.c (vect_determine_vectorization_factor)
5317 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5318 (vect_get_known_peeling_cost): Likewise.
5319 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5320 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5321 (vect_transform_loop): Likewise. Use the lowest possible VF when
5322 updating the upper bounds of the loop.
5323 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5325 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5326 polynomial unroll factors.
5327 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5328 (vect_make_slp_decision): Likewise.
5329 (vect_supported_load_permutation_p): Likewise, and polynomial
5331 (vect_analyze_slp_cost): Handle polynomial vf.
5332 (vect_slp_analyze_node_operations): Likewise.
5333 (vect_slp_analyze_bb_1): Likewise.
5334 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5335 than an unsigned HOST_WIDE_INT.
5336 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5337 (vectorizable_load): Handle polynomial vf.
5338 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5340 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5342 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5343 Alan Hayward <alan.hayward@arm.com>
5344 David Sherwood <david.sherwood@arm.com>
5346 * match.pd: Handle bit operations involving three constants
5347 and try to fold one pair.
5349 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5351 * tree-vect-loop-manip.c: Include gimple-fold.h.
5352 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5353 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5354 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5355 Add a path that uses a step of VF instead of 1, but disable it
5357 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5358 and niters_no_overflow parameters. Update calls to
5359 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5360 Create a new SSA name if the latter choses to use a ste other
5361 than zero, and return it via niters_vector_mult_vf_var.
5362 * tree-vect-loop.c (vect_transform_loop): Update calls to
5363 vect_do_peeling, vect_gen_vector_loop_niters and
5364 slpeel_make_loop_iterate_ntimes.
5365 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5366 (vect_gen_vector_loop_niters): Update declarations after above changes.
5368 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5370 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5371 128-bit round to integer instructions.
5372 (ceil<mode>2): Likewise.
5373 (btrunc<mode>2): Likewise.
5374 (round<mode>2): Likewise.
5376 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5378 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5379 unaligned VSX load/store on P8/P9.
5380 (expand_block_clear): Allow the use of unaligned VSX
5381 load/store on P8/P9.
5383 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5385 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5387 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5388 swap associated with both a load and a store.
5390 2018-01-02 Andrew Waterman <andrew@sifive.com>
5392 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5393 * config/riscv/riscv.md (clear_cache): Use it.
5395 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5397 * web.c: Remove out-of-date comment.
5399 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5401 * expr.c (fixup_args_size_notes): Check that any existing
5402 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5403 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5404 (emit_single_push_insn): ...here.
5406 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5408 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5409 (const_vector_encoded_nelts): New function.
5410 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5411 (const_vector_int_elt, const_vector_elt): Declare.
5412 * emit-rtl.c (const_vector_int_elt_1): New function.
5413 (const_vector_elt): Likewise.
5414 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5415 of CONST_VECTOR_ELT.
5417 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5419 * expr.c: Include rtx-vector-builder.h.
5420 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5421 directly on the tree encoding.
5422 (const_vector_from_tree): Likewise.
5423 * optabs.c: Include rtx-vector-builder.h.
5424 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5425 sequence of "u" values.
5426 * vec-perm-indices.c: Include rtx-vector-builder.h.
5427 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5428 directly on the vec_perm_indices encoding.
5430 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5432 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5433 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5434 * rtx-vector-builder.h: New file.
5435 * rtx-vector-builder.c: Likewise.
5436 * rtl.h (rtx_def::u2): Add a const_vector field.
5437 (CONST_VECTOR_NPATTERNS): New macro.
5438 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5439 (CONST_VECTOR_DUPLICATE_P): Likewise.
5440 (CONST_VECTOR_STEPPED_P): Likewise.
5441 (CONST_VECTOR_ENCODED_ELT): Likewise.
5442 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5443 (unwrap_const_vec_duplicate): Likewise.
5444 (const_vec_series_p): Check for a non-duplicated vector encoding.
5445 Say that the function only returns true for integer vectors.
5446 * emit-rtl.c: Include rtx-vector-builder.h.
5447 (gen_const_vec_duplicate_1): Delete.
