1 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
3 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
5 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
6 Remove all values except native, 8540 and 8548.
8 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
10 * config/i386/i386.c (ix86_output_function_return): Pass
11 INVALID_REGNUM, instead of -1, as invalid register number to
12 indirect_thunk_name and output_indirect_thunk.
14 2018-02-02 Julia Koval <julia.koval@intel.com>
16 * config.gcc: Add -march=icelake.
17 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
18 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
19 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
20 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
21 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
22 (processor_target_table): Add icelake.
23 (ix86_option_override_internal): Handle new PTAs.
24 (get_builtin_code_for_version): Handle icelake.
25 (M_INTEL_COREI7_ICELAKE): New.
26 (fold_builtin_cpu): Handle icelake.
27 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
28 * doc/invoke.texi: Add -march=icelake.
30 2018-02-02 Julia Koval <julia.koval@intel.com>
32 * config/i386/i386.c (ix86_option_override_internal): Change flags type
34 * wide-int-bitmask.h: New.
36 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
39 * config/i386/i386.md: Replace Pmode with word_mode in
40 builtin_setjmp_setup and builtin_longjmp to support x32.
42 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
46 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
48 (rs6000_supported_cpu_names): New static variable.
49 (linux_cpu_translation_table): Likewise.
50 (elf_platform) <cpu>: Define new static variable and use it.
51 Translate kernel AT_PLATFORM name to canonical name if needed.
52 Error if platform name is unknown.
54 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
57 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
59 2018-02-01 Jeff Law <law@redhat.com>
62 * config/i386/i386.c (release_scratch_register_on_entry): Add new
63 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
64 the scratch if RELEASE_VIA_POP is false.
65 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
66 If we have to save a temporary register, decrement SIZE appropriately.
67 Pass new arguments to release_scratch_register_on_entry.
68 (ix86_adjust_stack_and_probe): Likewise.
69 (ix86_emit_probe_stack_range): Pass new arguments to
70 release_scratch_register_on_entry.
72 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
74 PR rtl-optimization/84157
75 * combine.c (change_zero_ext): Use REG_P predicate in
76 front of HARD_REGISTER_P predicate.
78 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
80 * config/avr/avr.c (avr_option_override): Move disabling of
81 -fdelete-null-pointer-checks to...
82 * common/config/avr/avr-common.c (avr_option_optimization_table):
85 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
87 PR tree-optimization/81635
88 * tree-data-ref.c (split_constant_offset_1): For types that
89 wrap on overflow, try to use range info to prove that wrapping
92 2018-02-01 Renlin Li <renlin.li@arm.com>
95 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
97 (aarch64_register_move_cost): Likewise.
98 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
100 (REG_CLASS_NAMES): Likewise.
101 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
102 TAILCALL_ADDR_REGS. Remove IP registers.
103 * config/aarch64/aarch64.md (Ucs): Update register constraint.
105 2018-02-01 Richard Biener <rguenther@suse.de>
107 * domwalk.h (dom_walker::dom_walker): Add additional constructor
108 for specifying RPO order and allow NULL for that.
109 * domwalk.c (dom_walker::dom_walker): Likewise.
110 (dom_walker::walk): Handle NULL RPO order.
111 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
113 (rewrite_update_dom_walker): Likewise.
114 (mark_def_dom_walker): Likewise.
116 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
118 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
119 (aarch64_maybe_expand_sve_subreg_move): Declare.
120 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
121 * config/aarch64/predicates.md (aarch64_any_register_operand): New
123 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
124 that are semantically a reverse operation.
125 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
126 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
127 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
129 (aarch64_can_change_mode_class): For big-endian, forbid changes
130 between two SVE modes if they have different element sizes.
132 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
134 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
135 the TImode handling for big-endian targets.
137 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
139 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
140 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
142 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
143 Remove BSWAP handing for big-endian targets and use the form of
144 LD1RQ appropariate for the mode.
146 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
148 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
149 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
152 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
155 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
156 check for operands that need to go through aarch64_sve_reload_be.
158 2018-02-01 Jakub Jelinek <jakub@redhat.com>
160 PR tree-optimization/81661
161 PR tree-optimization/84117
162 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
163 * tree-eh.c: Include gimplify.h.
164 (find_trapping_overflow, replace_trapping_overflow,
165 rewrite_to_non_trapping_overflow): New functions.
166 * tree-vect-loop.c: Include tree-eh.h.
167 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
168 * tree-data-ref.c: Include tree-eh.h.
169 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
171 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
173 PR rtl-optimization/84123
174 * combine.c (change_zero_ext): Check if hard register satisfies
175 can_change_dest_mode before calling gen_lowpart_SUBREG.
177 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
180 * ira.c (ira_init_register_move_cost): Remove assert.
182 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
184 PR rtl-optimization/84071
185 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
186 * doc/tm.texi: Regenerate.
188 2018-01-31 Richard Biener <rguenther@suse.de>
190 PR tree-optimization/84132
191 * tree-data-ref.c (analyze_miv_subscript): Properly
192 check whether evolution_function_is_affine_multivariate_p
193 before calling gcd_of_steps_may_divide_p.
195 2018-01-31 Julia Koval <julia.koval@intel.com>
198 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
199 * config/i386/i386.md (rdpid_rex64) New.
200 (rdpid): Make 32bit only.
202 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
205 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
206 an IDENTIFIER_NODE for FUNCTION_TYPE's.
208 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
211 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
213 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
215 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
217 PR rtl-optimization/84071
218 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
219 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
221 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
223 * config/arc/arc.c (arc_handle_aux_attribute): New function.
224 (arc_attribute_table): Add 'aux' attribute.
225 (arc_in_small_data_p): Consider aux like variables.
226 (arc_is_aux_reg_p): New function.
227 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
228 (arc_get_aux_arg): New function.
229 (prepare_move_operands): Handle aux-register access.
230 (arc_handle_aux_attribute): New function.
231 * doc/extend.texi (ARC Variable attributes): Add subsection.
233 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
235 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
236 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
237 (arc_attribute_table): Add 'uncached' attribute.
238 (arc_print_operand): Print '.di' flag for uncached memory
240 (arc_in_small_data_p): Do not consider for small data the uncached
242 (arc_is_uncached_mem_p): New function.
243 * config/arc/predicates.md (compact_store_memory_operand): Check
244 for uncached memory accesses.
245 (nonvol_nonimm_operand): Likewise.
246 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
248 2018-01-31 Jakub Jelinek <jakub@redhat.com>
251 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
252 falign-loops=): Add Optimization flag.
254 2018-01-30 Jeff Law <law@redhat.com>
257 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
258 INT_REGISTERS_SAVED. Check it prior to calling
259 get_scratch_register_on_entry.
260 (ix86_adjust_stack_and_probe): Similarly.
261 (ix86_emit_probe_stack_range): Similarly.
262 (ix86_expand_prologue): Corresponding changes.
264 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
267 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
268 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
270 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
273 * lra-constraints.c (curr_insn_transform): Process AND in the
276 2018-01-30 Jakub Jelinek <jakub@redhat.com>
278 PR rtl-optimization/83986
279 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
280 dependence against last_pending_memory_flush in addition to
283 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
285 PR tree-optimization/81611
286 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
289 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
291 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
294 2018-01-30 Richard Biener <rguenther@suse.de>
295 Jakub Jelinek <jakub@redhat.com>
297 PR tree-optimization/84111
298 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
299 inner loops added during recursion, as they don't have up-to-date
302 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
305 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
306 (can_inline_edge_by_limits_p): ... here.
307 (can_early_inline_edge_p, check_callers,
308 update_caller_keys, update_callee_keys, recursive_inlining,
309 add_new_edges_to_heap, speculation_useful_p,
310 inline_small_functions,
311 inline_small_functions, flatten_function,
312 inline_to_all_callers_1): Update.
314 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
316 * profile-count.c (profile_count::combine_with_ipa_count): Handle
319 2018-01-30 Richard Biener <rguenther@suse.de>
321 PR tree-optimization/83008
322 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
323 invariant and constant vector uses in stmts when they need
326 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
329 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
330 * configure: Regenerate.
332 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
334 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
336 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
337 Use gen_rtx_REG rather than gen_lowpart.
339 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
341 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
342 rather than 0 when creating partial subregs.
344 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
346 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
349 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
352 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
353 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
354 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
355 flags. This restores the settings used before the 2017-07-24.
356 Turning off pre increment/decrement/modify allows IVOPTS to
357 optimize DF/SF loops where the index is an int.
359 2018-01-29 Richard Biener <rguenther@suse.de>
360 Kelvin Nilsen <kelvin@gcc.gnu.org>
363 * tree-vect-stmts.c (vectorizable_call): Don't call
364 targetm.vectorize_builtin_md_vectorized_function if callee is
367 2018-01-22 Carl Love <cel@us.ibm.com>
369 * doc/extend.tex: Fix typo in second arg in
370 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
372 2018-01-29 Richard Biener <rguenther@suse.de>
374 PR tree-optimization/84086
375 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
376 (flush_ssaname_freelist): When SSA names were released reset
379 2018-01-29 Richard Biener <rguenther@suse.de>
381 PR tree-optimization/84057
382 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
383 removed paths when removing edges.
385 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
387 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
388 -mfunction-return=@var{choice}.
390 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
393 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
395 (layout::print_source_line): Likewise.
396 (test_get_line_width_without_trailing_whitespace): Add test cases.
398 2018-01-27 Jakub Jelinek <jakub@redhat.com>
401 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
404 2018-01-26 Jim Wilson <jimw@sifive.com>
406 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
408 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
411 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
413 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
414 and CMP + SUB-immediate -> SUBS.
416 2018-01-26 Martin Sebor <msebor@redhat.com>
418 PR tree-optimization/83896
419 * tree-ssa-strlen.c (get_string_len): Rename...
420 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
421 Avoid assuming length is constant.
422 (handle_char_store): Use HOST_WIDE_INT for string length.
424 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
427 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
428 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
430 2018-01-26 Richard Biener <rguenther@suse.de>
432 PR rtl-optimization/84003
433 * dse.c (record_store): Only record redundant stores when
434 the earlier store aliases at least all accesses the later one does.
436 2018-01-26 Jakub Jelinek <jakub@redhat.com>
438 PR rtl-optimization/83985
439 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
440 REG_CFA_RESTORE insns.
441 (delete_unmarked_insns): Don't ignore separate shrink wrapping
442 REG_CFA_RESTORE insns here.
445 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
446 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
448 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
450 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
451 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
452 (arc_init): Likewise.
453 (arc_override_options): Likewise.
454 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
456 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
458 * config/arc/arc.h (TARGET_DBNZ): Define.
459 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
460 properly set the tune attribute.
461 (dbnz): Use TARGET_DBNZ guard.
462 * config/arc/arc.opt (mtune): Add core3 option.
464 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
466 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
467 recognize new pic like addresses.
468 (arc_delegitimize_address): Clean up.
470 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
472 * config/arc/arc-arches.def: Option mrf16 valid for all
474 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
475 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
476 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
477 * config/arc/arc-tables.opt: Regenerate.
478 * config/arc/arc.c (arc_conditional_register_usage): Handle
479 reduced register file case.
480 (arc_file_start): Set must have build attributes.
481 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
483 * config/arc/arc.opt (mrf16): Add new option.
484 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
485 * config/arc/genmultilib.awk: Handle new mrf16 option.
486 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
487 * config/arc/t-multilib: Regenerate.
488 * doc/invoke.texi (ARC Options): Document mrf16 option.
490 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
492 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
493 * config/arc/arc.c (arc_handle_secure_attribute): New function.
494 (arc_attribute_table): Add 'secure_call' attribute.
495 (arc_print_operand): Print secure call operand.
496 (arc_function_ok_for_sibcall): Don't optimize tail calls when
498 (arc_is_secure_call_p): New function. * config/arc/arc.md
499 (call_i): Add support for sjli instruction.
500 (call_value_i): Likewise.
501 * config/arc/constraints.md (Csc): New constraint.
503 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
504 John Eric Martin <John.Martin@emmicro-us.com>
506 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
507 * config/arc/arc.c (_arc_jli_section): New struct.
508 (arc_jli_section): New type.
509 (rc_jli_sections): New static variable.
510 (arc_handle_jli_attribute): New function.
511 (arc_attribute_table): Add jli_always and jli_fixed attribute.
512 (arc_file_end): New function.
513 (TARGET_ASM_FILE_END): Define.
514 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
515 (arc_add_jli_section): New function.
516 (jli_call_scan): Likewise.
517 (arc_reorg): Call jli_call_scan.
518 (arc_output_addsi): Remove 'S' from printing asm operand.
519 (arc_is_jli_call_p): New function.
520 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
522 (movhi_insn): Likewise.
523 (movsi_insn): Likewise.
524 (movsi_set_cc_insn): Likewise.
525 (loadqi_update): Likewise.
526 (load_zeroextendqisi_update): Likewise.
527 (load_signextendqisi_update): Likewise.
528 (loadhi_update): Likewise.
529 (load_zeroextendhisi_update): Likewise.
530 (load_signextendhisi_update): Likewise.
531 (loadsi_update): Likewise.
532 (loadsf_update): Likewise.
533 (movsicc_insn): Likewise.
534 (bset_insn): Likewise.
535 (bxor_insn): Likewise.
536 (bclr_insn): Likewise.
537 (bmsk_insn): Likewise.
538 (bicsi3_insn): Likewise.
539 (cmpsi_cc_c_insn): Likewise.
540 (movsi_ne): Likewise.
541 (movsi_cond_exec): Likewise.
542 (clrsbsi2): Likewise.
553 (call_i): Remove 'S' asm letter, add jli instruction.
554 (call_value_i): Likewise.
555 * config/arc/arc.op (mjli-always): New option.
556 * config/arc/constraints.md (Cji): New constraint.
557 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
559 (subsf3_fpx): Likewise.
560 (mulsf3_fpx): Likewise.
561 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
563 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
565 * doc/invoke.texi (ARC): Document mjli-always option.
567 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
569 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
570 and use incw and decw where possible
571 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
573 2018-01-26 Richard Biener <rguenther@suse.de>
575 PR tree-optimization/81082
576 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
577 association if it requires casting to unsigned.
578 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
579 from fold_plusminus_mult_expr to catch important cases late when
580 range info is available.
582 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
584 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
585 * configure.ac (hidden_linkonce): New test.
586 * configure: Regenerate.
587 * config.in: Regenerate.
589 2018-01-26 Julia Koval <julia.koval@intel.com>
591 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
592 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
593 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
594 _mm_mask_bitshuffle_epi64_mask): Fix type.
595 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
596 USI_FTYPE_V4DI_V4DI_USI): Remove.
597 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
598 __builtin_ia32_vpshufbitqmb256_mask,
599 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
600 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
601 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
603 2018-01-26 Alan Modra <amodra@gmail.com>
606 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
607 UNSPEC_VBPERMQ. Sort other unspecs.
609 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
611 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
613 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
616 * predict.c (drop_profile): Do not push/pop cfun; update also
618 (handle_missing_profiles): Fix logic looking for zero profiles.
620 2018-01-25 Jakub Jelinek <jakub@redhat.com>
623 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
624 on functions with #pragma omp declare simd or functions with simd
626 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
627 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
628 Remove trailing \n from warning_at calls.
630 2018-01-25 Tom de Vries <tom@codesourcery.com>
633 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
634 for neutered workers.
636 2018-01-24 Joseph Myers <joseph@codesourcery.com>
639 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
640 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
642 2018-01-24 Jeff Law <law@redhat.com>
645 * i386.c (get_probe_interval): Move to earlier point.
646 (ix86_compute_frame_layout): If -fstack-clash-protection and
647 the frame is larger than the probe interval, then use pushes
648 to save registers rather than reg->mem moves.
649 (ix86_expand_prologue): Remove conditional for int_registers_saved
652 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
655 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
656 min/max for never referenced object.
658 2018-01-24 Jakub Jelinek <jakub@redhat.com>
661 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
663 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
664 attributes from DECL_ATTRIBUTES (decl) without affecting
665 DECL_ATTRIBUTES (current_function_decl).
666 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
667 functions with non-NULL DECL_ABSTRACT_ORIGIN.
669 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
671 PR tree-optimization/83979
672 * fold-const.c (fold_comparison): Use constant_boolean_node
673 instead of boolean_{true,false}_node.
675 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
677 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
680 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
682 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
683 Simplify the clause that sets the length attribute.
684 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
685 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
686 clause that sets the length attribute.
687 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
689 2018-01-24 Tom de Vries <tom@codesourcery.com>
692 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
693 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
694 Add strict parameter.
695 (prevent_branch_around_nothing): Insert dummy insn between branch to
696 label and label with no ptx insn inbetween.
697 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
699 2018-01-24 Tom de Vries <tom@codesourcery.com>
702 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
703 for neutered threads in warp.
704 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
706 2018-01-24 Richard Biener <rguenther@suse.de>
708 PR tree-optimization/83176
709 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
712 2018-01-24 Richard Biener <rguenther@suse.de>
714 PR tree-optimization/82819
715 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
716 code generating pluses that are no-ops in the target precision.
718 2018-01-24 Richard Biener <rguenther@suse.de>
721 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
723 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
725 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
726 to merge probabilities.
727 * predict.c (probably_never_executed): Also mark as cold functions
728 with global 0 profile and guessed local profile.
729 * profile-count.c (profile_probability::combine_with_count): New
731 * profile-count.h (profile_probability::operator*,
732 profile_probability::operator*=, profile_probability::operator/,
733 profile_probability::operator/=): Reduce precision to adjusted
734 and set value to guessed on contradictory divisions.
735 (profile_probability::combine_with_freq): Remove.
736 (profile_probability::combine_wiht_count): Declare.
737 (profile_count::force_nonzero):: Set to adjusted.
738 (profile_count::probability_in):: Set quality to adjusted.
739 * tree-ssa-tail-merge.c (replace_block_by): Use
742 2018-01-23 Andrew Waterman <andrew@sifive.com>
743 Jim Wilson <jimw@sifive.com>
745 * config/riscv/riscv.c (riscv_stack_boundary): New.
746 (riscv_option_override): Set riscv_stack_boundary. Handle
747 riscv_preferred_stack_boundary_arg.
748 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
749 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
750 (STACK_BOUNDARY): Set to riscv_stack_boundary.
751 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
752 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
753 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
755 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
758 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
759 of struct ix86_frame.
760 (ix86_expand_epilogue): Likewise. Add a local variable for
761 the reg_save_offset field in struct ix86_frame.
763 2018-01-23 Bin Cheng <bin.cheng@arm.com>
765 PR tree-optimization/82604
766 * tree-loop-distribution.c (enum partition_kind): New enum item
767 PKIND_PARTIAL_MEMSET.
768 (partition_builtin_p): Support above new enum item.
769 (generate_code_for_partition): Ditto.
770 (compute_access_range): Differentiate cases that equality can be
771 proven at all loops, the innermost loops or no loops.
772 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
773 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
774 (finalize_partitions, distribute_loop): Don't fuse partition of
775 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
776 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
779 2018-01-23 Martin Liska <mliska@suse.cz>
781 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
782 order to ignore the predictor.
783 (PRED_POLYMORPHIC_CALL): Likewise.
784 (PRED_RECURSIVE_CALL): Likewise.
786 2018-01-23 Martin Liska <mliska@suse.cz>
788 * tree-profile.c (tree_profiling): Print function header to
789 aware reader which function we are working on.
790 * value-prof.c (gimple_find_values_to_profile): Do not print
791 not interesting value histograms.
793 2018-01-23 Martin Liska <mliska@suse.cz>
795 * profile-count.h (enum profile_quality): Add
796 profile_uninitialized as the first value. Do not number values
797 as they are zero based.
798 (profile_count::verify): Update sanity check.
799 (profile_probability::verify): Likewise.
801 2018-01-23 Nathan Sidwell <nathan@acm.org>
803 * doc/invoke.texi (ffor-scope): Deprecate.
805 2018-01-23 David Malcolm <dmalcolm@redhat.com>
807 PR tree-optimization/83510
808 * domwalk.c (set_all_edges_as_executable): New function.
809 (dom_walker::dom_walker): Convert bool param
810 "skip_unreachable_blocks" to enum reachability. Move setup of
811 edge flags to set_all_edges_as_executable and only do it when
812 reachability is REACHABLE_BLOCKS.
813 * domwalk.h (enum dom_walker::reachability): New enum.
814 (dom_walker::dom_walker): Convert bool param
815 "skip_unreachable_blocks" to enum reachability.
816 (set_all_edges_as_executable): New decl.
817 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
818 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
820 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
821 but converting true to REACHABLE_BLOCKS.
822 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
824 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
825 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
826 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
828 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
829 if check_all_array_refs will be called.
831 2018-01-23 David Malcolm <dmalcolm@redhat.com>
833 * tree.c (selftest::test_location_wrappers): Add more test
836 2018-01-23 David Malcolm <dmalcolm@redhat.com>
838 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
839 (selftest::test_bit_in_range): Likewise.
841 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
844 * doc/sourcebuild.texi (vect_float): Say that the selector
845 only describes the situation when -funsafe-math-optimizations is on.
846 (vect_float_strict): Document.
848 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
850 PR tree-optimization/83965
851 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
852 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
853 instead of checking only for a reduction.
854 (vect_recog_widen_sum_pattern): Likewise.
856 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
858 * predict.c (probably_never_executed): Only use precise profile info.
859 (compute_function_frequency): Skip after inlining hack since we now
860 have quality checking.
