1 2018-03-05 Jakub Jelinek <jakub@redhat.com>
4 * config/i386/i386.c (ix86_function_ok_for_sibcall): Check for
5 regparm >= 3 with no arg reg available also for calls with
6 flag_force_indirect_call. Pass decl to ix86_function_regparm.
9 * config/i386/sse.md (*<code><mode>3): Replace <mask_prefix3> with
11 (*<plusminus_insn><mode>3): Likewise. Remove <mask_operand3> uses.
13 2018-03-05 Peter Bergner <bergner@vnet.ibm.com>
16 * config/rs6000/vector.md (mov<mode>): Disallow altivec memory operands.
18 2018-03-05 Richard Biener <rguenther@suse.de>
20 PR tree-optimization/84486
21 * tree-ssa-pre.c (create_expression_by_pieces): Remove dead code.
22 When inserting a __builtin_assume_aligned call set the LHS
23 SSA name alignment info accordingly.
25 2018-03-05 Wilco Dijkstra <wdijkstr@arm.com>
27 PR tree-optimization/84114
28 * config/aarch64/aarch64.c (aarch64_reassociation_width)
29 Avoid reassociation of FLOAT_MODE addition.
31 2018-03-05 Olga Makhotina <olga.makhotina@intel.com>
33 * common/config/i386/i386-common.c (OPTION_MASK_ISA_PCONFIG_SET,
34 OPTION_MASK_ISA_PCONFIG_UNSET, OPTION_MASK_ISA_WBNOINVD_SET,
35 OPTION_MASK_ISA_WBNOINVD_UNSET): New definitions.
36 (ix86_handle_option): Handle -mpconfig and -mwbnoinvd.
37 * config.gcc (pconfigintrin.h, wbnoinvdintrin.h) : Add headers.
38 * config/i386/cpuid.h (bit_PCONFIG, bit_WBNOINVD): New.
39 * config/i386/driver-i386.c (host_detect_local_cpu): Detect -mpconfig
41 * config/i386/i386-builtin.def (__builtin_ia32_wbnoinvd,
42 __builtin_ia32_wbinvd): New builtins.
44 * config/i386/i386-c.c (__WBNOINVD__, __PCONFIG__): New.
46 * config/i386/i386.c (ix86_target_string): Add -mpconfig and -mwbnoinvd.
47 (ix86_valid_target_attribute_inner_p): Ditto.
48 (ix86_init_mmx_sse_builtins): Add special_args2.
49 * config/i386/i386.h (TARGET_PCONFIG, TARGET_PCONFIG_P, TARGET_WBNOINVD,
50 TARGET_WBNOINVD_P): New.
51 * config/i386/i386.md (UNSPECV_WBINVD, UNSPECV_WBNOINVD): New.
52 (define_insn "wbinvd", define_insn "wbnoinvd"): New.
53 * config/i386/i386.opt: Add -mpconfig and -mwbnoinvd.
54 * config/i386/immintrin.h (_wbinvd): New intrinsic.
55 * config/i386/pconfigintrin.h: New file.
56 * config/i386/wbnoinvdintrin.h: Ditto.
57 * config/i386/x86intrin.h: Add headers pconfigintrin.h and wbnoinvdintrin.h.
58 * doc/invoke.texi (-mpconfig, -mwbnoinvd): New.
60 2018-03-05 Richard Biener <rguenther@suse.de>
62 PR tree-optimization/84670
63 * tree-ssa-pre.c (struct bb_bitmap_sets): Add visited_with_visited_succs
65 (BB_VISITED_WITH_VISITED_SUCCS): New define.
66 (compute_antic): Initialize BB_VISITED_WITH_VISITED_SUCCS.
67 (compute_antic_aux): Only assert the number of values in ANTIC_IN
68 doesn't grow if all successors (recursively) were visited at least
71 2018-03-05 Richard Biener <rguenther@suse.de>
73 PR tree-optimization/84650
74 * tree-ssa-loop-im.c (pass_lim::execute): Reset the SCEV cache
75 if executed in the loop pipeline.
77 2018-03-05 Sandra Loosemore <sandra@codesourcery.com>
79 * doc/configfiles.texi (Configuration Files): Move info about
80 conditionalizing $target-protos.h to...
81 * doc/sourcebuild.texi (Back End): Here. Explain how $target.h
82 differs from $target-protos.h.
84 2018-03-05 Kito Cheng <kito.cheng@gmail.com>
85 Chung-Ju Wu <jasonwucj@gmail.com>
87 * config/nds32/nds32-protos.h (nds32_expand_setmem): Declare.
88 * config/nds32/nds32-multiple.md (setmemsi): Define.
89 * config/nds32/nds32-memory-manipulation.c
90 (nds32_gen_dup_4_byte_to_word_value): New.
91 (emit_setmem_word_loop): New.
92 (emit_setmem_byte_loop): New.
93 (nds32_expand_setmem_loop): New.
94 (nds32_expand_setmem_loop_v3m): New.
95 (nds32_expand_setmem_unroll): New.
96 (nds32_expand_setmem): New.
98 2018-03-04 Kito Cheng <kito.cheng@gmail.com>
99 Chung-Ju Wu <jasonwucj@gmail.com>
101 * config/nds32/nds32-memory-manipulation.c
102 (nds32_emit_load_store): New.
103 (nds32_emit_post_inc_load_store): New.
104 (nds32_emit_mem_move): New.
105 (nds32_emit_mem_move_block): New.
106 (nds32_expand_movmemsi_loop_unknown_size): New.
107 (nds32_expand_movmemsi_loop_known_size): New.
108 (nds32_expand_movmemsi_loop): New.
109 (nds32_expand_movmemsi_unroll): New.
110 (nds32_expand_movmemqi): Rename ...
111 (nds32_expand_movmemsi): ... to this.
112 * config/nds32/nds32-multiple.md (movmemqi): Rename ...
113 (movmemsi): ... to this.
114 * config/nds32/nds32-protos.h (nds32_expand_movmemqi): Rename ...
115 (nds32_expand_movmemsi): ... to this.
117 2018-03-04 Kito Cheng <kito.cheng@gmail.com>
118 Monk Chiang <sh.chiang04@gmail.com>
119 Chung-Ju Wu <jasonwucj@gmail.com>
121 * config/nds32/nds32-protos.h
122 (nds32_expand_load_multiple): New arguments.
123 (nds32_expand_store_multiple): Ditto.
124 (nds32_valid_multiple_load_store): Rename ...
125 (nds32_valid_multiple_load_store_p): ... to this.
126 * config/nds32/nds32-memory-manipulation.c
127 (nds32_expand_load_multiple): Refine implementation.
128 (nds32_expand_store_multiple): Ditto.
129 * config/nds32/nds32-multiple.md
130 (load_multiple): Update nds32_expand_load_multiple interface.
131 (store_multiple): Update nds32_expand_store_multiple interface.
132 * config/nds32/nds32-predicates.c
133 (nds32_valid_multiple_load_store): Rename ...
134 (nds32_valid_multiple_load_store_p): ... to this and refine
136 * config/nds32/predicates.md
137 (nds32_load_multiple_and_update_address_operation): New predicate.
138 (nds32_store_multiple_and_update_address_operation): New predicate.
140 2018-03-04 Kito Cheng <kito.cheng@gmail.com>
141 Chung-Ju Wu <jasonwucj@gmail.com>
143 * config/nds32/nds32.md (type): Add load_multiple and store_multiple.
144 (combo): New attribute.
145 * config/nds32/nds32-multiple.md: Refine patterns with new attributes.
147 2018-03-03 Chung-Ju Wu <jasonwucj@gmail.com>
149 * config/nds32/nds32.opt: Change -mcmodel= default value.
151 2018-03-03 Kito Cheng <kito.cheng@gmail.com>
152 Monk Chiang <sh.chiang04@gmail.com>
153 Chung-Ju Wu <jasonwucj@gmail.com>
155 * config/nds32/constants.md (unspec_element): New enum.
156 * config/nds32/constraints.md (Umw): New constraint.
157 * config/nds32/nds32-intrinsic.c: Add more builtin functions.
158 * config/nds32/nds32-intrinsic.md: Likewise.
159 * config/nds32/nds32-md-auxiliary.c (nds32_regno_to_enable4): New.
160 (nds32_valid_smw_lwm_base_p): New.
161 (nds32_output_smw_single_word): New.
162 (nds32_output_lmw_single_word): New.
163 (nds32_expand_unaligned_load): New.
164 (nds32_expand_unaligned_store): New.
165 * config/nds32/nds32-protos.h (nds32_valid_smw_lwm_base_p): Declare.
166 (nds32_output_smw_single_word): Declare.
167 (nds32_output_lmw_single_word): Declare.
168 (nds32_expand_unaligned_load): Declare.
169 (nds32_expand_unaligned_store): Declare.
170 * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_UALOAD_HW,
171 NDS32_BUILTIN_UALOAD_W, NDS32_BUILTIN_UALOAD_DW,
172 NDS32_BUILTIN_UASTORE_HW, NDS32_BUILTIN_UASTORE_W,
173 NDS32_BUILTIN_UASTORE_DW.
174 * config/nds32/predicates.md (nds32_lmw_smw_base_operand): New
177 2018-03-03 Monk Chiang <sh.chiang04@gmail.com>
178 Kito Cheng <kito.cheng@gmail.com>
179 Chung-Ju Wu <jasonwucj@gmail.com>
181 * config/nds32/nds32-intrinsic.c
182 (nds32_expand_builtin_null_ftype_reg): Delete.
183 (nds32_expand_builtin_reg_ftype_imm): Ditto.
184 (nds32_expand_builtin_null_ftype_reg_imm): Ditto.
185 (nds32_read_argument): New.
186 (nds32_legitimize_target): Ditto.
187 (nds32_legitimize_argument): Ditto.
188 (nds32_check_constant_argument): Ditto.
189 (nds32_expand_unop_builtin): Ditto.
190 (nds32_expand_unopimm_builtin): Ditto.
191 (nds32_expand_binop_builtin): Ditto.
192 (nds32_builtin_decl_impl): Ditto.
193 (builtin_description): Ditto.
194 (nds32_expand_builtin_impl): Rewrite with new infrastructure.
195 (nds32_init_builtins_impl): Ditto.
196 * config/nds32/nds32.c (TARGET_BUILTIN_DECL): Define.
197 (nds32_builtin_decl): New.
198 * config/nds32/nds32.h (nds32_builtins): Add NDS32_BUILTIN_COUNT.
199 * config/nds32/nds32-protos.h (nds32_builtin_decl_impl): Declare.
201 2018-03-02 Jeff Law <law@redhat.com>
203 * reorg.c (stop_search_p): Handle DEBUG_INSN.
204 (redundant_insn, fill_simple_delay_slots): Likewise.
205 (fill_slots_from_thread): Likewise.
206 * resource.c (mark_referenced_resources): Likewise.
207 (mark_set_resources, find_dead_or_set_registers): Likewise.
209 2018-03-02 Jakub Jelinek <jakub@redhat.com>
211 * substring-locations.h (format_warning_va): Formatting fix for
213 (format_warning_at_substring): Fix up ATTRIBUTE_GCC_DIAG second
215 (format_warning_n_va, format_warning_at_substring_n): New prototypes.
216 * substring-locations.c: Include intl.h.
217 (format_warning_va): Turned into small wrapper around
218 format_warning_n_va, renamed to ...
219 (format_warning_n_va): ... this, add N and PLURAL_GMSGID arguments,
220 rename GMSGID to SINGULAR_GMSGID, if SINGULAR_GMSGID != PLURAL_GMSGID,
222 (format_warning_at_substring_n): New function.
223 * gimple-ssa-sprintf.c: Remove GCC diagnostic ignored pragma.
224 (fmtwarn): Add ATTRIBUTE_GCC_DIAG. Turn into a copy of
225 format_warning_at_substring with just a shorter name instead of
226 const function pointer.
227 (fmtwarn_n): New function.
228 (maybe_warn, format_directive, parse_directive): Use fmtwarn_n where
229 appropriate, get rid of all the fmtstr temporaries, move conditionals
230 with G_() wrapped string literals directly into fmtwarn arguments,
231 cast dir.len to (int), formatting fixes.
233 2018-03-02 Thomas Schwinge <thomas@codesourcery.com>
235 * doc/invoke.texi: Remove "Cilk Plus" references.
237 2018-03-02 Jakub Jelinek <jakub@redhat.com>
238 Richard Biener <rguenther@suse.de>
241 * expr.c (expand_expr_real_1) <case CALL_EXPR>: Don't emit diagnostics
242 for error or warning attributes if CALL_FROM_THUNK_P is set.
245 2018-03-02 Jakub Jelinek <jakub@redhat.com>
248 * config/pa/pa.h (TARGET_CPU_CPP_BUILTINS): Predefine
249 __SIZEOF_128__ macro if HPUX_LONG_DOUBLE_LIBRARY.
252 * config/ia64/ia64.h (TARGET_CPU_CPP_BUILTINS): Predefine
253 __SIZEOF_{FPREG,FLOAT{80,128}}__ macros.
255 * predict.c (test_prediction_value_range): Use PROB_UNINITIALIZED
256 instead of -1U in last predictors element's probability member.
258 2018-03-02 Eric Botcazou <ebotcazou@adacore.com>
261 * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
262 arguments if they are comparable.
264 2018-03-02 Richard Sandiford <richard.sandiford@linaro.org>
266 PR tree-optimization/84634
267 * tree-vect-stmts.c (vectorizable_store, vectorizable_load): Replace
268 masks and masked_loop_p with a single loop_masks, making sure it's
269 null for bb vectorization.
271 2018-03-02 Richard Sandiford <richard.sandiford@linaro.org>
273 * tree-vect-data-refs.c (vect_analyze_data_ref_dependence)
274 (vect_analyze_data_ref_access): Use loop->safe_len rather than
275 loop->force_vectorize to check whether there is no alias.
277 2018-03-02 Jakub Jelinek <jakub@redhat.com>
280 * rtl.h (prev_real_nondebug_insn, next_real_nondebug_insn): New
282 * emit-rtl.c (next_real_insn, prev_real_insn): Fix up function
284 (next_real_nondebug_insn, prev_real_nondebug_insn): New functions.
285 * cfgcleanup.c (try_head_merge_bb): Use prev_real_nondebug_insn
286 instead of a loop around prev_real_insn.
287 * combine.c (move_deaths): Use prev_real_nondebug_insn instead of
291 * config/i386/i386.c (ix86_print_operand): Use conditional
292 output_operand_lossage instead of gcc_assert if CONST_VECTOR is not
295 2018-03-02 Richard Biener <rguenther@suse.de>
297 PR tree-optimization/84427
298 * tree-ssa-pre.c (bitmap_remove_expr_from_set): Remove.
299 (bitmap_set_subtract_values): Rewrite to handle multiple
302 (prune_clobbered_mems): Likewise.
303 (phi_translate): Take edge instead of pred/phiblock.
304 (phi_translate_1): Likewise.
305 (phi_translate_set): Likewise. Insert all translated
306 exprs for a value into the set, keeping possibly multiple
307 expressions per value.
308 (compute_antic_aux): Adjust for phi_translate changes.
309 When intersecting union the expressions and prune those
310 not in the final value set, keeping possibly multiple
311 expressions per value. Do not use value-insertion
312 for unioning ANTIC_OUT U EXP_GEN - TMP_GEN but merge
313 all expressions. Add verification that the value-sets
314 only shrink during iteration.
315 (compute_partial_antic_aux): Adjust for the phi_translate changes.
316 (do_pre_regular_insertion): Likewise.
317 (do_pre_partial_partial_insertion): Likewise.
319 2018-03-02 Richard Biener <rguenther@suse.de>
322 * config/darwin.c (saved_debug_info_level): New static global.
323 (darwin_asm_lto_start): Disable debug info generation for LTO out.
324 (darwin_asm_lto_end): Restore debug info generation settings.
326 2018-03-01 Martin Liska <mliska@suse.cz>
329 * sanopt.c (sanitize_rewrite_addressable_params): Do not handle
332 2018-03-01 Richard Biener <rguenther@suse.de>
335 * dwarf2out.c (gen_variable_die): Properly handle late VLA
336 type annotation with LTO when debug was disabled at compile-time.
338 2018-03-01 Matthew Fortune <mfortune@gmail.com>
340 * config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
342 (mips_final_postscan_insn): Likewise.
344 2018-03-01 Richard Sandiford <richard.sandiford@linaro.org>
346 PR rtl-optimization/84528
347 * alias.c (init_alias_target): Add commentary.
348 (init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
349 a unique base value if the frame pointer is not eliminated
350 to the stack pointer.
352 2018-03-01 Tom de Vries <tom@codesourcery.com>
354 PR rtl-optimization/83327
355 * lra-int.h (hard_regs_spilled_into): Declare.
356 * lra.c (hard_regs_spilled_into): Define.
357 (init_reg_info): Init hard_regs_spilled_into.
358 * lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
359 * lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
360 (process_bb_lives): Handle hard_regs_spilled_into.
361 (lra_create_live_ranges_1): Before doing liveness propagation, clear
362 regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.
364 2018-02-28 David Edelsohn <dje.gcc@gmail.com>
366 * config.gcc (powerpc-ibm-aix7.1.*): New stanza.
367 (powerpc-ibm-aix[789]*): Default to AIX 7.2.
368 * config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
369 * config/rs6000/aix72.h: New file.
371 2018-02-28 Jakub Jelinek <jakub@redhat.com>
373 * gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
374 instead of warning_at with conditional singular and plural messages
378 * stor-layout.c (update_alignment_for_field): For
379 targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
380 && !DECL_PACKED (field), do the alignment update, just use
381 only desired_align instead of MAX (type_align, desired_align)
383 (place_field): Don't do known_align < desired_align handling
384 early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
385 is non-NULL, instead do it after rli->prev_field handling and
386 only if not within a bitfield word. For DECL_PACKED (field)
387 use type_align of BITS_PER_UNIT.
389 2018-02-28 Eric Botcazou <ebotcazou@adacore.com>
391 * config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
392 superfluous parentheses and trailing spaces.
394 2018-02-28 Richard Biener <rguenther@suse.de>
396 PR tree-optimization/84584
397 * graphite-scop-detection.c (scop_detection::add_scop): Discard
398 SCoPs with fake exit edge.
400 2018-02-28 Martin Liska <mliska@suse.cz>
403 * timevar.c (timer::print): Fix format to properly print 100%
406 2018-02-28 Richard Biener <rguenther@suse.de>
409 * genmatch.c (capture_info::walk_match): Do not mark
410 captured expressions without operands as expr_p given
411 they act more like predicates and should be subject to
412 "lost tail" side-effect preserving.
414 2018-02-28 Alexandre Oliva <aoliva@redhat.com>
416 PR rtl-optimization/81611
417 * auto-inc-dec.c (attempt_change): Move dead note from
418 mem_insn if it's the next use of regno
419 (find_address): Take address use of reg holding
420 non-incremented value. Add parm to limit search to the named
422 (merge_in_block): Attempt to use a mem insn that is the next
423 use of the original regno.
425 2018-02-27 Martin Sebor <msebor@redhat.com>
428 * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
429 * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
431 2018-02-27 Martin Sebor <msebor@redhat.com>
434 * diagnostic-core.h (warning_n, error_n, inform_n): Change
435 n argument to unsigned HOST_WIDE_INT.
436 * diagnostic.c (warning_n, error_n, inform_n): Ditto.
437 (diagnostic_n_impl): Ditto. Handle arguments in excess of LONG_MAX.
438 * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
439 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
441 2018-02-27 Richard Biener <rguenther@suse.de>
443 PR tree-optimization/84512
444 * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
445 Do not use the estimate returned from record_stmt_cost for
446 the scalar iteration cost but sum properly using add_stmt_cost.
448 2018-02-27 Richard Biener <rguenther@suse.de>
450 PR tree-optimization/84466
451 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
452 Adjust last change to less strictly validate use operands.
454 2018-02-27 Martin Liska <mliska@suse.cz>
456 PR gcov-profile/84548
457 * gcov.c (process_file): Allow partial overlap and consider it
458 also as group functions.
459 (output_lines): Properly calculate range of lines for a group.
461 2018-02-27 Martin Liska <mliska@suse.cz>
463 * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
464 'ggc' suffixes. Change first column width.
465 (timer::print): Fix formatting of the column.
467 2018-02-27 Alexandre Oliva <aoliva@redhat.com>
469 * tree-ssa-live.c (remove_unused_scope_block_p): Do not
470 preserve inline entry blocks for the sake of debug inline
471 entry point markers alone.
472 (remove_unused_locals): Suggest in comments a better place to
473 force the preservation of inline entry blocks that are
474 otherwise unused, but do not preserve them.
476 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
478 * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
480 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
483 * config/i386/constraints.md (Bs): Replace
484 ix86_indirect_branch_register with
485 TARGET_INDIRECT_BRANCH_REGISTER.
487 * config/i386/i386.md (indirect_jump): Likewise.
488 (tablejump): Likewise.
489 (*sibcall_memory): Likewise.
490 (*sibcall_value_memory): Likewise.
491 Peepholes of indirect call and jump via memory: Likewise.
492 (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
493 (*sibcall_value_GOT_32): Likewise.
494 * config/i386/predicates.md (indirect_branch_operand): Likewise.
495 (GOT_memory_operand): Likewise.
496 (call_insn_operand): Likewise.
497 (sibcall_insn_operand): Likewise.
498 (GOT32_symbol_operand): Likewise.
499 * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
501 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
503 PR rtl-optimization/83496
504 * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
505 booleans to RTXes. Call fix_reg_dead_note on every non-null element.
506 (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
507 redundant insn, if any.
508 (relax_delay_slots): Likewise.
509 (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
511 2018-02-26 Richard Sandiford <richard.sandiford@linaro.org>
513 PR tree-optimization/83965
514 * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
515 that grouped statements are part of a reduction chain. Return
516 true if the statement is not marked as a reduction itself but
518 (vect_recog_dot_prod_pattern): Don't check whether the statement
519 is part of a group here.
520 (vect_recog_sad_pattern): Likewise.
521 (vect_recog_widen_sum_pattern): Likewise.
523 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
526 * final.c (rest_of_clean_state): Also look for calls inside sequences.
528 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
531 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
533 (ix86_output_indirect_function_return): New prototype.
534 (ix86_split_simple_return_pop_internal): Likewise.
535 * config/i386/i386.c (indirect_return_via_cx): New.
536 (indirect_return_via_cx_bnd): Likewise.
537 (indirect_thunk_name): Handle return va CX_REG.
538 (output_indirect_thunk_function): Create alias for
539 __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
540 (ix86_output_indirect_jmp): Remove the bool argument.
541 (ix86_output_indirect_function_return): New function.
542 (ix86_split_simple_return_pop_internal): Likewise.
543 * config/i386/i386.md (*indirect_jump): Don't pass false
544 to ix86_output_indirect_jmp.
545 (*tablejump_1): Likewise.
546 (simple_return_pop_internal): Change it to define_insn_and_split.
547 Call ix86_split_simple_return_pop_internal to split it for
549 (simple_return_indirect_internal): Call
550 ix86_output_indirect_function_return instead of
551 ix86_output_indirect_jmp.
553 2018-02-26 Jakub Jelinek <jakub@redhat.com>
556 * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
557 memset and value initialization afterwards.
559 2018-02-26 Christophe Lyon <christophe.lyon@linaro.org>
561 * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
563 2018-02-26 Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
566 * common/config/aarch64/aarch64-common.c
567 (aarch_option_optimization_table[]): Switch
568 off fomit-frame-pointer
570 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
571 Chung-Ju Wu <jasonwucj@gmail.com>
573 * config/nds32/nds32-multiple.md (load_multiple): Disallow
575 (store_multiple): Ditto.
577 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
579 * config.gcc: Add --with-cpu support for nds32 target.
580 * config/nds32/nds32-opts.h (nds32_cpu_type): New.
581 * config/nds32/nds32.opt: Add -mcpu= option.
583 2018-02-25 Segher Boessenkool <segher@kernel.crashing.org>
585 * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
586 isel=yes): Warn for these deprecated options.
588 2018-02-23 David Edelsohn <dje.gcc@gmail.com>
590 * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
591 ISA_2_5_MASKS_EMBEDDED.
593 2018-02-23 Jakub Jelinek <jakub@redhat.com>
595 * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
596 p->max as pointers rather than using iterative_hash_expr.
598 2018-02-23 Carl Love <cel@us.ibm.com>
600 * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
601 macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
603 * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
604 P8V_BUILTIN_VEC_VSIGNED2. Change VSX_BUILTIN_VEC_VUNSIGNED2 to
605 P8V_BUILTIN_VEC_VUNSIGNED2.
607 2018-02-22 Vladimir Makarov <vmakarov@redhat.com>
610 * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
611 * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
613 * lra-constraints.c (curr_insn_transform): Set up
614 LRA_NON_CLOBBERED_ALT for moves processed on the fast path. Use
616 (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
617 * lra-eliminations.c (spill_pseudos): Ditto.
618 (process_insn_for_elimination): Ditto.
619 * lra-lives.c (reg_early_clobber_p): Use the new macros.
620 * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
621 LRA_NON_CLOBBERED_ALT.
623 2018-02-22 Martin Sebor <msebor@redhat.com>
625 PR tree-optimization/84480
626 * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
627 to maybe_diag_stxncpy_trunc. Call it.
628 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
629 from gimple_fold_builtin_strcpy. Print inlining stack.
630 (handle_builtin_stxncpy): Print inlining stack.
631 * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
633 2018-02-22 H.J. Lu <hongjiu.lu@intel.com>
636 * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
637 error when -mindirect-branch=thunk-extern, -fcf-protection=branch
638 and -fcheck-pointer-bounds are used together.
639 (indirect_thunk_prefix): New enum.
640 (indirect_thunk_need_prefix): New function.
641 (indirect_thunk_name): Replace need_bnd_p with need_prefix. Use
642 "_nt" instead of "_bnd" for NOTRACK prefix.
643 (output_indirect_thunk): Replace need_bnd_p with need_prefix.
644 (output_indirect_thunk_function): Likewise.
646 (ix86_code_end): Update output_indirect_thunk_function calls.
647 (ix86_output_indirect_branch_via_reg): Replace
648 ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
649 (ix86_output_indirect_branch_via_push): Likewise.
650 (ix86_output_function_return): Likewise.
651 * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
652 incompatible with -fcf-protection=branch and
653 -fcheck-pointer-bounds.
655 2018-02-22 Steve Ellcey <sellcey@cavium.com>
658 * config/aarch64/aarch64.c (aarch64_print_address_internal):
659 Change gcc_assert call to output_operand_lossage.
661 2018-02-22 Steve Ellcey <sellcey@cavium.com>
663 * doc/extend.texi (__builtin_extend_pointer): Document builtin.