5448 (gen_const_vector): Call gen_const_vec_duplicate instead of
5449 gen_const_vec_duplicate_1.
5450 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5451 (gen_const_vec_duplicate): Use rtx_vector_builder.
5452 (gen_const_vec_series): Likewise.
5453 (gen_rtx_CONST_VECTOR): Likewise.
5454 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5455 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5456 Build a new vector rather than modifying a CONST_VECTOR in-place.
5457 (handle_special_swappables): Update call accordingly.
5458 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5459 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5460 Build a new vector rather than modifying a CONST_VECTOR in-place.
5461 (handle_special_swappables): Update call accordingly.
5463 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5465 * simplify-rtx.c (simplify_const_binary_operation): Use
5466 CONST_VECTOR_ELT instead of XVECEXP.
5468 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5470 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5471 the selector elements to be different from the data elements
5472 if the selector is a VECTOR_CST.
5473 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5474 ssizetype for the selector.
5476 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5478 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5479 before testing each element individually.
5480 * tree-vect-generic.c (lower_vec_perm): Likewise.
5482 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5484 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5485 * selftest-run-tests.c (selftest::run_tests): Call it.
5486 * vector-builder.h (vector_builder::operator ==): New function.
5487 (vector_builder::operator !=): Likewise.
5488 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5489 (vec_perm_indices::all_from_input_p): New function.
5490 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5491 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5492 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5493 instead of reading the VECTOR_CST directly. Detect whether both
5494 vector inputs are the same before constructing the vec_perm_indices,
5495 and update the number of inputs argument accordingly. Use the
5496 utility functions added above. Only construct sel2 if we need to.
5498 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5500 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5501 the broadcast of the low byte.
5502 (expand_mult_highpart): Use an explicit encoding for the permutes.
5503 * optabs-query.c (can_mult_highpart_p): Likewise.
5504 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5505 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5506 (vectorizable_bswap): Likewise.
5507 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5508 explicit encoding for the power-of-2 permutes.
5509 (vect_permute_store_chain): Likewise.
5510 (vect_grouped_load_supported): Likewise.
5511 (vect_permute_load_chain): Likewise.
5513 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5515 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5516 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5517 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5518 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5519 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5520 (vect_gen_perm_mask_any): Likewise.
5522 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5524 * int-vector-builder.h: New file.
5525 * vec-perm-indices.h: Include int-vector-builder.h.
5526 (vec_perm_indices): Redefine as an int_vector_builder.
5527 (auto_vec_perm_indices): Delete.
5528 (vec_perm_builder): Redefine as a stand-alone class.
5529 (vec_perm_indices::vec_perm_indices): New function.
5530 (vec_perm_indices::clamp): Likewise.
5531 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5532 (vec_perm_indices::new_vector): New function.
5533 (vec_perm_indices::new_expanded_vector): Update for new
5534 vec_perm_indices class.
5535 (vec_perm_indices::rotate_inputs): New function.
5536 (vec_perm_indices::all_in_range_p): Operate directly on the
5537 encoded form, without computing elided elements.
5538 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5539 encoding. Update for new vec_perm_indices class.
5540 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5541 the given vec_perm_builder.
5542 (expand_vec_perm_var): Update vec_perm_builder constructor.
5543 (expand_mult_highpart): Use vec_perm_builder instead of
5544 auto_vec_perm_indices.
5545 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5546 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5547 or double series encoding as appropriate.
5548 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5549 vec_perm_indices instead of auto_vec_perm_indices.
5550 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5551 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5552 (vect_permute_store_chain): Likewise.
5553 (vect_grouped_load_supported): Likewise.
5554 (vect_permute_load_chain): Likewise.
5555 (vect_shift_permute_load_chain): Likewise.
5556 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5557 (vect_transform_slp_perm_load): Likewise.
5558 (vect_schedule_slp_instance): Likewise.
5559 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5560 (vectorizable_mask_load_store): Likewise.
5561 (vectorizable_bswap): Likewise.
5562 (vectorizable_store): Likewise.
5563 (vectorizable_load): Likewise.