862 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
864 * profile-count.h (profile_probability::very_unlikely,
865 profile_probability::unlikely, profile_probability::even): Set
866 precision to guessed.
868 2018-01-23 Richard Biener <rguenther@suse.de>
870 PR tree-optimization/83963
871 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
872 Properly terminate dominator walk when crossing the exit edge not
873 when visiting its source block.
875 2018-01-23 Jakub Jelinek <jakub@redhat.com>
878 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
879 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
881 2018-01-22 Jakub Jelinek <jakub@redhat.com>
883 PR tree-optimization/83957
884 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
885 semicolon after for body surrounded by braces.
887 PR tree-optimization/83081
888 * profile-count.h (profile_probability::split): New method.
889 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
890 Use profile_probability::split.
891 (do_compare_rtx_and_jump): Fix adjustment of probabilities
892 when splitting a single conditional jump into 2.
894 2018-01-22 David Malcolm <dmalcolm@redhat.com>
896 PR tree-optimization/69452
897 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
900 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
902 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
903 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
904 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
906 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
908 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
909 * config/rl78/rl78.md: New define_expand "movdi"
910 * config/rl78/rl78.c: New function definition rl78_split_movdi
912 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
915 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
917 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
918 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
919 128-bit to produce an UNSPEC move to get the double word with the
920 signbit and then a shift directly to do signbit.
921 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
922 implementation with a new version that just does either a direct
923 move or a regular move. Move memory interface to separate insns.
924 Move insns so they are next to the expander.
925 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
926 with signbit move. Split big and little endian case.
927 (signbit<mode>2_dm_mem_le): Likewise.
928 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
929 (signbit<mode>2_dm2): Likewise.
931 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
933 * config/rl78/rl78.md: New define_expand "anddi3".
935 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
937 * config/rl78/rl78.md: New define_expand "umindi3".
939 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
941 * config/rl78/rl78.md: New define_expand "smindi3".
943 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
945 * config/rl78/rl78.md: New define_expand "smaxdi3".
947 2018-01-22 Carl Love <cel@us.ibm.com>
949 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
950 LVX_V1TI): Add macro expansion.
951 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
952 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
953 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
954 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
955 Change check to determine if the instruction is a byte reversing
956 entry. Fix typo in comment.
957 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
958 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
959 Add def_builtin calls for new builtins.
960 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
961 Add define_insn expansion.
963 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
965 * config/rl78/rl78.md: New define_expand "umaxdi3".
967 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
969 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
970 for non-QImode registers
972 2018-01-22 Richard Biener <rguenther@suse.de>
974 PR tree-optimization/83963
975 * graphite-scop-detection.c (scop_detection::get_sese): Delay
976 including the loop exit block.
977 (scop_detection::merge_sese): Likewise.
978 (scop_detection::add_scop): Do it here instead.
980 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
982 * doc/sourcebuild.texi (arm_softfloat): Document.
984 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
987 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
988 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
989 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
991 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
992 David Edelsohn <dje.gcc@gmail.com>
995 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
996 Change "crset eq" to "crset 2".
997 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
998 (*call_indirect_aix<mode>_nospec): Likewise.
999 (*call_value_indirect_aix<mode>_nospec): Likewise.
1000 (*call_indirect_elfv2<mode>_nospec): Likewise.
1001 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1002 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1003 change assembly output from . to $.
1004 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1005 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1006 (*tablejump<mode>_internal1_nospec): Likewise.
1008 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1011 * config/sh/sh_optimize_sett_clrt.cc:
1012 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1014 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1016 PR tree-optimization/83940
1017 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1018 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1019 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1020 use it to pass back the definition type.
1021 (vect_check_store_rhs): Likewise rhs_dt_out.
1022 (vect_build_gather_load_calls): Add a mask_dt argument and use
1023 it instead of a call to vect_is_simple_use.
1024 (vectorizable_store): Update calls to vect_check_load_store_mask
1025 and vect_check_store_rhs. Use the dt returned by the latter instead
1026 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1027 instead of calls to vect_is_simple_use. Pass the scalar rather
1028 than the vector operand to vect_is_simple_use when handling
1029 second and subsequent copies of an rhs value.
1030 (vectorizable_load): Update calls to vect_check_load_store_mask
1031 and vect_build_gather_load_calls. Use the cached mask_dt and
1032 gs_info.offset_dt instead of calls to vect_is_simple_use.
1034 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1037 * tree-emutls.c: Include gimplify.h.
1038 (lower_emutls_2): New function.
1039 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1040 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1041 it before further processing.
1044 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1045 UINTVAL (trueop1) instead of INTVAL (op1).
1047 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1051 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1052 INCOMING_FRAME_SP_OFFSET if not defined.
1053 (scan_trace): Add ENTRY argument. If true and
1054 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1055 emit a note to adjust the CFA offset.
1056 (create_cfi_notes): Adjust scan_trace callers.
1057 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1058 INCOMING_FRAME_SP_OFFSET in the CIE.
1059 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1060 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1062 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1063 * doc/tm.texi: Regenerated.
1065 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1067 PR rtl-optimization/83147
1068 * lra-constraints.c (remove_inheritance_pseudos): Use
1069 lra_substitute_pseudo_within_insn.
1071 2018-01-19 Tom de Vries <tom@codesourcery.com>
1072 Cesar Philippidis <cesar@codesourcery.com>
1075 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1077 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1080 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1081 spaces for function labels.
1083 2018-01-19 Martin Liska <mliska@suse.cz>
1085 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1086 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1087 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1088 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1089 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1090 (PRED_CONST_RETURN): Change from 69 to 65.
1091 (PRED_NULL_RETURN): Change from 91 to 71.
1092 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1093 (PRED_LOOP_GUARD): Change from 66 to 73.
1095 2018-01-19 Martin Liska <mliska@suse.cz>
1097 * predict.c (predict_insn_def): Add new assert.
1098 (struct branch_predictor): Change type to signed integer.
1099 (test_prediction_value_range): Amend test to cover
1101 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1102 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1103 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1104 (PRED_LOOP_IV_COMPARE): Likewise.
1105 * predict.h (PROB_UNINITIALIZED): Define new constant.
1107 2018-01-19 Martin Liska <mliska@suse.cz>
1109 * predict.c (dump_prediction): Add new format for
1110 analyze_brprob.py script which is enabled with -details
1112 * profile-count.h (precise_p): New function.
1114 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1116 PR tree-optimization/83922
1117 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1118 there are no statements that need masking.
1119 (vect_active_double_reduction_p): New function.
1120 (vect_analyze_loop_operations): Use it when handling phis that
1121 are not in the loop header.
1123 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1125 PR tree-optimization/83914
1126 * tree-vect-loop.c (vectorizable_induction): Don't convert
1127 init_expr or apply the peeling adjustment for inductions
1128 that are nested within the vectorized loop.
1130 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1132 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1135 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1139 * function.h (gimplify_parameters): Add gimple_seq * argument.
1140 * function.c: Include gimple.h and options.h.
1141 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1142 for the added local temporaries if needed.
1143 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1144 if there are any parameter cleanups, wrap whole body into a
1145 try/finally with the cleanups.
1147 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1150 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1151 Use GET_MODE_CLASS for scalar floating point.
1153 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1157 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1158 Fix call of call_cgraph_insertion_hooks.
1160 2018-01-18 Martin Sebor <msebor@redhat.com>
1162 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1164 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1167 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1170 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1173 * common.opt: (-ffile-prefix-map): New option.
1174 * opts.c (common_handle_option): Defer it.
1175 * opts-global.c (handle_common_deferred_options): Handle it.
1176 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1177 * file-prefix-map.h: New file.
1178 (remap_debug_filename, add_debug_prefix_map): ...here.
1179 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1180 * final.c (debug_prefix_map, add_debug_prefix_map
1181 remap_debug_filename): Move to...
1182 * file-prefix-map.c: New file.
1183 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1184 generalize, get rid of alloca(), use strrchr() instead of strchr().
1185 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1186 Implement in terms of add_prefix_map().
1187 (remap_macro_filename, remap_debug_filename): Implement in term of
1189 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1190 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1191 * dbxout.c: Include file-prefix-map.h.
1192 * varasm.c: Likewise.
1193 * vmsdbgout.c: Likewise.
1194 * xcoffout.c: Likewise.
1195 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1196 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1197 * doc/invoke.texi (-ffile-prefix-map): Document.
1198 (-fdebug-prefix-map): Update description.
1200 2018-01-18 Martin Liska <mliska@suse.cz>
1202 * config/i386/i386.c (indirect_thunk_name): Document that also
1204 (output_indirect_thunk): Document why both instructions
1205 (pause and lfence) are generated.
1207 2018-01-18 Richard Biener <rguenther@suse.de>
1209 PR tree-optimization/83887
1210 * graphite-scop-detection.c
1211 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1212 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1213 (scop_detection::merge_sese): Re-implement with a flood-fill
1214 algorithm that properly finds a SESE region if it exists.
1216 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1219 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1220 pointer_diff optimizations use view_convert instead of convert.
1222 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1224 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1225 Generate different code for -mno-speculate-indirect-jumps.
1226 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1227 (*call_indirect_aix<mode>): Disable for
1228 -mno-speculate-indirect-jumps.
1229 (*call_indirect_aix<mode>_nospec): New define_insn.
1230 (*call_value_indirect_aix<mode>): Disable for
1231 -mno-speculate-indirect-jumps.
1232 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1233 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1234 -mno-speculate-indirect-jumps.
1235 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1237 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1239 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1240 long double type, set the flags for noting the default long double
1241 type, even if we don't pass or return a long double type.
1243 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1246 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1249 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1251 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1252 support for merge[hl].
1253 (fold_mergehl_helper): New helper function.
1254 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1255 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1256 (altivec_vmrglw_direct): Add xxmrglw insn.
1258 2018-01-17 Andrew Waterman <andrew@sifive.com>
1260 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1261 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1263 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1266 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1267 call the lto_location_cache before reading the
1268 DECL_SOURCE_LOCATION of the types.
1270 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1271 Richard Sandiford <richard.sandiford@linaro.org>
1273 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1274 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1275 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1276 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1277 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1279 * config/aarch64/constraints.md (aarch64_movti_operand):
1281 * config/aarch64/predicates.md (Uti): Add new constraint.
1283 2018-01-17 Carl Love <cel@us.ibm.com>
1284 * config/rs6000/vsx.md (define_expand xl_len_r,
1285 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1286 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1288 (define_expand, define_insn): Move the shift left from the
1289 define_insn to the define_expand for lxvl and stxvl instructions.
1290 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1291 and XL_LEN_R definitions to PURE.
1293 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1295 * config/i386/i386.c (indirect_thunk_name): Declare regno
1296 as unsigned int. Compare regno with INVALID_REGNUM.
1297 (output_indirect_thunk): Ditto.
1298 (output_indirect_thunk_function): Ditto.
1299 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1300 in the call to output_indirect_thunk_function.
1302 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1305 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1306 rather than the size of inner_type to determine the stack slot size
1307 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1309 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1312 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1315 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1317 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1318 endian Linux systems to optionally enable multilibs for selecting
1319 the long double type if the user configured an explicit type.
1320 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1321 have no long double multilibs if not defined.
1322 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1323 warn if the user used -mabi={ieee,ibm}longdouble and we built
1324 multilibs for long double.
1325 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1326 appropriate multilib option.
1327 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1329 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1330 for building long double multilibs.
1331 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1333 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1335 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1338 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1340 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1343 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1346 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1349 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1351 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1352 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1354 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1356 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1357 different rtl trees depending on TARGET_64BIT.
1358 (rs6000_gen_lvx): Likewise.
1360 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1362 * config/visium/visium.md (nop): Tweak comment.
1363 (hazard_nop): Likewise.
1365 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1367 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1368 -mspeculate-indirect-jumps.
1369 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1370 for -mno-speculate-indirect-jumps.
1371 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1372 (*call_value_indirect_elfv2<mode>): Disable for
1373 -mno-speculate-indirect-jumps.
1374 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1375 (indirect_jump): Emit different RTL for
1376 -mno-speculate-indirect-jumps.
1377 (*indirect_jump<mode>): Disable for
1378 -mno-speculate-indirect-jumps.
1379 (*indirect_jump<mode>_nospec): New define_insn.
1380 (tablejump): Emit different RTL for
1381 -mno-speculate-indirect-jumps.
1382 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1383 (tablejumpsi_nospec): New define_expand.
1384 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1385 (tablejumpdi_nospec): New define_expand.
1386 (*tablejump<mode>_internal1): Disable for
1387 -mno-speculate-indirect-jumps.
1388 (*tablejump<mode>_internal1_nospec): New define_insn.
1389 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1392 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1394 * caller-save.c (insert_save): Drop unnecessary parameter. All
1397 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1398 Richard Biener <rguenth@suse.de>
1401 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1402 return early, inline manually is_gimple_sizepos. Make sure if we
1403 call gimplify_expr we don't end up with a gimple constant.
1404 * tree.c (variably_modified_type_p): Don't return true for
1405 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1406 * gimplify.h (is_gimple_sizepos): Remove.
1408 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1410 PR tree-optimization/83857
1411 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1412 vectorizable_live_operation for pure SLP statements.
1413 (vectorizable_live_operation): Handle PHIs.
1415 2018-01-16 Richard Biener <rguenther@suse.de>
1417 PR tree-optimization/83867
1418 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1419 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1421 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1424 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1425 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1426 If off is not INTEGER_CST, issue a may not be aligned warning
1427 rather than isn't aligned. Use isn%'t rather than isn't.
1428 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1430 <case MULT_EXPR>: Improve the case when bottom and one of the
1431 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1432 operand, in that case check if the other operand is multiple of
1433 bottom divided by the INTEGER_CST operand.
1435 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1438 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1439 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1440 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1441 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1442 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1443 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1444 (pa_function_arg_size): New function.
1446 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1448 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1449 in a separate statement.
1451 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1453 PR tree-optimization/83847
1454 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1455 group gathers and scatters.
1457 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1459 PR rtl-optimization/86620
1460 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1462 PR rtl-optimization/83213
1463 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1464 to last if both are JUMP_INSNs.
1466 PR tree-optimization/83843
1467 * gimple-ssa-store-merging.c
1468 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1469 store_immediate_info for bswap/nop orig_stores.
1471 2018-01-15 Andrew Waterman <andrew@sifive.com>
1473 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1475 <UDIV>: Increase cost if !TARGET_DIV.
1477 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1479 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1480 (define_attr "cr_logical_3op"): New.
1481 (cceq_ior_compare): Adjust.
1482 (cceq_ior_compare_complement): Adjust.
1483 (*cceq_rev_compare): Adjust.
1484 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1485 (is_cracked_insn): Adjust.
1486 (insn_must_be_first_in_group): Adjust.
1487 * config/rs6000/40x.md: Adjust.
1488 * config/rs6000/440.md: Adjust.
1489 * config/rs6000/476.md: Adjust.
1490 * config/rs6000/601.md: Adjust.
1491 * config/rs6000/603.md: Adjust.
1492 * config/rs6000/6xx.md: Adjust.
1493 * config/rs6000/7450.md: Adjust.
1494 * config/rs6000/7xx.md: Adjust.
1495 * config/rs6000/8540.md: Adjust.
1496 * config/rs6000/cell.md: Adjust.
1497 * config/rs6000/e300c2c3.md: Adjust.
1498 * config/rs6000/e500mc.md: Adjust.
1499 * config/rs6000/e500mc64.md: Adjust.
1500 * config/rs6000/e5500.md: Adjust.
1501 * config/rs6000/e6500.md: Adjust.
1502 * config/rs6000/mpc.md: Adjust.
1503 * config/rs6000/power4.md: Adjust.
1504 * config/rs6000/power5.md: Adjust.
1505 * config/rs6000/power6.md: Adjust.
1506 * config/rs6000/power7.md: Adjust.
1507 * config/rs6000/power8.md: Adjust.
1508 * config/rs6000/power9.md: Adjust.
1509 * config/rs6000/rs64.md: Adjust.
1510 * config/rs6000/titan.md: Adjust.
1512 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1514 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1515 ix86_indirect_branch_register logic.
1517 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1519 * config/i386/constraints.md (Bs): Update
1520 ix86_indirect_branch_register check. Don't check
1521 ix86_indirect_branch_register with GOT_memory_operand.
1523 * config/i386/predicates.md (GOT_memory_operand): Don't check
1524 ix86_indirect_branch_register here.
1525 (GOT32_symbol_operand): Likewise.
1527 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1529 * config/i386/predicates.md (constant_call_address_operand):
1530 Rewrite ix86_indirect_branch_register logic.
1531 (sibcall_insn_operand): Likewise.
1533 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1535 * config/i386/constraints.md (Bs): Replace
1536 ix86_indirect_branch_thunk_register with
1537 ix86_indirect_branch_register.
1539 * config/i386/i386.md (indirect_jump): Likewise.
1540 (tablejump): Likewise.
1541 (*sibcall_memory): Likewise.
1542 (*sibcall_value_memory): Likewise.
1543 Peepholes of indirect call and jump via memory: Likewise.
1544 * config/i386/i386.opt: Likewise.
1545 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1546 (GOT_memory_operand): Likewise.
1547 (call_insn_operand): Likewise.
1548 (sibcall_insn_operand): Likewise.
1549 (GOT32_symbol_operand): Likewise.
1551 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1554 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1555 type rather than type addr's type points to.
1556 (expand_omp_atomic_mutex): Likewise.
1557 (expand_omp_atomic): Likewise.
1559 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1562 * config/i386/i386.c (output_indirect_thunk_function): Use
1563 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1564 for __x86_return_thunk.
1566 2018-01-15 Richard Biener <rguenther@suse.de>
1569 * expmed.c (extract_bit_field_1): Fix typo.
1571 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1574 * config/arm/iterators.md (VF): New mode iterator.
1575 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1576 Remove integer-related logic from pattern.
1577 (neon_vabd<mode>_3): Likewise.
1579 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1582 * common.opt (fstrict-overflow): No longer an alias.
1583 (fwrapv-pointer): New option.
1584 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1585 also for pointer types based on flag_wrapv_pointer.
1586 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1587 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1588 opts->x_flag_wrapv got set.
1589 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1590 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1591 POINTER_TYPE_OVERFLOW_UNDEFINED.
1592 * match.pd: Likewise in address comparison pattern.
1593 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1595 2018-01-15 Richard Biener <rguenther@suse.de>
1598 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1599 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1600 Reset type names to their identifier if their TYPE_DECL doesn't
1601 have linkage (and thus is used for ODR and devirt).
1602 (save_debug_info_for_decl): Remove.
1603 (save_debug_info_for_type): Likewise.
1604 (add_tree_to_fld_list): Adjust.
1605 * tree-pretty-print.c (dump_generic_node): Make dumping of
1606 type names more robust.
1608 2018-01-15 Richard Biener <rguenther@suse.de>
1610 * BASE-VER: Bump to 8.0.1.
1612 2018-01-14 Martin Sebor <msebor@redhat.com>
1615 * builtins.c (check_access): Avoid warning when the no-warning bit
1618 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1620 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1621 * ira-color (allocno_hard_regs_compare): Likewise.
1623 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1626 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1627 Use .pushsection/.popsection.
1629 2018-01-14 Martin Sebor <msebor@redhat.com>
1632 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1634 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1636 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1637 entry from extra_headers.
1638 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1639 extra_headers, make the list bitwise identical to the i?86-*-* one.
1641 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1643 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1644 -mcmodel=large with -mindirect-branch=thunk,
1645 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1646 -mfunction-return=thunk-extern.
1647 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1648 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1649 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1651 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1653 * config/i386/i386.c (print_reg): Print the name of the full
1654 integer register without '%'.
1655 (ix86_print_operand): Handle 'V'.
1656 * doc/extend.texi: Document 'V' modifier.
1658 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1660 * config/i386/constraints.md (Bs): Disallow memory operand for
1661 -mindirect-branch-register.
1663 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1664 (GOT_memory_operand): Likewise.
1665 (call_insn_operand): Likewise.
1666 (sibcall_insn_operand): Likewise.
1667 (GOT32_symbol_operand): Likewise.
1668 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1669 for -mindirect-branch-register.
1670 (tablejump): Likewise.
1671 (*sibcall_memory): Likewise.
1672 (*sibcall_value_memory): Likewise.
1673 Disallow peepholes of indirect call and jump via memory for
1674 -mindirect-branch-register.
1675 (*call_pop): Replace m with Bw.
1676 (*call_value_pop): Likewise.
1677 (*sibcall_pop_memory): Replace m with Bs.
1678 * config/i386/i386.opt (mindirect-branch-register): New option.
1679 * doc/invoke.texi: Document -mindirect-branch-register option.
1681 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1683 * config/i386/i386-protos.h (ix86_output_function_return): New.
1684 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1685 set function_return_type.
1686 (indirect_thunk_name): Add ret_p to indicate thunk for function
1688 (output_indirect_thunk_function): Pass false to
1689 indirect_thunk_name.
1690 (ix86_output_indirect_branch_via_reg): Likewise.
1691 (ix86_output_indirect_branch_via_push): Likewise.
1692 (output_indirect_thunk_function): Create alias for function
1693 return thunk if regno < 0.
1694 (ix86_output_function_return): New function.
1695 (ix86_handle_fndecl_attribute): Handle function_return.
1696 (ix86_attribute_table): Add function_return.
1697 * config/i386/i386.h (machine_function): Add
1698 function_return_type.