665 2018-02-22 DJ Delorie <dj@redhat.com>
666 Sebastian Perta <sebastian.perta@renesas.com>
667 Oleg Endo <olegendo@gcc.gnu.org>
669 * config/rx/rx.c (rx_rtx_costs): New function.
670 (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
672 2018-02-22 Thomas Preud'homme <thomas.preudhomme@arm.com>
674 * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
676 2018-02-22 Martin Liska <mliska@suse.cz>
679 * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
680 Add "native" as a possible value.
682 2018-02-22 Martin Liska <mliska@suse.cz>
685 * config/i386/i386.c (ix86_option_override_internal):
686 Add "native" as a possible value for -march and -mtune.
688 2018-02-22 Jakub Jelinek <jakub@redhat.com>
691 * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
692 to all type variants.
694 PR tree-optimization/84503
695 * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
696 width as info->bitpos + info->bitsize - start.
697 (merged_store_group::merge_overlapping): Simplify width computation.
698 (check_no_overlap): New function.
699 (imm_store_chain_info::try_coalesce_bswap): Compute expected
700 start + width and last_order of the group, fail if check_no_overlap
702 (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
703 to group if check_no_overlap fails.
705 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
707 * config/rs6000/altivec.md: Delete contraint arguments to
708 define_expand, define_split, and define_peephole2, and in
709 define_insn_and_split if always unused.
710 * config/rs6000/darwin.md: Ditto.
711 * config/rs6000/dfp.md: Ditto.
712 * config/rs6000/rs6000.md: Ditto.
713 * config/rs6000/sync.md: Ditto.
714 * config/rs6000/vector.md: Ditto.
715 * config/rs6000/vsx.md: Ditto.
717 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
719 * config/rs6000/altivec.md: Write output control strings as braced
720 blocks instead of double-quoted strings.
721 * config/rs6000/darwin.md: Ditto.
722 * config/rs6000/rs6000.md: Ditto.
723 * config/rs6000/vector.md: Ditto.
724 * config/rs6000/vsx.md: Ditto.
726 2018-02-21 Jason Merrill <jason@redhat.com>
728 PR c++/84314 - ICE with templates and fastcall attribute.
729 * attribs.c (build_type_attribute_qual_variant): Remove assert.
731 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
733 * ipa-cp.c (determine_versionability): Fix comment typos.
735 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
738 * ipa-cp.c (determine_versionability): Do not version functions caling
741 2018-02-21 Martin Liska <mliska@suse.cz>
744 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
745 Add "native" as a possible value.
746 * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT): Define
747 the macro when native cpu detection is available.
749 2018-02-21 Martin Liska <mliska@suse.cz>
752 * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
753 Add "native" as a possible value.
754 * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
755 when native cpu detection is available.
757 2018-02-21 Jakub Jelinek <jakub@redhat.com>
758 Martin Sebor <msebor@redhat.com>
760 PR tree-optimization/84478
761 * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
763 * gimple-fold.c (get_range_strlen): Make minlen const and assume it
764 can't be NULL. Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
765 support which is conservatively correct, for 2 only stay conservative
766 for maxlen. Formatting and comment capitalization fixes. Add STRICT
767 argument to the 2 argument get_range_strlen, adjust 6 arg
768 get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
770 (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
771 (gimple_fold_builtin_strlen): Pass true as last argument to
774 2018-02-20 Martin Sebor <msebor@redhat.com>
777 * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
778 (builtin_memref::set_base_and_offset): Same. Handle inner references.
779 (builtin_memref::builtin_memref): Factor out parts into
780 set_base_and_offset and call it.
782 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
785 * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
786 is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
787 greater precision. If to_mode is a MODE_PARTIAL_INT, stop the
788 search at the associated MODE_INT.
790 2018-02-20 Jeff Law <law@redhat.com>
793 PR tree-optimization/81592
795 * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
796 for range data rather than using global data.
797 * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
798 range data rather than using global data.
799 * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
800 pass it to children as needed.
801 (struct directive::fmtresult): Similarly.
802 (struct directive::set_width): Similarly.
803 (struct directive::set_precision): Similarly.
804 (format_integer, format_directive, parse_directive): Similarly.
805 (format_none): Accept unnamed vr_values parameter.
806 (format_percent, format_floating, format_character): Similarly.
807 (format_string, format_plain): Similarly.
808 * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
809 the EVRP range analyzer for range data rather than using global data.
810 * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
811 gimple-ssa-evrp-analyze.h
812 (class sprintf_dom_walker): Add after_dom_children member function.
813 Add evrp_range_analyzer member.
814 (sprintf_dom_walker::before_dom_children): Call into the EVRP
815 range analyzer as needed.
816 (sprintf_dom_walker::after_dom_children): New member function.
817 * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
819 (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
820 (evrp_range_analyzer::pop_to_marker): Likewise.
822 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
824 PR tree-optimization/84419
825 * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
826 with the required type if its current type is compatible but
829 2018-02-20 Jakub Jelinek <jakub@redhat.com>
832 * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
835 2018-02-20 Martin Liska <mliska@suse.cz>
838 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
839 possible values if we don't have a hint.
841 2018-02-20 Martin Liska <mliska@suse.cz>
845 * final.c (shorten_branches): Build align_tab array with one
847 * opts.c (finish_options): Add alignment option limit check.
848 (MAX_CODE_ALIGN): Likewise.
849 (MAX_CODE_ALIGN_VALUE): Likewise.
850 * doc/invoke.texi: Document maximum allowed option value for
851 all -falign-* options.
853 2018-02-19 Jakub Jelinek <jakub@redhat.com>
856 * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
857 * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
858 * var-tracking.c (emit_note_insn_var_location): Remove all references
859 to NOTE_INSN_CALL_ARG_LOCATION.
860 (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
861 the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
862 Use copy_rtx_if_shared.
863 * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
864 NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
865 (dwarf2out_var_location): Remove handling of
866 NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
868 * final.c (final_scan_insn): Remove all references to
869 NOTE_INSN_CALL_ARG_LOCATION.
870 (rest_of_clean_state): Likewise. Remove REG_CALL_ARG_LOCATION notes
871 before dumping final insns.
872 * except.c (emit_note_eh_region_end): Remove all references to
873 NOTE_INSN_CALL_ARG_LOCATION.
874 * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
875 * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
876 * config/arc/arc.c (hwloop_optimize): Likewise.
877 * config/arm/arm.c (create_fix_barrier): Likewise.
878 * config/s390/s390.c (s390_chunkify_start): Likewise.
879 * config/sh/sh.c (find_barrier): Likewise.
880 * config/i386/i386.c (rest_of_insert_endbranch,
881 ix86_seh_fixup_eh_fallthru): Likewise.
882 * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
883 * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
884 * config/frv/frv.c (frv_function_prologue): Likewise.
885 * emit-rtl.c (try_split): Likewise. Copy over REG_CALL_ARG_LOCATION
887 (note_outside_basic_block_p): Remove all references to
888 NOTE_INSN_CALL_ARG_LOCATION.
889 * gengtype.c (adjust_field_rtx_def): Likewise.
890 * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
892 * jump.c (cleanup_barriers, delete_related_insns): Likewise.
893 * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
896 * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
899 PR tree-optimization/84452
900 * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
901 expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
904 2018-02-19 Martin Liska <mliska@suse.cz>
907 * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
909 2018-02-19 Martin Liska <mliska@suse.cz>
910 Richard Sandiford <richard.sandiford@linaro.org>
912 PR tree-optimization/82491
913 * gimple-fold.c (get_base_constructor): Make earlier bail out
916 2018-02-19 Carl Love <cel@us.ibm.com>
918 * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
919 BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
921 * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
924 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
926 * config/rl78/rl78.md (movdf): New define expand.
928 2018-02-19 Martin Liska <mliska@suse.cz>
931 * doc/invoke.texi: Fix typo.
932 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
934 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
936 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
937 handle rs6000_single_float and rs6000_double_float specially for
940 2018-02-16 Jeff Law <law@redhat.com>
942 * config/rx/rx.c (add_pop_cfi_notes): New function.;
945 2018-02-16 Jakub Jelinek <jakub@redhat.com>
948 * ipa-inline.c (inline_small_functions): Fix a typo.
950 2018-02-16 Nathan Sidwell <nathan@acm.org>
952 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
954 2018-02-16 Carl Love <cel@us.ibm.com>
956 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
957 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
958 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
959 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
960 expansion to P8V_BUILTIN_VEC_FLOAT2.
962 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
964 PR rtl-optimization/70023
965 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
966 src_regno into account.
968 2018-02-16 Carl Love <cel@us.ibm.com>
970 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
971 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
972 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
973 * config/rs6000/rs6000.c: Remove case statements for
974 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
975 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
976 and P9V_BUILTIN_VEC_VINSERT4B.
977 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
978 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
979 * config/rs6000/vsx.md:
980 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
983 2018-02-16 Carl Love <cel@us.ibm.com>
985 * config/rs6000/altivec.h: Add builtin names vec_extract4b
987 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
989 * config/rs6000/rs6000-c.c: Add the definitions for
990 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
991 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
992 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
993 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
994 definition for insert4b and define insn *insert3b_internal.
995 * doc/extend.texi: Add documentation for vec_extract4b.
997 2018-02-16 Nathan Sidwell <nathan@acm.org>
999 * doc/extend.texi (Backwards Compatibility): Mention friend
1000 injection. Note for-scope is deprecated.
1001 * doc/invoke.texi (-ffriend-injection): Deprecate.
1003 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
1005 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
1006 that moved to I2, also allow destinations that are a paradoxical
1007 subreg (instead of a normal reg).
1009 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
1012 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
1015 2018-02-16 Richard Biener <rguenther@suse.de>
1017 PR tree-optimization/84037
1018 PR tree-optimization/84016
1020 * config/i386/i386.c (ix86_builtin_vectorization_cost):
1021 Adjust vec_construct for the fact we need additional higher latency
1022 128bit inserts for AVX256 and AVX512 vector builds.
1023 (ix86_add_stmt_cost): Scale vector construction cost for
1026 2018-02-16 Richard Biener <rguenther@suse.de>
1028 PR tree-optimization/84417
1029 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
1030 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
1031 (non_rewritable_lvalue_p): Likewise, use poly-ints.
1033 2018-02-16 Martin Liska <mliska@suse.cz>
1036 * internal-fn.def (ASAN_CHECK): Set proper flags.
1037 (ASAN_MARK): Likewise.
1039 2018-02-16 Julia Koval <julia.koval@intel.com>
1041 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
1042 from PTA_CANNONLAKE.
1044 2018-02-16 Jakub Jelinek <jakub@redhat.com>
1047 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
1048 Use ++iter rather than iter++ for std::list iterators.
1049 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
1050 defer deleting them until all nodes in the forest are processed. Do
1051 free even leaf nodes. Change to_process into auto_vec.
1054 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
1055 * vec.h (vec_default_construct): Use memset instead of placement new
1056 if BROKEN_VALUE_INITIALIZATION is defined.
1057 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
1058 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
1061 PR rtl-optimization/83723
1062 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
1063 * lra.c (lra_substitute_pseudo): Likewise. If true, use
1064 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
1066 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
1068 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
1070 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
1072 PR rtl-optimization/81443
1073 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
1074 from inner REGs to paradoxical SUBREGs.
1076 2018-02-16 Richard Biener <rguenther@suse.de>
1078 PR tree-optimization/84399
1079 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
1080 For operands we can analyze at their definition make sure we can
1081 analyze them at each use as well.
1083 2018-02-16 Richard Biener <rguenther@suse.de>
1085 PR tree-optimization/84190
1086 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
1087 volatile accesses if the decl isn't volatile.
1089 2018-02-15 Jason Merrill <jason@redhat.com>
1091 PR c++/84314 - ICE with templates and fastcall attribute.
1092 * attribs.c (build_type_attribute_qual_variant): Don't clobber
1093 TYPE_CANONICAL on an existing type.
1095 2018-02-15 Jakub Jelinek <jakub@redhat.com>
1097 PR tree-optimization/84383
1098 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
1099 dstoff nor call operand_equal_p if dstbase is NULL.
1101 PR tree-optimization/84334
1102 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
1103 also a CONSTANT_CLASS_P, punt.
1105 2018-02-14 Jim Wilson <jimw@sifive.com>
1107 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
1108 first SMALL_OPERAND check. New local min_second_step. Move assert
1109 to where locals are set. Add TARGET_RVC support.
1110 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
1112 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
1114 * doc/invoke.texi: Correct -Wformat-overflow code sample.
1116 2018-02-14 Martin Sebor <msebor@redhat.com>
1118 PR tree-optimization/83698
1119 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
1120 arrays constrain the offset range to their bounds.
1121 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
1122 (builtin_access::overlap): Avoid setting the size of overlap if it's
1124 (maybe_diag_overlap): Also consider arrays when deciding what values
1125 of offsets to include in diagnostics.
1127 2018-02-14 Martin Sebor <msebor@redhat.com>
1130 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
1131 that correspond to the kind of a declaration.
1133 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
1136 * config/pa/pa.md: Load address of PIC label using the linkage table
1137 if the label is nonlocal.
1139 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
1141 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
1142 warning message if user requests -maltivec=be.
1143 * doc/invoke.texi: Document deprecation of -maltivec=be.
1145 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
1148 * config/rs6000/rs6000-c.c: Update definitions for
1149 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
1150 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
1152 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1155 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
1156 add _get_ssp intrinsics. Remove argument from
1157 __builtin_ia32_rdssp[d|q].
1158 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
1159 * config/i386/i386-builtin.def: Remove argument from
1160 __builtin_ia32_rdssp[d|q].
1161 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
1162 ix86_expand_special_args_builtin for _rdssp[d|q].
1163 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
1164 Clear register before usage.
1165 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
1166 Add documentation for new _get_ssp and _inc_ssp intrinsics.
1168 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
1170 PR tree-optimization/84357
1171 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
1172 operand 1 of an ARRAY_REF too.
1174 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
1177 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
1178 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
1180 (set_of_reg): New struct.
1181 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
1182 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
1183 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
1185 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
1186 Split into bitclr, bitset, bitinvert patterns if appropriate.
1187 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
1188 use rx_fuse_in_memory_bitop.
1189 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
1190 to named insn, correct maximum insn length.
1192 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
1195 * machmode.def: Define a complex mode for PARTIAL_INT.
1196 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
1198 * doc/rtl.texi: Document CSPImode.
1199 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
1201 (msp430_hard_regno_nregs_with_padding): Likewise.
1203 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
1206 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
1208 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
1210 PR rtl-optimization/84169
1211 * combine.c (try_combine): New variable split_i2i3. Set it to true if
1212 we generated a parallel as new i3 and we split that to new i2 and i3
1213 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
1214 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
1215 those to i2, not i1. Partially rewrite this scan code.
1217 2018-02-13 Jakub Jelinek <jakub@redhat.com>
1220 * stor-layout.c (place_field): For variable length fields, adjust
1221 offset_align afterwards not just based on the field's alignment,
1222 but also on the size.
1225 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
1226 of exps and logs in the use_exp2 case.
1228 2018-02-13 Jeff Law <law@redhat.com>
1230 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
1233 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
1236 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
1240 * common.opt (gas-loc-support, gas-locview-support): New.
1241 (ginline-points, ginternal-reset-location-views): New.
1242 * doc/invoke.texi: Document them. Use @itemx where intended.
1243 (gvariable-location-views): Adjust.
1244 * target.def (reset_location_view): New.
1245 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
1246 (TARGET_RESET_LOCATION_VIEW): New.
1247 * doc/tm.texi: Rebuilt.
1248 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
1249 (dwarf2out_default_as_locview_support): New.
1250 (output_asm_line_debug_info): Use option variables.
1251 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
1252 (output_loc_list): Likewise.
1253 (add_high_low_attributes): Check option variables.
1254 Don't output entry view attribute in strict mode.
1255 (gen_inlined_subroutine_die): Check option variables.
1256 (dwarf2out_inline_entry): Likewise.
1257 (init_sections_and_labels): Likewise.
1258 (dwarf2out_early_finish): Likewise.
1259 (maybe_reset_location_view): New, from...
1260 (dwarf2out_var_location): ... here. Call it.
1261 * debug.h (dwarf2out_default_as_loc_support): Declare.
1262 (dwarf2out_default_as_locview_support): Declare.
1263 * hooks.c (hook_int_rtx_insn_0): New.
1264 * hooks.h (hook_int_rtx_insn_0): Declare.
1265 * toplev.c (process_options): Take -gas-loc-support and
1266 -gas-locview-support from dwarf2out. Enable
1267 -gvariable-location-views by default only with locview
1268 assembler support. Enable -ginternal-reset-location-views by
1269 default only if the target defines the corresponding hook.
1270 Enable -ginline-points by default if location views are
1271 enabled; force it disabled if statement frontiers are
1273 * tree-inline.c (expand_call_inline): Check option variables.
1274 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
1276 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
1278 PR tree-optimization/84321
1279 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
1280 handling. Also check whether the anti-range contains any values
1281 that satisfy the mask; switch to a VR_RANGE if not.
1283 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
1286 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
1288 2018-02-13 Martin Jambor <mjambor@suse.cz>
1291 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
1292 of call statements, also set location of a load to a temporary.
1294 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
1296 * config/rl78/rl78.c (add_vector_labels): New function.
1297 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
1298 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
1299 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
1300 which checks that no arguments are passed.
1301 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
1302 * doc/extend.texi: Documentation for the new attribute.
1304 2018-02-13 Andreas Schwab <schwab@suse.de>
1306 * config/riscv/linux.h (CPP_SPEC): Define.
1308 2018-02-13 Jakub Jelinek <jakub@redhat.com>
1311 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
1312 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
1313 OPTION_MASK_ISA_AES as first argument to def_builtin_const
1314 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
1315 instead of OPTION_MASK_ISA_PCLMUL as first argument to
1316 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
1317 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
1318 temporarily for AES and PCLMUL builtins.
1320 PR tree-optimization/84339
1321 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
1322 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
1326 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
1327 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
1328 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
1330 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
1332 * omp-simd-clone.h: New file.
1333 * omp-simd-clone.c: Include omp-simd-clone.h.
1334 (expand_simd_clones): No longer static.
1335 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
1336 cgraph.h and omp-simd-clone.h.
1337 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
1338 (vect_recog_widen_shift_pattern): Formatting fix.
1339 (vect_pattern_recog_1): Don't check optab for calls.
1342 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
1343 operands[2] into a REG before using gen_lowpart on it.
1345 2018-02-12 Jeff Law <law@redhat.com>
1348 * config/sh/sh.c (find_barrier): Consider a sibling call
1351 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
1352 successfully back substituting a reg.
1354 2018-02-12 Richard Biener <rguenther@suse.de>
1356 PR tree-optimization/84037
1357 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
1358 parameter, move visited init to caller.
1359 (vect_slp_analyze_operations): Separate cost from validity
1360 check, initialize visited once for all instances.
1361 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
1363 * tree-vect-stmts.c (vect_model_simple_cost): Make early
1365 (vect_model_promotion_demotion_cost): Likewise.
1366 (vectorizable_bswap): Guard cost modeling with !slp_node
1367 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
1369 (vectorizable_call): Likewise.
1370 (vectorizable_conversion): Likewise.
1371 (vectorizable_assignment): Likewise.
1372 (vectorizable_shift): Likewise.
1373 (vectorizable_operation): Likewise.
1374 (vectorizable_store): Likewise.
1375 (vectorizable_load): Likewise.
1376 (vectorizable_condition): Likewise.
1377 (vectorizable_comparison): Likewise.
1379 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
1382 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
1383 (ASAN_MARK): Fix fnspec to account for return value, change pointer
1384 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
1386 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1389 * params.def (inline-min-speedup): Increase from 8 to 15.
1390 (max-inline-insns-auto): Decrease from 40 to 30.
1391 * ipa-split.c (consider_split): Add some buffer for function to
1392 be considered inlining candidate.
1393 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
1396 2018-02-12 Richard Biener <rguenther@suse.de>
1398 PR tree-optimization/84037
1399 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
1400 matched stmts if we cannot swap the non-matched ones.
1402 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
1404 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
1405 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
1406 _mm_maskz_scalef_round_ss): New intrinsics.
1407 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
1408 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
1409 __builtin_ia32_scalefss_round): Remove.
1410 (__builtin_ia32_scalefsd_mask_round,
1411 __builtin_ia32_scalefss_mask_round): New intrinsics.
1412 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
1413 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
1414 ((match_operand:VF_128 2 "<round_nimm_predicate>"
1415 "<round_constraint>")): Changed to ...
1416 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
1417 "<round_scalar_constraint>")): ... this.
1418 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
1419 %0, %1, %2<round_op3>}"): Changed to ...
1420 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
1421 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
1422 %2<round_scalar_mask_op3>}"): ... this.
1423 * config/i386/subst.md (round_scalar_nimm_predicate): New.
1425 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
1427 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
1428 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
1429 (_mm_maskz_sqrt_round_ss): New intrinsics.
1430 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
1431 (__builtin_ia32_sqrtsd_mask_round)
1432 (__builtin_ia32_sqrtss_mask_round): New builtins.
1433 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
1434 (__builtin_ia32_sqrtss_round): Remove.
1435 (__builtin_ia32_sqrtsd_mask_round)
1436 (__builtin_ia32_sqrtss_mask_round): New builtins.
1437 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
1438 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
1439 ((match_operand:VF_128 1 "vector_operand"
1440 "xBm,<round_constraint>")): Changed to ...
1441 ((match_operand:VF_128 1 "vector_operand"
1442 "xBm,<round_scalar_constraint>")): ... this.
1443 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
1444 %0, %2, %<iptr>1<round_op3>}): Changed to ...
1445 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
1446 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
1447 %<iptr>1<round_scalar_mask_op3>}): ... this.
1448 ((set_attr "prefix" "<round_prefix>")): Changed to ...
1449 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
1451 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
1454 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
1455 Cast vec_cmpeq result to correct type.
1456 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
1457 Cast vec_cmpgt result to correct type.
1459 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
1461 * final.c (final_scan_insn_1): Renamed from...
1462 (final_scan_insn): ... this. New wrapper, to recover
1463 seen from the outermost call in recursive ones.
1464 * config/sparc/sparc.c (output_return): Drop seen from call.
1465 (output_sibcall): Likewise.
1466 * config/visium/visium.c (output_branch): Likewise.
1468 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
1470 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
1473 2018-02-10 Alan Modra <amodra@gmail.com>
1476 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
1477 Specify LR as an input.
1479 2018-02-10 Jakub Jelinek <jakub@redhat.com>
1482 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
1483 remove_member_access_dummy_vars): New functions.
1484 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
1485 lower_omp_1, execute_lower_omp): Use them.
1487 PR rtl-optimization/84308
1488 * shrink-wrap.c (spread_components): Release todo vector.
1490 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
1492 PR rtl-optimization/57193
1493 * ira-color.c (struct allocno_color_data): Add member
1494 conflict_allocno_hard_prefs.
1495 (update_conflict_allocno_hard_prefs): New.
1496 (bucket_allocno_compare_func): Add a preference based on
1497 conflict_allocno_hard_prefs.
1498 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
1499 (color_allocnos): Remove a dead code. Initiate
1500 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
1502 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1505 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
1506 constraint from =wa to wa. Avoid a subreg on the output operand,
1507 instead use a pseudo and subreg it in a move.
1508 (p9_xxbrd_<mode>): Changed to ...
1509 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
1510 (p9_xxbrd_v2df): New expander.
1511 (p9_xxbrw_<mode>): Changed to ...
1512 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
1513 (p9_xxbrw_v4sf): New expander.
1515 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1517 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
1519 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
1522 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
1523 multiply in 32-bit mode.
1524 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
1525 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
1528 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1530 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
1531 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
1532 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
1533 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
1535 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
1538 * dwarf2out.c (is_trivial_indirect_ref): New function.
1539 (dwarf2out_late_global_decl): Do not generate a location
1540 attribute for variables that have a non-trivial DECL_VALUE_EXPR
1541 and that are not defined in the current unit.
1543 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1545 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
1546 instead of a libcall for UNORDERED.
1548 2018-02-09 Tamar Christina <tamar.christina@arm.com>
1551 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
1552 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
1554 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1557 * config/s390/s390.c (s390_set_current_function): Invoke
1558 s390_indirect_branch_settings also if fndecl didn't change.
1560 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1562 * config/rs6000/rs6000.md (blockage): Set length to zero.
1564 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1566 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
1568 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1571 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
1572 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
1576 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
1577 PARALLEL incoming that failed vt_get_decl_and_offset check.
1580 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
1581 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
1583 (get_variable_section): For decls in named .bss* sections pass true as
1584 second argument to bss_initializer_p.
1586 2018-02-09 Marek Polacek <polacek@redhat.com>
1587 Jakub Jelinek <jakub@redhat.com>
1590 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
1591 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
1592 Sync some changes from cxx_fold_indirect_ref.
1594 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1596 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
1598 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
1599 (BLOCK_INLINE_ENTRY_LABEL): New.
1600 (dwarf2out_var_location): Disregard inline entry markers.
1601 (inline_entry_data): New struct.
1602 (inline_entry_data_hasher): New hashtable type.
1603 (inline_entry_data_hasher::hash): New.
1604 (inline_entry_data_hasher::equal): New.
1605 (inline_entry_data_table): New variable.
1606 (add_high_low_attributes): Add DW_AT_entry_pc and
1607 DW_AT_GNU_entry_view attributes if a pending entry is found
1608 in inline_entry_data_table. Add old entry_pc attribute only
1609 if debug nonbinding markers are disabled.
1610 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
1611 markers are enabled.
1612 (block_within_block_p, dwarf2out_inline_entry): New.
1613 (dwarf2out_finish): Check that no entries remained in
1614 inline_entry_data_table.
1615 * final.c (reemit_insn_block_notes): Handle inline entry notes.
1616 (final_scan_insn, notice_source_line): Likewise.
1617 (rest_of_clean_state): Skip inline entry markers.
1618 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
1620 * gimple.c (gimple_build_debug_inline_entry): New.
1621 * gimple.h (enum gimple_debug_subcode): Add
1622 GIMPLE_DEBUG_INLINE_ENTRY.
1623 (gimple_build_debug_inline_entry): Declare.
1624 (gimple_debug_inline_entry_p): New.
1625 (gimple_debug_nonbind_marker_p): Adjust.
1626 * insn-notes.def (INLINE_ENTRY): New.
1627 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
1628 inline entry marker notes.
1629 (print_insn): Likewise.