5564 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5565 vec_perm_indices instead of auto_vec_perm_indices. Use
5566 tree_to_vec_perm_builder to read the vector from a tree.
5567 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5568 vec_perm_builder instead of a vec_perm_indices.
5569 (have_whole_vector_shift): Use vec_perm_builder and
5570 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5571 truncation to calc_vec_perm_mask_for_shift.
5572 (vect_create_epilog_for_reduction): Likewise.
5573 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5574 from auto_vec_perm_indices to vec_perm_indices.
5575 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5576 instead of changing individual elements.
5577 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5578 the vector in d.perm.
5579 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5580 from auto_vec_perm_indices to vec_perm_indices.
5581 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5582 instead of changing individual elements.
5583 (arm_vectorize_vec_perm_const): Use new_vector to install
5584 the vector in d.perm.
5585 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5586 Update vec_perm_builder constructor.
5587 (rs6000_expand_interleave): Likewise.
5588 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5589 (rs6000_expand_interleave): Likewise.
5591 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5593 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5594 to qimode could truncate the indices.
5595 * optabs.c (expand_vec_perm_var): Likewise.
5597 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5599 * Makefile.in (OBJS): Add vec-perm-indices.o.
5600 * vec-perm-indices.h: New file.
5601 * vec-perm-indices.c: Likewise.
5602 * target.h (vec_perm_indices): Replace with a forward class
5604 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5605 * optabs.h: Include vec-perm-indices.h.
5606 (expand_vec_perm): Delete.
5607 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5608 (expand_vec_perm_const): Declare.
5609 * target.def (vec_perm_const_ok): Replace with...
5610 (vec_perm_const): ...this new hook.
5611 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5612 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5613 * doc/tm.texi: Regenerate.
5614 * optabs.def (vec_perm_const): Delete.
5615 * doc/md.texi (vec_perm_const): Likewise.
5616 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5617 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5618 expand_vec_perm for constant permutation vectors. Assert that
5619 the mode of variable permutation vectors is the integer equivalent
5620 of the mode that is being permuted.
5621 * optabs-query.h (selector_fits_mode_p): Declare.
5622 * optabs-query.c: Include vec-perm-indices.h.
5623 (selector_fits_mode_p): New function.
5624 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5625 is defined, instead of checking whether the vec_perm_const_optab
5626 exists. Use targetm.vectorize.vec_perm_const instead of
5627 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5628 fit in the vector mode before using a variable permute.
5629 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5630 vec_perm_indices instead of an rtx.
5631 (expand_vec_perm): Replace with...
5632 (expand_vec_perm_const): ...this new function. Take the selector
5633 as a vec_perm_indices rather than an rtx. Also take the mode of
5634 the selector. Update call to shift_amt_for_vec_perm_mask.
5635 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5636 Use vec_perm_indices::new_expanded_vector to expand the original
5637 selector into bytes. Check whether the indices fit in the vector
5638 mode before using a variable permute.
5639 (expand_vec_perm_var): Make global.
5640 (expand_mult_highpart): Use expand_vec_perm_const.
5641 * fold-const.c: Includes vec-perm-indices.h.
5642 * tree-ssa-forwprop.c: Likewise.
5643 * tree-vect-data-refs.c: Likewise.
5644 * tree-vect-generic.c: Likewise.
5645 * tree-vect-loop.c: Likewise.
5646 * tree-vect-slp.c: Likewise.
5647 * tree-vect-stmts.c: Likewise.
5648 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5650 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5651 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5652 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5653 (aarch64_vectorize_vec_perm_const): ...this new function.
5654 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5655 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5656 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5657 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5658 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5659 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5660 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5662 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5663 check for NEON modes.
5664 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5665 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5666 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5667 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5669 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5670 the old VEC_PERM_CONST conditions.
5671 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5672 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5673 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5674 (ia64_vectorize_vec_perm_const_ok): Merge into...
5675 (ia64_vectorize_vec_perm_const): ...this new function.
5676 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5677 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5678 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5679 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5680 * config/mips/mips.c (mips_expand_vec_perm_const)
5681 (mips_vectorize_vec_perm_const_ok): Merge into...