1699 * config/i386/i386.md (simple_return_internal): Use
1700 ix86_output_function_return.
1701 (simple_return_internal_long): Likewise.
1702 * config/i386/i386.opt (mfunction-return=): New option.
1703 (indirect_branch): Mention -mfunction-return=.
1704 * doc/extend.texi: Document function_return function attribute.
1705 * doc/invoke.texi: Document -mfunction-return= option.
1707 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1709 * config/i386/i386-opts.h (indirect_branch): New.
1710 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1711 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1712 with local indirect jump when converting indirect call and jump.
1713 (ix86_set_indirect_branch_type): New.
1714 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1715 (indirectlabelno): New.
1716 (indirect_thunk_needed): Likewise.
1717 (indirect_thunk_bnd_needed): Likewise.
1718 (indirect_thunks_used): Likewise.
1719 (indirect_thunks_bnd_used): Likewise.
1720 (INDIRECT_LABEL): Likewise.
1721 (indirect_thunk_name): Likewise.
1722 (output_indirect_thunk): Likewise.
1723 (output_indirect_thunk_function): Likewise.
1724 (ix86_output_indirect_branch_via_reg): Likewise.
1725 (ix86_output_indirect_branch_via_push): Likewise.
1726 (ix86_output_indirect_branch): Likewise.
1727 (ix86_output_indirect_jmp): Likewise.
1728 (ix86_code_end): Call output_indirect_thunk_function if needed.
1729 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1731 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1732 (ix86_attribute_table): Add indirect_branch.
1733 * config/i386/i386.h (machine_function): Add indirect_branch_type
1734 and has_local_indirect_jump.
1735 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1737 (tablejump): Likewise.
1738 (*indirect_jump): Use ix86_output_indirect_jmp.
1739 (*tablejump_1): Likewise.
1740 (simple_return_indirect_internal): Likewise.
1741 * config/i386/i386.opt (mindirect-branch=): New option.
1742 (indirect_branch): New.
1745 (thunk-inline): Likewise.
1746 (thunk-extern): Likewise.
1747 * doc/extend.texi: Document indirect_branch function attribute.
1748 * doc/invoke.texi: Document -mindirect-branch= option.
1750 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1753 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1755 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1757 * ipa-inline.c (want_inline_small_function_p): Return false if
1758 inlining has already failed with CIF_FINAL_ERROR.
1759 (update_caller_keys): Call want_inline_small_function_p before
1761 (update_callee_keys): Likewise.
1763 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1765 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1767 (rs6000_quadword_masked_address_p): Likewise.
1768 (quad_aligned_load_p): Likewise.
1769 (quad_aligned_store_p): Likewise.
1770 (const_load_sequence_p): Add comment to describe the outer-most loop.
1771 (mimic_memory_attributes_and_flags): New function.
1772 (rs6000_gen_stvx): Likewise.
1773 (replace_swapped_aligned_store): Likewise.
1774 (rs6000_gen_lvx): Likewise.
1775 (replace_swapped_aligned_load): Likewise.
1776 (replace_swapped_load_constant): Capitalize argument name in
1777 comment describing this function.
1778 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1779 and stores that access quad-word aligned addresses and replace
1780 with stvx or lvx instructions when appropriate.
1781 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1782 New function prototype.
1783 (rs6000_quadword_masked_address_p): Likewise.
1784 (rs6000_gen_lvx): Likewise.
1785 (rs6000_gen_stvx): Likewise.
1786 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1787 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1788 when memory address is aligned.
1789 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1790 this split to select lvx instruction when memory address is aligned.
1791 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1792 instruction when memory address is aligned.
1793 (*vsx_le_perm_load_v16qi): Likewise.
1794 (four unnamed splitters): Modify to select the stvx instruction
1795 when memory is aligned.
1797 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1799 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1800 which appears in the queue multiple times.
1802 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1803 Alan Hayward <alan.hayward@arm.com>
1804 David Sherwood <david.sherwood@arm.com>
1806 * tree-vectorizer.h (vec_lower_bound): New structure.
1807 (_loop_vec_info): Add check_nonzero and lower_bounds.
1808 (LOOP_VINFO_CHECK_NONZERO): New macro.
1809 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1810 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1811 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1812 fields. Make seg_len the distance travelled, not including the
1814 (dr_direction_indicator): Declare.
1815 (dr_zero_step_indicator): Likewise.
1816 (dr_known_forward_stride_p): Likewise.
1817 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1819 (runtime_alias_check_p): Allow runtime alias checks with
1821 (operator ==): Compare access_size and align.
1822 (prune_runtime_alias_test_list): Rework for new distinction between
1823 the access_size and seg_len.
1824 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1826 (get_segment_min_max): New function.
1827 (create_intersect_range_checks): Use it.
1828 (dr_step_indicator): New function.
1829 (dr_direction_indicator): Likewise.
1830 (dr_zero_step_indicator): Likewise.
1831 (dr_known_forward_stride_p): Likewise.
1832 * tree-loop-distribution.c (data_ref_segment_size): Return
1833 DR_STEP * (niters - 1).
1834 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1836 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1837 (vect_preserves_scalar_order_p): New function, split out from...
1838 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1839 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1840 (vect_vfa_access_size): New function.
1841 (vect_vfa_align): Likewise.
1842 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1843 (dump_lower_bound): New function.
1844 (vect_check_lower_bound): Likewise.
1845 (vect_small_gap_p): Likewise.
1846 (vectorizable_with_step_bound_p): Likewise.
1847 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1848 depencies if the vectorization factor is 1. Convert the checks
1849 for nonzero steps into checks on the bounds of DR_STEP. Try using
1850 a bunds check for variable steps if the minimum required step is
1851 relatively small. Update calls to the dr_with_seg_len
1852 constructor and to vect_compile_time_alias.
1853 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1855 (vect_loop_versioning): Call it.
1856 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1858 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1860 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1861 Alan Hayward <alan.hayward@arm.com>
1862 David Sherwood <david.sherwood@arm.com>
1864 * doc/sourcebuild.texi (vect_scatter_store): Document.
1865 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1867 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1869 * genopinit.c (main): Add supports_vec_scatter_store and
1870 supports_vec_scatter_store_cached to target_optabs.
1871 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1872 IFN_MASK_SCATTER_STORE.
1873 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1875 * internal-fn.h (internal_store_fn_p): Declare.
1876 (internal_fn_stored_value_index): Likewise.
1877 * internal-fn.c (scatter_store_direct): New macro.
1878 (expand_scatter_store_optab_fn): New function.
1879 (direct_scatter_store_optab_supported_p): New macro.
1880 (internal_store_fn_p): New function.
1881 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1882 IFN_MASK_SCATTER_STORE.
1883 (internal_fn_mask_index): Likewise.
1884 (internal_fn_stored_value_index): New function.
1885 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1887 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1888 * optabs-query.c (supports_vec_scatter_store_p): New function.
1889 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1890 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1891 true for scatter stores.
1892 (vect_gather_scatter_fn_p): Handle scatter stores too.
1893 (vect_check_gather_scatter): Consider using scatter stores if
1894 supports_vec_scatter_store_p.
1895 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1897 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1898 internal_fn_stored_value_index.
1899 (check_load_store_masking): Handle scatter stores too.
1900 (vect_get_store_rhs): Make public.
1901 (vectorizable_call): Use internal_store_fn_p.
1902 (vectorizable_store): Handle scatter store internal functions.
1903 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1904 when deciding whether the end of the group has been reached.
1905 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1906 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1907 (mask_scatter_store<mode>): New insns.
1909 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1910 Alan Hayward <alan.hayward@arm.com>
1911 David Sherwood <david.sherwood@arm.com>
1913 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1914 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1915 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1917 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1918 Use vect_truncate_gather_scatter_offset if we can't treat the
1919 operation as a normal gather load or scatter store.
1920 (get_group_load_store_type): Take the gather_scatter_info
1921 as argument. Try using a gather load or scatter store for
1922 single-element groups.
1923 (get_load_store_type): Update calls to get_group_load_store_type
1924 and vect_use_strided_gather_scatters_p.
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-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1931 optional tree argument.
1932 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1934 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1935 but continue to use the current value as a fallback.
1936 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1937 to compare the updates.
1938 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1939 (get_load_store_type): Use it when handling a strided access.
1940 (vect_get_strided_load_store_ops): New function.
1941 (vect_get_data_ptr_increment): Likewise.
1942 (vectorizable_load): Handle strided gather loads. Always pass
1943 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1945 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1946 Alan Hayward <alan.hayward@arm.com>
1947 David Sherwood <david.sherwood@arm.com>
1949 * doc/md.texi (gather_load@var{m}): Document.
1950 (mask_gather_load@var{m}): Likewise.
1951 * genopinit.c (main): Add supports_vec_gather_load and
1952 supports_vec_gather_load_cached to target_optabs.
1953 * optabs-tree.c (init_tree_optimization_optabs): Use
1954 ggc_cleared_alloc to allocate target_optabs.
1955 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1956 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1958 * internal-fn.h (internal_load_fn_p): Declare.
1959 (internal_gather_scatter_fn_p): Likewise.
1960 (internal_fn_mask_index): Likewise.
1961 (internal_gather_scatter_fn_supported_p): Likewise.
1962 * internal-fn.c (gather_load_direct): New macro.
1963 (expand_gather_load_optab_fn): New function.
1964 (direct_gather_load_optab_supported_p): New macro.
1965 (direct_internal_fn_optab): New function.
1966 (internal_load_fn_p): Likewise.
1967 (internal_gather_scatter_fn_p): Likewise.
1968 (internal_fn_mask_index): Likewise.
1969 (internal_gather_scatter_fn_supported_p): Likewise.
1970 * optabs-query.c (supports_at_least_one_mode_p): New function.
1971 (supports_vec_gather_load_p): Likewise.
1972 * optabs-query.h (supports_vec_gather_load_p): Declare.
1973 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1974 and memory_type field.
1975 (NUM_PATTERNS): Bump to 15.
1976 * tree-vect-data-refs.c: Include internal-fn.h.
1977 (vect_gather_scatter_fn_p): New function.
1978 (vect_describe_gather_scatter_call): Likewise.
1979 (vect_check_gather_scatter): Try using internal functions for
1980 gather loads. Recognize existing calls to a gather load function.
1981 (vect_analyze_data_refs): Consider using gather loads if
1982 supports_vec_gather_load_p.
1983 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1984 (vect_get_gather_scatter_offset_type): Likewise.
1985 (vect_convert_mask_for_vectype): Likewise.
1986 (vect_add_conversion_to_patterm): Likewise.
1987 (vect_try_gather_scatter_pattern): Likewise.
1988 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1989 (vect_vect_recog_func_ptrs): Add it.
1990 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1991 internal_fn_mask_index and internal_gather_scatter_fn_p.
1992 (check_load_store_masking): Take the gather_scatter_info as an
1993 argument and handle gather loads.
1994 (vect_get_gather_scatter_ops): New function.
1995 (vectorizable_call): Check internal_load_fn_p.
1996 (vectorizable_load): Likewise. Handle gather load internal
1998 (vectorizable_store): Update call to check_load_store_masking.
1999 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2000 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2001 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2002 (aarch64_gather_scale_operand_d): New predicates.
2003 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2004 (mask_gather_load<mode>): New insns.
2006 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2007 Alan Hayward <alan.hayward@arm.com>
2008 David Sherwood <david.sherwood@arm.com>
2010 * optabs.def (fold_left_plus_optab): New optab.
2011 * doc/md.texi (fold_left_plus_@var{m}): Document.
2012 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2013 * internal-fn.c (fold_left_direct): Define.
2014 (expand_fold_left_optab_fn): Likewise.
2015 (direct_fold_left_optab_supported_p): Likewise.
2016 * fold-const-call.c (fold_const_fold_left): New function.
2017 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2018 * tree-parloops.c (valid_reduction_p): New function.
2019 (gather_scalar_reductions): Use it.
2020 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2021 (vect_finish_replace_stmt): Declare.
2022 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2023 (needs_fold_left_reduction_p): New function, split out from...
2024 (vect_is_simple_reduction): ...here. Accept reductions that
2025 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2026 (vect_force_simple_reduction): Also store the reduction type in
2027 the assignment's STMT_VINFO_REDUC_TYPE.
2028 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2029 (merge_with_identity): New function.
2030 (vect_expand_fold_left): Likewise.
2031 (vectorize_fold_left_reduction): Likewise.
2032 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2033 scalar phi in place for it. Check for target support and reject
2034 cases that would reassociate the operation. Defer the transform
2035 phase to vectorize_fold_left_reduction.
2036 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2037 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2038 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2040 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2042 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2043 call to ifc_temp_var.
2045 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2046 Alan Hayward <alan.hayward@arm.com>
2047 David Sherwood <david.sherwood@arm.com>
2049 * target.def (legitimize_address_displacement): Take the original
2050 offset as a poly_int.
2051 * targhooks.h (default_legitimize_address_displacement): Update
2053 * targhooks.c (default_legitimize_address_displacement): Likewise.
2054 * doc/tm.texi: Regenerate.
2055 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2056 as an argument, moving assert of ad->disp == ad->disp_term to...
2057 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2058 Try calling targetm.legitimize_address_displacement before expanding
2059 the address rather than afterwards, and adjust for the new interface.
2060 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2061 Match the new hook interface. Handle SVE addresses.
2062 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2065 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2067 * Makefile.in (OBJS): Add early-remat.o.
2068 * target.def (select_early_remat_modes): New hook.
2069 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2070 * doc/tm.texi: Regenerate.
2071 * targhooks.h (default_select_early_remat_modes): Declare.
2072 * targhooks.c (default_select_early_remat_modes): New function.
2073 * timevar.def (TV_EARLY_REMAT): New timevar.
2074 * passes.def (pass_early_remat): New pass.
2075 * tree-pass.h (make_pass_early_remat): Declare.
2076 * early-remat.c: New file.
2077 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2079 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2081 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2082 Alan Hayward <alan.hayward@arm.com>
2083 David Sherwood <david.sherwood@arm.com>
2085 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2086 vfm1 with a bound_epilog parameter.
2087 (vect_do_peeling): Update calls accordingly, and move the prologue
2088 call earlier in the function. Treat the base bound_epilog as 0 for
2089 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2090 this base when peeling for gaps.
2091 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2092 with fully-masked loops.
2093 (vect_estimate_min_profitable_iters): Handle the single peeled
2094 iteration in that case.
2096 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2097 Alan Hayward <alan.hayward@arm.com>
2098 David Sherwood <david.sherwood@arm.com>
2100 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2101 single-element interleaving even if the size is not a power of 2.
2102 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2103 accesses for single-element interleaving if the group size is
2106 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2107 Alan Hayward <alan.hayward@arm.com>
2108 David Sherwood <david.sherwood@arm.com>
2110 * doc/md.texi (fold_extract_last_@var{m}): Document.
2111 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2112 * optabs.def (fold_extract_last_optab): New optab.
2113 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2114 * internal-fn.c (fold_extract_direct): New macro.
2115 (expand_fold_extract_optab_fn): Likewise.
2116 (direct_fold_extract_optab_supported_p): Likewise.
2117 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2118 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2119 EXTRACT_LAST_REDUCTION.
2120 (get_initial_def_for_reduction): Do not create an initial vector
2121 for EXTRACT_LAST_REDUCTION reductions.
2122 (vectorizable_reduction): Leave the scalar phi in place for
2123 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2124 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2125 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2126 transform phase to vectorizable_condition.
2127 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2129 (vect_finish_stmt_generation): ...here.
2130 (vect_finish_replace_stmt): New function.
2131 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2132 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2134 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2136 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2137 Alan Hayward <alan.hayward@arm.com>
2138 David Sherwood <david.sherwood@arm.com>
2140 * doc/md.texi (extract_last_@var{m}): Document.
2141 * optabs.def (extract_last_optab): New optab.
2142 * internal-fn.def (EXTRACT_LAST): New internal function.
2143 * internal-fn.c (cond_unary_direct): New macro.
2144 (expand_cond_unary_optab_fn): Likewise.
2145 (direct_cond_unary_optab_supported_p): Likewise.
2146 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2147 loops using EXTRACT_LAST.
2148 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2149 (extract_last_<mode>): ...this optab.
2150 (vec_extract<mode><Vel>): Update accordingly.
2152 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2153 Alan Hayward <alan.hayward@arm.com>
2154 David Sherwood <david.sherwood@arm.com>
2156 * target.def (empty_mask_is_expensive): New hook.
2157 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2158 * doc/tm.texi: Regenerate.
2159 * targhooks.h (default_empty_mask_is_expensive): Declare.
2160 * targhooks.c (default_empty_mask_is_expensive): New function.
2161 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2162 if the target says that empty masks are expensive.
2163 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2165 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2167 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2168 Alan Hayward <alan.hayward@arm.com>
2169 David Sherwood <david.sherwood@arm.com>
2171 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2172 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2173 (vect_use_loop_mask_for_alignment_p): New function.
2174 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2175 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2176 niters_skip argument. Make sure that the first niters_skip elements
2177 of the first iteration are inactive.
2178 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2179 Update call to vect_set_loop_masks_directly.
2180 (get_misalign_in_elems): New function, split out from...
2181 (vect_gen_prolog_loop_niters): ...here.
2182 (vect_update_init_of_dr): Take a code argument that specifies whether
2183 the adjustment should be added or subtracted.
2184 (vect_update_init_of_drs): Likewise.
2185 (vect_prepare_for_masked_peels): New function.
2186 (vect_do_peeling): Skip prologue peeling if we're using a mask
2187 instead. Update call to vect_update_inits_of_drs.
2188 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2190 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2191 alignment. Do not include the number of peeled iterations in
2192 the minimum threshold in that case.
2193 (vectorizable_induction): Adjust the start value down by
2194 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2195 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2196 Take the number of skipped iterations into account when calculating
2198 * tree-vect-stmts.c (vect_gen_while_not): New function.
2200 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2201 Alan Hayward <alan.hayward@arm.com>
2202 David Sherwood <david.sherwood@arm.com>
2204 * doc/sourcebuild.texi (vect_fully_masked): Document.
2205 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2207 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2209 (vect_analyze_loop_2): ...here. Don't check the vectorization
2210 factor against the number of loop iterations if the loop is
2213 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2214 Alan Hayward <alan.hayward@arm.com>
2215 David Sherwood <david.sherwood@arm.com>
2217 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2218 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2219 (dump_groups): Update accordingly.
2220 (iv_use::mem_type): New member variable.
2221 (address_p): New function.
2222 (record_use): Add a mem_type argument and initialize the new
2224 (record_group_use): Add a mem_type argument. Use address_p.
2225 Remove obsolete null checks of base_object. Update call to record_use.
2226 (find_interesting_uses_op): Update call to record_group_use.
2227 (find_interesting_uses_cond): Likewise.
2228 (find_interesting_uses_address): Likewise.
2229 (get_mem_type_for_internal_fn): New function.
2230 (find_address_like_use): Likewise.
2231 (find_interesting_uses_stmt): Try find_address_like_use before
2232 calling find_interesting_uses_op.
2233 (addr_offset_valid_p): Use the iv mem_type field as the type
2234 of the addressed memory.
2235 (add_autoinc_candidates): Likewise.
2236 (get_address_cost): Likewise.
2237 (split_small_address_groups_p): Use address_p.
2238 (split_address_groups): Likewise.
2239 (add_iv_candidate_for_use): Likewise.
2240 (autoinc_possible_for_pair): Likewise.
2241 (rewrite_groups): Likewise.
2242 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2243 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2244 (get_alias_ptr_type_for_ptr_address): New function.
2245 (rewrite_use_address): Rewrite address uses in calls that were
2246 identified by find_address_like_use.
2248 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2249 Alan Hayward <alan.hayward@arm.com>
2250 David Sherwood <david.sherwood@arm.com>
2252 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2254 * gimple-expr.h (is_gimple_addressable: Likewise.
2255 * gimple-expr.c (is_gimple_address): Likewise.
2256 * internal-fn.c (expand_call_mem_ref): New function.
2257 (expand_mask_load_optab_fn): Use it.
2258 (expand_mask_store_optab_fn): Likewise.
2260 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2261 Alan Hayward <alan.hayward@arm.com>
2262 David Sherwood <david.sherwood@arm.com>
2264 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2265 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2266 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2267 (cond_umax@var{mode}): Document.
2268 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2269 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2270 (cond_umin_optab, cond_umax_optab): New optabs.
2271 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2272 (COND_IOR, COND_XOR): New internal functions.
2273 * internal-fn.h (get_conditional_internal_fn): Declare.
2274 * internal-fn.c (cond_binary_direct): New macro.
2275 (expand_cond_binary_optab_fn): Likewise.
2276 (direct_cond_binary_optab_supported_p): Likewise.
2277 (get_conditional_internal_fn): New function.
2278 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2279 Cope with reduction statements that are vectorized as calls rather
2281 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2282 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2283 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2284 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2285 (UNSPEC_COND_EOR): New unspecs.
2286 (optab): Add mappings for them.
2287 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2288 (sve_int_op, sve_fp_op): New int attributes.
2290 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2291 Alan Hayward <alan.hayward@arm.com>
2292 David Sherwood <david.sherwood@arm.com>
2294 * optabs.def (while_ult_optab): New optab.
2295 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2296 * internal-fn.def (WHILE_ULT): New internal function.
2297 * internal-fn.h (direct_internal_fn_supported_p): New override
2298 that takes two types as argument.