1630 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
1631 (INSN_DEBUG_MARKER_KIND): Likewise.
1632 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
1633 * tree-inline.c (expand_call_inline): Build and insert
1634 debug_inline_entry stmt.
1635 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
1636 inline entry blocks early, if nonbind markers are enabled.
1637 (dump_scope_block): Dump fragment info.
1638 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
1639 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
1640 (gimple_build_debug_inline_entry): New.
1641 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
1642 Enable/disable inline entry points too.
1643 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
1644 (DEBUG_INSN): Describe inline entry markers.
1646 * common.opt (gvariable-location-views): New.
1647 (gvariable-location-views=incompat5): New.
1648 * config.in: Rebuilt.
1649 * configure: Rebuilt.
1650 * configure.ac: Test assembler for view support.
1651 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
1652 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
1653 * dwarf2out.c (var_loc_view): New typedef.
1654 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
1655 (dwarf2out_locviews_in_attribute): New.
1656 (dwarf2out_locviews_in_loclist): New.
1657 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
1658 (enum dw_line_info_opcode): Add LI_adv_address.
1659 (struct dw_line_info_table): Add view.
1660 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
1661 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
1662 (zero_view_p): New variable.
1663 (ZERO_VIEW_P): New macro.
1664 (output_asm_line_debug_info): New.
1665 (struct var_loc_node): Add view.
1666 (add_AT_view_list, AT_loc_list): New.
1667 (add_var_loc_to_decl): Add view param. Test it against last.
1668 (new_loc_list): Add view params. Record them.
1669 (AT_loc_list_ptr): Handle loc and view lists.
1670 (view_list_to_loc_list_val_node): New.
1671 (print_dw_val): Handle dw_val_class_view_list.
1672 (size_of_die): Likewise.
1673 (value_format): Likewise.
1674 (loc_list_has_views): New.
1675 (gen_llsym): Set vl_symbol too.
1676 (maybe_gen_llsym, skip_loc_list_entry): New.
1677 (dwarf2out_maybe_output_loclist_view_pair): New.
1678 (output_loc_list): Output view list or entries too.
1679 (output_view_list_offset): New.
1680 (output_die): Handle dw_val_class_view_list.
1681 (output_dwarf_version): New.
1682 (output_compilation_unit_header): Use it.
1683 (output_skeleton_debug_sections): Likewise.
1684 (output_rnglists, output_line_info): Likewise.
1685 (output_pubnames, output_aranges): Update version comments.
1686 (output_one_line_info_table): Output view numbers in asm comments.
1687 (dw_loc_list): Determine current endview, pass it to new_loc_list.
1688 Call maybe_gen_llsym.
1689 (loc_list_from_tree_1): Adjust.
1690 (add_AT_location_description): Create view list attribute if
1691 needed, check it's absent otherwise.
1692 (convert_cfa_to_fb_loc_list): Adjust.
1693 (maybe_emit_file): Call output_asm_line_debug_info for test.
1694 (dwarf2out_var_location): Reset views as needed. Precompute
1695 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
1696 attribute. Set view.
1697 (new_line_info_table): Reset next view.
1698 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
1699 (dwarf2out_source_line): Likewise. Output view resets and labels to
1700 the assembler, or select appropriate line info opcodes.
1701 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
1702 (optimize_string_length): Catch it. Adjust.
1703 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
1704 dw_val_class_view_list, and remove it if no longer needed.
1705 (hash_loc_list): Hash view numbers.
1706 (loc_list_hasher::equal): Compare them.
1707 (optimize_location_lists): Check whether a view list symbol is
1708 needed, and whether the locview attribute is present, and
1709 whether they match. Remove the locview attribute if no longer
1711 (index_location_lists): Call skip_loc_list_entry for test.
1712 (dwarf2out_finish): Call output_asm_line_debug_info for test.
1713 Use output_dwarf_version.
1714 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
1715 (struct dw_val_node): Add val_view_list.
1716 * final.c (SEEN_NEXT_VIEW): New.
1717 (set_next_view_needed): New.
1718 (clear_next_view_needed): New.
1719 (maybe_output_next_view): New.
1720 (final_start_function): Rename to...
1721 (final_start_function_1): ... this. Take pointer to FIRST,
1722 add SEEN parameter. Emit param bindings in the initial view.
1723 (final_start_function): Reintroduce SEEN-less interface.
1724 (final): Rename to...
1725 (final_1): ... this. Take SEEN parameter. Output final pending
1726 next view at the end.
1727 (final): Reintroduce seen-less interface.
1728 (final_scan_insn): Output pending next view before switching
1729 sections or ending a block. Mark the next view as needed when
1730 outputting variable locations. Notify debug backend of section
1731 changes, and of location view changes.
1732 (rest_of_handle_final): Adjust.
1733 * toplev.c (process_options): Autodetect value for debug variable
1734 location views option. Warn on incompat5 without -gdwarf-5.
1735 * doc/invoke.texi (gvariable-location-views): New.
1736 (gvariable-location-views=incompat5): New.
1737 (gno-variable-location-views): New.
1739 2018-02-08 David Malcolm <dmalcolm@redhat.com>
1741 PR tree-optimization/84136
1742 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
1743 that the result of find_edge is non-NULL.
1745 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1748 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
1749 storing integer register in SImode. Fix cost of 256 and 512
1750 byte aligned SSE register store.
1752 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1754 * config/i386/i386.c (ix86_multiplication_cost): Fix
1755 multiplication cost for TARGET_AVX512DQ.
1757 2018-02-08 Marek Polacek <polacek@redhat.com>
1759 PR tree-optimization/84238
1760 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
1763 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1765 PR tree-optimization/84265
1766 * tree-vect-stmts.c (vectorizable_store): Don't treat
1767 VMAT_CONTIGUOUS accesses as grouped.
1768 (vectorizable_load): Likewise.
1770 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1772 PR tree-optimization/81635
1773 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
1774 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
1775 (test_round_for_mask): New functions.
1776 (wide_int_cc_tests): Call test_round_for_mask.
1777 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
1778 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
1779 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
1780 range returned by get_range_info.
1782 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1785 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
1786 * symtab.c: Include builtins.h
1787 (symtab_node::output_to_lto_symbol_table_p): Move here
1788 from lto-streamer-out.c:output_symbol_p.
1789 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
1790 (output_symbol_p): Move all logic to symtab.c
1791 (produce_symtab): Update.
1793 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1795 * config/s390/s390-opts.h (enum indirect_branch): Define.
1796 * config/s390/s390-protos.h (s390_return_addr_from_memory)
1797 (s390_indirect_branch_via_thunk)
1798 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
1799 (enum s390_indirect_branch_type): Define.
1800 * config/s390/s390.c (struct s390_frame_layout, struct
1801 machine_function): Remove.
1802 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
1803 (indirect_branch_table_label_no, indirect_branch_table_name):
1805 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
1806 (enum s390_indirect_branch_option): Define.
1807 (s390_return_addr_from_memory): New function.
1808 (s390_handle_string_attribute): New function.
1809 (s390_attribute_table): Add new attribute handler.
1810 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
1811 (s390_indirect_branch_via_thunk): New function.
1812 (s390_indirect_branch_via_inline_thunk): New function.
1813 (s390_function_ok_for_sibcall): When jumping via thunk disallow
1814 sibling call optimization for non z10 compiles.
1815 (s390_emit_call): Force indirect branch target to be a single
1816 register. Add r1 clobber for non-z10 compiles.
1817 (s390_emit_epilogue): Emit return jump via return_use expander.
1818 (s390_reorg): Handle JUMP_INSNs as execute targets.
1819 (s390_option_override_internal): Perform validity checks for the
1820 new command line options.
1821 (s390_indirect_branch_attrvalue): New function.
1822 (s390_indirect_branch_settings): New function.
1823 (s390_set_current_function): Invoke s390_indirect_branch_settings.
1824 (s390_output_indirect_thunk_function): New function.
1825 (s390_code_end): Implement target hook.
1826 (s390_case_values_threshold): Implement target hook.
1827 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
1829 * config/s390/s390.h (struct s390_frame_layout)
1830 (struct machine_function): Move here from s390.c.
1831 (TARGET_INDIRECT_BRANCH_NOBP_RET)
1832 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
1833 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
1834 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
1835 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
1836 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
1837 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
1838 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
1839 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
1840 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
1841 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
1842 (mnemonic attribute): Add values which aren't recognized
1844 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
1845 pattern for branch conversion. Fix mnemonic attribute.
1846 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
1847 indirect branch via thunk if requested.
1848 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
1849 ("*indirect_jump"): Disable for branch conversion using out of
1851 ("indirect_jump_via_thunk<mode>_z10")
1852 ("indirect_jump_via_thunk<mode>")
1853 ("indirect_jump_via_inlinethunk<mode>_z10")
1854 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
1855 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
1856 ("casesi_jump_via_inlinethunk<mode>_z10")
1857 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
1858 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
1859 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
1860 ("*indirect2_jump"): Disable for branch conversion.
1861 ("casesi_jump"): Turn into expander and expand patterns for branch
1863 ("return_use"): New expander.
1864 ("*return"): Emit return via thunk and rename it to ...
1865 ("*return<mode>"): ... this one.
1866 * config/s390/s390.opt: Add new options and and enum for the
1869 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1871 * lra-constraints.c (match_reload): Unconditionally use
1872 gen_lowpart_SUBREG, rather than selecting between that
1873 and equivalent gen_rtx_SUBREG code.
1875 2018-02-08 Richard Biener <rguenther@suse.de>
1877 PR tree-optimization/84233
1878 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
1879 changed flag instead of boguously re-using phi_inserted.
1881 2018-02-08 Martin Jambor <mjambor@suse.cz>
1883 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
1884 static local variables.
1886 2018-02-08 Richard Biener <rguenther@suse.de>
1888 PR tree-optimization/84278
1889 * tree-vect-stmts.c (vectorizable_store): When looking for
1890 smaller vector types to perform grouped strided loads/stores
1891 make sure the mode is supported by the target.
1892 (vectorizable_load): Likewise.
1894 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1896 * config/aarch64/aarch64.c (aarch64_components_for_bb):
1897 Increase LDP/STP opportunities by adding adjacent callee-saves.
1899 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1901 PR rtl-optimization/84068
1902 PR rtl-optimization/83459
1903 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
1905 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
1907 PR tree-optimization/84224
1908 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
1909 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
1912 2018-02-07 Iain Sandoe <iain@codesourcery.com>
1915 * config/rs6000/altivec.md (*restore_world): Remove LR use.
1916 * config/rs6000/predicates.md (restore_world_operation): Adjust op
1917 count, remove one USE.
1919 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
1921 * doc/install.texi (Configuration): Document the
1922 --with-long-double-format={ibm,ieee} PowerPC configuration
1926 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
1927 Convert from define_expand to be define_insn_and_split. Rework
1928 float/double/_Float128 conversions to QI/HI/SImode to work with
1929 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
1930 conversions to QI/HImode types did a store and then a load to
1931 truncate the value. For conversions to VSX registers, don't split
1932 the insn, instead emit the code directly. Use the code iterator
1933 any_fix to combine signed and unsigned conversions.
1934 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
1935 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
1936 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1937 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
1938 (fix_<mode>di2_hw): Likewise.
1939 (fixuns_<mode>di2_hw): Likewise.
1940 (fix_<mode>si2_hw): Likewise.
1941 (fixuns_<mode>si2_hw): Likewise.
1942 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
1943 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1944 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1945 fix<uns>_trunc<SFDF:mode>si2_p8.
1946 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1948 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1949 (fix<uns>_<mode>_mem): Likewise.
1950 (fctiw<u>z_<mode>_mem): Likewise.
1951 (fix<uns>_<mode>_mem): Likewise.
1952 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1953 the register allocator from doing a direct move to the GPRs to do
1954 a store, and instead use the ISA 3.0 store byte/half-word from
1955 vector register instruction. For IEEE 128-bit floating point,
1956 also optimize stores of 32-bit ints.
1957 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1959 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1961 * genextract.c (push_pathstr_operand): New function to support
1963 (walk_rtx): Call push_pathstr_operand.
1964 (print_path): Support [a-zA-Z].
1966 2018-02-07 Richard Biener <rguenther@suse.de>
1968 PR tree-optimization/84037
1969 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1970 (cse_and_gimplify_to_preheader): Declare.
1971 (vect_get_place_in_interleaving_chain): Likewise.
1972 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1974 (_loop_vec_info::~_loop_vec_info): Delete it.
1975 (cse_and_gimplify_to_preheader): New function.
1976 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1977 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1978 (vectorizable_load): Likewise. For grouped stores always base
1979 the IV on the first element.
1980 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1981 condition before gimplifying.
1983 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1985 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1986 *DIV_EXPR and *MOD_EXPR.
1988 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1991 * config/i386/i386.c (ix86_option_override_internal): Mask out
1992 the CF_SET bit when checking -fcf-protection.
1994 2018-02-07 Tom de Vries <tom@codesourcery.com>
1997 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
2000 2018-02-07 Richard Biener <rguenther@suse.de>
2002 PR tree-optimization/84204
2003 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
2006 PR tree-optimization/84205
2007 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
2008 special-case isl_ast_op_zdiv_r.
2010 PR tree-optimization/84223
2011 * graphite-scop-detection.c (gather_bbs::before_dom_children):
2012 Only add conditions from within the region.
2013 (gather_bbs::after_dom_children): Adjust.
2015 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
2018 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
2019 * config/avr/avr.md: Only post-reload split REG-REG moves if
2020 either register is REGERAL_REG_P.
2022 2018-02-07 Jakub Jelinek <jakub@redhat.com>
2024 PR tree-optimization/84235
2025 * tree-ssa-scopedtables.c
2026 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
2027 if the subtraction is performed in floating point type where NaNs are
2028 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
2029 build 1. Formatting fix.
2031 2018-02-06 Jakub Jelinek <jakub@redhat.com>
2034 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
2035 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
2036 and skip it regardless of bb boundaries. Use CALL_P macro,
2037 don't test INSN_P (insn) together with CALL_P or JUMP_P check
2038 unnecessarily, formatting fix.
2040 2018-02-06 Michael Collison <michael.collison@arm.com>
2042 * config/arm/thumb2.md:
2043 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
2044 (*thumb_mov_notscc): Ditto.
2046 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
2049 * config/rs6000/rs6000.md (su code attribute): Use "u" for
2050 unsigned_fix, not "s".
2052 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2054 * configure.ac (gcc_fn_eh_frame_ro): New function.
2055 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
2056 correct .eh_frame permissions.
2057 * configure: Regenerate.
2059 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
2061 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
2064 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2066 * config/rs6000/rs6000.c (rs6000_option_override_internal):
2067 Display warning message for -mno-speculate-indirect-jumps.
2069 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
2071 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
2073 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
2075 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
2077 PR tree-optimization/84225
2078 * tree-eh.c (find_trapping_overflow): Only call
2079 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
2081 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
2084 * config/i386/i386.c: Reimplement the check of possible options
2085 -mibt/-mshstk conbination. Change error messages.
2086 * doc/invoke.texi: Fix a typo: remove extra '='.
2088 2018-02-06 Marek Polacek <polacek@redhat.com>
2090 PR tree-optimization/84228
2091 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
2093 2018-02-06 Tamar Christina <tamar.christina@arm.com>
2096 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
2097 emitted arch directives.
2098 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
2099 __ARM_FEATURE_COPROC before changing architectures.
2101 2018-02-06 Richard Biener <rguenther@suse.de>
2103 * config/i386/i386.c (print_reg): Fix typo.
2104 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
2106 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
2108 * configure: Regenerate.
2110 2018-02-05 Martin Sebor <msebor@redhat.com>
2112 PR tree-optimization/83369
2113 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
2116 2018-02-05 Martin Liska <mliska@suse.cz>
2118 * doc/invoke.texi: Cherry-pick upstream r323995.
2120 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
2122 * ira.c (ira_init_register_move_cost): Adjust comment.
2124 2018-02-05 Martin Liska <mliska@suse.cz>
2126 PR gcov-profile/84137
2127 * doc/gcov.texi: Fix typo in documentation.
2129 2018-02-05 Martin Liska <mliska@suse.cz>
2131 PR gcov-profile/83879
2132 * doc/gcov.texi: Document necessity of --dynamic-list-data when
2133 using dlopen functionality.
2135 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
2137 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
2138 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
2139 _mm_maskz_range_ss, _mm_mask_range_round_ss,
2140 _mm_maskz_range_round_ss): New intrinsics.
2141 (__builtin_ia32_rangesd128_round)
2142 (__builtin_ia32_rangess128_round): Remove.
2143 (__builtin_ia32_rangesd128_mask_round,
2144 __builtin_ia32_rangess128_mask_round): New builtins.
2145 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
2146 __builtin_ia32_rangess128_round): Remove.
2147 (__builtin_ia32_rangesd128_mask_round,
2148 __builtin_ia32_rangess128_mask_round): New builtins.
2149 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
2150 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
2151 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
2152 "<round_saeonly_constraint>")): Changed to ...
2153 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
2154 "<round_saeonly_scalar_constraint>")): ... this.
2155 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
2156 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
2157 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
2158 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
2159 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
2161 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
2163 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
2165 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
2166 Remove all values except native, 8540 and 8548.
2168 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
2170 * config/i386/i386.c (ix86_output_function_return): Pass
2171 INVALID_REGNUM, instead of -1, as invalid register number to
2172 indirect_thunk_name and output_indirect_thunk.
2174 2018-02-02 Julia Koval <julia.koval@intel.com>
2176 * config.gcc: Add -march=icelake.
2177 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
2178 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
2179 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
2180 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
2181 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
2182 (processor_target_table): Add icelake.
2183 (ix86_option_override_internal): Handle new PTAs.
2184 (get_builtin_code_for_version): Handle icelake.
2185 (M_INTEL_COREI7_ICELAKE): New.
2186 (fold_builtin_cpu): Handle icelake.
2187 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
2188 * doc/invoke.texi: Add -march=icelake.
2190 2018-02-02 Julia Koval <julia.koval@intel.com>
2192 * config/i386/i386.c (ix86_option_override_internal): Change flags type
2193 to wide_int_bitmask.
2194 * wide-int-bitmask.h: New.
2196 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
2199 * config/i386/i386.md: Replace Pmode with word_mode in
2200 builtin_setjmp_setup and builtin_longjmp to support x32.
2202 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
2206 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
2208 (rs6000_supported_cpu_names): New static variable.
2209 (linux_cpu_translation_table): Likewise.
2210 (elf_platform) <cpu>: Define new static variable and use it.
2211 Translate kernel AT_PLATFORM name to canonical name if needed.
2212 Error if platform name is unknown.
2214 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
2217 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
2219 2018-02-01 Jeff Law <law@redhat.com>
2222 * config/i386/i386.c (release_scratch_register_on_entry): Add new
2223 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
2224 the scratch if RELEASE_VIA_POP is false.
2225 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
2226 If we have to save a temporary register, decrement SIZE appropriately.
2227 Pass new arguments to release_scratch_register_on_entry.
2228 (ix86_adjust_stack_and_probe): Likewise.
2229 (ix86_emit_probe_stack_range): Pass new arguments to
2230 release_scratch_register_on_entry.
2232 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
2234 PR rtl-optimization/84157
2235 * combine.c (change_zero_ext): Use REG_P predicate in
2236 front of HARD_REGISTER_P predicate.
2238 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
2240 * config/avr/avr.c (avr_option_override): Move disabling of
2241 -fdelete-null-pointer-checks to...
2242 * common/config/avr/avr-common.c (avr_option_optimization_table):
2245 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2247 PR tree-optimization/81635
2248 * tree-data-ref.c (split_constant_offset_1): For types that
2249 wrap on overflow, try to use range info to prove that wrapping
2252 2018-02-01 Renlin Li <renlin.li@arm.com>
2255 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
2257 (aarch64_register_move_cost): Likewise.
2258 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
2260 (REG_CLASS_NAMES): Likewise.
2261 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
2262 TAILCALL_ADDR_REGS. Remove IP registers.
2263 * config/aarch64/aarch64.md (Ucs): Update register constraint.
2265 2018-02-01 Richard Biener <rguenther@suse.de>
2267 * domwalk.h (dom_walker::dom_walker): Add additional constructor
2268 for specifying RPO order and allow NULL for that.
2269 * domwalk.c (dom_walker::dom_walker): Likewise.
2270 (dom_walker::walk): Handle NULL RPO order.
2271 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
2273 (rewrite_update_dom_walker): Likewise.
2274 (mark_def_dom_walker): Likewise.
2276 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2278 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
2279 (aarch64_maybe_expand_sve_subreg_move): Declare.
2280 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
2281 * config/aarch64/predicates.md (aarch64_any_register_operand): New
2283 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
2284 that are semantically a reverse operation.
2285 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
2286 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
2287 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
2289 (aarch64_can_change_mode_class): For big-endian, forbid changes
2290 between two SVE modes if they have different element sizes.
2292 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2294 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
2295 the TImode handling for big-endian targets.
2297 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2299 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
2300 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
2302 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
2303 Remove BSWAP handing for big-endian targets and use the form of
2304 LD1RQ appropariate for the mode.
2306 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2308 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
2309 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
2312 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2315 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
2316 check for operands that need to go through aarch64_sve_reload_be.
2318 2018-02-01 Jakub Jelinek <jakub@redhat.com>
2320 PR tree-optimization/81661
2321 PR tree-optimization/84117
2322 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
2323 * tree-eh.c: Include gimplify.h.
2324 (find_trapping_overflow, replace_trapping_overflow,
2325 rewrite_to_non_trapping_overflow): New functions.
2326 * tree-vect-loop.c: Include tree-eh.h.
2327 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
2328 * tree-data-ref.c: Include tree-eh.h.
2329 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
2331 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
2333 PR rtl-optimization/84123
2334 * combine.c (change_zero_ext): Check if hard register satisfies
2335 can_change_dest_mode before calling gen_lowpart_SUBREG.
2337 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
2340 * ira.c (ira_init_register_move_cost): Remove assert.
2342 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2344 PR rtl-optimization/84071
2345 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
2346 * doc/tm.texi: Regenerate.
2348 2018-01-31 Richard Biener <rguenther@suse.de>
2350 PR tree-optimization/84132
2351 * tree-data-ref.c (analyze_miv_subscript): Properly
2352 check whether evolution_function_is_affine_multivariate_p
2353 before calling gcd_of_steps_may_divide_p.
2355 2018-01-31 Julia Koval <julia.koval@intel.com>
2358 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
2359 * config/i386/i386.md (rdpid_rex64) New.
2360 (rdpid): Make 32bit only.
2362 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
2365 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
2366 an IDENTIFIER_NODE for FUNCTION_TYPE's.
2368 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2371 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2373 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2375 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2377 PR rtl-optimization/84071
2378 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
2379 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
2381 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
2383 * config/arc/arc.c (arc_handle_aux_attribute): New function.
2384 (arc_attribute_table): Add 'aux' attribute.
2385 (arc_in_small_data_p): Consider aux like variables.
2386 (arc_is_aux_reg_p): New function.
2387 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
2388 (arc_get_aux_arg): New function.
2389 (prepare_move_operands): Handle aux-register access.
2390 (arc_handle_aux_attribute): New function.
2391 * doc/extend.texi (ARC Variable attributes): Add subsection.
2393 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
2395 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
2396 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
2397 (arc_attribute_table): Add 'uncached' attribute.
2398 (arc_print_operand): Print '.di' flag for uncached memory
2400 (arc_in_small_data_p): Do not consider for small data the uncached
2402 (arc_is_uncached_mem_p): New function.
2403 * config/arc/predicates.md (compact_store_memory_operand): Check
2404 for uncached memory accesses.
2405 (nonvol_nonimm_operand): Likewise.
2406 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
2408 2018-01-31 Jakub Jelinek <jakub@redhat.com>
2411 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
2412 falign-loops=): Add Optimization flag.
2414 2018-01-30 Jeff Law <law@redhat.com>
2417 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
2418 INT_REGISTERS_SAVED. Check it prior to calling
2419 get_scratch_register_on_entry.
2420 (ix86_adjust_stack_and_probe): Similarly.
2421 (ix86_emit_probe_stack_range): Similarly.
2422 (ix86_expand_prologue): Corresponding changes.
2424 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2427 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
2428 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
2430 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
2433 * lra-constraints.c (curr_insn_transform): Process AND in the
2436 2018-01-30 Jakub Jelinek <jakub@redhat.com>
2438 PR rtl-optimization/83986
2439 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
2440 dependence against last_pending_memory_flush in addition to
2443 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
2445 PR tree-optimization/81611
2446 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
2449 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2452 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
2455 2018-01-30 Richard Biener <rguenther@suse.de>
2456 Jakub Jelinek <jakub@redhat.com>
2458 PR tree-optimization/84111
2459 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
2460 inner loops added during recursion, as they don't have up-to-date
2463 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2466 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
2467 (can_inline_edge_by_limits_p): ... here.
2468 (can_early_inline_edge_p, check_callers,
2469 update_caller_keys, update_callee_keys, recursive_inlining,
2470 add_new_edges_to_heap, speculation_useful_p,
2471 inline_small_functions,
2472 inline_small_functions, flatten_function,
2473 inline_to_all_callers_1): Update.
2475 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2477 * profile-count.c (profile_count::combine_with_ipa_count): Handle
2480 2018-01-30 Richard Biener <rguenther@suse.de>
2482 PR tree-optimization/83008
2483 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
2484 invariant and constant vector uses in stmts when they need
2487 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2490 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
2491 * configure: Regenerate.
2493 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2495 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
2497 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
2498 Use gen_rtx_REG rather than gen_lowpart.
2500 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2502 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
2503 rather than 0 when creating partial subregs.
2505 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2507 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
2510 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
2513 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
2514 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
2515 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
2516 flags. This restores the settings used before the 2017-07-24.
2517 Turning off pre increment/decrement/modify allows IVOPTS to
2518 optimize DF/SF loops where the index is an int.
2520 2018-01-29 Richard Biener <rguenther@suse.de>
2521 Kelvin Nilsen <kelvin@gcc.gnu.org>
2524 * tree-vect-stmts.c (vectorizable_call): Don't call
2525 targetm.vectorize_builtin_md_vectorized_function if callee is
2528 2018-01-22 Carl Love <cel@us.ibm.com>
2530 * doc/extend.tex: Fix typo in second arg in
2531 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
2533 2018-01-29 Richard Biener <rguenther@suse.de>
2535 PR tree-optimization/84086
2536 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
2537 (flush_ssaname_freelist): When SSA names were released reset
2538 the SCEV hash table.