5682 (mips_vectorize_vec_perm_const): ...this new function.
5683 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5684 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5685 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5686 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5687 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5688 (rs6000_expand_vec_perm_const): Delete.
5689 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5691 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5692 (altivec_expand_vec_perm_const_le): Take each operand individually.
5693 Operate on constant selectors rather than rtxes.
5694 (altivec_expand_vec_perm_const): Likewise. Update call to
5695 altivec_expand_vec_perm_const_le.
5696 (rs6000_expand_vec_perm_const): Delete.
5697 (rs6000_vectorize_vec_perm_const_ok): Delete.
5698 (rs6000_vectorize_vec_perm_const): New function.
5699 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5700 an element count and rtx array.
5701 (rs6000_expand_extract_even): Update call accordingly.
5702 (rs6000_expand_interleave): Likewise.
5703 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5704 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5705 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5706 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5707 (rs6000_expand_vec_perm_const): Delete.
5708 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5709 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5710 (altivec_expand_vec_perm_const_le): Take each operand individually.
5711 Operate on constant selectors rather than rtxes.
5712 (altivec_expand_vec_perm_const): Likewise. Update call to
5713 altivec_expand_vec_perm_const_le.
5714 (rs6000_expand_vec_perm_const): Delete.
5715 (rs6000_vectorize_vec_perm_const_ok): Delete.
5716 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5717 reference to the SPE evmerge intructions.
5718 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5719 an element count and rtx array.
5720 (rs6000_expand_extract_even): Update call accordingly.
5721 (rs6000_expand_interleave): Likewise.
5722 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5723 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5725 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5727 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5729 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5730 vector mode and that that mode matches the mode of the data
5732 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5733 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5734 directly using expand_vec_perm_1 when forcing selectors into
5736 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5738 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5740 * optabs-query.h (can_vec_perm_p): Delete.
5741 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5742 * optabs-query.c (can_vec_perm_p): Split into...
5743 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5744 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5745 particular selector is valid.
5746 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5747 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5748 (vect_grouped_load_supported): Likewise.
5749 (vect_shift_permute_load_chain): Likewise.
5750 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5751 (vect_transform_slp_perm_load): Likewise.
5752 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5753 (vectorizable_bswap): Likewise.
5754 (vect_gen_perm_mask_checked): Likewise.
5755 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5756 implementations of variable permutation vectors into account
5757 when deciding which selector to use.
5758 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5759 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5760 with a false third argument.
5761 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5762 to test whether the constant selector is valid and can_vec_perm_var_p
5763 to test whether a variable selector is valid.
5765 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5767 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5768 * optabs-query.c (can_vec_perm_p): Likewise.
5769 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5770 instead of vec_perm_indices.
5771 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5772 (vect_gen_perm_mask_checked): Likewise,
5773 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5774 (vect_gen_perm_mask_checked): Likewise,
5776 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5778 * optabs-query.h (qimode_for_vec_perm): Declare.
5779 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5780 (qimode_for_vec_perm): ...this new function.
5781 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5783 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5785 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5786 does not have a conditional at the top.
5788 2018-01-02 Richard Biener <rguenther@suse.de>
5790 * ipa-inline.c (big_speedup_p): Fix expression.
5792 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5795 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5798 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5802 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5803 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5804 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5805 cond_taken_branch_cost 3->4.
5807 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5809 PR tree-optimization/83581
5810 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5811 TODO_cleanup_cfg if any changes have been made.
5814 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5815 convert_modes if target mode has the right side, but different mode
5819 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5820 last argument when extracting from CONCAT. If either from_real or
5821 from_imag is NULL, use expansion through memory. If result is not
5822 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5823 the parts directly to inner mode, if even that fails, use expansion
5827 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5828 check for bswap in mode rather than HImode and use that in expand_unop
5831 Copyright (C) 2018 Free Software Foundation, Inc.
5833 Copying and distribution of this file, with or without modification,
5834 are permitted in any medium without royalty provided the copyright
5835 notice and this notice are preserved.
projects / platform / upstream / gcc.git / blob