2299 * internal-fn.c (while_direct): New macro.
2300 (expand_while_optab_fn): New function.
2301 (convert_optab_supported_p): Likewise.
2302 (direct_while_optab_supported_p): New macro.
2303 * wide-int.h (wi::udiv_ceil): New function.
2304 * tree-vectorizer.h (rgroup_masks): New structure.
2305 (vec_loop_masks): New typedef.
2306 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2308 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2309 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2310 (vect_max_vf): New function.
2311 (slpeel_make_loop_iterate_ntimes): Delete.
2312 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2313 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2314 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2315 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2316 internal-fn.h, stor-layout.h and optabs-query.h.
2317 (vect_set_loop_mask): New function.
2318 (add_preheader_seq): Likewise.
2319 (add_header_seq): Likewise.
2320 (interleave_supported_p): Likewise.
2321 (vect_maybe_permute_loop_masks): Likewise.
2322 (vect_set_loop_masks_directly): Likewise.
2323 (vect_set_loop_condition_masked): Likewise.
2324 (vect_set_loop_condition_unmasked): New function, split out from
2325 slpeel_make_loop_iterate_ntimes.
2326 (slpeel_make_loop_iterate_ntimes): Rename to..
2327 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2328 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2329 (vect_do_peeling): Update call accordingly.
2330 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2332 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2333 mask_compare_type, can_fully_mask_p and fully_masked_p.
2334 (release_vec_loop_masks): New function.
2335 (_loop_vec_info): Use it to free the loop masks.
2336 (can_produce_all_loop_masks_p): New function.
2337 (vect_get_max_nscalars_per_iter): Likewise.
2338 (vect_verify_full_masking): Likewise.
2339 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2340 retries, and free the mask rgroups before retrying. Check loop-wide
2341 reasons for disallowing fully-masked loops. Make the final decision
2342 about whether use a fully-masked loop or not.
2343 (vect_estimate_min_profitable_iters): Do not assume that peeling
2344 for the number of iterations will be needed for fully-masked loops.
2345 (vectorizable_reduction): Disable fully-masked loops.
2346 (vectorizable_live_operation): Likewise.
2347 (vect_halve_mask_nunits): New function.
2348 (vect_double_mask_nunits): Likewise.
2349 (vect_record_loop_mask): Likewise.
2350 (vect_get_loop_mask): Likewise.
2351 (vect_transform_loop): Handle the case in which the final loop
2352 iteration might handle a partial vector. Call vect_set_loop_condition
2353 instead of slpeel_make_loop_iterate_ntimes.
2354 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2355 (check_load_store_masking): New function.
2356 (prepare_load_store_mask): Likewise.
2357 (vectorizable_store): Handle fully-masked loops.
2358 (vectorizable_load): Likewise.
2359 (supportable_widening_operation): Use vect_halve_mask_nunits for
2361 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2362 (vect_gen_while): New function.
2363 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2364 (aarch64_uqdec<mode>): New insn.
2366 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2367 Alan Hayward <alan.hayward@arm.com>
2368 David Sherwood <david.sherwood@arm.com>
2370 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2371 (reduc_xor_scal_optab): New optabs.
2372 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2373 (reduc_xor_scal_@var{m}): Document.
2374 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2375 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2377 * fold-const-call.c (fold_const_call): Handle them.
2378 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2379 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2380 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2381 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2382 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2383 (UNSPEC_XORV): New unspecs.
2384 (optab): Add entries for them.
2385 (BITWISEV): New int iterator.
2386 (bit_reduc_op): New int attributes.
2388 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2389 Alan Hayward <alan.hayward@arm.com>
2390 David Sherwood <david.sherwood@arm.com>
2392 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2393 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2394 * optabs.def (vec_shl_insert_optab): New optab.
2395 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2396 (duplicate_and_interleave): Likewise.
2397 * tree-vect-loop.c: Include internal-fn.h.
2398 (neutral_op_for_slp_reduction): New function, split out from
2399 get_initial_defs_for_reduction.
2400 (get_initial_def_for_reduction): Handle option 2 for variable-length
2401 vectors by loading the neutral value into a vector and then shifting
2402 the initial value into element 0.
2403 (get_initial_defs_for_reduction): Replace the code argument with
2404 the neutral value calculated by neutral_op_for_slp_reduction.
2405 Use gimple_build_vector for constant-length vectors.
2406 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2407 but the first group_size elements have a neutral value.
2408 Use duplicate_and_interleave otherwise.
2409 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2410 Update call to get_initial_defs_for_reduction. Handle SLP
2411 reductions for variable-length vectors by creating one vector
2412 result for each scalar result, with the elements associated
2413 with other scalar results stubbed out with the neutral value.
2414 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2415 Require IFN_VEC_SHL_INSERT for double reductions on
2416 variable-length vectors, or SLP reductions that have
2417 a neutral value. Require can_duplicate_and_interleave_p
2418 support for variable-length unchained SLP reductions if there
2419 is no neutral value, such as for MIN/MAX reductions. Also require
2420 the number of vector elements to be a multiple of the number of
2421 SLP statements when doing variable-length unchained SLP reductions.
2422 Update call to vect_create_epilog_for_reduction.
2423 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2424 and remove initial values.
2425 (duplicate_and_interleave): Make public.
2426 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2427 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2429 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2430 Alan Hayward <alan.hayward@arm.com>
2431 David Sherwood <david.sherwood@arm.com>
2433 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2434 (can_duplicate_and_interleave_p): New function.
2435 (vect_get_and_check_slp_defs): Take the vector of statements
2436 rather than just the current one. Remove excess parentheses.
2437 Restriction rejectinon of vect_constant_def and vect_external_def
2438 for variable-length vectors to boolean types, or types for which
2439 can_duplicate_and_interleave_p is false.
2440 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2441 (duplicate_and_interleave): New function.
2442 (vect_get_constant_vectors): Use gimple_build_vector for
2443 constant-length vectors and suitable variable-length constant
2444 vectors. Use duplicate_and_interleave for other variable-length
2445 vectors. Don't defer the update when inserting new statements.
2447 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2448 Alan Hayward <alan.hayward@arm.com>
2449 David Sherwood <david.sherwood@arm.com>
2451 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2452 min_profitable_iters doesn't go negative.
2454 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2455 Alan Hayward <alan.hayward@arm.com>
2456 David Sherwood <david.sherwood@arm.com>
2458 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2459 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2460 * optabs.def (vec_mask_load_lanes_optab): New optab.
2461 (vec_mask_store_lanes_optab): Likewise.
2462 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2463 (MASK_STORE_LANES): Likewise.
2464 * internal-fn.c (mask_load_lanes_direct): New macro.
2465 (mask_store_lanes_direct): Likewise.
2466 (expand_mask_load_optab_fn): Handle masked operations.
2467 (expand_mask_load_lanes_optab_fn): New macro.
2468 (expand_mask_store_optab_fn): Handle masked operations.
2469 (expand_mask_store_lanes_optab_fn): New macro.
2470 (direct_mask_load_lanes_optab_supported_p): Likewise.
2471 (direct_mask_store_lanes_optab_supported_p): Likewise.
2472 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2474 (vect_load_lanes_supported): Likewise.
2475 * tree-vect-data-refs.c (strip_conversion): New function.
2476 (can_group_stmts_p): Likewise.
2477 (vect_analyze_data_ref_accesses): Use it instead of checking
2478 for a pair of assignments.
2479 (vect_store_lanes_supported): Take a masked_p parameter.
2480 (vect_load_lanes_supported): Likewise.
2481 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2482 vect_store_lanes_supported and vect_load_lanes_supported.
2483 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2484 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2485 parameter. Don't allow gaps for masked accesses.
2486 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2487 and vect_load_lanes_supported.
2488 (get_load_store_type): Take a masked_p parameter and update
2489 call to get_group_load_store_type.
2490 (vectorizable_store): Update call to get_load_store_type.
2491 Handle IFN_MASK_STORE_LANES.
2492 (vectorizable_load): Update call to get_load_store_type.
2493 Handle IFN_MASK_LOAD_LANES.
2495 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2496 Alan Hayward <alan.hayward@arm.com>
2497 David Sherwood <david.sherwood@arm.com>
2499 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2501 * config/aarch64/aarch64-protos.h
2502 (aarch64_sve_struct_memory_operand_p): Declare.
2503 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2504 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2505 (VPRED, vpred): Handle SVE structure modes.
2506 * config/aarch64/constraints.md (Utx): New constraint.
2507 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2508 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2509 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2510 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2511 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2512 structure modes. Split into pieces after RA.
2513 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2514 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2516 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2517 SVE structure modes.
2518 (aarch64_classify_address): Likewise.
2519 (sizetochar): Move earlier in file.
2520 (aarch64_print_operand): Handle SVE register lists.
2521 (aarch64_array_mode): New function.
2522 (aarch64_sve_struct_memory_operand_p): Likewise.
2523 (TARGET_ARRAY_MODE): Redefine.
2525 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2526 Alan Hayward <alan.hayward@arm.com>
2527 David Sherwood <david.sherwood@arm.com>
2529 * target.def (array_mode): New target hook.
2530 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2531 * doc/tm.texi: Regenerate.
2532 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2533 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2534 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2536 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2539 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2540 Alan Hayward <alan.hayward@arm.com>
2541 David Sherwood <david.sherwood@arm.com>
2543 * fold-const.c (fold_binary_loc): Check the argument types
2544 rather than the result type when testing for a vector operation.
2546 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2548 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2549 * doc/tm.texi: Regenerate.
2551 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2552 Alan Hayward <alan.hayward@arm.com>
2553 David Sherwood <david.sherwood@arm.com>
2555 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2556 (sve): Document new AArch64 extension.
2557 * doc/md.texi (w): Extend the description of the AArch64
2558 constraint to include SVE vectors.
2559 (Upl, Upa): Document new AArch64 predicate constraints.
2560 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2562 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2563 (msve-vector-bits=): New option.
2564 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2565 SVE when these are disabled.
2566 (sve): New extension.
2567 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2568 modes. Adjust their number of units based on aarch64_sve_vg.
2569 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2570 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2571 aarch64_addr_query_type.
2572 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2573 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2574 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2575 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2576 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2577 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2578 (aarch64_simd_imm_zero_p): Delete.
2579 (aarch64_check_zero_based_sve_index_immediate): Declare.
2580 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2581 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2582 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2583 (aarch64_sve_float_mul_immediate_p): Likewise.
2584 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2586 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2587 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2588 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2589 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2590 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2591 (aarch64_regmode_natural_size): Likewise.
2592 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2593 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2595 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2596 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2597 for VG and the SVE predicate registers.
2598 (V_ALIASES): Add a "z"-prefixed alias.
2599 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2600 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2601 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2602 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2603 (REG_CLASS_NAMES): Add entries for them.
2604 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2605 and the predicate registers.
2606 (aarch64_sve_vg): Declare.
2607 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2608 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2609 (REGMODE_NATURAL_SIZE): Define.
2610 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2612 * config/aarch64/aarch64.c: Include cfgrtl.h.
2613 (simd_immediate_info): Add a constructor for series vectors,
2614 and an associated step field.
2615 (aarch64_sve_vg): New variable.
2616 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2617 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2618 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2619 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2620 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2621 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2622 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2623 (aarch64_get_mask_mode): New functions.
2624 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2625 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2626 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2627 predicate modes and predicate registers. Explicitly restrict
2628 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2629 to store a vector mode if it is recognized by
2630 aarch64_classify_vector_mode.
2631 (aarch64_regmode_natural_size): New function.
2632 (aarch64_hard_regno_caller_save_mode): Return the original mode
2634 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2635 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2636 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2637 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2639 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2640 does not overlap dest if the function is frame-related. Handle
2642 (aarch64_split_add_offset): New function.
2643 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2644 them aarch64_add_offset.
2645 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2646 and update call to aarch64_sub_sp.
2647 (aarch64_add_cfa_expression): New function.
2648 (aarch64_expand_prologue): Pass extra temporary registers to the
2649 functions above. Handle the case in which we need to emit new
2650 DW_CFA_expressions for registers that were originally saved
2651 relative to the stack pointer, but now have to be expressed
2652 relative to the frame pointer.
2653 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2655 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2656 IP0 and IP1 values for SVE frames.
2657 (aarch64_expand_vec_series): New function.
2658 (aarch64_expand_sve_widened_duplicate): Likewise.
2659 (aarch64_expand_sve_const_vector): Likewise.
2660 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2661 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2662 into the register, rather than emitting a SET directly.
2663 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2664 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2665 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2666 (offset_9bit_signed_scaled_p): New functions.
2667 (aarch64_replicate_bitmask_imm): New function.
2668 (aarch64_bitmask_imm): Use it.
2669 (aarch64_cannot_force_const_mem): Reject expressions involving
2670 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2671 (aarch64_classify_index): Handle SVE indices, by requiring
2672 a plain register index with a scale that matches the element size.
2673 (aarch64_classify_address): Handle SVE addresses. Assert that
2674 the mode of the address is VOIDmode or an integer mode.
2675 Update call to aarch64_classify_symbol.
2676 (aarch64_classify_symbolic_expression): Update call to
2677 aarch64_classify_symbol.
2678 (aarch64_const_vec_all_in_range_p): New function.
2679 (aarch64_print_vector_float_operand): Likewise.
2680 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2681 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2682 and the FP immediates 1.0 and 0.5.
2683 (aarch64_print_address_internal): Handle SVE addresses.
2684 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2685 (aarch64_regno_regclass): Handle predicate registers.
2686 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2688 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2689 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2690 (aarch64_convert_sve_vector_bits): New function.
2691 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2692 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2694 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2695 Handle SVE vector and predicate modes. Accept VL-based constants
2696 that need only one temporary register, and VL offsets that require
2697 no temporary registers.
2698 (aarch64_conditional_register_usage): Mark the predicate registers
2699 as fixed if SVE isn't available.
2700 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2701 Return true for SVE vector and predicate modes.
2702 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2703 rather than an unsigned int. Handle SVE modes.
2704 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2706 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2708 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2709 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2710 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2711 (aarch64_sve_float_mul_immediate_p): New functions.
2712 (aarch64_sve_valid_immediate): New function.
2713 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2714 Explicitly reject structure modes. Check for INDEX constants.
2715 Handle PTRUE and PFALSE constants.
2716 (aarch64_check_zero_based_sve_index_immediate): New function.
2717 (aarch64_simd_imm_zero_p): Delete.
2718 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2719 vector modes. Accept constants in the range of CNT[BHWD].
2720 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2721 ask for an Advanced SIMD mode.
2722 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2723 (aarch64_simd_vector_alignment): Handle SVE predicates.
2724 (aarch64_vectorize_preferred_vector_alignment): New function.
2725 (aarch64_simd_vector_alignment_reachable): Use it instead of
2727 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2728 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2730 (MAX_VECT_LEN): Delete.
2731 (expand_vec_perm_d): Add a vec_flags field.
2732 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2733 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2734 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2736 (aarch64_evpc_rev): Rename to...
2737 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2738 (aarch64_evpc_rev_global): New function.
2739 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2740 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2742 (aarch64_evpc_sve_tbl): New function.
2743 (aarch64_expand_vec_perm_const_1): Update after rename of
2744 aarch64_evpc_rev. Handle SVE permutes too, trying
2745 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2746 than aarch64_evpc_tbl.
2747 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2748 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2749 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2750 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2751 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2752 (aarch64_expand_sve_vcond): New functions.
2753 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2754 of aarch64_vector_mode_p.
2755 (aarch64_dwarf_poly_indeterminate_value): New function.
2756 (aarch64_compute_pressure_classes): Likewise.
2757 (aarch64_can_change_mode_class): Likewise.
2758 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2759 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2760 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2761 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2762 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2763 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2764 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2765 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2767 (Dn, Dl, Dr): Accept const as well as const_vector.
2768 (Dz): Likewise. Compare against CONST0_RTX.
2769 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2770 of "vector" where appropriate.
2771 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2772 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2773 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2774 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2775 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2776 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2777 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2778 (v_int_equiv): Extend to SVE modes.
2779 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2781 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2782 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2783 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2784 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2785 (SVE_COND_FP_CMP): New int iterators.
2786 (perm_hilo): Handle the new unpack unspecs.
2787 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2789 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2790 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2791 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2792 (aarch64_equality_operator, aarch64_constant_vector_operand)
2793 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2794 (aarch64_sve_nonimmediate_operand): Likewise.
2795 (aarch64_sve_general_operand): Likewise.
2796 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2797 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2798 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2799 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2800 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2801 (aarch64_sve_float_arith_immediate): Likewise.
2802 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2803 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2804 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2805 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2806 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2807 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2808 (aarch64_sve_float_arith_operand): Likewise.
2809 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2810 (aarch64_sve_float_mul_operand): Likewise.
2811 (aarch64_sve_vec_perm_operand): Likewise.
2812 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2813 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2814 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2815 as well as const_vector.
2816 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2817 in file. Use CONST0_RTX and CONSTM1_RTX.
2818 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2819 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2820 Use aarch64_simd_imm_zero.
2821 * config/aarch64/aarch64-sve.md: New file.
2822 * config/aarch64/aarch64.md: Include it.
2823 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2824 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2825 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2826 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2827 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2828 (sve): New attribute.
2829 (enabled): Disable instructions with the sve attribute unless
2831 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2832 aarch64_expand_mov_immediate.
2833 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2834 CNT[BHSD] immediates.
2835 (movti): Split CONST_POLY_INT moves into two halves.
2836 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2837 Split additions that need a temporary here if the destination
2838 is the stack pointer.
2839 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2840 (*add<mode>3_poly_1): New instruction.
2841 (set_clobber_cc): New expander.
2843 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2845 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2846 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2847 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2848 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2849 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2850 Change innermode from fixed_mode_size to machine_mode.
2851 (simplify_subreg): Update call accordingly. Handle a constant-sized
2852 subreg of a variable-length CONST_VECTOR.
2854 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2855 Alan Hayward <alan.hayward@arm.com>
2856 David Sherwood <david.sherwood@arm.com>
2858 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2859 (add_offset_to_base): New function, split out from...
2860 (create_mem_ref): ...here. When handling a scale other than 1,
2861 check first whether the address is valid without the offset.
2862 Add it into the base if so, leaving the index and scale as-is.
2864 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2867 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2868 fold_for_warn before checking if arg2 is INTEGER_CST.
2870 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2872 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2873 (store_multiple_operation): Delete.
2874 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2875 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2876 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2877 guarded by TARGET_STRING.
2878 (rs6000_output_load_multiple): Delete.
2879 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2880 OPTION_MASK_STRING / TARGET_STRING handling.
2881 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2882 (const rs6000_opt_masks) <"string">: Change mask to 0.
2883 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2884 (MASK_STRING): Delete.
2885 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2887 (load_multiple): Delete.
2894 (store_multiple): Delete.
2901 (movmemsi_8reg): Delete.
2902 (corresponding unnamed define_insn): Delete.
2903 (movmemsi_6reg): Delete.
2904 (corresponding unnamed define_insn): Delete.
2905 (movmemsi_4reg): Delete.
2906 (corresponding unnamed define_insn): Delete.
2907 (movmemsi_2reg): Delete.
2908 (corresponding unnamed define_insn): Delete.
2909 (movmemsi_1reg): Delete.
2910 (corresponding unnamed define_insn): Delete.
2911 * config/rs6000/rs6000.opt (mno-string): New.
2912 (mstring): Replace by deprecation warning stub.
2913 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2915 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2917 * regrename.c (regrename_do_replace): If replacing the same
2918 reg multiple times, try to reuse last created gen_raw_REG.
2921 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2922 main to workaround a bug in GDB.
2924 2018-01-12 Tom de Vries <tom@codesourcery.com>
2927 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2929 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2931 PR rtl-optimization/80481
2932 * ira-color.c (get_cap_member): New function.
2933 (allocnos_conflict_by_live_ranges_p): Use it.
2934 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2935 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2937 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2940 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2941 (*saddl_se_1): Ditto.
2943 (*ssubl_se_1): Ditto.
2945 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2947 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2948 rather than wi::to_widest for DR_INITs.
2949 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2950 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2951 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2953 (vect_analyze_group_access_1): Note that here.
2955 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2957 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2958 polynomial type sizes.
2960 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2962 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2963 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2964 (gimple_add_tmp_var): Likewise.
2966 2018-01-12 Martin Liska <mliska@suse.cz>
2968 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2969 (gimple_alloc_sizes): Likewise.
2970 (dump_gimple_statistics): Use PRIu64 in printf format.
2971 * gimple.h: Change uint64_t to int.
2973 2018-01-12 Martin Liska <mliska@suse.cz>
2975 * tree-core.h: Use uint64_t instead of int.
2976 * tree.c (tree_node_counts): Likewise.
2977 (tree_node_sizes): Likewise.
2978 (dump_tree_statistics): Use PRIu64 in printf format.
2980 2018-01-12 Martin Liska <mliska@suse.cz>
2982 * Makefile.in: As qsort_chk is implemented in vec.c, add
2983 vec.o to linkage of gencfn-macros.
2984 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2985 passing the info to record_node_allocation_statistics.
2986 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2988 * ggc-common.c (struct ggc_usage): Add operator== and use
2989 it in operator< and compare function.
2990 * mem-stats.h (struct mem_usage): Likewise.
2991 * vec.c (struct vec_usage): Remove operator< and compare
2992 function. Can be simply inherited.