2540 2018-01-29 Richard Biener <rguenther@suse.de>
2542 PR tree-optimization/84057
2543 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
2544 removed paths when removing edges.
2546 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
2548 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
2549 -mfunction-return=@var{choice}.
2551 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
2554 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
2556 (layout::print_source_line): Likewise.
2557 (test_get_line_width_without_trailing_whitespace): Add test cases.
2559 2018-01-27 Jakub Jelinek <jakub@redhat.com>
2562 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
2565 2018-01-26 Jim Wilson <jimw@sifive.com>
2567 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
2569 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
2572 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2574 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
2575 and CMP + SUB-immediate -> SUBS.
2577 2018-01-26 Martin Sebor <msebor@redhat.com>
2579 PR tree-optimization/83896
2580 * tree-ssa-strlen.c (get_string_len): Rename...
2581 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
2582 Avoid assuming length is constant.
2583 (handle_char_store): Use HOST_WIDE_INT for string length.
2585 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
2588 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
2589 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
2591 2018-01-26 Richard Biener <rguenther@suse.de>
2593 PR rtl-optimization/84003
2594 * dse.c (record_store): Only record redundant stores when
2595 the earlier store aliases at least all accesses the later one does.
2597 2018-01-26 Jakub Jelinek <jakub@redhat.com>
2599 PR rtl-optimization/83985
2600 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
2601 REG_CFA_RESTORE insns.
2602 (delete_unmarked_insns): Don't ignore separate shrink wrapping
2603 REG_CFA_RESTORE insns here.
2606 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
2607 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
2609 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2611 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
2612 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
2613 (arc_init): Likewise.
2614 (arc_override_options): Likewise.
2615 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
2617 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
2619 * config/arc/arc.h (TARGET_DBNZ): Define.
2620 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
2621 properly set the tune attribute.
2622 (dbnz): Use TARGET_DBNZ guard.
2623 * config/arc/arc.opt (mtune): Add core3 option.
2625 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2627 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
2628 recognize new pic like addresses.
2629 (arc_delegitimize_address): Clean up.
2631 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2633 * config/arc/arc-arches.def: Option mrf16 valid for all
2635 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
2636 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
2637 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
2638 * config/arc/arc-tables.opt: Regenerate.
2639 * config/arc/arc.c (arc_conditional_register_usage): Handle
2640 reduced register file case.
2641 (arc_file_start): Set must have build attributes.
2642 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
2644 * config/arc/arc.opt (mrf16): Add new option.
2645 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
2646 * config/arc/genmultilib.awk: Handle new mrf16 option.
2647 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
2648 * config/arc/t-multilib: Regenerate.
2649 * doc/invoke.texi (ARC Options): Document mrf16 option.
2651 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2653 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
2654 * config/arc/arc.c (arc_handle_secure_attribute): New function.
2655 (arc_attribute_table): Add 'secure_call' attribute.
2656 (arc_print_operand): Print secure call operand.
2657 (arc_function_ok_for_sibcall): Don't optimize tail calls when
2659 (arc_is_secure_call_p): New function. * config/arc/arc.md
2660 (call_i): Add support for sjli instruction.
2661 (call_value_i): Likewise.
2662 * config/arc/constraints.md (Csc): New constraint.
2664 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2665 John Eric Martin <John.Martin@emmicro-us.com>
2667 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
2668 * config/arc/arc.c (_arc_jli_section): New struct.
2669 (arc_jli_section): New type.
2670 (rc_jli_sections): New static variable.
2671 (arc_handle_jli_attribute): New function.
2672 (arc_attribute_table): Add jli_always and jli_fixed attribute.
2673 (arc_file_end): New function.
2674 (TARGET_ASM_FILE_END): Define.
2675 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
2676 (arc_add_jli_section): New function.
2677 (jli_call_scan): Likewise.
2678 (arc_reorg): Call jli_call_scan.
2679 (arc_output_addsi): Remove 'S' from printing asm operand.
2680 (arc_is_jli_call_p): New function.
2681 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
2683 (movhi_insn): Likewise.
2684 (movsi_insn): Likewise.
2685 (movsi_set_cc_insn): Likewise.
2686 (loadqi_update): Likewise.
2687 (load_zeroextendqisi_update): Likewise.
2688 (load_signextendqisi_update): Likewise.
2689 (loadhi_update): Likewise.
2690 (load_zeroextendhisi_update): Likewise.
2691 (load_signextendhisi_update): Likewise.
2692 (loadsi_update): Likewise.
2693 (loadsf_update): Likewise.
2694 (movsicc_insn): Likewise.
2695 (bset_insn): Likewise.
2696 (bxor_insn): Likewise.
2697 (bclr_insn): Likewise.
2698 (bmsk_insn): Likewise.
2699 (bicsi3_insn): Likewise.
2700 (cmpsi_cc_c_insn): Likewise.
2701 (movsi_ne): Likewise.
2702 (movsi_cond_exec): Likewise.
2703 (clrsbsi2): Likewise.
2714 (call_i): Remove 'S' asm letter, add jli instruction.
2715 (call_value_i): Likewise.
2716 * config/arc/arc.op (mjli-always): New option.
2717 * config/arc/constraints.md (Cji): New constraint.
2718 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
2720 (subsf3_fpx): Likewise.
2721 (mulsf3_fpx): Likewise.
2722 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
2724 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
2726 * doc/invoke.texi (ARC): Document mjli-always option.
2728 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
2730 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
2731 avoid addition with 0 and use incw and decw where possible.
2733 2018-01-26 Richard Biener <rguenther@suse.de>
2735 PR tree-optimization/81082
2736 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
2737 association if it requires casting to unsigned.
2738 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
2739 from fold_plusminus_mult_expr to catch important cases late when
2740 range info is available.
2742 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2744 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
2745 * configure.ac (hidden_linkonce): New test.
2746 * configure: Regenerate.
2747 * config.in: Regenerate.
2749 2018-01-26 Julia Koval <julia.koval@intel.com>
2751 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
2752 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
2753 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
2754 _mm_mask_bitshuffle_epi64_mask): Fix type.
2755 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
2756 USI_FTYPE_V4DI_V4DI_USI): Remove.
2757 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
2758 __builtin_ia32_vpshufbitqmb256_mask,
2759 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
2760 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
2761 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
2763 2018-01-26 Alan Modra <amodra@gmail.com>
2766 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
2767 UNSPEC_VBPERMQ. Sort other unspecs.
2769 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
2771 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
2773 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
2776 * predict.c (drop_profile): Do not push/pop cfun; update also
2778 (handle_missing_profiles): Fix logic looking for zero profiles.
2780 2018-01-25 Jakub Jelinek <jakub@redhat.com>
2783 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
2784 on functions with #pragma omp declare simd or functions with simd
2786 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
2787 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
2788 Remove trailing \n from warning_at calls.
2790 2018-01-25 Tom de Vries <tom@codesourcery.com>
2793 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2794 for neutered workers.
2796 2018-01-24 Joseph Myers <joseph@codesourcery.com>
2799 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
2800 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
2802 2018-01-24 Jeff Law <law@redhat.com>
2805 * i386.c (get_probe_interval): Move to earlier point.
2806 (ix86_compute_frame_layout): If -fstack-clash-protection and
2807 the frame is larger than the probe interval, then use pushes
2808 to save registers rather than reg->mem moves.
2809 (ix86_expand_prologue): Remove conditional for int_registers_saved
2812 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
2815 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
2816 min/max for never referenced object.
2818 2018-01-24 Jakub Jelinek <jakub@redhat.com>
2821 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
2823 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
2824 attributes from DECL_ATTRIBUTES (decl) without affecting
2825 DECL_ATTRIBUTES (current_function_decl).
2826 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
2827 functions with non-NULL DECL_ABSTRACT_ORIGIN.
2829 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
2831 PR tree-optimization/83979
2832 * fold-const.c (fold_comparison): Use constant_boolean_node
2833 instead of boolean_{true,false}_node.
2835 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
2837 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
2840 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2842 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2843 Simplify the clause that sets the length attribute.
2844 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2845 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
2846 clause that sets the length attribute.
2847 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2849 2018-01-24 Tom de Vries <tom@codesourcery.com>
2852 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
2853 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
2854 Add strict parameter.
2855 (prevent_branch_around_nothing): Insert dummy insn between branch to
2856 label and label with no ptx insn inbetween.
2857 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
2859 2018-01-24 Tom de Vries <tom@codesourcery.com>
2862 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2863 for neutered threads in warp.
2864 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
2866 2018-01-24 Richard Biener <rguenther@suse.de>
2868 PR tree-optimization/83176
2869 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
2872 2018-01-24 Richard Biener <rguenther@suse.de>
2874 PR tree-optimization/82819
2875 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
2876 code generating pluses that are no-ops in the target precision.
2878 2018-01-24 Richard Biener <rguenther@suse.de>
2881 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
2883 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2885 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
2886 to merge probabilities.
2887 * predict.c (probably_never_executed): Also mark as cold functions
2888 with global 0 profile and guessed local profile.
2889 * profile-count.c (profile_probability::combine_with_count): New
2891 * profile-count.h (profile_probability::operator*,
2892 profile_probability::operator*=, profile_probability::operator/,
2893 profile_probability::operator/=): Reduce precision to adjusted
2894 and set value to guessed on contradictory divisions.
2895 (profile_probability::combine_with_freq): Remove.
2896 (profile_probability::combine_wiht_count): Declare.
2897 (profile_count::force_nonzero):: Set to adjusted.
2898 (profile_count::probability_in):: Set quality to adjusted.
2899 * tree-ssa-tail-merge.c (replace_block_by): Use
2902 2018-01-23 Andrew Waterman <andrew@sifive.com>
2903 Jim Wilson <jimw@sifive.com>
2905 * config/riscv/riscv.c (riscv_stack_boundary): New.
2906 (riscv_option_override): Set riscv_stack_boundary. Handle
2907 riscv_preferred_stack_boundary_arg.
2908 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
2909 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
2910 (STACK_BOUNDARY): Set to riscv_stack_boundary.
2911 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
2912 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
2913 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
2915 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
2918 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
2919 of struct ix86_frame.
2920 (ix86_expand_epilogue): Likewise. Add a local variable for
2921 the reg_save_offset field in struct ix86_frame.
2923 2018-01-23 Bin Cheng <bin.cheng@arm.com>
2925 PR tree-optimization/82604
2926 * tree-loop-distribution.c (enum partition_kind): New enum item
2927 PKIND_PARTIAL_MEMSET.
2928 (partition_builtin_p): Support above new enum item.
2929 (generate_code_for_partition): Ditto.
2930 (compute_access_range): Differentiate cases that equality can be
2931 proven at all loops, the innermost loops or no loops.
2932 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
2933 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
2934 (finalize_partitions, distribute_loop): Don't fuse partition of
2935 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
2936 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
2939 2018-01-23 Martin Liska <mliska@suse.cz>
2941 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
2942 order to ignore the predictor.
2943 (PRED_POLYMORPHIC_CALL): Likewise.
2944 (PRED_RECURSIVE_CALL): Likewise.
2946 2018-01-23 Martin Liska <mliska@suse.cz>
2948 * tree-profile.c (tree_profiling): Print function header to
2949 aware reader which function we are working on.
2950 * value-prof.c (gimple_find_values_to_profile): Do not print
2951 not interesting value histograms.
2953 2018-01-23 Martin Liska <mliska@suse.cz>
2955 * profile-count.h (enum profile_quality): Add
2956 profile_uninitialized as the first value. Do not number values
2957 as they are zero based.
2958 (profile_count::verify): Update sanity check.
2959 (profile_probability::verify): Likewise.
2961 2018-01-23 Nathan Sidwell <nathan@acm.org>
2963 * doc/invoke.texi (ffor-scope): Deprecate.
2965 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2967 PR tree-optimization/83510
2968 * domwalk.c (set_all_edges_as_executable): New function.
2969 (dom_walker::dom_walker): Convert bool param
2970 "skip_unreachable_blocks" to enum reachability. Move setup of
2971 edge flags to set_all_edges_as_executable and only do it when
2972 reachability is REACHABLE_BLOCKS.
2973 * domwalk.h (enum dom_walker::reachability): New enum.
2974 (dom_walker::dom_walker): Convert bool param
2975 "skip_unreachable_blocks" to enum reachability.
2976 (set_all_edges_as_executable): New decl.
2977 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2978 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2980 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2981 but converting true to REACHABLE_BLOCKS.
2982 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2984 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2985 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2986 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2988 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2989 if check_all_array_refs will be called.
2991 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2993 * tree.c (selftest::test_location_wrappers): Add more test
2996 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2998 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2999 (selftest::test_bit_in_range): Likewise.
3001 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
3004 * doc/sourcebuild.texi (vect_float): Say that the selector
3005 only describes the situation when -funsafe-math-optimizations is on.
3006 (vect_float_strict): Document.
3008 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
3010 PR tree-optimization/83965
3011 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
3012 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
3013 instead of checking only for a reduction.
3014 (vect_recog_widen_sum_pattern): Likewise.
3016 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
3018 * predict.c (probably_never_executed): Only use precise profile info.
3019 (compute_function_frequency): Skip after inlining hack since we now
3020 have quality checking.
3022 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
3024 * profile-count.h (profile_probability::very_unlikely,
3025 profile_probability::unlikely, profile_probability::even): Set
3026 precision to guessed.
3028 2018-01-23 Richard Biener <rguenther@suse.de>
3030 PR tree-optimization/83963
3031 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
3032 Properly terminate dominator walk when crossing the exit edge not
3033 when visiting its source block.
3035 2018-01-23 Jakub Jelinek <jakub@redhat.com>
3038 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
3039 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
3041 2018-01-22 Jakub Jelinek <jakub@redhat.com>
3043 PR tree-optimization/83957
3044 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
3045 semicolon after for body surrounded by braces.
3047 PR tree-optimization/83081
3048 * profile-count.h (profile_probability::split): New method.
3049 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
3050 Use profile_probability::split.
3051 (do_compare_rtx_and_jump): Fix adjustment of probabilities
3052 when splitting a single conditional jump into 2.
3054 2018-01-22 David Malcolm <dmalcolm@redhat.com>
3056 PR tree-optimization/69452
3057 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
3060 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3062 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
3063 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
3064 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
3066 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3068 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
3069 * config/rl78/rl78.md (movdi): New define_expand.
3070 * config/rl78/rl78.c (rl78_split_movdi): New function.
3072 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
3075 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
3077 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
3078 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
3079 128-bit to produce an UNSPEC move to get the double word with the
3080 signbit and then a shift directly to do signbit.
3081 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
3082 implementation with a new version that just does either a direct
3083 move or a regular move. Move memory interface to separate insns.
3084 Move insns so they are next to the expander.
3085 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
3086 with signbit move. Split big and little endian case.
3087 (signbit<mode>2_dm_mem_le): Likewise.
3088 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
3089 (signbit<mode>2_dm2): Likewise.
3091 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3093 * config/rl78/rl78.md (anddi3): New define_expand.
3095 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3097 * config/rl78/rl78.md (umindi3): New define_expand.
3099 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3101 * config/rl78/rl78.md (smindi3): New define_expand.
3103 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3105 * config/rl78/rl78.md (smaxdi3): New define_expand.
3107 2018-01-22 Carl Love <cel@us.ibm.com>
3109 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
3110 LVX_V1TI): Add macro expansion.
3111 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
3112 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
3113 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
3114 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
3115 Change check to determine if the instruction is a byte reversing
3116 entry. Fix typo in comment.
3117 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
3118 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
3119 Add def_builtin calls for new builtins.
3120 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
3121 Add define_insn expansion.
3123 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3125 * config/rl78/rl78.md (umaxdi3): New define_expand.
3127 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
3129 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
3130 for non-QImode registers.
3132 2018-01-22 Richard Biener <rguenther@suse.de>
3134 PR tree-optimization/83963
3135 * graphite-scop-detection.c (scop_detection::get_sese): Delay
3136 including the loop exit block.
3137 (scop_detection::merge_sese): Likewise.
3138 (scop_detection::add_scop): Do it here instead.
3140 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3142 * doc/sourcebuild.texi (arm_softfloat): Document.
3144 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
3147 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
3148 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
3149 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
3151 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3152 David Edelsohn <dje.gcc@gmail.com>
3155 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
3156 Change "crset eq" to "crset 2".
3157 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
3158 (*call_indirect_aix<mode>_nospec): Likewise.
3159 (*call_value_indirect_aix<mode>_nospec): Likewise.
3160 (*call_indirect_elfv2<mode>_nospec): Likewise.
3161 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
3162 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
3163 change assembly output from . to $.
3164 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
3165 (indirect_jump<mode>_nospec): Change assembly output from . to $.
3166 (*tablejump<mode>_internal1_nospec): Likewise.
3168 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
3171 * config/sh/sh_optimize_sett_clrt.cc:
3172 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
3174 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
3176 PR tree-optimization/83940
3177 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
3178 offset_dt to vect_constant_def rather than vect_unknown_def_type.
3179 (vect_check_load_store_mask): Add a mask_dt_out parameter and
3180 use it to pass back the definition type.
3181 (vect_check_store_rhs): Likewise rhs_dt_out.
3182 (vect_build_gather_load_calls): Add a mask_dt argument and use
3183 it instead of a call to vect_is_simple_use.
3184 (vectorizable_store): Update calls to vect_check_load_store_mask
3185 and vect_check_store_rhs. Use the dt returned by the latter instead
3186 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
3187 instead of calls to vect_is_simple_use. Pass the scalar rather
3188 than the vector operand to vect_is_simple_use when handling
3189 second and subsequent copies of an rhs value.
3190 (vectorizable_load): Update calls to vect_check_load_store_mask
3191 and vect_build_gather_load_calls. Use the cached mask_dt and
3192 gs_info.offset_dt instead of calls to vect_is_simple_use.
3194 2018-01-20 Jakub Jelinek <jakub@redhat.com>
3197 * tree-emutls.c: Include gimplify.h.
3198 (lower_emutls_2): New function.
3199 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
3200 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
3201 it before further processing.
3204 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
3205 UINTVAL (trueop1) instead of INTVAL (op1).
3207 2018-01-19 Jakub Jelinek <jakub@redhat.com>
3211 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
3212 INCOMING_FRAME_SP_OFFSET if not defined.
3213 (scan_trace): Add ENTRY argument. If true and
3214 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
3215 emit a note to adjust the CFA offset.
3216 (create_cfi_notes): Adjust scan_trace callers.
3217 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
3218 INCOMING_FRAME_SP_OFFSET in the CIE.
3219 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
3220 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
3222 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
3223 * doc/tm.texi: Regenerated.
3225 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
3227 PR rtl-optimization/83147
3228 * lra-constraints.c (remove_inheritance_pseudos): Use
3229 lra_substitute_pseudo_within_insn.
3231 2018-01-19 Tom de Vries <tom@codesourcery.com>
3232 Cesar Philippidis <cesar@codesourcery.com>
3235 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
3237 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
3240 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
3241 spaces for function labels.
3243 2018-01-19 Martin Liska <mliska@suse.cz>
3245 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
3246 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
3247 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
3248 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
3249 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
3250 (PRED_CONST_RETURN): Change from 69 to 65.
3251 (PRED_NULL_RETURN): Change from 91 to 71.
3252 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
3253 (PRED_LOOP_GUARD): Change from 66 to 73.
3255 2018-01-19 Martin Liska <mliska@suse.cz>
3257 * predict.c (predict_insn_def): Add new assert.
3258 (struct branch_predictor): Change type to signed integer.
3259 (test_prediction_value_range): Amend test to cover
3261 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
3262 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
3263 (PRED_LOOP_ITERATIONS_MAX): Likewise.
3264 (PRED_LOOP_IV_COMPARE): Likewise.
3265 * predict.h (PROB_UNINITIALIZED): Define new constant.
3267 2018-01-19 Martin Liska <mliska@suse.cz>
3269 * predict.c (dump_prediction): Add new format for
3270 analyze_brprob.py script which is enabled with -details
3272 * profile-count.h (precise_p): New function.
3274 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
3276 PR tree-optimization/83922
3277 * tree-vect-loop.c (vect_verify_full_masking): Return false if
3278 there are no statements that need masking.
3279 (vect_active_double_reduction_p): New function.
3280 (vect_analyze_loop_operations): Use it when handling phis that
3281 are not in the loop header.
3283 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
3285 PR tree-optimization/83914
3286 * tree-vect-loop.c (vectorizable_induction): Don't convert
3287 init_expr or apply the peeling adjustment for inductions
3288 that are nested within the vectorized loop.
3290 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3292 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
3295 2018-01-18 Jakub Jelinek <jakub@redhat.com>
3299 * function.h (gimplify_parameters): Add gimple_seq * argument.
3300 * function.c: Include gimple.h and options.h.
3301 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
3302 for the added local temporaries if needed.
3303 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
3304 if there are any parameter cleanups, wrap whole body into a
3305 try/finally with the cleanups.
3307 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
3310 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
3311 Use GET_MODE_CLASS for scalar floating point.
3313 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
3317 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
3318 Fix call of call_cgraph_insertion_hooks.
3320 2018-01-18 Martin Sebor <msebor@redhat.com>
3322 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
3324 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
3327 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
3330 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
3333 * common.opt: (-ffile-prefix-map): New option.
3334 * opts.c (common_handle_option): Defer it.
3335 * opts-global.c (handle_common_deferred_options): Handle it.
3336 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
3337 * file-prefix-map.h: New file.
3338 (remap_debug_filename, add_debug_prefix_map): ...here.
3339 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
3340 * final.c (debug_prefix_map, add_debug_prefix_map
3341 remap_debug_filename): Move to...
3342 * file-prefix-map.c: New file.
3343 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
3344 generalize, get rid of alloca(), use strrchr() instead of strchr().
3345 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
3346 Implement in terms of add_prefix_map().
3347 (remap_macro_filename, remap_debug_filename): Implement in term of
3349 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
3350 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
3351 * dbxout.c: Include file-prefix-map.h.
3352 * varasm.c: Likewise.
3353 * vmsdbgout.c: Likewise.
3354 * xcoffout.c: Likewise.
3355 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
3356 * doc/cppopts.texi (-fmacro-prefix-map): Document.
3357 * doc/invoke.texi (-ffile-prefix-map): Document.
3358 (-fdebug-prefix-map): Update description.
3360 2018-01-18 Martin Liska <mliska@suse.cz>
3362 * config/i386/i386.c (indirect_thunk_name): Document that also
3364 (output_indirect_thunk): Document why both instructions
3365 (pause and lfence) are generated.
3367 2018-01-18 Richard Biener <rguenther@suse.de>
3369 PR tree-optimization/83887
3370 * graphite-scop-detection.c
3371 (scop_detection::get_nearest_dom_with_single_entry): Remove.
3372 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
3373 (scop_detection::merge_sese): Re-implement with a flood-fill
3374 algorithm that properly finds a SESE region if it exists.
3376 2018-01-18 Jakub Jelinek <jakub@redhat.com>
3379 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
3380 pointer_diff optimizations use view_convert instead of convert.
3382 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3384 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
3385 Generate different code for -mno-speculate-indirect-jumps.
3386 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
3387 (*call_indirect_aix<mode>): Disable for
3388 -mno-speculate-indirect-jumps.
3389 (*call_indirect_aix<mode>_nospec): New define_insn.
3390 (*call_value_indirect_aix<mode>): Disable for
3391 -mno-speculate-indirect-jumps.
3392 (*call_value_indirect_aix<mode>_nospec): New define_insn.
3393 (*sibcall_nonlocal_sysv<mode>): Generate different code for
3394 -mno-speculate-indirect-jumps.
3395 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
3397 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
3399 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
3400 long double type, set the flags for noting the default long double
3401 type, even if we don't pass or return a long double type.
3403 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
3406 * ipa-inline.c (flatten_function): Do not overwrite final inlining
3409 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
3411 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
3412 support for merge[hl].
3413 (fold_mergehl_helper): New helper function.
3414 (tree-vector-builder.h): New #include for tree_vector_builder usage.
3415 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
3416 (altivec_vmrglw_direct): Add xxmrglw insn.
3418 2018-01-17 Andrew Waterman <andrew@sifive.com>
3420 * config/riscv/riscv.c (riscv_conditional_register_usage): If
3421 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
3423 2018-01-17 David Malcolm <dmalcolm@redhat.com>
3426 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
3427 call the lto_location_cache before reading the
3428 DECL_SOURCE_LOCATION of the types.
3430 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
3431 Richard Sandiford <richard.sandiford@linaro.org>
3433 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
3434 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
3435 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
3436 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
3437 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
3439 * config/aarch64/constraints.md (aarch64_movti_operand):
3441 * config/aarch64/predicates.md (Uti): Add new constraint.
3443 2018-01-17 Carl Love <cel@us.ibm.com>
3445 * config/rs6000/vsx.md (define_expand xl_len_r,
3446 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
3447 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
3449 (define_expand, define_insn): Move the shift left from the
3450 define_insn to the define_expand for lxvl and stxvl instructions.
3451 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
3452 and XL_LEN_R definitions to PURE.
3454 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
3456 * config/i386/i386.c (indirect_thunk_name): Declare regno
3457 as unsigned int. Compare regno with INVALID_REGNUM.
3458 (output_indirect_thunk): Ditto.
3459 (output_indirect_thunk_function): Ditto.
3460 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
3461 in the call to output_indirect_thunk_function.
3463 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
3466 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
3467 rather than the size of inner_type to determine the stack slot size
3468 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
3470 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
3473 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
3476 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
3478 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
3479 endian Linux systems to optionally enable multilibs for selecting
3480 the long double type if the user configured an explicit type.
3481 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
3482 have no long double multilibs if not defined.
3483 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
3484 warn if the user used -mabi={ieee,ibm}longdouble and we built
3485 multilibs for long double.
3486 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
3487 appropriate multilib option.
3488 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
3490 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
3491 for building long double multilibs.
3492 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
3494 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
3496 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
3499 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
3501 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
3504 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
3507 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
3510 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3512 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
3513 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
3515 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
3517 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
3518 different rtl trees depending on TARGET_64BIT.
3519 (rs6000_gen_lvx): Likewise.
3521 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3523 * config/visium/visium.md (nop): Tweak comment.
3524 (hazard_nop): Likewise.
3526 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3528 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
3529 -mspeculate-indirect-jumps.
3530 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
3531 for -mno-speculate-indirect-jumps.
3532 (*call_indirect_elfv2<mode>_nospec): New define_insn.
3533 (*call_value_indirect_elfv2<mode>): Disable for
3534 -mno-speculate-indirect-jumps.
3535 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
3536 (indirect_jump): Emit different RTL for
3537 -mno-speculate-indirect-jumps.