2994 2018-01-12 Martin Jambor <mjambor@suse.cz>
2997 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2998 * tree-ssa-math-opts.c: Include domwalk.h.
2999 (convert_mult_to_fma_1): New function.
3000 (fma_transformation_info): New type.
3001 (fma_deferring_state): Likewise.
3002 (cancel_fma_deferring): New function.
3003 (result_of_phi): Likewise.
3004 (last_fma_candidate_feeds_initial_phi): Likewise.
3005 (convert_mult_to_fma): Added deferring logic, split actual
3006 transformation to convert_mult_to_fma_1.
3007 (math_opts_dom_walker): New type.
3008 (math_opts_dom_walker::after_dom_children): New method, body moved
3009 here from pass_optimize_widening_mul::execute, added deferring logic
3011 (pass_optimize_widening_mul::execute): Moved most of code to
3012 math_opts_dom_walker::after_dom_children.
3013 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3014 * config/i386/i386.c (ix86_option_override_internal): Added
3015 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3017 2018-01-12 Richard Biener <rguenther@suse.de>
3020 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3021 inline instance vars.
3023 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3026 * config/rx/rx.c (rx_is_restricted_memory_address):
3029 2018-01-12 Richard Biener <rguenther@suse.de>
3031 PR tree-optimization/80846
3032 * target.def (split_reduction): New target hook.
3033 * targhooks.c (default_split_reduction): New function.
3034 * targhooks.h (default_split_reduction): Declare.
3035 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3036 target requests first reduce vectors by combining low and high
3038 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3039 (get_vectype_for_scalar_type_and_size): Export.
3040 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3041 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3042 * doc/tm.texi: Regenerate.
3043 * config/i386/i386.c (ix86_split_reduction): Implement
3044 TARGET_VECTORIZE_SPLIT_REDUCTION.
3046 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3049 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3050 in PIC mode except for TARGET_VXWORKS_RTP.
3051 * config/sparc/sparc.c: Include cfgrtl.h.
3052 (TARGET_INIT_PIC_REG): Define.
3053 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3054 (sparc_pic_register_p): New predicate.
3055 (sparc_legitimate_address_p): Use it.
3056 (sparc_legitimize_pic_address): Likewise.
3057 (sparc_delegitimize_address): Likewise.
3058 (sparc_mode_dependent_address_p): Likewise.
3059 (gen_load_pcrel_sym): Remove 4th parameter.
3060 (load_got_register): Adjust call to above. Remove obsolete stuff.
3061 (sparc_expand_prologue): Do not call load_got_register here.
3062 (sparc_flat_expand_prologue): Likewise.
3063 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3064 (sparc_use_pseudo_pic_reg): New function.
3065 (sparc_init_pic_reg): Likewise.
3066 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3067 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3069 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3071 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3072 Add item for branch_cost.
3074 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3076 PR rtl-optimization/83565
3077 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3078 not extend the result to a larger mode for rotate operations.
3079 (num_sign_bit_copies1): Likewise.
3081 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3084 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3086 Use values-Xc.o for -pedantic.
3087 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3089 2018-01-12 Martin Liska <mliska@suse.cz>
3092 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3094 (possible_polymorphic_call_targets): Use it.
3095 (ipa_devirt): Likewise.
3097 2018-01-12 Martin Liska <mliska@suse.cz>
3099 * profile-count.h (enum profile_quality): Use 0 as invalid
3100 enum value of profile_quality.
3102 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3104 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3105 -mext-string options.
3107 2018-01-12 Richard Biener <rguenther@suse.de>
3109 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3110 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3111 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3113 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3115 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3117 * configure.ac (--with-long-double-format): Add support for the
3118 configuration option to change the default long double format on
3120 * config.gcc (powerpc*-linux*-*): Likewise.
3121 * configure: Regenerate.
3122 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3123 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3124 used without modification.
3126 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3128 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3129 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3130 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3131 MISC_BUILTIN_SPEC_BARRIER.
3132 (rs6000_init_builtins): Likewise.
3133 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3135 (speculation_barrier): New define_insn.
3136 * doc/extend.texi: Document __builtin_speculation_barrier.
3138 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3141 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3142 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3143 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3145 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3146 integral modes instead of "ss" and "sd".
3147 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3148 vectors with 32-bit and 64-bit elements.
3149 (vecdupssescalarmodesuffix): New mode attribute.
3150 (vec_dup<mode>): Use it.
3152 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3155 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3156 frame if argument is passed on stack.
3158 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3161 * ree.c (combine_reaching_defs): Optimize also
3162 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3163 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3165 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3168 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3170 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3173 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3174 after they are computed.
3176 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3178 PR tree-optimization/83695
3179 * gimple-loop-linterchange.cc
3180 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3181 reset cached scev information after interchange.
3182 (pass_linterchange::execute): Remove call to scev_reset_htab.
3184 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3186 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3187 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3188 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3189 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3190 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3191 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3192 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3193 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3194 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3195 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3196 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3197 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3198 (V_lane_reg): Likewise.
3199 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3201 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3202 (vfmal_lane_low<mode>_intrinsic,
3203 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3204 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3205 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3206 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3207 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3208 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3210 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3212 * config/arm/arm-cpus.in (fp16fml): New feature.
3213 (ALL_SIMD): Add fp16fml.
3214 (armv8.2-a): Add fp16fml as an option.
3215 (armv8.3-a): Likewise.
3216 (armv8.4-a): Add fp16fml as part of fp16.
3217 * config/arm/arm.h (TARGET_FP16FML): Define.
3218 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3220 * config/arm/arm-modes.def (V2HF): Define.
3221 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3222 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3223 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3224 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3225 vfmsl_low, vfmsl_high): New set of builtins.
3226 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3227 (vfml_op): New code attribute.
3228 (VFMLHALVES): New int iterator.
3229 (VFML, VFMLSEL): New mode attributes.
3230 (V_reg): Define mapping for V2HF.
3231 (V_hi, V_lo): New mode attributes.
3232 (VF_constraint): Likewise.
3233 (vfml_half, vfml_half_selector): New int attributes.
3234 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3236 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3237 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3239 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3240 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3241 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3242 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3244 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3245 Document new effective target and option set.
3247 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3249 * config/arm/arm-cpus.in (armv8_4): New feature.
3250 (ARMv8_4a): New fgroup.
3251 (armv8.4-a): New arch.
3252 * config/arm/arm-tables.opt: Regenerate.
3253 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3254 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3255 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3256 Add matching rules for -march=armv8.4-a and extensions.
3257 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3259 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3262 * config/rx/rx.md (BW): New mode attribute.
3263 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3265 2018-01-11 Richard Biener <rguenther@suse.de>
3267 PR tree-optimization/83435
3268 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3269 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3270 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3272 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3273 Alan Hayward <alan.hayward@arm.com>
3274 David Sherwood <david.sherwood@arm.com>
3276 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3278 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3279 (aarch64_print_address_internal): Use it to check for a zero offset.
3281 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3282 Alan Hayward <alan.hayward@arm.com>
3283 David Sherwood <david.sherwood@arm.com>
3285 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3286 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3287 Return a poly_int64 rather than a HOST_WIDE_INT.
3288 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3289 rather than a HOST_WIDE_INT.
3290 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3291 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3292 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3293 final_offset from HOST_WIDE_INT to poly_int64.
3294 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3295 to_constant when getting the number of units in an Advanced SIMD
3297 (aarch64_builtin_vectorized_function): Check for a constant number
3299 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3301 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3302 attribute instead of GET_MODE_NUNITS.
3303 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3304 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3305 GET_MODE_SIZE for fixed-size registers.
3306 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3307 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3308 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3309 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3310 (aarch64_print_operand, aarch64_print_address_internal)
3311 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3312 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3313 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3314 Handle polynomial GET_MODE_SIZE.
3315 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3316 wider than SImode without modification.
3317 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3318 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3319 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3320 passing and returning SVE modes.
3321 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3322 rather than GEN_INT.
3323 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3324 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3325 (aarch64_allocate_and_probe_stack_space): Likewise.
3326 (aarch64_layout_frame): Cope with polynomial offsets.
3327 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3328 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3330 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3331 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3332 poly_int64 rather than a HOST_WIDE_INT.
3333 (aarch64_get_separate_components, aarch64_process_components)
3334 (aarch64_expand_prologue, aarch64_expand_epilogue)
3335 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3336 (aarch64_anchor_offset): New function, split out from...
3337 (aarch64_legitimize_address): ...here.
3338 (aarch64_builtin_vectorization_cost): Handle polynomial
3339 TYPE_VECTOR_SUBPARTS.
3340 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3342 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3343 number of elements from the PARALLEL rather than the mode.
3344 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3345 rather than GET_MODE_BITSIZE.
3346 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3347 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3348 (aarch64_expand_vec_perm_const_1): Handle polynomial
3349 d->perm.length () and d->perm elements.
3350 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3351 Apply to_constant to d->perm elements.
3352 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3353 polynomial CONST_VECTOR_NUNITS.
3354 (aarch64_move_pointer): Take amount as a poly_int64 rather
3356 (aarch64_progress_pointer): Avoid temporary variable.
3357 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3358 the mode attribute instead of GET_MODE.
3360 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3361 Alan Hayward <alan.hayward@arm.com>
3362 David Sherwood <david.sherwood@arm.com>
3364 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3365 x exists before using it.
3366 (aarch64_add_constant_internal): Rename to...
3367 (aarch64_add_offset_1): ...this. Replace regnum with separate
3368 src and dest rtxes. Handle the case in which they're different,
3369 including when the offset is zero. Replace scratchreg with an rtx.
3370 Use 2 additions if there is no spare register into which we can
3371 move a 16-bit constant.
3372 (aarch64_add_constant): Delete.
3373 (aarch64_add_offset): Replace reg with separate src and dest
3374 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3375 Use aarch64_add_offset_1.
3376 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3377 an rtx rather than an int. Take the delta as a poly_int64
3378 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3379 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3380 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3381 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3382 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3384 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3385 aarch64_add_constant.
3387 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3389 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3390 Use scalar_float_mode.
3392 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3394 * config/aarch64/aarch64-simd.md
3395 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3396 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3397 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3398 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3399 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3400 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3401 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3402 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3403 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3404 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3406 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3409 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3410 targ_options->x_arm_arch_string is non NULL.
3412 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3414 * config/aarch64/aarch64.h
3415 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3417 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3420 * expmed.c (emit_store_flag_force): Swap if const op0
3421 and change VOIDmode to mode of op0.
3423 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3425 PR rtl-optimization/83761
3426 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3427 than bytes to mode_for_size.
3429 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3432 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3433 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3436 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3439 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3440 when in layout mode.
3441 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3442 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3445 2018-01-10 Michael Collison <michael.collison@arm.com>
3447 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3448 * config/aarch64/aarch64-option-extension.def: Add
3449 AARCH64_OPT_EXTENSION of 'fp16fml'.
3450 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3451 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3452 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3453 * config/aarch64/constraints.md (Ui7): New constraint.
3454 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3455 (VFMLA_SEL_W): Ditto.
3458 (VFMLA16_LOW): New int iterator.
3459 (VFMLA16_HIGH): Ditto.
3460 (UNSPEC_FMLAL): New unspec.
3461 (UNSPEC_FMLSL): Ditto.
3462 (UNSPEC_FMLAL2): Ditto.
3463 (UNSPEC_FMLSL2): Ditto.
3464 (f16mac): New code attribute.
3465 * config/aarch64/aarch64-simd-builtins.def
3466 (aarch64_fmlal_lowv2sf): Ditto.
3467 (aarch64_fmlsl_lowv2sf): Ditto.
3468 (aarch64_fmlalq_lowv4sf): Ditto.
3469 (aarch64_fmlslq_lowv4sf): Ditto.
3470 (aarch64_fmlal_highv2sf): Ditto.
3471 (aarch64_fmlsl_highv2sf): Ditto.
3472 (aarch64_fmlalq_highv4sf): Ditto.
3473 (aarch64_fmlslq_highv4sf): Ditto.
3474 (aarch64_fmlal_lane_lowv2sf): Ditto.
3475 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3476 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3477 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3478 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3479 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3480 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3481 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3482 (aarch64_fmlal_lane_highv2sf): Ditto.
3483 (aarch64_fmlsl_lane_highv2sf): Ditto.
3484 (aarch64_fmlal_laneq_highv2sf): Ditto.
3485 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3486 (aarch64_fmlalq_lane_highv4sf): Ditto.
3487 (aarch64_fmlsl_lane_highv4sf): Ditto.
3488 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3489 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3490 * config/aarch64/aarch64-simd.md:
3491 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3492 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3493 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3494 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3495 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3496 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3497 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3498 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3499 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3500 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3501 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3502 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3503 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3504 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3505 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3506 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3507 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3508 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3509 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3510 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3511 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3512 (vfmlsl_low_u32): Ditto.
3513 (vfmlalq_low_u32): Ditto.
3514 (vfmlslq_low_u32): Ditto.
3515 (vfmlal_high_u32): Ditto.
3516 (vfmlsl_high_u32): Ditto.
3517 (vfmlalq_high_u32): Ditto.
3518 (vfmlslq_high_u32): Ditto.
3519 (vfmlal_lane_low_u32): Ditto.
3520 (vfmlsl_lane_low_u32): Ditto.
3521 (vfmlal_laneq_low_u32): Ditto.
3522 (vfmlsl_laneq_low_u32): Ditto.
3523 (vfmlalq_lane_low_u32): Ditto.
3524 (vfmlslq_lane_low_u32): Ditto.
3525 (vfmlalq_laneq_low_u32): Ditto.
3526 (vfmlslq_laneq_low_u32): Ditto.
3527 (vfmlal_lane_high_u32): Ditto.
3528 (vfmlsl_lane_high_u32): Ditto.
3529 (vfmlal_laneq_high_u32): Ditto.
3530 (vfmlsl_laneq_high_u32): Ditto.
3531 (vfmlalq_lane_high_u32): Ditto.
3532 (vfmlslq_lane_high_u32): Ditto.
3533 (vfmlalq_laneq_high_u32): Ditto.
3534 (vfmlslq_laneq_high_u32): Ditto.
3535 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3536 (AARCH64_FL_FOR_ARCH8_4): New.
3537 (AARCH64_ISA_F16FML): New ISA flag.
3538 (TARGET_F16FML): New feature flag for fp16fml.
3539 (doc/invoke.texi): Document new fp16fml option.
3541 2018-01-10 Michael Collison <michael.collison@arm.com>
3543 * config/aarch64/aarch64-builtins.c:
3544 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3545 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3546 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3547 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3548 (AARCH64_ISA_SHA3): New ISA flag.
3549 (TARGET_SHA3): New feature flag for sha3.
3550 * config/aarch64/iterators.md (sha512_op): New int attribute.
3551 (CRYPTO_SHA512): New int iterator.
3552 (UNSPEC_SHA512H): New unspec.
3553 (UNSPEC_SHA512H2): Ditto.
3554 (UNSPEC_SHA512SU0): Ditto.
3555 (UNSPEC_SHA512SU1): Ditto.
3556 * config/aarch64/aarch64-simd-builtins.def
3557 (aarch64_crypto_sha512hqv2di): New builtin.
3558 (aarch64_crypto_sha512h2qv2di): Ditto.
3559 (aarch64_crypto_sha512su0qv2di): Ditto.
3560 (aarch64_crypto_sha512su1qv2di): Ditto.
3561 (aarch64_eor3qv8hi): Ditto.
3562 (aarch64_rax1qv2di): Ditto.
3563 (aarch64_xarqv2di): Ditto.
3564 (aarch64_bcaxqv8hi): Ditto.
3565 * config/aarch64/aarch64-simd.md:
3566 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3567 (aarch64_crypto_sha512su0qv2di): Ditto.
3568 (aarch64_crypto_sha512su1qv2di): Ditto.
3569 (aarch64_eor3qv8hi): Ditto.
3570 (aarch64_rax1qv2di): Ditto.
3571 (aarch64_xarqv2di): Ditto.
3572 (aarch64_bcaxqv8hi): Ditto.
3573 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3574 (vsha512h2q_u64): Ditto.
3575 (vsha512su0q_u64): Ditto.
3576 (vsha512su1q_u64): Ditto.
3577 (veor3q_u16): Ditto.
3578 (vrax1q_u64): Ditto.
3580 (vbcaxq_u16): Ditto.
3581 * config/arm/types.md (crypto_sha512): New type attribute.
3582 (crypto_sha3): Ditto.
3583 (doc/invoke.texi): Document new sha3 option.
3585 2018-01-10 Michael Collison <michael.collison@arm.com>
3587 * config/aarch64/aarch64-builtins.c:
3588 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3589 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3590 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3591 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3592 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3593 (AARCH64_ISA_SM4): New ISA flag.
3594 (TARGET_SM4): New feature flag for sm4.
3595 * config/aarch64/aarch64-simd-builtins.def
3596 (aarch64_sm3ss1qv4si): Ditto.
3597 (aarch64_sm3tt1aq4si): Ditto.
3598 (aarch64_sm3tt1bq4si): Ditto.
3599 (aarch64_sm3tt2aq4si): Ditto.
3600 (aarch64_sm3tt2bq4si): Ditto.
3601 (aarch64_sm3partw1qv4si): Ditto.
3602 (aarch64_sm3partw2qv4si): Ditto.
3603 (aarch64_sm4eqv4si): Ditto.
3604 (aarch64_sm4ekeyqv4si): Ditto.
3605 * config/aarch64/aarch64-simd.md:
3606 (aarch64_sm3ss1qv4si): Ditto.
3607 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3608 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3609 (aarch64_sm4eqv4si): Ditto.
3610 (aarch64_sm4ekeyqv4si): Ditto.
3611 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3612 (sm3part_op): Ditto.
3613 (CRYPTO_SM3TT): Ditto.
3614 (CRYPTO_SM3PART): Ditto.
3615 (UNSPEC_SM3SS1): New unspec.
3616 (UNSPEC_SM3TT1A): Ditto.
3617 (UNSPEC_SM3TT1B): Ditto.
3618 (UNSPEC_SM3TT2A): Ditto.
3619 (UNSPEC_SM3TT2B): Ditto.
3620 (UNSPEC_SM3PARTW1): Ditto.
3621 (UNSPEC_SM3PARTW2): Ditto.
3622 (UNSPEC_SM4E): Ditto.
3623 (UNSPEC_SM4EKEY): Ditto.
3624 * config/aarch64/constraints.md (Ui2): New constraint.
3625 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3626 * config/arm/types.md (crypto_sm3): New type attribute.
3627 (crypto_sm4): Ditto.
3628 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3629 (vsm3tt1aq_u32): Ditto.
3630 (vsm3tt1bq_u32): Ditto.
3631 (vsm3tt2aq_u32): Ditto.
3632 (vsm3tt2bq_u32): Ditto.
3633 (vsm3partw1q_u32): Ditto.
3634 (vsm3partw2q_u32): Ditto.
3635 (vsm4eq_u32): Ditto.
3636 (vsm4ekeyq_u32): Ditto.
3637 (doc/invoke.texi): Document new sm4 option.
3639 2018-01-10 Michael Collison <michael.collison@arm.com>
3641 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3642 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3643 (AARCH64_FL_FOR_ARCH8_4): New.
3644 (AARCH64_FL_V8_4): New flag.
3645 (doc/invoke.texi): Document new armv8.4-a option.
3647 2018-01-10 Michael Collison <michael.collison@arm.com>
3649 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3650 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3651 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3652 * config/aarch64/aarch64-option-extension.def: Add
3653 AARCH64_OPT_EXTENSION of 'sha2'.
3654 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3655 (crypto): Disable sha2 and aes if crypto disabled.
3656 (crypto): Enable aes and sha2 if enabled.
3657 (simd): Disable sha2 and aes if simd disabled.
3658 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3660 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3661 (TARGET_SHA2): New feature flag for sha2.
3662 (TARGET_AES): New feature flag for aes.
3663 * config/aarch64/aarch64-simd.md:
3664 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3665 conditional on TARGET_AES.
3666 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3667 (aarch64_crypto_sha1hsi): Make pattern conditional
3669 (aarch64_crypto_sha1hv4si): Ditto.
3670 (aarch64_be_crypto_sha1hv4si): Ditto.
3671 (aarch64_crypto_sha1su1v4si): Ditto.
3672 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3673 (aarch64_crypto_sha1su0v4si): Ditto.
3674 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3675 (aarch64_crypto_sha256su0v4si): Ditto.
3676 (aarch64_crypto_sha256su1v4si): Ditto.
3677 (doc/invoke.texi): Document new aes and sha2 options.
3679 2018-01-10 Martin Sebor <msebor@redhat.com>
3681 PR tree-optimization/83781
3682 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3685 2018-01-11 Martin Sebor <msebor@gmail.com>
3686 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3688 PR tree-optimization/83501
3689 PR tree-optimization/81703
3691 * tree-ssa-strlen.c (get_string_cst): Rename...
3692 (get_string_len): ...to this. Handle global constants.
3693 (handle_char_store): Adjust.
3695 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3696 Jim Wilson <jimw@sifive.com>
3698 * config/riscv/riscv-protos.h (riscv_output_return): New.
3699 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3700 (riscv_attribute_table, riscv_output_return),
3701 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3702 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3703 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3705 (riscv_expand_prologue): Add early return for naked function.
3706 (riscv_expand_epilogue): Likewise.
3707 (riscv_function_ok_for_sibcall): Return false for naked function.
3708 (riscv_set_current_function): New.
3709 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3710 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3711 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3712 * doc/extend.texi (RISC-V Function Attributes): New.
3714 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3716 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3717 check for 128-bit long double before checking TCmode.
3718 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3719 128-bit long doubles before checking TFmode or TCmode.
3720 (FLOAT128_IBM_P): Likewise.
3722 2018-01-10 Martin Sebor <msebor@redhat.com>
3724 PR tree-optimization/83671
3725 * builtins.c (c_strlen): Unconditionally return zero for the empty
3727 Use -Warray-bounds for warnings.
3728 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3729 for non-constant array indices with COMPONENT_REF, arrays of
3730 arrays, and pointers to arrays.
3731 (gimple_fold_builtin_strlen): Determine and set length range for
3732 non-constant character arrays.
3734 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3737 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3740 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3742 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3744 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3747 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3748 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3749 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3750 indexed_or_indirect_operand predicate.
3751 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3752 (*vsx_le_perm_load_v8hi): Likewise.
3753 (*vsx_le_perm_load_v16qi): Likewise.
3754 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3755 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3756 (*vsx_le_perm_store_v8hi): Likewise.
3757 (*vsx_le_perm_store_v16qi): Likewise.
3758 (eight unnamed splitters): Likewise.
3760 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3762 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3763 * config/rs6000/emmintrin.h: Likewise.
3764 * config/rs6000/mmintrin.h: Likewise.
3765 * config/rs6000/xmmintrin.h: Likewise.
3767 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3770 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3772 * tree.c (tree_nop_conversion): Return true for location wrapper
3774 (maybe_wrap_with_location): New function.
3775 (selftest::check_strip_nops): New function.
3776 (selftest::test_location_wrappers): New function.
3777 (selftest::tree_c_tests): Call it.
3778 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3779 (maybe_wrap_with_location): New decl.
3780 (EXPR_LOCATION_WRAPPER_P): New macro.
3781 (location_wrapper_p): New inline function.
3782 (tree_strip_any_location_wrapper): New inline function.
3784 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3787 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3788 stack_realign_offset for the largest alignment of stack slot
3790 (ix86_find_max_used_stack_alignment): New function.
3791 (ix86_finalize_stack_frame_flags): Use it. Set
3792 max_used_stack_alignment if we don't realign stack.
3793 * config/i386/i386.h (machine_function): Add
3794 max_used_stack_alignment.
3796 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3798 * config/arm/arm.opt (-mbranch-cost): New option.
3799 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3802 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3805 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3806 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3808 2018-01-10 Richard Biener <rguenther@suse.de>
3811 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3812 early out so it also covers the case where we have a non-NULL
3815 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3817 PR tree-optimization/83753
3818 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3819 for non-strided grouped accesses if the number of elements is 1.
3821 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3824 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3825 * i386.h (TARGET_USE_GATHER): Define.
3826 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3828 2018-01-10 Martin Liska <mliska@suse.cz>
3831 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3832 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3834 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3835 CLEANUP_NO_PARTITIONING is not set.
3837 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3839 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3840 for vectors, as a partial revert of r254296.
3841 * rtl.h (const_vec_p): Delete.
3842 (const_vec_duplicate_p): Don't test for vector CONSTs.
3843 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3844 * expmed.c (make_tree): Likewise.
3847 * common.md (E, F): Use CONSTANT_P instead of checking for
3849 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3850 checking for CONST_VECTOR.
3852 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3855 * predict.c (force_edge_cold): Handle in more sane way edges
3858 2018-01-09 Carl Love <cel@us.ibm.com>
3860 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3862 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3863 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3864 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3865 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3866 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3867 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3868 * config/rs6000/rs6000-protos.h: Add extern defition for
3869 rs6000_generate_float2_double_code.
3870 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3872 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3873 (float2_v2df): Add define_expand.
3875 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3878 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3879 op_mode in the force_to_mode call.
3881 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3883 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3884 instead of checking each element individually.
3885 (aarch64_evpc_uzp): Likewise.
3886 (aarch64_evpc_zip): Likewise.
3887 (aarch64_evpc_ext): Likewise.
3888 (aarch64_evpc_rev): Likewise.
3889 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3890 instead of checking each element individually. Return true without
3892 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3893 whether all selected elements come from the same input, instead of
3894 checking each element individually. Remove calls to gen_rtx_REG,
3895 start_sequence and end_sequence and instead assert that no rtl is
3898 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3900 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3901 order of HIGH and CONST checks.
3903 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3905 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3906 if the destination isn't an SSA_NAME.
3908 2018-01-09 Richard Biener <rguenther@suse.de>
3910 PR tree-optimization/83668
3911 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3913 (canonicalize_loop_form): ... here, renamed from ...
3914 (canonicalize_loop_closed_ssa_form): ... this and amended to
3915 swap successor edges for loop exit blocks to make us use
3916 the RPO order we need for initial schedule generation.
3918 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3920 PR tree-optimization/64811
3921 * match.pd: When optimizing comparisons with Inf, avoid
3922 introducing or losing exceptions from comparisons with NaN.
3924 2018-01-09 Martin Liska <mliska@suse.cz>
3927 * asan.c (shadow_mem_size): Add gcc_assert.
3929 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3931 Don't save registers in main().
3934 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3935 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3936 * config/avr/avr.c (avr_set_current_function): Don't error if
3937 naked, OS_task or OS_main are specified at the same time.
3938 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3940 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3942 * common/config/avr/avr-common.c (avr_option_optimization_table):
3943 Switch on -mmain-is-OS_task for optimizing compilations.
3945 2018-01-09 Richard Biener <rguenther@suse.de>
3947 PR tree-optimization/83572
3948 * graphite.c: Include cfganal.h.
3949 (graphite_transform_loops): Connect infinite loops to exit
3950 and remove fake edges at the end.
3952 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3954 * ipa-inline.c (edge_badness): Revert accidental checkin.
3956 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3959 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3960 symbols; not inline clones.
3962 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3965 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3966 hard registers. Formatting fixes.
3968 PR preprocessor/83722
3969 * gcc.c (try_generate_repro): Pass
3970 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3971 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3974 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3975 Kito Cheng <kito.cheng@gmail.com>
3977 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3978 (riscv_leaf_function_p): Delete.
3979 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3981 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3983 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3985 (do_ifelse): New function.
3986 (do_isel): New function.
3987 (do_sub3): New function.
3988 (do_add3): New function.
3989 (do_load_mask_compare): New function.
3990 (do_overlap_load_compare): New function.
3991 (expand_compare_loop): New function.
3992 (expand_block_compare): Call expand_compare_loop() when appropriate.
3993 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3995 (-mblock-compare-inline-loop-limit): New option.
3997 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4000 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4001 Reverse order of second and third operands in first alternative.
4002 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4003 of first and second elements in UNSPEC_VPERMR vector.
4004 (altivec_expand_vec_perm_le): Likewise.
4006 2018-01-08 Jeff Law <law@redhat.com>
4008 PR rtl-optimizatin/81308
4009 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4010 (process_switch): If group_case_labels makes a change, then set
4012 (pass_convert_switch::execute): If a switch is converted, then
4013 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4015 PR rtl-optimization/81308
4016 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4019 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4021 PR target/83663 - Revert r255946
4022 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4023 generation for cases where splatting a value is not useful.
4024 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4025 across a vec_duplicate and a paradoxical subreg forming a vector
4026 mode to a vec_concat.
4028 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4030 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4031 -march=armv8.3-a variants.
4032 * config/arm/t-multilib: Likewise.
4033 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4035 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4037 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4039 (cceq_ior_compare_complement): Give it a name so I can use it, and
4040 change boolean_or_operator predicate to boolean_operator so it can
4041 be used to generate a crand.
4042 (eqne): New code iterator.
4043 (bd/bd_neg): New code_attrs.
4044 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4045 a single define_insn.
4046 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4047 decrement (bdnzt/bdnzf/bdzt/bdzf).
4048 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4049 with the new names of the branch decrement patterns, and added the
4050 names of the branch decrement conditional patterns.
4052 2018-01-08 Richard Biener <rguenther@suse.de>
4054 PR tree-optimization/83563
4055 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4058 2018-01-08 Richard Biener <rguenther@suse.de>
4061 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4063 2018-01-08 Richard Biener <rguenther@suse.de>
4065 PR tree-optimization/83685
4066 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4067 references to abnormals.
4069 2018-01-08 Richard Biener <rguenther@suse.de>
4072 * dwarf2out.c (output_indirect_strings): Handle empty
4073 skeleton_debug_str_hash.
4074 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4076 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4078 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4079 (emit_store_direct): Likewise.
4080 (arc_trampoline_adjust_address): Likewise.
4081 (arc_asm_trampoline_template): New function.
4082 (arc_initialize_trampoline): Use asm_trampoline_template.
4083 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4084 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4085 * config/arc/arc.md (flush_icache): Delete pattern.
4087 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4089 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4090 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4093 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4096 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4097 by not USED_FOR_TARGET.
4098 (make_pass_resolve_sw_modes): Likewise.
4100 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4102 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4105 2018-01-08 Richard Biener <rguenther@suse.de>
4108 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4110 2018-01-08 Richard Biener <rguenther@suse.de>
4113 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4115 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4118 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4119 basic blocks with a small number of successors.
4120 (convert_control_dep_chain_into_preds): Improve handling of
4122 (dump_predicates): Split apart into...
4123 (dump_pred_chain): ...here...
4124 (dump_pred_info): ...and here.
4125 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4126 (can_chain_union_be_invalidated_p): Improve check for invalidation
4128 (uninit_uses_cannot_happen): Avoid unnecessary if
4129 convert_control_dep_chain_into_preds yielded nothing.
4131 2018-01-06 Martin Sebor <msebor@redhat.com>
4133 PR tree-optimization/83640
4134 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4135 subtracting negative offset from size.
4136 (builtin_access::overlap): Adjust offset bounds of the access to fall
4137 within the size of the object if possible.
4139 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4141 PR rtl-optimization/83699
4142 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4143 extract_bit_field_as_subreg to cases in which the extracted
4144 value is also a vector.
4146 * lra-constraints.c (process_alt_operands): Test for the equivalence
4147 substitutions when detecting a possible reload cycle.
4149 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4152 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4153 by default if flag_selective_schedling{,2}. Formatting fixes.
4155 PR rtl-optimization/83682
4156 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4157 if it has non-VECTOR_MODE element mode.
4158 (vec_duplicate_p): Likewise.
4161 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4162 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4164 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4167 * config/i386/i386-builtin.def
4168 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4169 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4170 Require also OPTION_MASK_ISA_AVX512F in addition to
4171 OPTION_MASK_ISA_GFNI.
4172 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4173 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4174 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4175 to OPTION_MASK_ISA_GFNI.
4176 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4177 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4178 OPTION_MASK_ISA_AVX512BW.
4179 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4180 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4181 addition to OPTION_MASK_ISA_GFNI.
4182 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4183 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4184 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4185 to OPTION_MASK_ISA_GFNI.
4186 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4187 a requirement for all ISAs rather than any of them with a few
4189 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4191 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4192 bitmasks to be enabled with 3 exceptions, instead of requiring any
4193 enabled ISA with lots of exceptions.
4194 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4195 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4196 Change avx512bw in isa attribute to avx512f.
4197 * config/i386/sgxintrin.h: Add license boilerplate.
4198 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4199 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4200 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4201 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4203 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4204 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4205 temporarily sse2 rather than sse if not enabled already.
4208 * config/i386/sse.md (VI248_VLBW): Rename to ...
4209 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4210 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4211 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4212 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4213 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4214 mode iterator instead of VI248_VLBW.
4216 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4218 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4219 (record_modified): Skip clobbers; add debug output.
4220 (param_change_prob): Use sreal frequencies.
4222 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4224 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4225 punt for user-aligned variables.
4227 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4229 * tree-chrec.c (chrec_contains_symbols): Return true for
4232 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4235 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4236 of (x|y) == x for BICS pattern.
4238 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4240 PR tree-optimization/83605
4241 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4242 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4245 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4247 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4248 * config/epiphany/rtems.h: New file.
4250 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4251 Uros Bizjak <ubizjak@gmail.com>
4254 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4255 QIreg_operand instead of register_operand predicate.
4256 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4257 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4258 comments instead of -fmitigate[-_]rop.
4260 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4263 * cgraphunit.c (symbol_table::compile): Switch to text_section
4264 before calling assembly_start debug hook.
4265 * run-rtl-passes.c (run_rtl_passes): Likewise.
4268 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4270 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4271 range_int_cst_p rather than !symbolic_range_p before calling
4272 extract_range_from_multiplicative_op_1.
4274 2018-01-04 Jeff Law <law@redhat.com>
4276 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4277 redundant test in assertion.
4279 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4281 * doc/rtl.texi: Document machine_mode wrapper classes.
4283 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4285 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4288 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4290 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4291 the VEC_PERM_EXPR fold to fail.
4293 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4296 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4297 to switched_sections.
4299 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4302 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4305 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4308 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4309 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4311 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4314 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4315 is BLKmode and bitpos not zero or mode change is needed.
4317 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4320 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4323 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4326 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4327 instead of MULT rtx. Update all corresponding splitters.
4329 (*ssub<modesuffix>): Ditto.
4331 (*cmp_sadd_di): Update split patterns.
4332 (*cmp_sadd_si): Ditto.
4333 (*cmp_sadd_sidi): Ditto.
4334 (*cmp_ssub_di): Ditto.
4335 (*cmp_ssub_si): Ditto.
4336 (*cmp_ssub_sidi): Ditto.
4337 * config/alpha/predicates.md (const23_operand): New predicate.
4338 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4339 Look for ASHIFT, not MULT inner operand.
4340 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4342 2018-01-04 Martin Liska <mliska@suse.cz>
4344 PR gcov-profile/83669
4345 * gcov.c (output_intermediate_file): Add version to intermediate
4347 * doc/gcov.texi: Document new field 'version' in intermediate
4348 file format. Fix location of '-k' option of gcov command.
4350 2018-01-04 Martin Liska <mliska@suse.cz>
4353 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4355 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4357 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4359 2018-01-03 Martin Sebor <msebor@redhat.com>
4361 PR tree-optimization/83655
4362 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4363 checking calls with invalid arguments.
4365 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4367 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4368 (vectorizable_mask_load_store): Delete.
4369 (vectorizable_call): Return false for masked loads and stores.
4370 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4371 instead of gimple_assign_rhs1.
4372 (vectorizable_load): Handle IFN_MASK_LOAD.
4373 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4375 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4377 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4379 (vectorizable_mask_load_store): ...here.
4380 (vectorizable_load): ...and here.
4382 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4384 * tree-vect-stmts.c (vect_build_all_ones_mask)
4385 (vect_build_zero_merge_argument): New functions, split out from...
4386 (vectorizable_load): ...here.
4388 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4390 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4392 (vectorizable_mask_load_store): ...here.
4393 (vectorizable_store): ...and here.
4395 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4397 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4399 (vectorizable_mask_load_store): ...here.
4401 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4403 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4404 (vect_model_store_cost): Take a vec_load_store_type instead of a
4406 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4407 (vect_model_store_cost): Take a vec_load_store_type instead of a
4409 (vectorizable_mask_load_store): Update accordingly.
4410 (vectorizable_store): Likewise.
4411 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4413 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4415 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4416 IFN_MASK_LOAD calls here rather than...
4417 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4419 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4420 Alan Hayward <alan.hayward@arm.com>
4421 David Sherwood <david.sherwood@arm.com>
4423 * expmed.c (extract_bit_field_1): For vector extracts,
4424 fall back to extract_bit_field_as_subreg if vec_extract
4427 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4428 Alan Hayward <alan.hayward@arm.com>
4429 David Sherwood <david.sherwood@arm.com>
4431 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4432 they are variable or constant sized.
4433 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4434 slots for constant-sized data.
4436 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4437 Alan Hayward <alan.hayward@arm.com>
4438 David Sherwood <david.sherwood@arm.com>
4440 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4441 handling COND_EXPRs with boolean comparisons, try to find a better
4442 basis for the mask type than the boolean itself.
4444 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4446 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4447 is calculated and how it can be overridden.
4448 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4449 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4451 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4454 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4455 Alan Hayward <alan.hayward@arm.com>
4456 David Sherwood <david.sherwood@arm.com>
4458 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4459 Remove the mode argument.
4460 (aarch64_simd_valid_immediate): Remove the mode and inverse
4462 * config/aarch64/iterators.md (bitsize): New iterator.
4463 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4464 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4465 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4466 aarch64_simd_valid_immediate.
4467 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4468 (aarch64_reg_or_bic_imm): Likewise.
4469 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4470 with an insn_type enum and msl with a modifier_type enum.
4471 Replace element_width with a scalar_mode. Change the shift
4472 to unsigned int. Add constructors for scalar_float_mode and
4473 scalar_int_mode elements.
4474 (aarch64_vect_float_const_representable_p): Delete.
4475 (aarch64_can_const_movi_rtx_p)
4476 (aarch64_simd_scalar_immediate_valid_for_move)
4477 (aarch64_simd_make_constant): Update call to
4478 aarch64_simd_valid_immediate.
4479 (aarch64_advsimd_valid_immediate_hs): New function.
4480 (aarch64_advsimd_valid_immediate): Likewise.
4481 (aarch64_simd_valid_immediate): Remove mode and inverse
4482 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4483 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4484 and aarch64_float_const_representable_p on the result.
4485 (aarch64_output_simd_mov_immediate): Remove mode argument.
4486 Update call to aarch64_simd_valid_immediate and use of
4487 simd_immediate_info.
4488 (aarch64_output_scalar_simd_mov_immediate): Update call
4491 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4492 Alan Hayward <alan.hayward@arm.com>
4493 David Sherwood <david.sherwood@arm.com>
4495 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4496 (mode_nunits): Likewise CONST_MODE_NUNITS.
4497 * machmode.def (ADJUST_NUNITS): Document.
4498 * genmodes.c (mode_data::need_nunits_adj): New field.
4499 (blank_mode): Update accordingly.
4500 (adj_nunits): New variable.
4501 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4503 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4504 listed in adj_nunits.
4505 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4506 listed in adj_nunits. Don't emit case statements for such modes.
4507 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4508 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4509 nothing if adj_nunits is nonnull.
4510 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4511 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4512 (emit_mode_fbit): Update use of print_maybe_const_decl.
4513 (emit_move_size): Likewise. Treat the array as non-const
4515 (emit_mode_adjustments): Handle adj_nunits.
4517 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4519 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4520 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4521 (VECTOR_MODES): Use it.
4522 (make_vector_modes): Take the prefix as an argument.
4524 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4525 Alan Hayward <alan.hayward@arm.com>
4526 David Sherwood <david.sherwood@arm.com>
4528 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4529 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4530 for MODE_VECTOR_BOOL.
4531 * machmode.def (VECTOR_BOOL_MODE): Document.
4532 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4533 (make_vector_bool_mode): New function.
4534 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4536 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4537 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4539 * stor-layout.c (int_mode_for_mode): Likewise.
4540 * tree.c (build_vector_type_for_mode): Likewise.
4541 * varasm.c (output_constant_pool_2): Likewise.
4542 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4543 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4544 for MODE_VECTOR_BOOL.
4545 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4546 of mode class checks.
4547 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4548 instead of a list of mode class checks.
4549 (expand_vector_scalar_condition): Likewise.
4550 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4552 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4553 Alan Hayward <alan.hayward@arm.com>
4554 David Sherwood <david.sherwood@arm.com>
4556 * machmode.h (mode_size): Change from unsigned short to
4558 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4559 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4560 or if measurement_type is not polynomial.
4561 (fixed_size_mode::includes_p): Check for constant-sized modes.
4562 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4563 return a poly_uint16 rather than an unsigned short.
4564 (emit_mode_size): Change the type of mode_size from unsigned short
4565 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4566 (emit_mode_adjustments): Cope with polynomial vector sizes.
4567 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4569 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4571 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4572 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4573 * caller-save.c (setup_save_areas): Likewise.
4574 (replace_reg_with_saved_mem): Likewise.
4575 * calls.c (emit_library_call_value_1): Likewise.
4576 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4577 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4578 (gen_lowpart_for_combine): Likewise.
4579 * convert.c (convert_to_integer_1): Likewise.
4580 * cse.c (equiv_constant, cse_insn): Likewise.
4581 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4582 (cselib_subst_to_values): Likewise.
4583 * dce.c (word_dce_process_block): Likewise.
4584 * df-problems.c (df_word_lr_mark_ref): Likewise.
4585 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4586 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4587 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4588 (rtl_for_decl_location): Likewise.
4589 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4590 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4591 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4592 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4593 (expand_expr_real_1): Likewise.
4594 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4595 (pad_below): Likewise.
4596 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4597 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4598 * ira.c (get_subreg_tracking_sizes): Likewise.
4599 * ira-build.c (ira_create_allocno_objects): Likewise.
4600 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4601 (ira_sort_regnos_for_alter_reg): Likewise.
4602 * ira-costs.c (record_operand_costs): Likewise.
4603 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4604 (resolve_simple_move): Likewise.
4605 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4606 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4607 (lra_constraints): Likewise.
4608 (CONST_POOL_OK_P): Reject variable-sized modes.
4609 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4610 (add_pseudo_to_slot, lra_spill): Likewise.
4611 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4612 * optabs-query.c (get_best_extraction_insn): Likewise.
4613 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4614 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4615 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4616 * recog.c (offsettable_address_addr_space_p): Likewise.