3538 (*indirect_jump<mode>): Disable for
3539 -mno-speculate-indirect-jumps.
3540 (*indirect_jump<mode>_nospec): New define_insn.
3541 (tablejump): Emit different RTL for
3542 -mno-speculate-indirect-jumps.
3543 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
3544 (tablejumpsi_nospec): New define_expand.
3545 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
3546 (tablejumpdi_nospec): New define_expand.
3547 (*tablejump<mode>_internal1): Disable for
3548 -mno-speculate-indirect-jumps.
3549 (*tablejump<mode>_internal1_nospec): New define_insn.
3550 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
3553 2018-01-16 Artyom Skrobov tyomitch@gmail.com
3555 * caller-save.c (insert_save): Drop unnecessary parameter. All
3558 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3559 Richard Biener <rguenth@suse.de>
3562 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
3563 return early, inline manually is_gimple_sizepos. Make sure if we
3564 call gimplify_expr we don't end up with a gimple constant.
3565 * tree.c (variably_modified_type_p): Don't return true for
3566 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
3567 * gimplify.h (is_gimple_sizepos): Remove.
3569 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3571 PR tree-optimization/83857
3572 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
3573 vectorizable_live_operation for pure SLP statements.
3574 (vectorizable_live_operation): Handle PHIs.
3576 2018-01-16 Richard Biener <rguenther@suse.de>
3578 PR tree-optimization/83867
3579 * tree-vect-stmts.c (vect_transform_stmt): Precompute
3580 nested_in_vect_loop_p since the scalar stmt may get invalidated.
3582 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3585 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
3586 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
3587 If off is not INTEGER_CST, issue a may not be aligned warning
3588 rather than isn't aligned. Use isn%'t rather than isn't.
3589 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
3591 <case MULT_EXPR>: Improve the case when bottom and one of the
3592 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
3593 operand, in that case check if the other operand is multiple of
3594 bottom divided by the INTEGER_CST operand.
3596 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3599 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
3600 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
3601 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
3602 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
3603 * config/pa/pa.c (pa_function_arg_advance): Likewise.
3604 (pa_function_arg, pa_arg_partial_bytes): Likewise.
3605 (pa_function_arg_size): New function.
3607 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3609 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
3610 in a separate statement.
3612 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3614 PR tree-optimization/83847
3615 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
3616 group gathers and scatters.
3618 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3620 PR rtl-optimization/86620
3621 * params.def (max-sched-ready-insns): Bump minimum value to 1.
3623 PR rtl-optimization/83213
3624 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
3625 to last if both are JUMP_INSNs.
3627 PR tree-optimization/83843
3628 * gimple-ssa-store-merging.c
3629 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
3630 store_immediate_info for bswap/nop orig_stores.
3632 2018-01-15 Andrew Waterman <andrew@sifive.com>
3634 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
3636 <UDIV>: Increase cost if !TARGET_DIV.
3638 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
3640 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
3641 (define_attr "cr_logical_3op"): New.
3642 (cceq_ior_compare): Adjust.
3643 (cceq_ior_compare_complement): Adjust.
3644 (*cceq_rev_compare): Adjust.
3645 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
3646 (is_cracked_insn): Adjust.
3647 (insn_must_be_first_in_group): Adjust.
3648 * config/rs6000/40x.md: Adjust.
3649 * config/rs6000/440.md: Adjust.
3650 * config/rs6000/476.md: Adjust.
3651 * config/rs6000/601.md: Adjust.
3652 * config/rs6000/603.md: Adjust.
3653 * config/rs6000/6xx.md: Adjust.
3654 * config/rs6000/7450.md: Adjust.
3655 * config/rs6000/7xx.md: Adjust.
3656 * config/rs6000/8540.md: Adjust.
3657 * config/rs6000/cell.md: Adjust.
3658 * config/rs6000/e300c2c3.md: Adjust.
3659 * config/rs6000/e500mc.md: Adjust.
3660 * config/rs6000/e500mc64.md: Adjust.
3661 * config/rs6000/e5500.md: Adjust.
3662 * config/rs6000/e6500.md: Adjust.
3663 * config/rs6000/mpc.md: Adjust.
3664 * config/rs6000/power4.md: Adjust.
3665 * config/rs6000/power5.md: Adjust.
3666 * config/rs6000/power6.md: Adjust.
3667 * config/rs6000/power7.md: Adjust.
3668 * config/rs6000/power8.md: Adjust.
3669 * config/rs6000/power9.md: Adjust.
3670 * config/rs6000/rs64.md: Adjust.
3671 * config/rs6000/titan.md: Adjust.
3673 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3675 * config/i386/predicates.md (indirect_branch_operand): Rewrite
3676 ix86_indirect_branch_register logic.
3678 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3680 * config/i386/constraints.md (Bs): Update
3681 ix86_indirect_branch_register check. Don't check
3682 ix86_indirect_branch_register with GOT_memory_operand.
3684 * config/i386/predicates.md (GOT_memory_operand): Don't check
3685 ix86_indirect_branch_register here.
3686 (GOT32_symbol_operand): Likewise.
3688 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3690 * config/i386/predicates.md (constant_call_address_operand):
3691 Rewrite ix86_indirect_branch_register logic.
3692 (sibcall_insn_operand): Likewise.
3694 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3696 * config/i386/constraints.md (Bs): Replace
3697 ix86_indirect_branch_thunk_register with
3698 ix86_indirect_branch_register.
3700 * config/i386/i386.md (indirect_jump): Likewise.
3701 (tablejump): Likewise.
3702 (*sibcall_memory): Likewise.
3703 (*sibcall_value_memory): Likewise.
3704 Peepholes of indirect call and jump via memory: Likewise.
3705 * config/i386/i386.opt: Likewise.
3706 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3707 (GOT_memory_operand): Likewise.
3708 (call_insn_operand): Likewise.
3709 (sibcall_insn_operand): Likewise.
3710 (GOT32_symbol_operand): Likewise.
3712 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3715 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
3716 type rather than type addr's type points to.
3717 (expand_omp_atomic_mutex): Likewise.
3718 (expand_omp_atomic): Likewise.
3720 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3723 * config/i386/i386.c (output_indirect_thunk_function): Use
3724 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
3725 for __x86_return_thunk.
3727 2018-01-15 Richard Biener <rguenther@suse.de>
3730 * expmed.c (extract_bit_field_1): Fix typo.
3732 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3735 * config/arm/iterators.md (VF): New mode iterator.
3736 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
3737 Remove integer-related logic from pattern.
3738 (neon_vabd<mode>_3): Likewise.
3740 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3743 * common.opt (fstrict-overflow): No longer an alias.
3744 (fwrapv-pointer): New option.
3745 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
3746 also for pointer types based on flag_wrapv_pointer.
3747 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
3748 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
3749 opts->x_flag_wrapv got set.
3750 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
3751 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
3752 POINTER_TYPE_OVERFLOW_UNDEFINED.
3753 * match.pd: Likewise in address comparison pattern.
3754 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
3756 2018-01-15 Richard Biener <rguenther@suse.de>
3759 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
3760 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
3761 Reset type names to their identifier if their TYPE_DECL doesn't
3762 have linkage (and thus is used for ODR and devirt).
3763 (save_debug_info_for_decl): Remove.
3764 (save_debug_info_for_type): Likewise.
3765 (add_tree_to_fld_list): Adjust.
3766 * tree-pretty-print.c (dump_generic_node): Make dumping of
3767 type names more robust.
3769 2018-01-15 Richard Biener <rguenther@suse.de>
3771 * BASE-VER: Bump to 8.0.1.
3773 2018-01-14 Martin Sebor <msebor@redhat.com>
3776 * builtins.c (check_access): Avoid warning when the no-warning bit
3779 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
3781 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
3782 * ira-color (allocno_hard_regs_compare): Likewise.
3784 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
3787 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
3788 Use .pushsection/.popsection.
3790 2018-01-14 Martin Sebor <msebor@redhat.com>
3793 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
3795 2018-01-14 Jakub Jelinek <jakub@redhat.com>
3797 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
3798 entry from extra_headers.
3799 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
3800 extra_headers, make the list bitwise identical to the i?86-*-* one.
3802 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3804 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
3805 -mcmodel=large with -mindirect-branch=thunk,
3806 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
3807 -mfunction-return=thunk-extern.
3808 * doc/invoke.texi: Document -mcmodel=large is incompatible with
3809 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
3810 -mfunction-return=thunk and -mfunction-return=thunk-extern.
3812 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3814 * config/i386/i386.c (print_reg): Print the name of the full
3815 integer register without '%'.
3816 (ix86_print_operand): Handle 'V'.
3817 * doc/extend.texi: Document 'V' modifier.
3819 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3821 * config/i386/constraints.md (Bs): Disallow memory operand for
3822 -mindirect-branch-register.
3824 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3825 (GOT_memory_operand): Likewise.
3826 (call_insn_operand): Likewise.
3827 (sibcall_insn_operand): Likewise.
3828 (GOT32_symbol_operand): Likewise.
3829 * config/i386/i386.md (indirect_jump): Call convert_memory_address
3830 for -mindirect-branch-register.
3831 (tablejump): Likewise.
3832 (*sibcall_memory): Likewise.
3833 (*sibcall_value_memory): Likewise.
3834 Disallow peepholes of indirect call and jump via memory for
3835 -mindirect-branch-register.
3836 (*call_pop): Replace m with Bw.
3837 (*call_value_pop): Likewise.
3838 (*sibcall_pop_memory): Replace m with Bs.
3839 * config/i386/i386.opt (mindirect-branch-register): New option.
3840 * doc/invoke.texi: Document -mindirect-branch-register option.
3842 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3844 * config/i386/i386-protos.h (ix86_output_function_return): New.
3845 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
3846 set function_return_type.
3847 (indirect_thunk_name): Add ret_p to indicate thunk for function
3849 (output_indirect_thunk_function): Pass false to
3850 indirect_thunk_name.
3851 (ix86_output_indirect_branch_via_reg): Likewise.
3852 (ix86_output_indirect_branch_via_push): Likewise.
3853 (output_indirect_thunk_function): Create alias for function
3854 return thunk if regno < 0.
3855 (ix86_output_function_return): New function.
3856 (ix86_handle_fndecl_attribute): Handle function_return.
3857 (ix86_attribute_table): Add function_return.
3858 * config/i386/i386.h (machine_function): Add
3859 function_return_type.
3860 * config/i386/i386.md (simple_return_internal): Use
3861 ix86_output_function_return.
3862 (simple_return_internal_long): Likewise.
3863 * config/i386/i386.opt (mfunction-return=): New option.
3864 (indirect_branch): Mention -mfunction-return=.
3865 * doc/extend.texi: Document function_return function attribute.
3866 * doc/invoke.texi: Document -mfunction-return= option.
3868 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3870 * config/i386/i386-opts.h (indirect_branch): New.
3871 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
3872 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
3873 with local indirect jump when converting indirect call and jump.
3874 (ix86_set_indirect_branch_type): New.
3875 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
3876 (indirectlabelno): New.
3877 (indirect_thunk_needed): Likewise.
3878 (indirect_thunk_bnd_needed): Likewise.
3879 (indirect_thunks_used): Likewise.
3880 (indirect_thunks_bnd_used): Likewise.
3881 (INDIRECT_LABEL): Likewise.
3882 (indirect_thunk_name): Likewise.
3883 (output_indirect_thunk): Likewise.
3884 (output_indirect_thunk_function): Likewise.
3885 (ix86_output_indirect_branch_via_reg): Likewise.
3886 (ix86_output_indirect_branch_via_push): Likewise.
3887 (ix86_output_indirect_branch): Likewise.
3888 (ix86_output_indirect_jmp): Likewise.
3889 (ix86_code_end): Call output_indirect_thunk_function if needed.
3890 (ix86_output_call_insn): Call ix86_output_indirect_branch if
3892 (ix86_handle_fndecl_attribute): Handle indirect_branch.
3893 (ix86_attribute_table): Add indirect_branch.
3894 * config/i386/i386.h (machine_function): Add indirect_branch_type
3895 and has_local_indirect_jump.
3896 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
3898 (tablejump): Likewise.
3899 (*indirect_jump): Use ix86_output_indirect_jmp.
3900 (*tablejump_1): Likewise.
3901 (simple_return_indirect_internal): Likewise.
3902 * config/i386/i386.opt (mindirect-branch=): New option.
3903 (indirect_branch): New.
3906 (thunk-inline): Likewise.
3907 (thunk-extern): Likewise.
3908 * doc/extend.texi: Document indirect_branch function attribute.
3909 * doc/invoke.texi: Document -mindirect-branch= option.
3911 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
3914 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
3916 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
3918 * ipa-inline.c (want_inline_small_function_p): Return false if
3919 inlining has already failed with CIF_FINAL_ERROR.
3920 (update_caller_keys): Call want_inline_small_function_p before
3922 (update_callee_keys): Likewise.
3924 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
3926 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
3928 (rs6000_quadword_masked_address_p): Likewise.
3929 (quad_aligned_load_p): Likewise.
3930 (quad_aligned_store_p): Likewise.
3931 (const_load_sequence_p): Add comment to describe the outer-most loop.
3932 (mimic_memory_attributes_and_flags): New function.
3933 (rs6000_gen_stvx): Likewise.
3934 (replace_swapped_aligned_store): Likewise.
3935 (rs6000_gen_lvx): Likewise.
3936 (replace_swapped_aligned_load): Likewise.
3937 (replace_swapped_load_constant): Capitalize argument name in
3938 comment describing this function.
3939 (rs6000_analyze_swaps): Add a third pass to search for vector loads
3940 and stores that access quad-word aligned addresses and replace
3941 with stvx or lvx instructions when appropriate.
3942 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
3943 New function prototype.
3944 (rs6000_quadword_masked_address_p): Likewise.
3945 (rs6000_gen_lvx): Likewise.
3946 (rs6000_gen_stvx): Likewise.
3947 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3948 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3949 when memory address is aligned.
3950 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3951 this split to select lvx instruction when memory address is aligned.
3952 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3953 instruction when memory address is aligned.
3954 (*vsx_le_perm_load_v16qi): Likewise.
3955 (four unnamed splitters): Modify to select the stvx instruction
3956 when memory is aligned.
3958 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3960 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3961 which appears in the queue multiple times.
3963 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3964 Alan Hayward <alan.hayward@arm.com>
3965 David Sherwood <david.sherwood@arm.com>
3967 * tree-vectorizer.h (vec_lower_bound): New structure.
3968 (_loop_vec_info): Add check_nonzero and lower_bounds.
3969 (LOOP_VINFO_CHECK_NONZERO): New macro.
3970 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3971 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3972 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3973 fields. Make seg_len the distance travelled, not including the
3975 (dr_direction_indicator): Declare.
3976 (dr_zero_step_indicator): Likewise.
3977 (dr_known_forward_stride_p): Likewise.
3978 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3980 (runtime_alias_check_p): Allow runtime alias checks with
3982 (operator ==): Compare access_size and align.
3983 (prune_runtime_alias_test_list): Rework for new distinction between
3984 the access_size and seg_len.
3985 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3987 (get_segment_min_max): New function.
3988 (create_intersect_range_checks): Use it.
3989 (dr_step_indicator): New function.
3990 (dr_direction_indicator): Likewise.
3991 (dr_zero_step_indicator): Likewise.
3992 (dr_known_forward_stride_p): Likewise.
3993 * tree-loop-distribution.c (data_ref_segment_size): Return
3994 DR_STEP * (niters - 1).
3995 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3997 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3998 (vect_preserves_scalar_order_p): New function, split out from...
3999 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
4000 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
4001 (vect_vfa_access_size): New function.
4002 (vect_vfa_align): Likewise.
4003 (vect_compile_time_alias): Take access_size_a and access_b arguments.
4004 (dump_lower_bound): New function.
4005 (vect_check_lower_bound): Likewise.
4006 (vect_small_gap_p): Likewise.
4007 (vectorizable_with_step_bound_p): Likewise.
4008 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
4009 depencies if the vectorization factor is 1. Convert the checks
4010 for nonzero steps into checks on the bounds of DR_STEP. Try using
4011 a bunds check for variable steps if the minimum required step is
4012 relatively small. Update calls to the dr_with_seg_len
4013 constructor and to vect_compile_time_alias.
4014 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
4016 (vect_loop_versioning): Call it.
4017 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
4019 (vect_estimate_min_profitable_iters): Account for any bounds checks.
4021 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4022 Alan Hayward <alan.hayward@arm.com>
4023 David Sherwood <david.sherwood@arm.com>
4025 * doc/sourcebuild.texi (vect_scatter_store): Document.
4026 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
4028 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
4030 * genopinit.c (main): Add supports_vec_scatter_store and
4031 supports_vec_scatter_store_cached to target_optabs.
4032 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
4033 IFN_MASK_SCATTER_STORE.
4034 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
4036 * internal-fn.h (internal_store_fn_p): Declare.
4037 (internal_fn_stored_value_index): Likewise.
4038 * internal-fn.c (scatter_store_direct): New macro.
4039 (expand_scatter_store_optab_fn): New function.
4040 (direct_scatter_store_optab_supported_p): New macro.
4041 (internal_store_fn_p): New function.
4042 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
4043 IFN_MASK_SCATTER_STORE.
4044 (internal_fn_mask_index): Likewise.
4045 (internal_fn_stored_value_index): New function.
4046 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
4048 * optabs-query.h (supports_vec_scatter_store_p): Declare.
4049 * optabs-query.c (supports_vec_scatter_store_p): New function.
4050 * tree-vectorizer.h (vect_get_store_rhs): Declare.
4051 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
4052 true for scatter stores.
4053 (vect_gather_scatter_fn_p): Handle scatter stores too.
4054 (vect_check_gather_scatter): Consider using scatter stores if
4055 supports_vec_scatter_store_p.
4056 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
4058 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
4059 internal_fn_stored_value_index.
4060 (check_load_store_masking): Handle scatter stores too.
4061 (vect_get_store_rhs): Make public.
4062 (vectorizable_call): Use internal_store_fn_p.
4063 (vectorizable_store): Handle scatter store internal functions.
4064 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
4065 when deciding whether the end of the group has been reached.
4066 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
4067 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
4068 (mask_scatter_store<mode>): New insns.
4070 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4071 Alan Hayward <alan.hayward@arm.com>
4072 David Sherwood <david.sherwood@arm.com>
4074 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
4075 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
4076 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
4078 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
4079 Use vect_truncate_gather_scatter_offset if we can't treat the
4080 operation as a normal gather load or scatter store.
4081 (get_group_load_store_type): Take the gather_scatter_info
4082 as argument. Try using a gather load or scatter store for
4083 single-element groups.
4084 (get_load_store_type): Update calls to get_group_load_store_type
4085 and vect_use_strided_gather_scatters_p.
4087 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4088 Alan Hayward <alan.hayward@arm.com>
4089 David Sherwood <david.sherwood@arm.com>
4091 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
4092 optional tree argument.
4093 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
4095 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
4096 but continue to use the current value as a fallback.
4097 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
4098 to compare the updates.
4099 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
4100 (get_load_store_type): Use it when handling a strided access.
4101 (vect_get_strided_load_store_ops): New function.
4102 (vect_get_data_ptr_increment): Likewise.
4103 (vectorizable_load): Handle strided gather loads. Always pass
4104 a step to vect_create_data_ref_ptr and bump_vector_ptr.
4106 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4107 Alan Hayward <alan.hayward@arm.com>
4108 David Sherwood <david.sherwood@arm.com>
4110 * doc/md.texi (gather_load@var{m}): Document.
4111 (mask_gather_load@var{m}): Likewise.
4112 * genopinit.c (main): Add supports_vec_gather_load and
4113 supports_vec_gather_load_cached to target_optabs.
4114 * optabs-tree.c (init_tree_optimization_optabs): Use
4115 ggc_cleared_alloc to allocate target_optabs.
4116 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
4117 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
4119 * internal-fn.h (internal_load_fn_p): Declare.
4120 (internal_gather_scatter_fn_p): Likewise.
4121 (internal_fn_mask_index): Likewise.
4122 (internal_gather_scatter_fn_supported_p): Likewise.
4123 * internal-fn.c (gather_load_direct): New macro.
4124 (expand_gather_load_optab_fn): New function.
4125 (direct_gather_load_optab_supported_p): New macro.
4126 (direct_internal_fn_optab): New function.
4127 (internal_load_fn_p): Likewise.
4128 (internal_gather_scatter_fn_p): Likewise.
4129 (internal_fn_mask_index): Likewise.
4130 (internal_gather_scatter_fn_supported_p): Likewise.
4131 * optabs-query.c (supports_at_least_one_mode_p): New function.
4132 (supports_vec_gather_load_p): Likewise.
4133 * optabs-query.h (supports_vec_gather_load_p): Declare.
4134 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
4135 and memory_type field.
4136 (NUM_PATTERNS): Bump to 15.
4137 * tree-vect-data-refs.c: Include internal-fn.h.
4138 (vect_gather_scatter_fn_p): New function.
4139 (vect_describe_gather_scatter_call): Likewise.
4140 (vect_check_gather_scatter): Try using internal functions for
4141 gather loads. Recognize existing calls to a gather load function.
4142 (vect_analyze_data_refs): Consider using gather loads if
4143 supports_vec_gather_load_p.
4144 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
4145 (vect_get_gather_scatter_offset_type): Likewise.
4146 (vect_convert_mask_for_vectype): Likewise.
4147 (vect_add_conversion_to_patterm): Likewise.
4148 (vect_try_gather_scatter_pattern): Likewise.
4149 (vect_recog_gather_scatter_pattern): New pattern recognizer.
4150 (vect_vect_recog_func_ptrs): Add it.
4151 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
4152 internal_fn_mask_index and internal_gather_scatter_fn_p.
4153 (check_load_store_masking): Take the gather_scatter_info as an
4154 argument and handle gather loads.
4155 (vect_get_gather_scatter_ops): New function.
4156 (vectorizable_call): Check internal_load_fn_p.
4157 (vectorizable_load): Likewise. Handle gather load internal
4159 (vectorizable_store): Update call to check_load_store_masking.
4160 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
4161 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
4162 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
4163 (aarch64_gather_scale_operand_d): New predicates.
4164 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
4165 (mask_gather_load<mode>): New insns.
4167 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4168 Alan Hayward <alan.hayward@arm.com>
4169 David Sherwood <david.sherwood@arm.com>
4171 * optabs.def (fold_left_plus_optab): New optab.
4172 * doc/md.texi (fold_left_plus_@var{m}): Document.
4173 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
4174 * internal-fn.c (fold_left_direct): Define.
4175 (expand_fold_left_optab_fn): Likewise.
4176 (direct_fold_left_optab_supported_p): Likewise.
4177 * fold-const-call.c (fold_const_fold_left): New function.
4178 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
4179 * tree-parloops.c (valid_reduction_p): New function.
4180 (gather_scalar_reductions): Use it.
4181 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
4182 (vect_finish_replace_stmt): Declare.
4183 * tree-vect-loop.c (fold_left_reduction_fn): New function.
4184 (needs_fold_left_reduction_p): New function, split out from...
4185 (vect_is_simple_reduction): ...here. Accept reductions that
4186 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
4187 (vect_force_simple_reduction): Also store the reduction type in
4188 the assignment's STMT_VINFO_REDUC_TYPE.
4189 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
4190 (merge_with_identity): New function.
4191 (vect_expand_fold_left): Likewise.
4192 (vectorize_fold_left_reduction): Likewise.
4193 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
4194 scalar phi in place for it. Check for target support and reject
4195 cases that would reassociate the operation. Defer the transform
4196 phase to vectorize_fold_left_reduction.
4197 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
4198 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
4199 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
4201 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4203 * tree-if-conv.c (predicate_mem_writes): Remove redundant
4204 call to ifc_temp_var.
4206 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4207 Alan Hayward <alan.hayward@arm.com>
4208 David Sherwood <david.sherwood@arm.com>
4210 * target.def (legitimize_address_displacement): Take the original
4211 offset as a poly_int.
4212 * targhooks.h (default_legitimize_address_displacement): Update
4214 * targhooks.c (default_legitimize_address_displacement): Likewise.
4215 * doc/tm.texi: Regenerate.
4216 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
4217 as an argument, moving assert of ad->disp == ad->disp_term to...
4218 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
4219 Try calling targetm.legitimize_address_displacement before expanding
4220 the address rather than afterwards, and adjust for the new interface.
4221 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
4222 Match the new hook interface. Handle SVE addresses.
4223 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
4226 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4228 * Makefile.in (OBJS): Add early-remat.o.
4229 * target.def (select_early_remat_modes): New hook.
4230 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
4231 * doc/tm.texi: Regenerate.
4232 * targhooks.h (default_select_early_remat_modes): Declare.
4233 * targhooks.c (default_select_early_remat_modes): New function.
4234 * timevar.def (TV_EARLY_REMAT): New timevar.
4235 * passes.def (pass_early_remat): New pass.
4236 * tree-pass.h (make_pass_early_remat): Declare.
4237 * early-remat.c: New file.
4238 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
4240 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
4242 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4243 Alan Hayward <alan.hayward@arm.com>
4244 David Sherwood <david.sherwood@arm.com>
4246 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
4247 vfm1 with a bound_epilog parameter.
4248 (vect_do_peeling): Update calls accordingly, and move the prologue
4249 call earlier in the function. Treat the base bound_epilog as 0 for
4250 fully-masked loops and retain vf - 1 for other loops. Add 1 to
4251 this base when peeling for gaps.
4252 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
4253 with fully-masked loops.
4254 (vect_estimate_min_profitable_iters): Handle the single peeled
4255 iteration in that case.
4257 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4258 Alan Hayward <alan.hayward@arm.com>
4259 David Sherwood <david.sherwood@arm.com>
4261 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
4262 single-element interleaving even if the size is not a power of 2.
4263 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
4264 accesses for single-element interleaving if the group size is
4267 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4268 Alan Hayward <alan.hayward@arm.com>
4269 David Sherwood <david.sherwood@arm.com>
4271 * doc/md.texi (fold_extract_last_@var{m}): Document.
4272 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
4273 * optabs.def (fold_extract_last_optab): New optab.
4274 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
4275 * internal-fn.c (fold_extract_direct): New macro.
4276 (expand_fold_extract_optab_fn): Likewise.
4277 (direct_fold_extract_optab_supported_p): Likewise.
4278 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
4279 * tree-vect-loop.c (vect_model_reduction_cost): Handle
4280 EXTRACT_LAST_REDUCTION.
4281 (get_initial_def_for_reduction): Do not create an initial vector
4282 for EXTRACT_LAST_REDUCTION reductions.
4283 (vectorizable_reduction): Leave the scalar phi in place for
4284 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
4285 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
4286 epilogue code for EXTRACT_LAST_REDUCTION and defer the
4287 transform phase to vectorizable_condition.
4288 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
4290 (vect_finish_stmt_generation): ...here.
4291 (vect_finish_replace_stmt): New function.
4292 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
4293 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
4295 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
4297 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4298 Alan Hayward <alan.hayward@arm.com>
4299 David Sherwood <david.sherwood@arm.com>
4301 * doc/md.texi (extract_last_@var{m}): Document.
4302 * optabs.def (extract_last_optab): New optab.
4303 * internal-fn.def (EXTRACT_LAST): New internal function.
4304 * internal-fn.c (cond_unary_direct): New macro.
4305 (expand_cond_unary_optab_fn): Likewise.
4306 (direct_cond_unary_optab_supported_p): Likewise.
4307 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
4308 loops using EXTRACT_LAST.
4309 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
4310 (extract_last_<mode>): ...this optab.
4311 (vec_extract<mode><Vel>): Update accordingly.
4313 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4314 Alan Hayward <alan.hayward@arm.com>
4315 David Sherwood <david.sherwood@arm.com>
4317 * target.def (empty_mask_is_expensive): New hook.
4318 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
4319 * doc/tm.texi: Regenerate.
4320 * targhooks.h (default_empty_mask_is_expensive): Declare.
4321 * targhooks.c (default_empty_mask_is_expensive): New function.
4322 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
4323 if the target says that empty masks are expensive.
4324 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
4326 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
4328 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4329 Alan Hayward <alan.hayward@arm.com>
4330 David Sherwood <david.sherwood@arm.com>
4332 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
4333 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
4334 (vect_use_loop_mask_for_alignment_p): New function.
4335 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
4336 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
4337 niters_skip argument. Make sure that the first niters_skip elements
4338 of the first iteration are inactive.
4339 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
4340 Update call to vect_set_loop_masks_directly.
4341 (get_misalign_in_elems): New function, split out from...
4342 (vect_gen_prolog_loop_niters): ...here.
4343 (vect_update_init_of_dr): Take a code argument that specifies whether
4344 the adjustment should be added or subtracted.
4345 (vect_update_init_of_drs): Likewise.
4346 (vect_prepare_for_masked_peels): New function.
4347 (vect_do_peeling): Skip prologue peeling if we're using a mask
4348 instead. Update call to vect_update_inits_of_drs.
4349 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4351 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
4352 alignment. Do not include the number of peeled iterations in
4353 the minimum threshold in that case.
4354 (vectorizable_induction): Adjust the start value down by
4355 LOOP_VINFO_MASK_SKIP_NITERS iterations.
4356 (vect_transform_loop): Call vect_prepare_for_masked_peels.
4357 Take the number of skipped iterations into account when calculating
4359 * tree-vect-stmts.c (vect_gen_while_not): New function.
4361 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4362 Alan Hayward <alan.hayward@arm.com>
4363 David Sherwood <david.sherwood@arm.com>
4365 * doc/sourcebuild.texi (vect_fully_masked): Document.
4366 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
4368 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
4370 (vect_analyze_loop_2): ...here. Don't check the vectorization
4371 factor against the number of loop iterations if the loop is
4374 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4375 Alan Hayward <alan.hayward@arm.com>
4376 David Sherwood <david.sherwood@arm.com>
4378 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
4379 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
4380 (dump_groups): Update accordingly.
4381 (iv_use::mem_type): New member variable.
4382 (address_p): New function.
4383 (record_use): Add a mem_type argument and initialize the new
4385 (record_group_use): Add a mem_type argument. Use address_p.
4386 Remove obsolete null checks of base_object. Update call to record_use.
4387 (find_interesting_uses_op): Update call to record_group_use.
4388 (find_interesting_uses_cond): Likewise.
4389 (find_interesting_uses_address): Likewise.
4390 (get_mem_type_for_internal_fn): New function.
4391 (find_address_like_use): Likewise.
4392 (find_interesting_uses_stmt): Try find_address_like_use before
4393 calling find_interesting_uses_op.
4394 (addr_offset_valid_p): Use the iv mem_type field as the type
4395 of the addressed memory.
4396 (add_autoinc_candidates): Likewise.
4397 (get_address_cost): Likewise.
4398 (split_small_address_groups_p): Use address_p.
4399 (split_address_groups): Likewise.
4400 (add_iv_candidate_for_use): Likewise.
4401 (autoinc_possible_for_pair): Likewise.
4402 (rewrite_groups): Likewise.
4403 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
4404 (determine_group_iv_cost): Update after split of USE_ADDRESS.
4405 (get_alias_ptr_type_for_ptr_address): New function.
4406 (rewrite_use_address): Rewrite address uses in calls that were
4407 identified by find_address_like_use.
4409 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4410 Alan Hayward <alan.hayward@arm.com>
4411 David Sherwood <david.sherwood@arm.com>
4413 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
4415 * gimple-expr.h (is_gimple_addressable: Likewise.
4416 * gimple-expr.c (is_gimple_address): Likewise.
4417 * internal-fn.c (expand_call_mem_ref): New function.
4418 (expand_mask_load_optab_fn): Use it.
4419 (expand_mask_store_optab_fn): Likewise.
4421 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4422 Alan Hayward <alan.hayward@arm.com>
4423 David Sherwood <david.sherwood@arm.com>
4425 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
4426 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
4427 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
4428 (cond_umax@var{mode}): Document.
4429 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
4430 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
4431 (cond_umin_optab, cond_umax_optab): New optabs.
4432 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
4433 (COND_IOR, COND_XOR): New internal functions.
4434 * internal-fn.h (get_conditional_internal_fn): Declare.
4435 * internal-fn.c (cond_binary_direct): New macro.
4436 (expand_cond_binary_optab_fn): Likewise.
4437 (direct_cond_binary_optab_supported_p): Likewise.
4438 (get_conditional_internal_fn): New function.
4439 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
4440 Cope with reduction statements that are vectorized as calls rather
4442 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
4443 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
4444 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
4445 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
4446 (UNSPEC_COND_EOR): New unspecs.
4447 (optab): Add mappings for them.
4448 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
4449 (sve_int_op, sve_fp_op): New int attributes.
4451 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4452 Alan Hayward <alan.hayward@arm.com>
4453 David Sherwood <david.sherwood@arm.com>
4455 * optabs.def (while_ult_optab): New optab.
4456 * doc/md.texi (while_ult@var{m}@var{n}): Document.
4457 * internal-fn.def (WHILE_ULT): New internal function.
4458 * internal-fn.h (direct_internal_fn_supported_p): New override
4459 that takes two types as argument.
4460 * internal-fn.c (while_direct): New macro.
4461 (expand_while_optab_fn): New function.
4462 (convert_optab_supported_p): Likewise.
4463 (direct_while_optab_supported_p): New macro.
4464 * wide-int.h (wi::udiv_ceil): New function.
4465 * tree-vectorizer.h (rgroup_masks): New structure.
4466 (vec_loop_masks): New typedef.
4467 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
4469 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
4470 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
4471 (vect_max_vf): New function.
4472 (slpeel_make_loop_iterate_ntimes): Delete.
4473 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
4474 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
4475 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
4476 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
4477 internal-fn.h, stor-layout.h and optabs-query.h.
4478 (vect_set_loop_mask): New function.
4479 (add_preheader_seq): Likewise.
4480 (add_header_seq): Likewise.
4481 (interleave_supported_p): Likewise.
4482 (vect_maybe_permute_loop_masks): Likewise.
4483 (vect_set_loop_masks_directly): Likewise.
4484 (vect_set_loop_condition_masked): Likewise.
4485 (vect_set_loop_condition_unmasked): New function, split out from
4486 slpeel_make_loop_iterate_ntimes.
4487 (slpeel_make_loop_iterate_ntimes): Rename to..
4488 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
4489 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
4490 (vect_do_peeling): Update call accordingly.
4491 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
4493 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4494 mask_compare_type, can_fully_mask_p and fully_masked_p.
4495 (release_vec_loop_masks): New function.
4496 (_loop_vec_info): Use it to free the loop masks.
4497 (can_produce_all_loop_masks_p): New function.
4498 (vect_get_max_nscalars_per_iter): Likewise.
4499 (vect_verify_full_masking): Likewise.
4500 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
4501 retries, and free the mask rgroups before retrying. Check loop-wide
4502 reasons for disallowing fully-masked loops. Make the final decision
4503 about whether use a fully-masked loop or not.
4504 (vect_estimate_min_profitable_iters): Do not assume that peeling
4505 for the number of iterations will be needed for fully-masked loops.
4506 (vectorizable_reduction): Disable fully-masked loops.
4507 (vectorizable_live_operation): Likewise.
4508 (vect_halve_mask_nunits): New function.
4509 (vect_double_mask_nunits): Likewise.
4510 (vect_record_loop_mask): Likewise.
4511 (vect_get_loop_mask): Likewise.
4512 (vect_transform_loop): Handle the case in which the final loop
4513 iteration might handle a partial vector. Call vect_set_loop_condition
4514 instead of slpeel_make_loop_iterate_ntimes.
4515 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
4516 (check_load_store_masking): New function.
4517 (prepare_load_store_mask): Likewise.
4518 (vectorizable_store): Handle fully-masked loops.
4519 (vectorizable_load): Likewise.
4520 (supportable_widening_operation): Use vect_halve_mask_nunits for
4522 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
4523 (vect_gen_while): New function.
4524 * config/aarch64/aarch64.md (umax<mode>3): New expander.
4525 (aarch64_uqdec<mode>): New insn.
4527 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4528 Alan Hayward <alan.hayward@arm.com>
4529 David Sherwood <david.sherwood@arm.com>
4531 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
4532 (reduc_xor_scal_optab): New optabs.
4533 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
4534 (reduc_xor_scal_@var{m}): Document.
4535 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
4536 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
4538 * fold-const-call.c (fold_const_call): Handle them.
4539 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
4540 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
4541 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
4542 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
4543 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
4544 (UNSPEC_XORV): New unspecs.
4545 (optab): Add entries for them.
4546 (BITWISEV): New int iterator.
4547 (bit_reduc_op): New int attributes.
4549 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4550 Alan Hayward <alan.hayward@arm.com>
4551 David Sherwood <david.sherwood@arm.com>
4553 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
4554 * internal-fn.def (VEC_SHL_INSERT): New internal function.
4555 * optabs.def (vec_shl_insert_optab): New optab.
4556 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
4557 (duplicate_and_interleave): Likewise.
4558 * tree-vect-loop.c: Include internal-fn.h.
4559 (neutral_op_for_slp_reduction): New function, split out from
4560 get_initial_defs_for_reduction.
4561 (get_initial_def_for_reduction): Handle option 2 for variable-length
4562 vectors by loading the neutral value into a vector and then shifting
4563 the initial value into element 0.
4564 (get_initial_defs_for_reduction): Replace the code argument with
4565 the neutral value calculated by neutral_op_for_slp_reduction.
4566 Use gimple_build_vector for constant-length vectors.
4567 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
4568 but the first group_size elements have a neutral value.
4569 Use duplicate_and_interleave otherwise.
4570 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
4571 Update call to get_initial_defs_for_reduction. Handle SLP
4572 reductions for variable-length vectors by creating one vector
4573 result for each scalar result, with the elements associated
4574 with other scalar results stubbed out with the neutral value.
4575 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
4576 Require IFN_VEC_SHL_INSERT for double reductions on
4577 variable-length vectors, or SLP reductions that have
4578 a neutral value. Require can_duplicate_and_interleave_p
4579 support for variable-length unchained SLP reductions if there
4580 is no neutral value, such as for MIN/MAX reductions. Also require
4581 the number of vector elements to be a multiple of the number of
4582 SLP statements when doing variable-length unchained SLP reductions.
4583 Update call to vect_create_epilog_for_reduction.
4584 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
4585 and remove initial values.
4586 (duplicate_and_interleave): Make public.
4587 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
4588 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
4590 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4591 Alan Hayward <alan.hayward@arm.com>
4592 David Sherwood <david.sherwood@arm.com>
4594 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
4595 (can_duplicate_and_interleave_p): New function.
4596 (vect_get_and_check_slp_defs): Take the vector of statements
4597 rather than just the current one. Remove excess parentheses.
4598 Restriction rejectinon of vect_constant_def and vect_external_def
4599 for variable-length vectors to boolean types, or types for which
4600 can_duplicate_and_interleave_p is false.
4601 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
4602 (duplicate_and_interleave): New function.
4603 (vect_get_constant_vectors): Use gimple_build_vector for
4604 constant-length vectors and suitable variable-length constant
4605 vectors. Use duplicate_and_interleave for other variable-length
4606 vectors. Don't defer the update when inserting new statements.
4608 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4609 Alan Hayward <alan.hayward@arm.com>
4610 David Sherwood <david.sherwood@arm.com>
4612 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
4613 min_profitable_iters doesn't go negative.
4615 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4616 Alan Hayward <alan.hayward@arm.com>
4617 David Sherwood <david.sherwood@arm.com>
4619 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
4620 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
4621 * optabs.def (vec_mask_load_lanes_optab): New optab.
4622 (vec_mask_store_lanes_optab): Likewise.
4623 * internal-fn.def (MASK_LOAD_LANES): New internal function.
4624 (MASK_STORE_LANES): Likewise.
4625 * internal-fn.c (mask_load_lanes_direct): New macro.
4626 (mask_store_lanes_direct): Likewise.
4627 (expand_mask_load_optab_fn): Handle masked operations.
4628 (expand_mask_load_lanes_optab_fn): New macro.
4629 (expand_mask_store_optab_fn): Handle masked operations.
4630 (expand_mask_store_lanes_optab_fn): New macro.
4631 (direct_mask_load_lanes_optab_supported_p): Likewise.
4632 (direct_mask_store_lanes_optab_supported_p): Likewise.
4633 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
4635 (vect_load_lanes_supported): Likewise.
4636 * tree-vect-data-refs.c (strip_conversion): New function.
4637 (can_group_stmts_p): Likewise.
4638 (vect_analyze_data_ref_accesses): Use it instead of checking
4639 for a pair of assignments.
4640 (vect_store_lanes_supported): Take a masked_p parameter.
4641 (vect_load_lanes_supported): Likewise.
4642 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
4643 vect_store_lanes_supported and vect_load_lanes_supported.
4644 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
4645 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
4646 parameter. Don't allow gaps for masked accesses.
4647 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
4648 and vect_load_lanes_supported.
4649 (get_load_store_type): Take a masked_p parameter and update
4650 call to get_group_load_store_type.
4651 (vectorizable_store): Update call to get_load_store_type.
4652 Handle IFN_MASK_STORE_LANES.
4653 (vectorizable_load): Update call to get_load_store_type.
4654 Handle IFN_MASK_LOAD_LANES.
4656 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4657 Alan Hayward <alan.hayward@arm.com>
4658 David Sherwood <david.sherwood@arm.com>
4660 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
4662 * config/aarch64/aarch64-protos.h
4663 (aarch64_sve_struct_memory_operand_p): Declare.
4664 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
4665 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
4666 (VPRED, vpred): Handle SVE structure modes.
4667 * config/aarch64/constraints.md (Utx): New constraint.
4668 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
4669 (aarch64_sve_struct_nonimmediate_operand): New predicates.
4670 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
4671 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
4672 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
4673 structure modes. Split into pieces after RA.
4674 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
4675 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
4677 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
4678 SVE structure modes.
4679 (aarch64_classify_address): Likewise.
4680 (sizetochar): Move earlier in file.
4681 (aarch64_print_operand): Handle SVE register lists.
4682 (aarch64_array_mode): New function.
4683 (aarch64_sve_struct_memory_operand_p): Likewise.
4684 (TARGET_ARRAY_MODE): Redefine.
4686 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4687 Alan Hayward <alan.hayward@arm.com>
4688 David Sherwood <david.sherwood@arm.com>
4690 * target.def (array_mode): New target hook.
4691 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
4692 * doc/tm.texi: Regenerate.
4693 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
4694 * hooks.c (hook_optmode_mode_uhwi_none): New function.
4695 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
4697 * stor-layout.c (mode_for_array): Likewise. Support polynomial
4700 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4701 Alan Hayward <alan.hayward@arm.com>
4702 David Sherwood <david.sherwood@arm.com>
4704 * fold-const.c (fold_binary_loc): Check the argument types
4705 rather than the result type when testing for a vector operation.
4707 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4709 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
4710 * doc/tm.texi: Regenerate.
4712 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4713 Alan Hayward <alan.hayward@arm.com>
4714 David Sherwood <david.sherwood@arm.com>
4716 * doc/invoke.texi (-msve-vector-bits=): Document new option.
4717 (sve): Document new AArch64 extension.
4718 * doc/md.texi (w): Extend the description of the AArch64
4719 constraint to include SVE vectors.
4720 (Upl, Upa): Document new AArch64 predicate constraints.
4721 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
4723 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
4724 (msve-vector-bits=): New option.
4725 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
4726 SVE when these are disabled.
4727 (sve): New extension.
4728 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
4729 modes. Adjust their number of units based on aarch64_sve_vg.
4730 (MAX_BITSIZE_MODE_ANY_MODE): Define.
4731 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
4732 aarch64_addr_query_type.
4733 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
4734 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
4735 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
4736 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
4737 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
4738 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
4739 (aarch64_simd_imm_zero_p): Delete.
4740 (aarch64_check_zero_based_sve_index_immediate): Declare.
4741 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4742 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4743 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4744 (aarch64_sve_float_mul_immediate_p): Likewise.
4745 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4747 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
4748 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
4749 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
4750 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
4751 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
4752 (aarch64_regmode_natural_size): Likewise.
4753 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
4754 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
4756 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
4757 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
4758 for VG and the SVE predicate registers.
4759 (V_ALIASES): Add a "z"-prefixed alias.
4760 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
4761 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
4762 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
4763 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
4764 (REG_CLASS_NAMES): Add entries for them.
4765 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
4766 and the predicate registers.
4767 (aarch64_sve_vg): Declare.
4768 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
4769 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
4770 (REGMODE_NATURAL_SIZE): Define.
4771 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
4773 * config/aarch64/aarch64.c: Include cfgrtl.h.
4774 (simd_immediate_info): Add a constructor for series vectors,
4775 and an associated step field.
4776 (aarch64_sve_vg): New variable.
4777 (aarch64_dbx_register_number): Handle VG and the predicate registers.
4778 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
4779 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
4780 (VEC_ANY_DATA, VEC_STRUCT): New constants.
4781 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
4782 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
4783 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
4784 (aarch64_get_mask_mode): New functions.
4785 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
4786 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
4787 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
4788 predicate modes and predicate registers. Explicitly restrict
4789 GPRs to modes of 16 bytes or smaller. Only allow FP registers
4790 to store a vector mode if it is recognized by
4791 aarch64_classify_vector_mode.
4792 (aarch64_regmode_natural_size): New function.
4793 (aarch64_hard_regno_caller_save_mode): Return the original mode
4795 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
4796 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
4797 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
4798 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
4800 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
4801 does not overlap dest if the function is frame-related. Handle
4803 (aarch64_split_add_offset): New function.
4804 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
4805 them aarch64_add_offset.
4806 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
4807 and update call to aarch64_sub_sp.
4808 (aarch64_add_cfa_expression): New function.
4809 (aarch64_expand_prologue): Pass extra temporary registers to the
4810 functions above. Handle the case in which we need to emit new
4811 DW_CFA_expressions for registers that were originally saved
4812 relative to the stack pointer, but now have to be expressed
4813 relative to the frame pointer.
4814 (aarch64_output_mi_thunk): Pass extra temporary registers to the
4816 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
4817 IP0 and IP1 values for SVE frames.
4818 (aarch64_expand_vec_series): New function.
4819 (aarch64_expand_sve_widened_duplicate): Likewise.
4820 (aarch64_expand_sve_const_vector): Likewise.
4821 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
4822 Handle SVE constants. Use emit_move_insn to move a force_const_mem
4823 into the register, rather than emitting a SET directly.
4824 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
4825 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
4826 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
4827 (offset_9bit_signed_scaled_p): New functions.
4828 (aarch64_replicate_bitmask_imm): New function.
4829 (aarch64_bitmask_imm): Use it.
4830 (aarch64_cannot_force_const_mem): Reject expressions involving
4831 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
4832 (aarch64_classify_index): Handle SVE indices, by requiring
4833 a plain register index with a scale that matches the element size.
4834 (aarch64_classify_address): Handle SVE addresses. Assert that
4835 the mode of the address is VOIDmode or an integer mode.
4836 Update call to aarch64_classify_symbol.
4837 (aarch64_classify_symbolic_expression): Update call to
4838 aarch64_classify_symbol.
4839 (aarch64_const_vec_all_in_range_p): New function.
4840 (aarch64_print_vector_float_operand): Likewise.
4841 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
4842 "vN" for FP registers with SVE modes. Handle (const ...) vectors
4843 and the FP immediates 1.0 and 0.5.
4844 (aarch64_print_address_internal): Handle SVE addresses.
4845 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
4846 (aarch64_regno_regclass): Handle predicate registers.
4847 (aarch64_secondary_reload): Handle big-endian reloads of SVE
4849 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
4850 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
4851 (aarch64_convert_sve_vector_bits): New function.
4852 (aarch64_override_options): Use it to handle -msve-vector-bits=.
4853 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4855 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
4856 Handle SVE vector and predicate modes. Accept VL-based constants
4857 that need only one temporary register, and VL offsets that require
4858 no temporary registers.
4859 (aarch64_conditional_register_usage): Mark the predicate registers
4860 as fixed if SVE isn't available.
4861 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
4862 Return true for SVE vector and predicate modes.
4863 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
4864 rather than an unsigned int. Handle SVE modes.
4865 (aarch64_preferred_simd_mode): Update call accordingly. Handle
4867 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
4869 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4870 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4871 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4872 (aarch64_sve_float_mul_immediate_p): New functions.
4873 (aarch64_sve_valid_immediate): New function.
4874 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
4875 Explicitly reject structure modes. Check for INDEX constants.
4876 Handle PTRUE and PFALSE constants.
4877 (aarch64_check_zero_based_sve_index_immediate): New function.
4878 (aarch64_simd_imm_zero_p): Delete.
4879 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
4880 vector modes. Accept constants in the range of CNT[BHWD].
4881 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
4882 ask for an Advanced SIMD mode.
4883 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
4884 (aarch64_simd_vector_alignment): Handle SVE predicates.
4885 (aarch64_vectorize_preferred_vector_alignment): New function.
4886 (aarch64_simd_vector_alignment_reachable): Use it instead of
4888 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
4889 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
4891 (MAX_VECT_LEN): Delete.
4892 (expand_vec_perm_d): Add a vec_flags field.
4893 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
4894 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
4895 (aarch64_evpc_ext): Don't apply a big-endian lane correction
4897 (aarch64_evpc_rev): Rename to...
4898 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
4899 (aarch64_evpc_rev_global): New function.
4900 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
4901 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
4903 (aarch64_evpc_sve_tbl): New function.
4904 (aarch64_expand_vec_perm_const_1): Update after rename of
4905 aarch64_evpc_rev. Handle SVE permutes too, trying
4906 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
4907 than aarch64_evpc_tbl.
4908 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
4909 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
4910 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
4911 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
4912 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
4913 (aarch64_expand_sve_vcond): New functions.
4914 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
4915 of aarch64_vector_mode_p.
4916 (aarch64_dwarf_poly_indeterminate_value): New function.
4917 (aarch64_compute_pressure_classes): Likewise.
4918 (aarch64_can_change_mode_class): Likewise.
4919 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
4920 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
4921 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
4922 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
4923 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
4924 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
4925 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
4926 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
4928 (Dn, Dl, Dr): Accept const as well as const_vector.
4929 (Dz): Likewise. Compare against CONST0_RTX.
4930 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
4931 of "vector" where appropriate.
4932 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
4933 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
4934 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
4935 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
4936 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
4937 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
4938 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
4939 (v_int_equiv): Extend to SVE modes.
4940 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
4942 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
4943 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4944 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4945 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4946 (SVE_COND_FP_CMP): New int iterators.
4947 (perm_hilo): Handle the new unpack unspecs.
4948 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4950 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4951 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4952 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4953 (aarch64_equality_operator, aarch64_constant_vector_operand)
4954 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4955 (aarch64_sve_nonimmediate_operand): Likewise.
4956 (aarch64_sve_general_operand): Likewise.
4957 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4958 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4959 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4960 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4961 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4962 (aarch64_sve_float_arith_immediate): Likewise.
4963 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4964 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4965 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4966 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4967 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4968 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4969 (aarch64_sve_float_arith_operand): Likewise.
4970 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4971 (aarch64_sve_float_mul_operand): Likewise.
4972 (aarch64_sve_vec_perm_operand): Likewise.
4973 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4974 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4975 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4976 as well as const_vector.
4977 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4978 in file. Use CONST0_RTX and CONSTM1_RTX.
4979 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4980 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4981 Use aarch64_simd_imm_zero.
4982 * config/aarch64/aarch64-sve.md: New file.
4983 * config/aarch64/aarch64.md: Include it.
4984 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4985 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4986 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4987 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4988 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4989 (sve): New attribute.
4990 (enabled): Disable instructions with the sve attribute unless
4992 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4993 aarch64_expand_mov_immediate.
4994 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4995 CNT[BHSD] immediates.
4996 (movti): Split CONST_POLY_INT moves into two halves.
4997 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4998 Split additions that need a temporary here if the destination
4999 is the stack pointer.
5000 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
5001 (*add<mode>3_poly_1): New instruction.
5002 (set_clobber_cc): New expander.
5004 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
5006 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
5007 parameter and use it instead of GET_MODE_SIZE (innermode). Use
5008 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
5009 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
5010 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
5011 Change innermode from fixed_mode_size to machine_mode.
5012 (simplify_subreg): Update call accordingly. Handle a constant-sized
5013 subreg of a variable-length CONST_VECTOR.
5015 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
5016 Alan Hayward <alan.hayward@arm.com>
5017 David Sherwood <david.sherwood@arm.com>
5019 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
5020 (add_offset_to_base): New function, split out from...
5021 (create_mem_ref): ...here. When handling a scale other than 1,
5022 check first whether the address is valid without the offset.
5023 Add it into the base if so, leaving the index and scale as-is.
5025 2018-01-12 Jakub Jelinek <jakub@redhat.com>
5028 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
5029 fold_for_warn before checking if arg2 is INTEGER_CST.
5031 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
5033 * config/rs6000/predicates.md (load_multiple_operation): Delete.
5034 (store_multiple_operation): Delete.
5035 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
5036 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
5037 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
5038 guarded by TARGET_STRING.