4617 * regcprop.c (maybe_mode_change): Likewise.
4618 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4619 * regrename.c (build_def_use): Likewise.
4620 * regstat.c (dump_reg_info): Likewise.
4621 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4622 (find_reloads, find_reloads_subreg_address): Likewise.
4623 * reload1.c (eliminate_regs_1): Likewise.
4624 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4625 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4626 (simplify_binary_operation_1, simplify_subreg): Likewise.
4627 * targhooks.c (default_function_arg_padding): Likewise.
4628 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4629 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4630 (verify_gimple_assign_ternary): Likewise.
4631 * tree-inline.c (estimate_move_cost): Likewise.
4632 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4633 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4634 (get_address_cost_ainc): Likewise.
4635 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4636 (vect_supportable_dr_alignment): Likewise.
4637 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4638 (vectorizable_reduction): Likewise.
4639 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4640 (vectorizable_operation, vectorizable_load): Likewise.
4641 * tree.c (build_same_sized_truth_vector_type): Likewise.
4642 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4643 * var-tracking.c (emit_note_insn_var_location): Likewise.
4644 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4645 (ADDR_VEC_ALIGN): Likewise.
4647 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4648 Alan Hayward <alan.hayward@arm.com>
4649 David Sherwood <david.sherwood@arm.com>
4651 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4653 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4654 or if measurement_type is polynomial.
4655 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4656 * combine.c (make_extraction): Likewise.
4657 * dse.c (find_shift_sequence): Likewise.
4658 * dwarf2out.c (mem_loc_descriptor): Likewise.
4659 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4660 (extract_bit_field, extract_low_bits): Likewise.
4661 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4662 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4663 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4664 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4665 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4666 * reload.c (find_reloads): Likewise.
4667 * reload1.c (alter_reg): Likewise.
4668 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4669 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4670 * tree-if-conv.c (predicate_mem_writes): Likewise.
4671 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4672 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4673 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4674 * valtrack.c (dead_debug_insert_temp): Likewise.
4675 * varasm.c (mergeable_constant_section): Likewise.
4676 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4678 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4679 Alan Hayward <alan.hayward@arm.com>
4680 David Sherwood <david.sherwood@arm.com>
4682 * expr.c (expand_assignment): Cope with polynomial mode sizes
4683 when assigning to a CONCAT.
4685 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4686 Alan Hayward <alan.hayward@arm.com>
4687 David Sherwood <david.sherwood@arm.com>
4689 * machmode.h (mode_precision): Change from unsigned short to
4691 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4693 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4694 or if measurement_type is not polynomial.
4695 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4696 in which the mode is already known to be a scalar_int_mode.
4697 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4698 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4700 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4701 for GET_MODE_PRECISION.
4702 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4703 for GET_MODE_PRECISION.
4704 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4706 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4707 (expand_field_assignment, make_extraction): Likewise.
4708 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4709 (get_last_value): Likewise.
4710 * convert.c (convert_to_integer_1): Likewise.
4711 * cse.c (cse_insn): Likewise.
4712 * expr.c (expand_expr_real_1): Likewise.
4713 * lra-constraints.c (simplify_operand_subreg): Likewise.
4714 * optabs-query.c (can_atomic_load_p): Likewise.
4715 * optabs.c (expand_atomic_load): Likewise.
4716 (expand_atomic_store): Likewise.
4717 * ree.c (combine_reaching_defs): Likewise.
4718 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4719 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4720 * tree.h (type_has_mode_precision_p): Likewise.
4721 * ubsan.c (instrument_si_overflow): Likewise.
4723 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4724 Alan Hayward <alan.hayward@arm.com>
4725 David Sherwood <david.sherwood@arm.com>
4727 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4728 polynomial numbers of units.
4729 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4730 (valid_vector_subparts_p): New function.
4731 (build_vector_type): Remove temporary shim and take the number
4732 of units as a poly_uint64 rather than an int.
4733 (build_opaque_vector_type): Take the number of units as a
4734 poly_uint64 rather than an int.
4735 * tree.c (build_vector_from_ctor): Handle polynomial
4736 TYPE_VECTOR_SUBPARTS.
4737 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4738 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4739 (build_vector_from_val): If the number of units is variable,
4740 use build_vec_duplicate_cst for constant operands and
4741 VEC_DUPLICATE_EXPR otherwise.
4742 (make_vector_type): Remove temporary is_constant ().
4743 (build_vector_type, build_opaque_vector_type): Take the number of
4744 units as a poly_uint64 rather than an int.
4745 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4747 * cfgexpand.c (expand_debug_expr): Likewise.
4748 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4749 (store_constructor, expand_expr_real_1): Likewise.
4750 (const_scalar_mask_from_tree): Likewise.
4751 * fold-const-call.c (fold_const_reduction): Likewise.
4752 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4753 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4754 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4755 (fold_relational_const): Likewise.
4756 (native_interpret_vector): Likewise. Change the size from an
4757 int to an unsigned int.
4758 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4759 TYPE_VECTOR_SUBPARTS.
4760 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4761 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4762 duplicating a non-constant operand into a variable-length vector.
4763 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4764 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4765 * ipa-icf.c (sem_variable::equals): Likewise.
4766 * match.pd: Likewise.
4767 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4768 * print-tree.c (print_node): Likewise.
4769 * stor-layout.c (layout_type): Likewise.
4770 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4771 * tree-cfg.c (verify_gimple_comparison): Likewise.
4772 (verify_gimple_assign_binary): Likewise.
4773 (verify_gimple_assign_ternary): Likewise.
4774 (verify_gimple_assign_single): Likewise.
4775 * tree-pretty-print.c (dump_generic_node): Likewise.
4776 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4777 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4778 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4779 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4780 (vect_shift_permute_load_chain): Likewise.
4781 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4782 (expand_vector_condition, optimize_vector_constructor): Likewise.
4783 (lower_vec_perm, get_compute_type): Likewise.
4784 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4785 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4786 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4787 (vect_recog_mask_conversion_pattern): Likewise.
4788 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4789 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4790 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4791 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4792 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4793 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4794 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4795 (supportable_widening_operation): Likewise.
4796 (supportable_narrowing_operation): Likewise.
4797 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4799 * varasm.c (output_constant): Likewise.
4801 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4802 Alan Hayward <alan.hayward@arm.com>
4803 David Sherwood <david.sherwood@arm.com>
4805 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4806 so that both the length == 3 and length != 3 cases set up their
4807 own permute vectors. Add comments explaining why we know the
4808 number of elements is constant.
4809 (vect_permute_load_chain): Likewise.
4811 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4812 Alan Hayward <alan.hayward@arm.com>
4813 David Sherwood <david.sherwood@arm.com>
4815 * machmode.h (mode_nunits): Change from unsigned char to
4817 (ONLY_FIXED_SIZE_MODES): New macro.
4818 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4819 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4820 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4822 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4823 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4824 or if measurement_type is not polynomial.
4825 * genmodes.c (ZERO_COEFFS): New macro.
4826 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4828 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4829 Use ZERO_COEFFS when emitting initializers.
4830 * data-streamer.h (bp_pack_poly_value): New function.
4831 (bp_unpack_poly_value): Likewise.
4832 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4833 for GET_MODE_NUNITS.
4834 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4835 for GET_MODE_NUNITS.
4836 * tree.c (make_vector_type): Remove temporary shim and make
4837 the real function take the number of units as a poly_uint64
4839 (build_vector_type_for_mode): Handle polynomial nunits.
4840 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4841 * emit-rtl.c (const_vec_series_p_1): Likewise.
4842 (gen_rtx_CONST_VECTOR): Likewise.
4843 * fold-const.c (test_vec_duplicate_folding): Likewise.
4844 * genrecog.c (validate_pattern): Likewise.
4845 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4846 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4847 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4848 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4849 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4850 * rtlanal.c (subreg_get_info): Likewise.
4851 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4852 (vect_grouped_load_supported): Likewise.
4853 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4854 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4855 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4856 (simplify_const_unary_operation, simplify_binary_operation_1)
4857 (simplify_const_binary_operation, simplify_ternary_operation)
4858 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4859 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4860 instead of CONST_VECTOR_NUNITS.
4861 * varasm.c (output_constant_pool_2): Likewise.
4862 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4863 explicit-encoded elements in the XVEC for variable-length vectors.
4865 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4867 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4869 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4870 Alan Hayward <alan.hayward@arm.com>
4871 David Sherwood <david.sherwood@arm.com>
4873 * coretypes.h (fixed_size_mode): Declare.
4874 (fixed_size_mode_pod): New typedef.
4875 * builtins.h (target_builtins::x_apply_args_mode)
4876 (target_builtins::x_apply_result_mode): Change type to
4877 fixed_size_mode_pod.
4878 * builtins.c (apply_args_size, apply_result_size, result_vector)
4879 (expand_builtin_apply_args_1, expand_builtin_apply)
4880 (expand_builtin_return): Update accordingly.
4882 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4884 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4885 * cselib.c (cselib_hash_rtx): Likewise.
4886 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4887 CONST_VECTOR encoding.
4889 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4890 Jeff Law <law@redhat.com>
4893 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4894 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4895 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4896 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4899 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4900 explicitly probe *sp in a noreturn function if there were any callee
4901 register saves or frame pointer is needed.
4903 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4906 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4907 BLKmode for ternary, binary or unary expressions.
4910 * var-tracking.c (delete_vta_debug_insn): New inline function.
4911 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4912 insns from get_insns () to NULL instead of each bb separately.
4913 Use delete_vta_debug_insn. No longer static.
4914 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4915 delete_vta_debug_insns callers.
4916 * rtl.h (delete_vta_debug_insns): Declare.
4917 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4918 instead of variable_tracking_main.
4920 2018-01-03 Martin Sebor <msebor@redhat.com>
4922 PR tree-optimization/83603
4923 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4924 arguments past the endof the argument list in functions declared
4925 without a prototype.
4926 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4927 Avoid checking when arguments are null.
4929 2018-01-03 Martin Sebor <msebor@redhat.com>
4932 * doc/extend.texi (attribute const): Fix a typo.
4933 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4934 issuing -Wsuggest-attribute for void functions.
4936 2018-01-03 Martin Sebor <msebor@redhat.com>
4938 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4939 offset_int::from instead of wide_int::to_shwi.
4940 (maybe_diag_overlap): Remove assertion.
4941 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4942 * gimple-ssa-sprintf.c (format_directive): Same.
4943 (parse_directive): Same.
4944 (sprintf_dom_walker::compute_format_length): Same.
4945 (try_substitute_return_value): Same.
4947 2018-01-03 Jeff Law <law@redhat.com>
4950 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4951 non-constant residual for zero at runtime and avoid probing in
4952 that case. Reorganize code for trailing problem to mirror handling
4955 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4957 PR tree-optimization/83501
4958 * tree-ssa-strlen.c (get_string_cst): New.
4959 (handle_char_store): Call get_string_cst.
4961 2018-01-03 Martin Liska <mliska@suse.cz>
4963 PR tree-optimization/83593
4964 * tree-ssa-strlen.c: Include tree-cfg.h.
4965 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4966 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4967 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4969 (strlen_dom_walker::before_dom_children): Call
4970 gimple_purge_dead_eh_edges. Dump tranformation with details
4972 (strlen_dom_walker::before_dom_children): Update call by adding
4973 new argument cleanup_eh.
4974 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4976 2018-01-03 Martin Liska <mliska@suse.cz>
4979 * cif-code.def (VARIADIC_THUNK): New enum value.
4980 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4983 2018-01-03 Jan Beulich <jbeulich@suse.com>
4985 * sse.md (mov<mode>_internal): Tighten condition for when to use
4986 vmovdqu<ssescalarsize> for TI and OI modes.
4988 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4990 Update copyright years.
4992 2018-01-03 Martin Liska <mliska@suse.cz>
4995 * ipa-visibility.c (function_and_variable_visibility): Skip
4996 functions with noipa attribure.
4998 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5000 * gcc.c (process_command): Update copyright notice dates.
5001 * gcov-dump.c (print_version): Ditto.
5002 * gcov.c (print_version): Ditto.
5003 * gcov-tool.c (print_version): Ditto.
5004 * gengtype.c (create_file): Ditto.
5005 * doc/cpp.texi: Bump @copying's copyright year.
5006 * doc/cppinternals.texi: Ditto.
5007 * doc/gcc.texi: Ditto.
5008 * doc/gccint.texi: Ditto.
5009 * doc/gcov.texi: Ditto.
5010 * doc/install.texi: Ditto.
5011 * doc/invoke.texi: Ditto.
5013 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5015 * vector-builder.h (vector_builder::m_full_nelts): Change from
5016 unsigned int to poly_uint64.
5017 (vector_builder::full_nelts): Update prototype accordingly.
5018 (vector_builder::new_vector): Likewise.
5019 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5020 (vector_builder::operator ==): Likewise.
5021 (vector_builder::finalize): Likewise.
5022 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5023 Take the number of elements as a poly_uint64 rather than an
5025 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5026 from unsigned int to poly_uint64.
5027 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5028 (vec_perm_indices::new_vector): Likewise.
5029 (vec_perm_indices::length): Likewise.
5030 (vec_perm_indices::nelts_per_input): Likewise.
5031 (vec_perm_indices::input_nelts): Likewise.
5032 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5033 number of elements per input as a poly_uint64 rather than an
5034 unsigned int. Use the original encoding for variable-length
5035 vectors, rather than clamping each individual element.
5036 For the second and subsequent elements in each pattern,
5037 clamp the step and base before clamping their sum.
5038 (vec_perm_indices::series_p): Handle polynomial element counts.
5039 (vec_perm_indices::all_in_range_p): Likewise.
5040 (vec_perm_indices_to_tree): Likewise.
5041 (vec_perm_indices_to_rtx): Likewise.
5042 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5043 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5044 (tree_vector_builder::new_binary_operation): Handle polynomial
5045 element counts. Return false if we need to know the number
5046 of elements at compile time.
5047 * fold-const.c (fold_vec_perm): Punt if the number of elements
5048 isn't known at compile time.
5050 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5052 * vec-perm-indices.h (vec_perm_builder): Change element type
5053 from HOST_WIDE_INT to poly_int64.
5054 (vec_perm_indices::element_type): Update accordingly.
5055 (vec_perm_indices::clamp): Handle polynomial element_types.
5056 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5057 (vec_perm_indices::all_in_range_p): Likewise.
5058 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5060 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5061 polynomial vec_perm_indices element types.
5062 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5063 * fold-const.c (fold_vec_perm): Likewise.
5064 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5065 * tree-vect-generic.c (lower_vec_perm): Likewise.
5066 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5067 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5068 element type to HOST_WIDE_INT.
5070 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5071 Alan Hayward <alan.hayward@arm.com>
5072 David Sherwood <david.sherwood@arm.com>
5074 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5075 rather than an int. Use plus_constant.
5076 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5077 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5079 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5080 Alan Hayward <alan.hayward@arm.com>
5081 David Sherwood <david.sherwood@arm.com>
5083 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5084 a HOST_WIDE_INT to a poly_int64.
5086 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5087 Alan Hayward <alan.hayward@arm.com>
5088 David Sherwood <david.sherwood@arm.com>
5090 * calls.c (load_register_parameters): Cope with polynomial
5091 mode sizes. Require a constant size for BLKmode parameters
5092 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5093 forces a parameter to be padded at the lsb end in order to
5094 fill a complete number of words, require the parameter size
5095 to be ordered wrt UNITS_PER_WORD.
5097 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5098 Alan Hayward <alan.hayward@arm.com>
5099 David Sherwood <david.sherwood@arm.com>
5101 * reload1.c (spill_stack_slot_width): Change element type
5102 from unsigned int to poly_uint64_pod.
5103 (alter_reg): Treat mode sizes as polynomial.
5105 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5106 Alan Hayward <alan.hayward@arm.com>
5107 David Sherwood <david.sherwood@arm.com>
5109 * reload.c (complex_word_subreg_p): New function.
5110 (reload_inner_reg_of_subreg, push_reload): Use it.
5112 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5113 Alan Hayward <alan.hayward@arm.com>
5114 David Sherwood <david.sherwood@arm.com>
5116 * lra-constraints.c (process_alt_operands): Reject matched
5117 operands whose sizes aren't ordered.
5118 (match_reload): Refer to this check here.
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 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5125 that the mode size is in the set {1, 2, 4, 8, 16}.
5127 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5128 Alan Hayward <alan.hayward@arm.com>
5129 David Sherwood <david.sherwood@arm.com>
5131 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5132 Use plus_constant instead of gen_rtx_PLUS.
5134 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5135 Alan Hayward <alan.hayward@arm.com>
5136 David Sherwood <david.sherwood@arm.com>
5138 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5139 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5140 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5141 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5142 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5143 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5144 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5145 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5146 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5147 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5149 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5150 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5151 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5152 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5153 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5154 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5155 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5156 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5157 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5158 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5160 * expr.c (emit_move_resolve_push): Treat the input and result
5161 of PUSH_ROUNDING as a poly_int64.
5162 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5163 (emit_push_insn): Likewise.
5164 * lra-eliminations.c (mark_not_eliminable): Likewise.
5165 * recog.c (push_operand): Likewise.
5166 * reload1.c (elimination_effects): Likewise.
5167 * rtlanal.c (nonzero_bits1): Likewise.
5168 * calls.c (store_one_arg): Likewise. Require the padding to be
5169 known at compile time.
5171 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5172 Alan Hayward <alan.hayward@arm.com>
5173 David Sherwood <david.sherwood@arm.com>
5175 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5176 Use plus_constant instead of gen_rtx_PLUS.
5178 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5179 Alan Hayward <alan.hayward@arm.com>
5180 David Sherwood <david.sherwood@arm.com>
5182 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5185 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5186 Alan Hayward <alan.hayward@arm.com>
5187 David Sherwood <david.sherwood@arm.com>
5189 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5190 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5191 via stack temporaries. Treat the mode size as polynomial too.
5193 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5194 Alan Hayward <alan.hayward@arm.com>
5195 David Sherwood <david.sherwood@arm.com>
5197 * expr.c (expand_expr_real_2): When handling conversions involving
5198 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5199 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5200 as a poly_uint64 too.
5202 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5203 Alan Hayward <alan.hayward@arm.com>
5204 David Sherwood <david.sherwood@arm.com>
5206 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5208 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5209 Alan Hayward <alan.hayward@arm.com>
5210 David Sherwood <david.sherwood@arm.com>
5212 * combine.c (can_change_dest_mode): Handle polynomial
5213 REGMODE_NATURAL_SIZE.
5214 * expmed.c (store_bit_field_1): Likewise.
5215 * expr.c (store_constructor): Likewise.
5216 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5217 and polynomial REGMODE_NATURAL_SIZE.
5218 (gen_lowpart_common): Likewise.
5219 * reginfo.c (record_subregs_of_mode): Likewise.
5220 * rtlanal.c (read_modify_subreg_p): Likewise.
5222 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5223 Alan Hayward <alan.hayward@arm.com>
5224 David Sherwood <david.sherwood@arm.com>
5226 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5227 numbers of elements.
5229 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5230 Alan Hayward <alan.hayward@arm.com>
5231 David Sherwood <david.sherwood@arm.com>
5233 * match.pd: Cope with polynomial numbers of vector elements.
5235 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5236 Alan Hayward <alan.hayward@arm.com>
5237 David Sherwood <david.sherwood@arm.com>
5239 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5240 in a POINTER_PLUS_EXPR.
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-simd-clone.c (simd_clone_subparts): New function.
5247 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5248 (ipa_simd_modify_function_body): Likewise.
5250 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5251 Alan Hayward <alan.hayward@arm.com>
5252 David Sherwood <david.sherwood@arm.com>
5254 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5255 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5256 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5257 (expand_vector_condition, vector_element): Likewise.
5258 (subparts_gt): New function.
5259 (get_compute_type): Use subparts_gt.
5260 (count_type_subparts): Delete.
5261 (expand_vector_operations_1): Use subparts_gt instead of
5262 count_type_subparts.
5264 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5265 Alan Hayward <alan.hayward@arm.com>
5266 David Sherwood <david.sherwood@arm.com>
5268 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5269 (vect_compile_time_alias): ...this new function. Do the calculation
5270 on poly_ints rather than trees.
5271 (vect_prune_runtime_alias_test_list): Update call accordingly.
5273 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5274 Alan Hayward <alan.hayward@arm.com>
5275 David Sherwood <david.sherwood@arm.com>
5277 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5279 (vect_schedule_slp_instance): Likewise.
5281 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5282 Alan Hayward <alan.hayward@arm.com>
5283 David Sherwood <david.sherwood@arm.com>
5285 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5286 constant and extern definitions for variable-length vectors.
5287 (vect_get_constant_vectors): Note that the number of units
5288 is known to be constant.
5290 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5291 Alan Hayward <alan.hayward@arm.com>
5292 David Sherwood <david.sherwood@arm.com>
5294 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5295 of units as polynomial. Choose between WIDE and NARROW based
5298 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5299 Alan Hayward <alan.hayward@arm.com>
5300 David Sherwood <david.sherwood@arm.com>
5302 * tree-vect-stmts.c (simd_clone_subparts): New function.
5303 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5305 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5306 Alan Hayward <alan.hayward@arm.com>
5307 David Sherwood <david.sherwood@arm.com>
5309 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5310 vectors as polynomial. Use build_index_vector for
5313 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5314 Alan Hayward <alan.hayward@arm.com>
5315 David Sherwood <david.sherwood@arm.com>
5317 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5318 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5319 for variable-length vectors.