5039 (rs6000_output_load_multiple): Delete.
5040 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
5041 OPTION_MASK_STRING / TARGET_STRING handling.
5042 (print_operand) <'N', 'O'>: Add comment that these are unused now.
5043 (const rs6000_opt_masks) <"string">: Change mask to 0.
5044 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
5045 (MASK_STRING): Delete.
5046 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
5048 (load_multiple): Delete.
5055 (store_multiple): Delete.
5062 (movmemsi_8reg): Delete.
5063 (corresponding unnamed define_insn): Delete.
5064 (movmemsi_6reg): Delete.
5065 (corresponding unnamed define_insn): Delete.
5066 (movmemsi_4reg): Delete.
5067 (corresponding unnamed define_insn): Delete.
5068 (movmemsi_2reg): Delete.
5069 (corresponding unnamed define_insn): Delete.
5070 (movmemsi_1reg): Delete.
5071 (corresponding unnamed define_insn): Delete.
5072 * config/rs6000/rs6000.opt (mno-string): New.
5073 (mstring): Replace by deprecation warning stub.
5074 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
5076 2018-01-12 Jakub Jelinek <jakub@redhat.com>
5078 * regrename.c (regrename_do_replace): If replacing the same
5079 reg multiple times, try to reuse last created gen_raw_REG.
5082 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
5083 main to workaround a bug in GDB.
5085 2018-01-12 Tom de Vries <tom@codesourcery.com>
5088 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
5090 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
5092 PR rtl-optimization/80481
5093 * ira-color.c (get_cap_member): New function.
5094 (allocnos_conflict_by_live_ranges_p): Use it.
5095 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
5096 (setup_slot_coalesced_allocno_live_ranges): Ditto.
5098 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
5101 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
5102 (*saddl_se_1): Ditto.
5104 (*ssubl_se_1): Ditto.
5106 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
5108 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
5109 rather than wi::to_widest for DR_INITs.
5110 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
5111 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
5112 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
5114 (vect_analyze_group_access_1): Note that here.
5116 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
5118 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
5119 polynomial type sizes.
5121 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
5123 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
5124 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
5125 (gimple_add_tmp_var): Likewise.
5127 2018-01-12 Martin Liska <mliska@suse.cz>
5129 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
5130 (gimple_alloc_sizes): Likewise.
5131 (dump_gimple_statistics): Use PRIu64 in printf format.
5132 * gimple.h: Change uint64_t to int.
5134 2018-01-12 Martin Liska <mliska@suse.cz>
5136 * tree-core.h: Use uint64_t instead of int.
5137 * tree.c (tree_node_counts): Likewise.
5138 (tree_node_sizes): Likewise.
5139 (dump_tree_statistics): Use PRIu64 in printf format.
5141 2018-01-12 Martin Liska <mliska@suse.cz>
5143 * Makefile.in: As qsort_chk is implemented in vec.c, add
5144 vec.o to linkage of gencfn-macros.
5145 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
5146 passing the info to record_node_allocation_statistics.
5147 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
5149 * ggc-common.c (struct ggc_usage): Add operator== and use
5150 it in operator< and compare function.
5151 * mem-stats.h (struct mem_usage): Likewise.
5152 * vec.c (struct vec_usage): Remove operator< and compare
5153 function. Can be simply inherited.
5155 2018-01-12 Martin Jambor <mjambor@suse.cz>
5158 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
5159 * tree-ssa-math-opts.c: Include domwalk.h.
5160 (convert_mult_to_fma_1): New function.
5161 (fma_transformation_info): New type.
5162 (fma_deferring_state): Likewise.
5163 (cancel_fma_deferring): New function.
5164 (result_of_phi): Likewise.
5165 (last_fma_candidate_feeds_initial_phi): Likewise.
5166 (convert_mult_to_fma): Added deferring logic, split actual
5167 transformation to convert_mult_to_fma_1.
5168 (math_opts_dom_walker): New type.
5169 (math_opts_dom_walker::after_dom_children): New method, body moved
5170 here from pass_optimize_widening_mul::execute, added deferring logic
5172 (pass_optimize_widening_mul::execute): Moved most of code to
5173 math_opts_dom_walker::after_dom_children.
5174 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
5175 * config/i386/i386.c (ix86_option_override_internal): Added
5176 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
5178 2018-01-12 Richard Biener <rguenther@suse.de>
5181 * dwarf2out.c (gen_variable_die): Do not reset old_die for
5182 inline instance vars.
5184 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
5187 * config/rx/rx.c (rx_is_restricted_memory_address):
5190 2018-01-12 Richard Biener <rguenther@suse.de>
5192 PR tree-optimization/80846
5193 * target.def (split_reduction): New target hook.
5194 * targhooks.c (default_split_reduction): New function.
5195 * targhooks.h (default_split_reduction): Declare.
5196 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
5197 target requests first reduce vectors by combining low and high
5199 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
5200 (get_vectype_for_scalar_type_and_size): Export.
5201 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
5202 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
5203 * doc/tm.texi: Regenerate.
5204 * config/i386/i386.c (ix86_split_reduction): Implement
5205 TARGET_VECTORIZE_SPLIT_REDUCTION.
5207 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
5210 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
5211 in PIC mode except for TARGET_VXWORKS_RTP.
5212 * config/sparc/sparc.c: Include cfgrtl.h.
5213 (TARGET_INIT_PIC_REG): Define.
5214 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
5215 (sparc_pic_register_p): New predicate.
5216 (sparc_legitimate_address_p): Use it.
5217 (sparc_legitimize_pic_address): Likewise.
5218 (sparc_delegitimize_address): Likewise.
5219 (sparc_mode_dependent_address_p): Likewise.
5220 (gen_load_pcrel_sym): Remove 4th parameter.
5221 (load_got_register): Adjust call to above. Remove obsolete stuff.
5222 (sparc_expand_prologue): Do not call load_got_register here.
5223 (sparc_flat_expand_prologue): Likewise.
5224 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
5225 (sparc_use_pseudo_pic_reg): New function.
5226 (sparc_init_pic_reg): Likewise.
5227 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
5228 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
5230 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
5232 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
5233 Add item for branch_cost.
5235 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
5237 PR rtl-optimization/83565
5238 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
5239 not extend the result to a larger mode for rotate operations.
5240 (num_sign_bit_copies1): Likewise.
5242 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5245 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
5247 Use values-Xc.o for -pedantic.
5248 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
5250 2018-01-12 Martin Liska <mliska@suse.cz>
5253 * ipa-devirt.c (final_warning_record::grow_type_warnings):
5255 (possible_polymorphic_call_targets): Use it.
5256 (ipa_devirt): Likewise.
5258 2018-01-12 Martin Liska <mliska@suse.cz>
5260 * profile-count.h (enum profile_quality): Use 0 as invalid
5261 enum value of profile_quality.
5263 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
5265 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
5266 -mext-string options.
5268 2018-01-12 Richard Biener <rguenther@suse.de>
5270 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
5271 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
5272 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
5274 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
5276 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
5278 * configure.ac (--with-long-double-format): Add support for the
5279 configuration option to change the default long double format on
5281 * config.gcc (powerpc*-linux*-*): Likewise.
5282 * configure: Regenerate.
5283 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
5284 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
5285 used without modification.
5287 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5289 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
5290 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
5291 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
5292 MISC_BUILTIN_SPEC_BARRIER.
5293 (rs6000_init_builtins): Likewise.
5294 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
5296 (speculation_barrier): New define_insn.
5297 * doc/extend.texi: Document __builtin_speculation_barrier.
5299 2018-01-11 Jakub Jelinek <jakub@redhat.com>
5302 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
5303 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
5304 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
5306 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
5307 integral modes instead of "ss" and "sd".
5308 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
5309 vectors with 32-bit and 64-bit elements.
5310 (vecdupssescalarmodesuffix): New mode attribute.
5311 (vec_dup<mode>): Use it.
5313 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
5316 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
5317 frame if argument is passed on stack.
5319 2018-01-11 Jakub Jelinek <jakub@redhat.com>
5322 * ree.c (combine_reaching_defs): Optimize also
5323 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
5324 reg2=any_extend(exp); reg1=reg2;, formatting fix.
5326 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
5329 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
5331 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
5334 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
5335 after they are computed.
5337 2018-01-11 Bin Cheng <bin.cheng@arm.com>
5339 PR tree-optimization/83695
5340 * gimple-loop-linterchange.cc
5341 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
5342 reset cached scev information after interchange.
5343 (pass_linterchange::execute): Remove call to scev_reset_htab.
5345 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5347 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
5348 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
5349 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
5350 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
5351 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
5352 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
5353 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
5354 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
5355 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
5356 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
5357 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
5358 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
5359 (V_lane_reg): Likewise.
5360 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
5362 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
5363 (vfmal_lane_low<mode>_intrinsic,
5364 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
5365 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
5366 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
5367 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
5368 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
5369 vfmsl_lane_high<mode>_intrinsic): New define_insns.
5371 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5373 * config/arm/arm-cpus.in (fp16fml): New feature.
5374 (ALL_SIMD): Add fp16fml.
5375 (armv8.2-a): Add fp16fml as an option.
5376 (armv8.3-a): Likewise.
5377 (armv8.4-a): Add fp16fml as part of fp16.
5378 * config/arm/arm.h (TARGET_FP16FML): Define.
5379 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
5381 * config/arm/arm-modes.def (V2HF): Define.
5382 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
5383 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
5384 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
5385 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
5386 vfmsl_low, vfmsl_high): New set of builtins.
5387 * config/arm/iterators.md (PLUSMINUS): New code iterator.
5388 (vfml_op): New code attribute.
5389 (VFMLHALVES): New int iterator.
5390 (VFML, VFMLSEL): New mode attributes.
5391 (V_reg): Define mapping for V2HF.
5392 (V_hi, V_lo): New mode attributes.
5393 (VF_constraint): Likewise.
5394 (vfml_half, vfml_half_selector): New int attributes.
5395 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
5397 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
5398 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
5400 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
5401 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
5402 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
5403 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
5405 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
5406 Document new effective target and option set.
5408 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5410 * config/arm/arm-cpus.in (armv8_4): New feature.
5411 (ARMv8_4a): New fgroup.
5412 (armv8.4-a): New arch.
5413 * config/arm/arm-tables.opt: Regenerate.
5414 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
5415 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
5416 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
5417 Add matching rules for -march=armv8.4-a and extensions.
5418 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
5420 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
5423 * config/rx/rx.md (BW): New mode attribute.
5424 (sync_lock_test_and_setsi): Add mode suffix to insn output.
5426 2018-01-11 Richard Biener <rguenther@suse.de>
5428 PR tree-optimization/83435
5429 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
5430 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
5431 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
5433 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5434 Alan Hayward <alan.hayward@arm.com>
5435 David Sherwood <david.sherwood@arm.com>
5437 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
5439 (aarch64_classify_address): Initialize it. Track polynomial offsets.
5440 (aarch64_print_address_internal): Use it to check for a zero offset.
5442 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5443 Alan Hayward <alan.hayward@arm.com>
5444 David Sherwood <david.sherwood@arm.com>
5446 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
5447 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
5448 Return a poly_int64 rather than a HOST_WIDE_INT.
5449 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
5450 rather than a HOST_WIDE_INT.
5451 * config/aarch64/aarch64.h (aarch64_frame): Protect with
5452 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
5453 hard_fp_offset, frame_size, initial_adjust, callee_offset and
5454 final_offset from HOST_WIDE_INT to poly_int64.
5455 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
5456 to_constant when getting the number of units in an Advanced SIMD
5458 (aarch64_builtin_vectorized_function): Check for a constant number
5460 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
5462 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
5463 attribute instead of GET_MODE_NUNITS.
5464 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
5465 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
5466 GET_MODE_SIZE for fixed-size registers.
5467 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
5468 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
5469 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
5470 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
5471 (aarch64_print_operand, aarch64_print_address_internal)
5472 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
5473 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
5474 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
5475 Handle polynomial GET_MODE_SIZE.
5476 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
5477 wider than SImode without modification.
5478 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
5479 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
5480 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
5481 passing and returning SVE modes.
5482 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
5483 rather than GEN_INT.
5484 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
5485 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
5486 (aarch64_allocate_and_probe_stack_space): Likewise.
5487 (aarch64_layout_frame): Cope with polynomial offsets.
5488 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
5489 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
5491 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
5492 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
5493 poly_int64 rather than a HOST_WIDE_INT.
5494 (aarch64_get_separate_components, aarch64_process_components)
5495 (aarch64_expand_prologue, aarch64_expand_epilogue)
5496 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
5497 (aarch64_anchor_offset): New function, split out from...
5498 (aarch64_legitimize_address): ...here.
5499 (aarch64_builtin_vectorization_cost): Handle polynomial
5500 TYPE_VECTOR_SUBPARTS.
5501 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
5503 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
5504 number of elements from the PARALLEL rather than the mode.
5505 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
5506 rather than GET_MODE_BITSIZE.
5507 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
5508 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
5509 (aarch64_expand_vec_perm_const_1): Handle polynomial
5510 d->perm.length () and d->perm elements.
5511 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
5512 Apply to_constant to d->perm elements.
5513 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
5514 polynomial CONST_VECTOR_NUNITS.
5515 (aarch64_move_pointer): Take amount as a poly_int64 rather
5517 (aarch64_progress_pointer): Avoid temporary variable.
5518 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
5519 the mode attribute instead of GET_MODE.
5521 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5522 Alan Hayward <alan.hayward@arm.com>
5523 David Sherwood <david.sherwood@arm.com>
5525 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
5526 x exists before using it.
5527 (aarch64_add_constant_internal): Rename to...
5528 (aarch64_add_offset_1): ...this. Replace regnum with separate
5529 src and dest rtxes. Handle the case in which they're different,
5530 including when the offset is zero. Replace scratchreg with an rtx.
5531 Use 2 additions if there is no spare register into which we can
5532 move a 16-bit constant.
5533 (aarch64_add_constant): Delete.
5534 (aarch64_add_offset): Replace reg with separate src and dest
5535 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
5536 Use aarch64_add_offset_1.
5537 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
5538 an rtx rather than an int. Take the delta as a poly_int64
5539 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
5540 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
5541 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
5542 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
5543 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
5545 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
5546 aarch64_add_constant.
5548 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5550 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
5551 Use scalar_float_mode.
5553 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5555 * config/aarch64/aarch64-simd.md
5556 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
5557 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
5558 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
5559 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
5560 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
5561 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
5562 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
5563 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
5564 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
5565 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
5567 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5570 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
5571 targ_options->x_arm_arch_string is non NULL.
5573 2018-01-11 Tamar Christina <tamar.christina@arm.com>
5575 * config/aarch64/aarch64.h
5576 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
5578 2018-01-11 Sudakshina Das <sudi.das@arm.com>
5581 * expmed.c (emit_store_flag_force): Swap if const op0
5582 and change VOIDmode to mode of op0.
5584 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5586 PR rtl-optimization/83761
5587 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
5588 than bytes to mode_for_size.
5590 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5593 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
5594 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
5597 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5600 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
5601 when in layout mode.
5602 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
5603 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
5606 2018-01-10 Michael Collison <michael.collison@arm.com>
5608 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
5609 * config/aarch64/aarch64-option-extension.def: Add
5610 AARCH64_OPT_EXTENSION of 'fp16fml'.
5611 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5612 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
5613 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
5614 * config/aarch64/constraints.md (Ui7): New constraint.
5615 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
5616 (VFMLA_SEL_W): Ditto.
5619 (VFMLA16_LOW): New int iterator.
5620 (VFMLA16_HIGH): Ditto.
5621 (UNSPEC_FMLAL): New unspec.
5622 (UNSPEC_FMLSL): Ditto.
5623 (UNSPEC_FMLAL2): Ditto.
5624 (UNSPEC_FMLSL2): Ditto.
5625 (f16mac): New code attribute.
5626 * config/aarch64/aarch64-simd-builtins.def
5627 (aarch64_fmlal_lowv2sf): Ditto.
5628 (aarch64_fmlsl_lowv2sf): Ditto.
5629 (aarch64_fmlalq_lowv4sf): Ditto.
5630 (aarch64_fmlslq_lowv4sf): Ditto.
5631 (aarch64_fmlal_highv2sf): Ditto.
5632 (aarch64_fmlsl_highv2sf): Ditto.
5633 (aarch64_fmlalq_highv4sf): Ditto.
5634 (aarch64_fmlslq_highv4sf): Ditto.
5635 (aarch64_fmlal_lane_lowv2sf): Ditto.
5636 (aarch64_fmlsl_lane_lowv2sf): Ditto.
5637 (aarch64_fmlal_laneq_lowv2sf): Ditto.
5638 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
5639 (aarch64_fmlalq_lane_lowv4sf): Ditto.
5640 (aarch64_fmlsl_lane_lowv4sf): Ditto.
5641 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
5642 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
5643 (aarch64_fmlal_lane_highv2sf): Ditto.
5644 (aarch64_fmlsl_lane_highv2sf): Ditto.
5645 (aarch64_fmlal_laneq_highv2sf): Ditto.
5646 (aarch64_fmlsl_laneq_highv2sf): Ditto.
5647 (aarch64_fmlalq_lane_highv4sf): Ditto.
5648 (aarch64_fmlsl_lane_highv4sf): Ditto.
5649 (aarch64_fmlalq_laneq_highv4sf): Ditto.
5650 (aarch64_fmlsl_laneq_highv4sf): Ditto.
5651 * config/aarch64/aarch64-simd.md:
5652 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
5653 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5654 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
5655 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5656 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
5657 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
5658 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
5659 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
5660 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
5661 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
5662 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
5663 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
5664 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
5665 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
5666 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
5667 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
5668 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
5669 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
5670 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
5671 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
5672 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
5673 (vfmlsl_low_u32): Ditto.
5674 (vfmlalq_low_u32): Ditto.
5675 (vfmlslq_low_u32): Ditto.
5676 (vfmlal_high_u32): Ditto.
5677 (vfmlsl_high_u32): Ditto.
5678 (vfmlalq_high_u32): Ditto.
5679 (vfmlslq_high_u32): Ditto.
5680 (vfmlal_lane_low_u32): Ditto.
5681 (vfmlsl_lane_low_u32): Ditto.
5682 (vfmlal_laneq_low_u32): Ditto.
5683 (vfmlsl_laneq_low_u32): Ditto.
5684 (vfmlalq_lane_low_u32): Ditto.
5685 (vfmlslq_lane_low_u32): Ditto.
5686 (vfmlalq_laneq_low_u32): Ditto.
5687 (vfmlslq_laneq_low_u32): Ditto.
5688 (vfmlal_lane_high_u32): Ditto.
5689 (vfmlsl_lane_high_u32): Ditto.
5690 (vfmlal_laneq_high_u32): Ditto.
5691 (vfmlsl_laneq_high_u32): Ditto.
5692 (vfmlalq_lane_high_u32): Ditto.
5693 (vfmlslq_lane_high_u32): Ditto.
5694 (vfmlalq_laneq_high_u32): Ditto.
5695 (vfmlslq_laneq_high_u32): Ditto.
5696 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
5697 (AARCH64_FL_FOR_ARCH8_4): New.
5698 (AARCH64_ISA_F16FML): New ISA flag.
5699 (TARGET_F16FML): New feature flag for fp16fml.
5700 (doc/invoke.texi): Document new fp16fml option.
5702 2018-01-10 Michael Collison <michael.collison@arm.com>
5704 * config/aarch64/aarch64-builtins.c:
5705 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
5706 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5707 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
5708 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
5709 (AARCH64_ISA_SHA3): New ISA flag.
5710 (TARGET_SHA3): New feature flag for sha3.
5711 * config/aarch64/iterators.md (sha512_op): New int attribute.
5712 (CRYPTO_SHA512): New int iterator.
5713 (UNSPEC_SHA512H): New unspec.
5714 (UNSPEC_SHA512H2): Ditto.
5715 (UNSPEC_SHA512SU0): Ditto.
5716 (UNSPEC_SHA512SU1): Ditto.
5717 * config/aarch64/aarch64-simd-builtins.def
5718 (aarch64_crypto_sha512hqv2di): New builtin.
5719 (aarch64_crypto_sha512h2qv2di): Ditto.
5720 (aarch64_crypto_sha512su0qv2di): Ditto.
5721 (aarch64_crypto_sha512su1qv2di): Ditto.
5722 (aarch64_eor3qv8hi): Ditto.
5723 (aarch64_rax1qv2di): Ditto.
5724 (aarch64_xarqv2di): Ditto.
5725 (aarch64_bcaxqv8hi): Ditto.
5726 * config/aarch64/aarch64-simd.md:
5727 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
5728 (aarch64_crypto_sha512su0qv2di): Ditto.
5729 (aarch64_crypto_sha512su1qv2di): Ditto.
5730 (aarch64_eor3qv8hi): Ditto.
5731 (aarch64_rax1qv2di): Ditto.
5732 (aarch64_xarqv2di): Ditto.
5733 (aarch64_bcaxqv8hi): Ditto.
5734 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
5735 (vsha512h2q_u64): Ditto.
5736 (vsha512su0q_u64): Ditto.
5737 (vsha512su1q_u64): Ditto.
5738 (veor3q_u16): Ditto.
5739 (vrax1q_u64): Ditto.
5741 (vbcaxq_u16): Ditto.
5742 * config/arm/types.md (crypto_sha512): New type attribute.
5743 (crypto_sha3): Ditto.
5744 (doc/invoke.texi): Document new sha3 option.
5746 2018-01-10 Michael Collison <michael.collison@arm.com>
5748 * config/aarch64/aarch64-builtins.c:
5749 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
5750 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5751 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
5752 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
5753 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
5754 (AARCH64_ISA_SM4): New ISA flag.
5755 (TARGET_SM4): New feature flag for sm4.
5756 * config/aarch64/aarch64-simd-builtins.def
5757 (aarch64_sm3ss1qv4si): Ditto.
5758 (aarch64_sm3tt1aq4si): Ditto.
5759 (aarch64_sm3tt1bq4si): Ditto.
5760 (aarch64_sm3tt2aq4si): Ditto.
5761 (aarch64_sm3tt2bq4si): Ditto.
5762 (aarch64_sm3partw1qv4si): Ditto.
5763 (aarch64_sm3partw2qv4si): Ditto.
5764 (aarch64_sm4eqv4si): Ditto.
5765 (aarch64_sm4ekeyqv4si): Ditto.
5766 * config/aarch64/aarch64-simd.md:
5767 (aarch64_sm3ss1qv4si): Ditto.
5768 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
5769 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
5770 (aarch64_sm4eqv4si): Ditto.
5771 (aarch64_sm4ekeyqv4si): Ditto.
5772 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
5773 (sm3part_op): Ditto.
5774 (CRYPTO_SM3TT): Ditto.
5775 (CRYPTO_SM3PART): Ditto.
5776 (UNSPEC_SM3SS1): New unspec.
5777 (UNSPEC_SM3TT1A): Ditto.
5778 (UNSPEC_SM3TT1B): Ditto.
5779 (UNSPEC_SM3TT2A): Ditto.
5780 (UNSPEC_SM3TT2B): Ditto.
5781 (UNSPEC_SM3PARTW1): Ditto.
5782 (UNSPEC_SM3PARTW2): Ditto.
5783 (UNSPEC_SM4E): Ditto.
5784 (UNSPEC_SM4EKEY): Ditto.
5785 * config/aarch64/constraints.md (Ui2): New constraint.
5786 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
5787 * config/arm/types.md (crypto_sm3): New type attribute.
5788 (crypto_sm4): Ditto.
5789 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
5790 (vsm3tt1aq_u32): Ditto.
5791 (vsm3tt1bq_u32): Ditto.
5792 (vsm3tt2aq_u32): Ditto.
5793 (vsm3tt2bq_u32): Ditto.
5794 (vsm3partw1q_u32): Ditto.
5795 (vsm3partw2q_u32): Ditto.
5796 (vsm4eq_u32): Ditto.
5797 (vsm4ekeyq_u32): Ditto.
5798 (doc/invoke.texi): Document new sm4 option.
5800 2018-01-10 Michael Collison <michael.collison@arm.com>
5802 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
5803 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
5804 (AARCH64_FL_FOR_ARCH8_4): New.
5805 (AARCH64_FL_V8_4): New flag.
5806 (doc/invoke.texi): Document new armv8.4-a option.
5808 2018-01-10 Michael Collison <michael.collison@arm.com>
5810 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5811 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
5812 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
5813 * config/aarch64/aarch64-option-extension.def: Add
5814 AARCH64_OPT_EXTENSION of 'sha2'.
5815 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
5816 (crypto): Disable sha2 and aes if crypto disabled.
5817 (crypto): Enable aes and sha2 if enabled.
5818 (simd): Disable sha2 and aes if simd disabled.
5819 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
5821 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
5822 (TARGET_SHA2): New feature flag for sha2.
5823 (TARGET_AES): New feature flag for aes.
5824 * config/aarch64/aarch64-simd.md:
5825 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
5826 conditional on TARGET_AES.
5827 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
5828 (aarch64_crypto_sha1hsi): Make pattern conditional
5830 (aarch64_crypto_sha1hv4si): Ditto.
5831 (aarch64_be_crypto_sha1hv4si): Ditto.
5832 (aarch64_crypto_sha1su1v4si): Ditto.
5833 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
5834 (aarch64_crypto_sha1su0v4si): Ditto.
5835 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
5836 (aarch64_crypto_sha256su0v4si): Ditto.
5837 (aarch64_crypto_sha256su1v4si): Ditto.
5838 (doc/invoke.texi): Document new aes and sha2 options.
5840 2018-01-10 Martin Sebor <msebor@redhat.com>
5842 PR tree-optimization/83781
5843 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
5846 2018-01-11 Martin Sebor <msebor@gmail.com>
5847 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5849 PR tree-optimization/83501
5850 PR tree-optimization/81703
5852 * tree-ssa-strlen.c (get_string_cst): Rename...
5853 (get_string_len): ...to this. Handle global constants.
5854 (handle_char_store): Adjust.
5856 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
5857 Jim Wilson <jimw@sifive.com>
5859 * config/riscv/riscv-protos.h (riscv_output_return): New.
5860 * config/riscv/riscv.c (struct machine_function): New naked_p field.
5861 (riscv_attribute_table, riscv_output_return),
5862 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
5863 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
5864 (riscv_compute_frame_info): Only compute frame->mask if not a naked
5866 (riscv_expand_prologue): Add early return for naked function.
5867 (riscv_expand_epilogue): Likewise.
5868 (riscv_function_ok_for_sibcall): Return false for naked function.
5869 (riscv_set_current_function): New.