5320 (vectorizable_mask_load_store): Treat the number of units as
5321 polynomial, asserting that it is constant if the condition has
5322 already been enforced.
5323 (vectorizable_store, vectorizable_load): Likewise.
5325 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5326 Alan Hayward <alan.hayward@arm.com>
5327 David Sherwood <david.sherwood@arm.com>
5329 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5330 of units as polynomial. Punt if we can't tell at compile time
5331 which vector contains the final result.
5333 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5334 Alan Hayward <alan.hayward@arm.com>
5335 David Sherwood <david.sherwood@arm.com>
5337 * tree-vect-loop.c (vectorizable_induction): Treat the number
5338 of units as polynomial. Punt on SLP inductions. Use an integer
5339 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5340 cast of such a series for variable-length floating-point
5343 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5344 Alan Hayward <alan.hayward@arm.com>
5345 David Sherwood <david.sherwood@arm.com>
5347 * tree.h (build_index_vector): Declare.
5348 * tree.c (build_index_vector): New function.
5349 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5350 of units as polynomial, forcibly converting it to a constant if
5351 vectorizable_reduction has already enforced the condition.
5352 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5353 to create a {1,2,3,...} vector.
5354 (vectorizable_reduction): Treat the number of units as polynomial.
5355 Choose vectype_in based on the largest scalar element size rather
5356 than the smallest number of units. Enforce the restrictions
5359 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5360 Alan Hayward <alan.hayward@arm.com>
5361 David Sherwood <david.sherwood@arm.com>
5363 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5364 number of units as polynomial.
5366 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5367 Alan Hayward <alan.hayward@arm.com>
5368 David Sherwood <david.sherwood@arm.com>
5370 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5371 * target.def (autovectorize_vector_sizes): Return the vector sizes
5372 by pointer, using vector_sizes rather than a bitmask.
5373 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5374 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5375 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5377 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5378 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5379 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5380 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5381 * omp-general.c (omp_max_vf): Likewise.
5382 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5383 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5384 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5385 * tree-vect-slp.c (vect_slp_bb): Likewise.
5386 * doc/tm.texi: Regenerate.
5387 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5389 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5390 the vector size as a poly_uint64 rather than an unsigned int.
5391 (current_vector_size): Change from an unsigned int to a poly_uint64.
5392 (get_vectype_for_scalar_type): Update accordingly.
5393 * tree.h (build_truth_vector_type): Take the size and number of
5394 units as a poly_uint64 rather than an unsigned int.
5395 (build_vector_type): Add a temporary overload that takes
5396 the number of units as a poly_uint64 rather than an unsigned int.
5397 * tree.c (make_vector_type): Likewise.
5398 (build_truth_vector_type): Take the number of units as a poly_uint64
5399 rather than an unsigned int.
5401 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5402 Alan Hayward <alan.hayward@arm.com>
5403 David Sherwood <david.sherwood@arm.com>
5405 * target.def (get_mask_mode): Take the number of units and length
5406 as poly_uint64s rather than unsigned ints.
5407 * targhooks.h (default_get_mask_mode): Update accordingly.
5408 * targhooks.c (default_get_mask_mode): Likewise.
5409 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5410 * doc/tm.texi: Regenerate.
5412 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5413 Alan Hayward <alan.hayward@arm.com>
5414 David Sherwood <david.sherwood@arm.com>
5416 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5417 * omp-general.c (omp_max_vf): Likewise.
5418 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5419 (expand_omp_simd): Handle polynomial safelen.
5420 * omp-low.c (omplow_simd_context): Add a default constructor.
5421 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5422 (lower_rec_simd_input_clauses): Update accordingly.
5423 (lower_rec_input_clauses): Likewise.
5425 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5426 Alan Hayward <alan.hayward@arm.com>
5427 David Sherwood <david.sherwood@arm.com>
5429 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5430 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5431 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5432 (vect_analyze_slp_cost): Likewise.
5433 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5434 (vect_model_load_cost): Likewise.
5436 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5437 Alan Hayward <alan.hayward@arm.com>
5438 David Sherwood <david.sherwood@arm.com>
5440 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5441 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5442 from an unsigned int * to a poly_uint64_pod *.
5443 (calculate_unrolling_factor): New function.
5444 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5446 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5447 Alan Hayward <alan.hayward@arm.com>
5448 David Sherwood <david.sherwood@arm.com>
5450 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5451 from an unsigned int to a poly_uint64.
5452 (_loop_vec_info::slp_unrolling_factor): Likewise.
5453 (_loop_vec_info::vectorization_factor): Change from an int
5455 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5456 (vect_get_num_vectors): New function.
5457 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5458 (vect_get_num_copies): Use vect_get_num_vectors.
5459 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5460 to an unsigned int *.
5461 (vect_analyze_data_refs): Change min_vf from an int * to a
5463 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5464 than an unsigned HOST_WIDE_INT.
5465 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5466 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5467 to an unsigned int *.
5468 (vect_analyze_data_ref_dependences): Likewise.
5469 (vect_compute_data_ref_alignment): Handle polynomial vf.
5470 (vect_enhance_data_refs_alignment): Likewise.
5471 (vect_prune_runtime_alias_test_list): Likewise.
5472 (vect_shift_permute_load_chain): Likewise.
5473 (vect_supportable_dr_alignment): Likewise.
5474 (dependence_distance_ge_vf): Take the vectorization factor as a
5475 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5476 (vect_analyze_data_refs): Change min_vf from an int * to a
5478 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5479 vfm1 as a poly_uint64 rather than an int. Make the same change
5480 for the returned bound_scalar.
5481 (vect_gen_vector_loop_niters): Handle polynomial vf.
5482 (vect_do_peeling): Likewise. Update call to
5483 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5484 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5486 * tree-vect-loop.c (vect_determine_vectorization_factor)
5487 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5488 (vect_get_known_peeling_cost): Likewise.
5489 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5490 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5491 (vect_transform_loop): Likewise. Use the lowest possible VF when
5492 updating the upper bounds of the loop.
5493 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5495 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5496 polynomial unroll factors.
5497 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5498 (vect_make_slp_decision): Likewise.
5499 (vect_supported_load_permutation_p): Likewise, and polynomial
5501 (vect_analyze_slp_cost): Handle polynomial vf.
5502 (vect_slp_analyze_node_operations): Likewise.
5503 (vect_slp_analyze_bb_1): Likewise.
5504 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5505 than an unsigned HOST_WIDE_INT.
5506 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5507 (vectorizable_load): Handle polynomial vf.
5508 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5510 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5512 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5513 Alan Hayward <alan.hayward@arm.com>
5514 David Sherwood <david.sherwood@arm.com>
5516 * match.pd: Handle bit operations involving three constants
5517 and try to fold one pair.
5519 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5521 * tree-vect-loop-manip.c: Include gimple-fold.h.
5522 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5523 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5524 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5525 Add a path that uses a step of VF instead of 1, but disable it
5527 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5528 and niters_no_overflow parameters. Update calls to
5529 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5530 Create a new SSA name if the latter choses to use a ste other
5531 than zero, and return it via niters_vector_mult_vf_var.
5532 * tree-vect-loop.c (vect_transform_loop): Update calls to
5533 vect_do_peeling, vect_gen_vector_loop_niters and
5534 slpeel_make_loop_iterate_ntimes.
5535 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5536 (vect_gen_vector_loop_niters): Update declarations after above changes.
5538 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5540 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5541 128-bit round to integer instructions.
5542 (ceil<mode>2): Likewise.
5543 (btrunc<mode>2): Likewise.
5544 (round<mode>2): Likewise.
5546 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5548 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5549 unaligned VSX load/store on P8/P9.
5550 (expand_block_clear): Allow the use of unaligned VSX
5551 load/store on P8/P9.
5553 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5555 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5557 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5558 swap associated with both a load and a store.
5560 2018-01-02 Andrew Waterman <andrew@sifive.com>
5562 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5563 * config/riscv/riscv.md (clear_cache): Use it.
5565 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5567 * web.c: Remove out-of-date comment.
5569 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5571 * expr.c (fixup_args_size_notes): Check that any existing
5572 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5573 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5574 (emit_single_push_insn): ...here.
5576 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5578 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5579 (const_vector_encoded_nelts): New function.
5580 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5581 (const_vector_int_elt, const_vector_elt): Declare.
5582 * emit-rtl.c (const_vector_int_elt_1): New function.
5583 (const_vector_elt): Likewise.
5584 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5585 of CONST_VECTOR_ELT.
5587 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5589 * expr.c: Include rtx-vector-builder.h.
5590 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5591 directly on the tree encoding.
5592 (const_vector_from_tree): Likewise.
5593 * optabs.c: Include rtx-vector-builder.h.
5594 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5595 sequence of "u" values.
5596 * vec-perm-indices.c: Include rtx-vector-builder.h.
5597 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5598 directly on the vec_perm_indices encoding.
5600 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5602 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5603 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5604 * rtx-vector-builder.h: New file.
5605 * rtx-vector-builder.c: Likewise.
5606 * rtl.h (rtx_def::u2): Add a const_vector field.
5607 (CONST_VECTOR_NPATTERNS): New macro.
5608 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5609 (CONST_VECTOR_DUPLICATE_P): Likewise.
5610 (CONST_VECTOR_STEPPED_P): Likewise.
5611 (CONST_VECTOR_ENCODED_ELT): Likewise.
5612 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5613 (unwrap_const_vec_duplicate): Likewise.
5614 (const_vec_series_p): Check for a non-duplicated vector encoding.
5615 Say that the function only returns true for integer vectors.
5616 * emit-rtl.c: Include rtx-vector-builder.h.
5617 (gen_const_vec_duplicate_1): Delete.
5618 (gen_const_vector): Call gen_const_vec_duplicate instead of
5619 gen_const_vec_duplicate_1.
5620 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5621 (gen_const_vec_duplicate): Use rtx_vector_builder.
5622 (gen_const_vec_series): Likewise.
5623 (gen_rtx_CONST_VECTOR): Likewise.
5624 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5625 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5626 Build a new vector rather than modifying a CONST_VECTOR in-place.
5627 (handle_special_swappables): Update call accordingly.
5628 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5629 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5630 Build a new vector rather than modifying a CONST_VECTOR in-place.
5631 (handle_special_swappables): Update call accordingly.
5633 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5635 * simplify-rtx.c (simplify_const_binary_operation): Use
5636 CONST_VECTOR_ELT instead of XVECEXP.
5638 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5640 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5641 the selector elements to be different from the data elements
5642 if the selector is a VECTOR_CST.
5643 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5644 ssizetype for the selector.
5646 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5648 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5649 before testing each element individually.
5650 * tree-vect-generic.c (lower_vec_perm): Likewise.
5652 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5654 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5655 * selftest-run-tests.c (selftest::run_tests): Call it.
5656 * vector-builder.h (vector_builder::operator ==): New function.
5657 (vector_builder::operator !=): Likewise.
5658 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5659 (vec_perm_indices::all_from_input_p): New function.
5660 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5661 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5662 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5663 instead of reading the VECTOR_CST directly. Detect whether both
5664 vector inputs are the same before constructing the vec_perm_indices,
5665 and update the number of inputs argument accordingly. Use the
5666 utility functions added above. Only construct sel2 if we need to.
5668 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5670 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5671 the broadcast of the low byte.
5672 (expand_mult_highpart): Use an explicit encoding for the permutes.
5673 * optabs-query.c (can_mult_highpart_p): Likewise.
5674 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5675 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5676 (vectorizable_bswap): Likewise.
5677 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5678 explicit encoding for the power-of-2 permutes.
5679 (vect_permute_store_chain): Likewise.
5680 (vect_grouped_load_supported): Likewise.
5681 (vect_permute_load_chain): Likewise.
5683 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5685 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5686 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5687 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5688 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5689 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5690 (vect_gen_perm_mask_any): Likewise.
5692 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5694 * int-vector-builder.h: New file.
5695 * vec-perm-indices.h: Include int-vector-builder.h.
5696 (vec_perm_indices): Redefine as an int_vector_builder.
5697 (auto_vec_perm_indices): Delete.
5698 (vec_perm_builder): Redefine as a stand-alone class.
5699 (vec_perm_indices::vec_perm_indices): New function.
5700 (vec_perm_indices::clamp): Likewise.
5701 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5702 (vec_perm_indices::new_vector): New function.
5703 (vec_perm_indices::new_expanded_vector): Update for new
5704 vec_perm_indices class.
5705 (vec_perm_indices::rotate_inputs): New function.
5706 (vec_perm_indices::all_in_range_p): Operate directly on the
5707 encoded form, without computing elided elements.
5708 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5709 encoding. Update for new vec_perm_indices class.
5710 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5711 the given vec_perm_builder.
5712 (expand_vec_perm_var): Update vec_perm_builder constructor.
5713 (expand_mult_highpart): Use vec_perm_builder instead of
5714 auto_vec_perm_indices.
5715 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5716 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5717 or double series encoding as appropriate.
5718 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5719 vec_perm_indices instead of auto_vec_perm_indices.
5720 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5721 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5722 (vect_permute_store_chain): Likewise.
5723 (vect_grouped_load_supported): Likewise.
5724 (vect_permute_load_chain): Likewise.
5725 (vect_shift_permute_load_chain): Likewise.
5726 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5727 (vect_transform_slp_perm_load): Likewise.
5728 (vect_schedule_slp_instance): Likewise.
5729 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5730 (vectorizable_mask_load_store): Likewise.
5731 (vectorizable_bswap): Likewise.
5732 (vectorizable_store): Likewise.
5733 (vectorizable_load): Likewise.
5734 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5735 vec_perm_indices instead of auto_vec_perm_indices. Use
5736 tree_to_vec_perm_builder to read the vector from a tree.
5737 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5738 vec_perm_builder instead of a vec_perm_indices.
5739 (have_whole_vector_shift): Use vec_perm_builder and
5740 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5741 truncation to calc_vec_perm_mask_for_shift.
5742 (vect_create_epilog_for_reduction): Likewise.
5743 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5744 from auto_vec_perm_indices to vec_perm_indices.
5745 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5746 instead of changing individual elements.
5747 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5748 the vector in d.perm.
5749 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5750 from auto_vec_perm_indices to vec_perm_indices.
5751 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5752 instead of changing individual elements.
5753 (arm_vectorize_vec_perm_const): Use new_vector to install
5754 the vector in d.perm.
5755 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5756 Update vec_perm_builder constructor.
5757 (rs6000_expand_interleave): Likewise.
5758 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5759 (rs6000_expand_interleave): Likewise.
5761 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5763 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5764 to qimode could truncate the indices.
5765 * optabs.c (expand_vec_perm_var): Likewise.
5767 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5769 * Makefile.in (OBJS): Add vec-perm-indices.o.
5770 * vec-perm-indices.h: New file.
5771 * vec-perm-indices.c: Likewise.
5772 * target.h (vec_perm_indices): Replace with a forward class
5774 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5775 * optabs.h: Include vec-perm-indices.h.
5776 (expand_vec_perm): Delete.
5777 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5778 (expand_vec_perm_const): Declare.
5779 * target.def (vec_perm_const_ok): Replace with...
5780 (vec_perm_const): ...this new hook.
5781 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5782 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5783 * doc/tm.texi: Regenerate.
5784 * optabs.def (vec_perm_const): Delete.
5785 * doc/md.texi (vec_perm_const): Likewise.
5786 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5787 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5788 expand_vec_perm for constant permutation vectors. Assert that
5789 the mode of variable permutation vectors is the integer equivalent
5790 of the mode that is being permuted.
5791 * optabs-query.h (selector_fits_mode_p): Declare.
5792 * optabs-query.c: Include vec-perm-indices.h.
5793 (selector_fits_mode_p): New function.
5794 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5795 is defined, instead of checking whether the vec_perm_const_optab
5796 exists. Use targetm.vectorize.vec_perm_const instead of
5797 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5798 fit in the vector mode before using a variable permute.
5799 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5800 vec_perm_indices instead of an rtx.
5801 (expand_vec_perm): Replace with...
5802 (expand_vec_perm_const): ...this new function. Take the selector
5803 as a vec_perm_indices rather than an rtx. Also take the mode of
5804 the selector. Update call to shift_amt_for_vec_perm_mask.
5805 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5806 Use vec_perm_indices::new_expanded_vector to expand the original
5807 selector into bytes. Check whether the indices fit in the vector
5808 mode before using a variable permute.
5809 (expand_vec_perm_var): Make global.
5810 (expand_mult_highpart): Use expand_vec_perm_const.
5811 * fold-const.c: Includes vec-perm-indices.h.
5812 * tree-ssa-forwprop.c: Likewise.
5813 * tree-vect-data-refs.c: Likewise.
5814 * tree-vect-generic.c: Likewise.
5815 * tree-vect-loop.c: Likewise.
5816 * tree-vect-slp.c: Likewise.
5817 * tree-vect-stmts.c: Likewise.
5818 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5820 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5821 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5822 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5823 (aarch64_vectorize_vec_perm_const): ...this new function.
5824 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5825 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5826 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5827 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5828 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5829 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5830 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5832 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5833 check for NEON modes.
5834 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5835 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5836 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5837 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5839 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5840 the old VEC_PERM_CONST conditions.
5841 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5842 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5843 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5844 (ia64_vectorize_vec_perm_const_ok): Merge into...
5845 (ia64_vectorize_vec_perm_const): ...this new function.
5846 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5847 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5848 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5849 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5850 * config/mips/mips.c (mips_expand_vec_perm_const)
5851 (mips_vectorize_vec_perm_const_ok): Merge into...
5852 (mips_vectorize_vec_perm_const): ...this new function.
5853 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5854 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5855 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5856 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5857 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5858 (rs6000_expand_vec_perm_const): Delete.
5859 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5861 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5862 (altivec_expand_vec_perm_const_le): Take each operand individually.
5863 Operate on constant selectors rather than rtxes.
5864 (altivec_expand_vec_perm_const): Likewise. Update call to
5865 altivec_expand_vec_perm_const_le.
5866 (rs6000_expand_vec_perm_const): Delete.
5867 (rs6000_vectorize_vec_perm_const_ok): Delete.
5868 (rs6000_vectorize_vec_perm_const): New function.
5869 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5870 an element count and rtx array.
5871 (rs6000_expand_extract_even): Update call accordingly.
5872 (rs6000_expand_interleave): Likewise.
5873 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5874 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5875 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5876 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5877 (rs6000_expand_vec_perm_const): Delete.
5878 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5879 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5880 (altivec_expand_vec_perm_const_le): Take each operand individually.
5881 Operate on constant selectors rather than rtxes.
5882 (altivec_expand_vec_perm_const): Likewise. Update call to
5883 altivec_expand_vec_perm_const_le.
5884 (rs6000_expand_vec_perm_const): Delete.
5885 (rs6000_vectorize_vec_perm_const_ok): Delete.
5886 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5887 reference to the SPE evmerge intructions.
5888 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5889 an element count and rtx array.
5890 (rs6000_expand_extract_even): Update call accordingly.
5891 (rs6000_expand_interleave): Likewise.
5892 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5893 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5895 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5897 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5899 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5900 vector mode and that that mode matches the mode of the data
5902 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5903 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5904 directly using expand_vec_perm_1 when forcing selectors into
5906 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5908 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5910 * optabs-query.h (can_vec_perm_p): Delete.
5911 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5912 * optabs-query.c (can_vec_perm_p): Split into...
5913 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5914 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5915 particular selector is valid.
5916 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5917 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5918 (vect_grouped_load_supported): Likewise.
5919 (vect_shift_permute_load_chain): Likewise.
5920 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5921 (vect_transform_slp_perm_load): Likewise.
5922 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5923 (vectorizable_bswap): Likewise.
5924 (vect_gen_perm_mask_checked): Likewise.
5925 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5926 implementations of variable permutation vectors into account
5927 when deciding which selector to use.
5928 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5929 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5930 with a false third argument.
5931 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5932 to test whether the constant selector is valid and can_vec_perm_var_p
5933 to test whether a variable selector is valid.
5935 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5937 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5938 * optabs-query.c (can_vec_perm_p): Likewise.
5939 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5940 instead of vec_perm_indices.
5941 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5942 (vect_gen_perm_mask_checked): Likewise,
5943 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5944 (vect_gen_perm_mask_checked): Likewise,
5946 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5948 * optabs-query.h (qimode_for_vec_perm): Declare.
5949 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5950 (qimode_for_vec_perm): ...this new function.
5951 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5953 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5955 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5956 does not have a conditional at the top.
5958 2018-01-02 Richard Biener <rguenther@suse.de>
5960 * ipa-inline.c (big_speedup_p): Fix expression.
5962 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5965 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5968 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5972 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5973 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5974 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5975 cond_taken_branch_cost 3->4.
5977 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5979 PR tree-optimization/83581
5980 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5981 TODO_cleanup_cfg if any changes have been made.
5984 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5985 convert_modes if target mode has the right side, but different mode
5989 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5990 last argument when extracting from CONCAT. If either from_real or
5991 from_imag is NULL, use expansion through memory. If result is not
5992 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5993 the parts directly to inner mode, if even that fails, use expansion
5997 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5998 check for bswap in mode rather than HImode and use that in expand_unop
6001 Copyright (C) 2018 Free Software Foundation, Inc.
6003 Copying and distribution of this file, with or without modification,
6004 are permitted in any medium without royalty provided the copyright
6005 notice and this notice are preserved.