5870 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
5871 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
5872 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
5873 * doc/extend.texi (RISC-V Function Attributes): New.
5875 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
5877 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
5878 check for 128-bit long double before checking TCmode.
5879 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
5880 128-bit long doubles before checking TFmode or TCmode.
5881 (FLOAT128_IBM_P): Likewise.
5883 2018-01-10 Martin Sebor <msebor@redhat.com>
5885 PR tree-optimization/83671
5886 * builtins.c (c_strlen): Unconditionally return zero for the empty
5888 Use -Warray-bounds for warnings.
5889 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
5890 for non-constant array indices with COMPONENT_REF, arrays of
5891 arrays, and pointers to arrays.
5892 (gimple_fold_builtin_strlen): Determine and set length range for
5893 non-constant character arrays.
5895 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
5898 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
5901 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
5903 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
5905 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5908 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
5909 VECTOR_MEM_ALTIVEC_OR_VSX_P.
5910 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
5911 indexed_or_indirect_operand predicate.
5912 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
5913 (*vsx_le_perm_load_v8hi): Likewise.
5914 (*vsx_le_perm_load_v16qi): Likewise.
5915 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
5916 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
5917 (*vsx_le_perm_store_v8hi): Likewise.
5918 (*vsx_le_perm_store_v16qi): Likewise.
5919 (eight unnamed splitters): Likewise.
5921 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5923 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
5924 * config/rs6000/emmintrin.h: Likewise.
5925 * config/rs6000/mmintrin.h: Likewise.
5926 * config/rs6000/xmmintrin.h: Likewise.
5928 2018-01-10 David Malcolm <dmalcolm@redhat.com>
5931 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
5933 * tree.c (tree_nop_conversion): Return true for location wrapper
5935 (maybe_wrap_with_location): New function.
5936 (selftest::check_strip_nops): New function.
5937 (selftest::test_location_wrappers): New function.
5938 (selftest::tree_c_tests): Call it.
5939 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
5940 (maybe_wrap_with_location): New decl.
5941 (EXPR_LOCATION_WRAPPER_P): New macro.
5942 (location_wrapper_p): New inline function.
5943 (tree_strip_any_location_wrapper): New inline function.
5945 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5948 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5949 stack_realign_offset for the largest alignment of stack slot
5951 (ix86_find_max_used_stack_alignment): New function.
5952 (ix86_finalize_stack_frame_flags): Use it. Set
5953 max_used_stack_alignment if we don't realign stack.
5954 * config/i386/i386.h (machine_function): Add
5955 max_used_stack_alignment.
5957 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5959 * config/arm/arm.opt (-mbranch-cost): New option.
5960 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5963 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5966 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5967 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5969 2018-01-10 Richard Biener <rguenther@suse.de>
5972 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5973 early out so it also covers the case where we have a non-NULL
5976 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5978 PR tree-optimization/83753
5979 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5980 for non-strided grouped accesses if the number of elements is 1.
5982 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5985 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5986 * i386.h (TARGET_USE_GATHER): Define.
5987 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5989 2018-01-10 Martin Liska <mliska@suse.cz>
5992 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5993 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5995 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5996 CLEANUP_NO_PARTITIONING is not set.
5998 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
6000 * doc/rtl.texi: Remove documentation of (const ...) wrappers
6001 for vectors, as a partial revert of r254296.
6002 * rtl.h (const_vec_p): Delete.
6003 (const_vec_duplicate_p): Don't test for vector CONSTs.
6004 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
6005 * expmed.c (make_tree): Likewise.
6008 * common.md (E, F): Use CONSTANT_P instead of checking for
6010 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
6011 checking for CONST_VECTOR.
6013 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
6016 * predict.c (force_edge_cold): Handle in more sane way edges
6019 2018-01-09 Carl Love <cel@us.ibm.com>
6021 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
6023 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
6024 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
6025 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
6026 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
6027 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
6028 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
6029 * config/rs6000/rs6000-protos.h: Add extern defition for
6030 rs6000_generate_float2_double_code.
6031 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
6033 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
6034 (float2_v2df): Add define_expand.
6036 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
6039 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
6040 op_mode in the force_to_mode call.
6042 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
6044 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
6045 instead of checking each element individually.
6046 (aarch64_evpc_uzp): Likewise.
6047 (aarch64_evpc_zip): Likewise.
6048 (aarch64_evpc_ext): Likewise.
6049 (aarch64_evpc_rev): Likewise.
6050 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
6051 instead of checking each element individually. Return true without
6053 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
6054 whether all selected elements come from the same input, instead of
6055 checking each element individually. Remove calls to gen_rtx_REG,
6056 start_sequence and end_sequence and instead assert that no rtl is
6059 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
6061 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
6062 order of HIGH and CONST checks.
6064 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
6066 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
6067 if the destination isn't an SSA_NAME.
6069 2018-01-09 Richard Biener <rguenther@suse.de>
6071 PR tree-optimization/83668
6072 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
6074 (canonicalize_loop_form): ... here, renamed from ...
6075 (canonicalize_loop_closed_ssa_form): ... this and amended to
6076 swap successor edges for loop exit blocks to make us use
6077 the RPO order we need for initial schedule generation.
6079 2018-01-09 Joseph Myers <joseph@codesourcery.com>
6081 PR tree-optimization/64811
6082 * match.pd: When optimizing comparisons with Inf, avoid
6083 introducing or losing exceptions from comparisons with NaN.
6085 2018-01-09 Martin Liska <mliska@suse.cz>
6088 * asan.c (shadow_mem_size): Add gcc_assert.
6090 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
6092 Don't save registers in main().
6095 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
6096 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
6097 * config/avr/avr.c (avr_set_current_function): Don't error if
6098 naked, OS_task or OS_main are specified at the same time.
6099 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
6101 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
6103 * common/config/avr/avr-common.c (avr_option_optimization_table):
6104 Switch on -mmain-is-OS_task for optimizing compilations.
6106 2018-01-09 Richard Biener <rguenther@suse.de>
6108 PR tree-optimization/83572
6109 * graphite.c: Include cfganal.h.
6110 (graphite_transform_loops): Connect infinite loops to exit
6111 and remove fake edges at the end.
6113 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
6115 * ipa-inline.c (edge_badness): Revert accidental checkin.
6117 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
6120 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
6121 symbols; not inline clones.
6123 2018-01-09 Jakub Jelinek <jakub@redhat.com>
6126 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
6127 hard registers. Formatting fixes.
6129 PR preprocessor/83722
6130 * gcc.c (try_generate_repro): Pass
6131 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
6132 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
6135 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
6136 Kito Cheng <kito.cheng@gmail.com>
6138 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
6139 (riscv_leaf_function_p): Delete.
6140 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
6142 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6144 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
6146 (do_ifelse): New function.
6147 (do_isel): New function.
6148 (do_sub3): New function.
6149 (do_add3): New function.
6150 (do_load_mask_compare): New function.
6151 (do_overlap_load_compare): New function.
6152 (expand_compare_loop): New function.
6153 (expand_block_compare): Call expand_compare_loop() when appropriate.
6154 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
6156 (-mblock-compare-inline-loop-limit): New option.
6158 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6161 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
6162 Reverse order of second and third operands in first alternative.
6163 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
6164 of first and second elements in UNSPEC_VPERMR vector.
6165 (altivec_expand_vec_perm_le): Likewise.
6167 2018-01-08 Jeff Law <law@redhat.com>
6169 PR rtl-optimizatin/81308
6170 * tree-switch-conversion.c (cfg_altered): New file scoped static.
6171 (process_switch): If group_case_labels makes a change, then set
6173 (pass_convert_switch::execute): If a switch is converted, then
6174 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
6176 PR rtl-optimization/81308
6177 * recog.c (split_all_insns): Conditionally cleanup the CFG after
6180 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
6182 PR target/83663 - Revert r255946
6183 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
6184 generation for cases where splatting a value is not useful.
6185 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
6186 across a vec_duplicate and a paradoxical subreg forming a vector
6187 mode to a vec_concat.
6189 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
6191 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
6192 -march=armv8.3-a variants.
6193 * config/arm/t-multilib: Likewise.
6194 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
6196 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6198 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
6200 (cceq_ior_compare_complement): Give it a name so I can use it, and
6201 change boolean_or_operator predicate to boolean_operator so it can
6202 be used to generate a crand.
6203 (eqne): New code iterator.
6204 (bd/bd_neg): New code_attrs.
6205 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
6206 a single define_insn.
6207 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
6208 decrement (bdnzt/bdnzf/bdzt/bdzf).
6209 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
6210 with the new names of the branch decrement patterns, and added the
6211 names of the branch decrement conditional patterns.
6213 2018-01-08 Richard Biener <rguenther@suse.de>
6215 PR tree-optimization/83563
6216 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
6219 2018-01-08 Richard Biener <rguenther@suse.de>
6222 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
6224 2018-01-08 Richard Biener <rguenther@suse.de>
6226 PR tree-optimization/83685
6227 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
6228 references to abnormals.
6230 2018-01-08 Richard Biener <rguenther@suse.de>
6233 * dwarf2out.c (output_indirect_strings): Handle empty
6234 skeleton_debug_str_hash.
6235 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
6237 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
6239 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
6240 (emit_store_direct): Likewise.
6241 (arc_trampoline_adjust_address): Likewise.
6242 (arc_asm_trampoline_template): New function.
6243 (arc_initialize_trampoline): Use asm_trampoline_template.
6244 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
6245 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
6246 * config/arc/arc.md (flush_icache): Delete pattern.
6248 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
6250 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
6251 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
6254 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
6257 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
6258 by not USED_FOR_TARGET.
6259 (make_pass_resolve_sw_modes): Likewise.
6261 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
6263 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
6266 2018-01-08 Richard Biener <rguenther@suse.de>
6269 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
6271 2018-01-08 Richard Biener <rguenther@suse.de>
6274 * match.pd ((t * 2) / 2) -> t): Add missing :c.
6276 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
6279 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
6280 basic blocks with a small number of successors.
6281 (convert_control_dep_chain_into_preds): Improve handling of
6283 (dump_predicates): Split apart into...
6284 (dump_pred_chain): ...here...
6285 (dump_pred_info): ...and here.
6286 (can_one_predicate_be_invalidated_p): Add debugging printfs.
6287 (can_chain_union_be_invalidated_p): Improve check for invalidation
6289 (uninit_uses_cannot_happen): Avoid unnecessary if
6290 convert_control_dep_chain_into_preds yielded nothing.
6292 2018-01-06 Martin Sebor <msebor@redhat.com>
6294 PR tree-optimization/83640
6295 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
6296 subtracting negative offset from size.
6297 (builtin_access::overlap): Adjust offset bounds of the access to fall
6298 within the size of the object if possible.
6300 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
6302 PR rtl-optimization/83699
6303 * expmed.c (extract_bit_field_1): Restrict the vector usage of
6304 extract_bit_field_as_subreg to cases in which the extracted
6305 value is also a vector.
6307 * lra-constraints.c (process_alt_operands): Test for the equivalence
6308 substitutions when detecting a possible reload cycle.
6310 2018-01-06 Jakub Jelinek <jakub@redhat.com>
6313 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
6314 by default if flag_selective_schedling{,2}. Formatting fixes.
6316 PR rtl-optimization/83682
6317 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
6318 if it has non-VECTOR_MODE element mode.
6319 (vec_duplicate_p): Likewise.
6322 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
6323 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
6325 2018-01-05 Jakub Jelinek <jakub@redhat.com>
6328 * config/i386/i386-builtin.def
6329 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
6330 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
6331 Require also OPTION_MASK_ISA_AVX512F in addition to
6332 OPTION_MASK_ISA_GFNI.
6333 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
6334 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
6335 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
6336 to OPTION_MASK_ISA_GFNI.
6337 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
6338 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
6339 OPTION_MASK_ISA_AVX512BW.
6340 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
6341 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
6342 addition to OPTION_MASK_ISA_GFNI.
6343 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
6344 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
6345 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
6346 to OPTION_MASK_ISA_GFNI.
6347 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
6348 a requirement for all ISAs rather than any of them with a few
6350 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
6352 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
6353 bitmasks to be enabled with 3 exceptions, instead of requiring any
6354 enabled ISA with lots of exceptions.
6355 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
6356 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
6357 Change avx512bw in isa attribute to avx512f.
6358 * config/i386/sgxintrin.h: Add license boilerplate.
6359 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
6360 to __AVX512F__ and __AVX512VL to __AVX512VL__.
6361 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
6362 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
6364 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
6365 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
6366 temporarily sse2 rather than sse if not enabled already.
6369 * config/i386/sse.md (VI248_VLBW): Rename to ...
6370 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
6371 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
6372 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
6373 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
6374 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
6375 mode iterator instead of VI248_VLBW.
6377 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
6379 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
6380 (record_modified): Skip clobbers; add debug output.
6381 (param_change_prob): Use sreal frequencies.
6383 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
6385 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
6386 punt for user-aligned variables.
6388 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
6390 * tree-chrec.c (chrec_contains_symbols): Return true for
6393 2018-01-05 Sudakshina Das <sudi.das@arm.com>
6396 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
6397 of (x|y) == x for BICS pattern.
6399 2018-01-05 Jakub Jelinek <jakub@redhat.com>
6401 PR tree-optimization/83605
6402 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
6403 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
6406 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
6408 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
6409 * config/epiphany/rtems.h: New file.
6411 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6412 Uros Bizjak <ubizjak@gmail.com>
6415 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
6416 QIreg_operand instead of register_operand predicate.
6417 * config/i386/i386.c (ix86_rop_should_change_byte_p,
6418 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
6419 comments instead of -fmitigate[-_]rop.
6421 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
6424 * cgraphunit.c (symbol_table::compile): Switch to text_section
6425 before calling assembly_start debug hook.
6426 * run-rtl-passes.c (run_rtl_passes): Likewise.
6429 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6431 * tree-vrp.c (extract_range_from_binary_expr_1): Check
6432 range_int_cst_p rather than !symbolic_range_p before calling
6433 extract_range_from_multiplicative_op_1.
6435 2018-01-04 Jeff Law <law@redhat.com>
6437 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
6438 redundant test in assertion.
6440 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6442 * doc/rtl.texi: Document machine_mode wrapper classes.
6444 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6446 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
6449 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6451 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
6452 the VEC_PERM_EXPR fold to fail.
6454 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6457 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
6458 to switched_sections.
6460 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6463 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
6466 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
6469 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
6470 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
6472 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6475 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
6476 is BLKmode and bitpos not zero or mode change is needed.
6478 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6481 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
6484 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
6487 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
6488 instead of MULT rtx. Update all corresponding splitters.
6490 (*ssub<modesuffix>): Ditto.
6492 (*cmp_sadd_di): Update split patterns.
6493 (*cmp_sadd_si): Ditto.
6494 (*cmp_sadd_sidi): Ditto.
6495 (*cmp_ssub_di): Ditto.
6496 (*cmp_ssub_si): Ditto.
6497 (*cmp_ssub_sidi): Ditto.
6498 * config/alpha/predicates.md (const23_operand): New predicate.
6499 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
6500 Look for ASHIFT, not MULT inner operand.
6501 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
6503 2018-01-04 Martin Liska <mliska@suse.cz>
6505 PR gcov-profile/83669
6506 * gcov.c (output_intermediate_file): Add version to intermediate
6508 * doc/gcov.texi: Document new field 'version' in intermediate
6509 file format. Fix location of '-k' option of gcov command.
6511 2018-01-04 Martin Liska <mliska@suse.cz>
6514 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
6516 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6518 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
6520 2018-01-03 Martin Sebor <msebor@redhat.com>
6522 PR tree-optimization/83655
6523 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
6524 checking calls with invalid arguments.
6526 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6528 * tree-vect-stmts.c (vect_get_store_rhs): New function.
6529 (vectorizable_mask_load_store): Delete.
6530 (vectorizable_call): Return false for masked loads and stores.
6531 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
6532 instead of gimple_assign_rhs1.
6533 (vectorizable_load): Handle IFN_MASK_LOAD.
6534 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
6536 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6538 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
6540 (vectorizable_mask_load_store): ...here.
6541 (vectorizable_load): ...and here.
6543 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6545 * tree-vect-stmts.c (vect_build_all_ones_mask)
6546 (vect_build_zero_merge_argument): New functions, split out from...
6547 (vectorizable_load): ...here.
6549 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6551 * tree-vect-stmts.c (vect_check_store_rhs): New function,
6553 (vectorizable_mask_load_store): ...here.
6554 (vectorizable_store): ...and here.
6556 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6558 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
6560 (vectorizable_mask_load_store): ...here.
6562 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6564 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
6565 (vect_model_store_cost): Take a vec_load_store_type instead of a
6567 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
6568 (vect_model_store_cost): Take a vec_load_store_type instead of a
6570 (vectorizable_mask_load_store): Update accordingly.
6571 (vectorizable_store): Likewise.
6572 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
6574 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6576 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
6577 IFN_MASK_LOAD calls here rather than...
6578 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
6580 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6581 Alan Hayward <alan.hayward@arm.com>
6582 David Sherwood <david.sherwood@arm.com>
6584 * expmed.c (extract_bit_field_1): For vector extracts,
6585 fall back to extract_bit_field_as_subreg if vec_extract
6588 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6589 Alan Hayward <alan.hayward@arm.com>
6590 David Sherwood <david.sherwood@arm.com>
6592 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
6593 they are variable or constant sized.
6594 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
6595 slots for constant-sized data.
6597 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6598 Alan Hayward <alan.hayward@arm.com>
6599 David Sherwood <david.sherwood@arm.com>
6601 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
6602 handling COND_EXPRs with boolean comparisons, try to find a better
6603 basis for the mask type than the boolean itself.
6605 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6607 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
6608 is calculated and how it can be overridden.
6609 * genmodes.c (max_bitsize_mode_any_mode): New variable.
6610 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
6612 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
6615 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6616 Alan Hayward <alan.hayward@arm.com>
6617 David Sherwood <david.sherwood@arm.com>
6619 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
6620 Remove the mode argument.
6621 (aarch64_simd_valid_immediate): Remove the mode and inverse
6623 * config/aarch64/iterators.md (bitsize): New iterator.
6624 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
6625 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
6626 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
6627 aarch64_simd_valid_immediate.
6628 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
6629 (aarch64_reg_or_bic_imm): Likewise.
6630 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
6631 with an insn_type enum and msl with a modifier_type enum.
6632 Replace element_width with a scalar_mode. Change the shift
6633 to unsigned int. Add constructors for scalar_float_mode and
6634 scalar_int_mode elements.
6635 (aarch64_vect_float_const_representable_p): Delete.
6636 (aarch64_can_const_movi_rtx_p)
6637 (aarch64_simd_scalar_immediate_valid_for_move)
6638 (aarch64_simd_make_constant): Update call to
6639 aarch64_simd_valid_immediate.
6640 (aarch64_advsimd_valid_immediate_hs): New function.
6641 (aarch64_advsimd_valid_immediate): Likewise.
6642 (aarch64_simd_valid_immediate): Remove mode and inverse
6643 arguments. Rewrite to use the above. Use const_vec_duplicate_p
6644 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
6645 and aarch64_float_const_representable_p on the result.
6646 (aarch64_output_simd_mov_immediate): Remove mode argument.
6647 Update call to aarch64_simd_valid_immediate and use of
6648 simd_immediate_info.
6649 (aarch64_output_scalar_simd_mov_immediate): Update call
6652 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6653 Alan Hayward <alan.hayward@arm.com>
6654 David Sherwood <david.sherwood@arm.com>
6656 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
6657 (mode_nunits): Likewise CONST_MODE_NUNITS.
6658 * machmode.def (ADJUST_NUNITS): Document.
6659 * genmodes.c (mode_data::need_nunits_adj): New field.
6660 (blank_mode): Update accordingly.
6661 (adj_nunits): New variable.
6662 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
6664 (emit_mode_size_inline): Set need_bytesize_adj for all modes
6665 listed in adj_nunits.
6666 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
6667 listed in adj_nunits. Don't emit case statements for such modes.
6668 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
6669 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
6670 nothing if adj_nunits is nonnull.
6671 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
6672 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
6673 (emit_mode_fbit): Update use of print_maybe_const_decl.
6674 (emit_move_size): Likewise. Treat the array as non-const
6676 (emit_mode_adjustments): Handle adj_nunits.
6678 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6680 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
6681 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
6682 (VECTOR_MODES): Use it.
6683 (make_vector_modes): Take the prefix as an argument.
6685 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6686 Alan Hayward <alan.hayward@arm.com>
6687 David Sherwood <david.sherwood@arm.com>
6689 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
6690 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
6691 for MODE_VECTOR_BOOL.
6692 * machmode.def (VECTOR_BOOL_MODE): Document.
6693 * genmodes.c (VECTOR_BOOL_MODE): New macro.
6694 (make_vector_bool_mode): New function.
6695 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
6697 * lto-streamer-in.c (lto_input_mode_table): Likewise.
6698 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
6700 * stor-layout.c (int_mode_for_mode): Likewise.
6701 * tree.c (build_vector_type_for_mode): Likewise.
6702 * varasm.c (output_constant_pool_2): Likewise.
6703 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
6704 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
6705 for MODE_VECTOR_BOOL.
6706 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
6707 of mode class checks.
6708 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
6709 instead of a list of mode class checks.
6710 (expand_vector_scalar_condition): Likewise.
6711 (type_for_widest_vector_mode): Handle BImode as an inner mode.
6713 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6714 Alan Hayward <alan.hayward@arm.com>
6715 David Sherwood <david.sherwood@arm.com>
6717 * machmode.h (mode_size): Change from unsigned short to
6719 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
6720 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6721 or if measurement_type is not polynomial.
6722 (fixed_size_mode::includes_p): Check for constant-sized modes.
6723 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
6724 return a poly_uint16 rather than an unsigned short.
6725 (emit_mode_size): Change the type of mode_size from unsigned short
6726 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
6727 (emit_mode_adjustments): Cope with polynomial vector sizes.
6728 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6730 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6732 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
6733 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
6734 * caller-save.c (setup_save_areas): Likewise.
6735 (replace_reg_with_saved_mem): Likewise.
6736 * calls.c (emit_library_call_value_1): Likewise.
6737 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
6738 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
6739 (gen_lowpart_for_combine): Likewise.
6740 * convert.c (convert_to_integer_1): Likewise.
6741 * cse.c (equiv_constant, cse_insn): Likewise.
6742 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
6743 (cselib_subst_to_values): Likewise.
6744 * dce.c (word_dce_process_block): Likewise.
6745 * df-problems.c (df_word_lr_mark_ref): Likewise.
6746 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
6747 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
6748 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
6749 (rtl_for_decl_location): Likewise.
6750 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
6751 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
6752 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
6753 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
6754 (expand_expr_real_1): Likewise.
6755 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
6756 (pad_below): Likewise.
6757 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6758 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
6759 * ira.c (get_subreg_tracking_sizes): Likewise.
6760 * ira-build.c (ira_create_allocno_objects): Likewise.
6761 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
6762 (ira_sort_regnos_for_alter_reg): Likewise.
6763 * ira-costs.c (record_operand_costs): Likewise.
6764 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
6765 (resolve_simple_move): Likewise.
6766 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
6767 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
6768 (lra_constraints): Likewise.
6769 (CONST_POOL_OK_P): Reject variable-sized modes.
6770 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
6771 (add_pseudo_to_slot, lra_spill): Likewise.
6772 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6773 * optabs-query.c (get_best_extraction_insn): Likewise.
6774 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6775 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
6776 (expand_mult_highpart, valid_multiword_target_p): Likewise.
6777 * recog.c (offsettable_address_addr_space_p): Likewise.
6778 * regcprop.c (maybe_mode_change): Likewise.
6779 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
6780 * regrename.c (build_def_use): Likewise.
6781 * regstat.c (dump_reg_info): Likewise.
6782 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
6783 (find_reloads, find_reloads_subreg_address): Likewise.
6784 * reload1.c (eliminate_regs_1): Likewise.
6785 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
6786 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
6787 (simplify_binary_operation_1, simplify_subreg): Likewise.
6788 * targhooks.c (default_function_arg_padding): Likewise.
6789 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
6790 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
6791 (verify_gimple_assign_ternary): Likewise.
6792 * tree-inline.c (estimate_move_cost): Likewise.
6793 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6794 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
6795 (get_address_cost_ainc): Likewise.
6796 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
6797 (vect_supportable_dr_alignment): Likewise.
6798 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6799 (vectorizable_reduction): Likewise.
6800 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
6801 (vectorizable_operation, vectorizable_load): Likewise.
6802 * tree.c (build_same_sized_truth_vector_type): Likewise.
6803 * valtrack.c (cleanup_auto_inc_dec): Likewise.
6804 * var-tracking.c (emit_note_insn_var_location): Likewise.
6805 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
6806 (ADDR_VEC_ALIGN): Likewise.
6808 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6809 Alan Hayward <alan.hayward@arm.com>
6810 David Sherwood <david.sherwood@arm.com>
6812 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
6814 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6815 or if measurement_type is polynomial.
6816 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
6817 * combine.c (make_extraction): Likewise.
6818 * dse.c (find_shift_sequence): Likewise.
6819 * dwarf2out.c (mem_loc_descriptor): Likewise.
6820 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
6821 (extract_bit_field, extract_low_bits): Likewise.
6822 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
6823 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
6824 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
6825 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
6826 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6827 * reload.c (find_reloads): Likewise.
6828 * reload1.c (alter_reg): Likewise.
6829 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
6830 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
6831 * tree-if-conv.c (predicate_mem_writes): Likewise.
6832 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
6833 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
6834 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
6835 * valtrack.c (dead_debug_insert_temp): Likewise.
6836 * varasm.c (mergeable_constant_section): Likewise.
6837 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
6839 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6840 Alan Hayward <alan.hayward@arm.com>
6841 David Sherwood <david.sherwood@arm.com>
6843 * expr.c (expand_assignment): Cope with polynomial mode sizes
6844 when assigning to a CONCAT.
6846 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6847 Alan Hayward <alan.hayward@arm.com>
6848 David Sherwood <david.sherwood@arm.com>
6850 * machmode.h (mode_precision): Change from unsigned short to
6852 (mode_to_precision): Return a poly_uint16 rather than an unsigned
6854 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
6855 or if measurement_type is not polynomial.
6856 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
6857 in which the mode is already known to be a scalar_int_mode.
6858 * genmodes.c (emit_mode_precision): Change the type of mode_precision
6859 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
6861 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6862 for GET_MODE_PRECISION.
6863 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6864 for GET_MODE_PRECISION.
6865 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
6867 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
6868 (expand_field_assignment, make_extraction): Likewise.
6869 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
6870 (get_last_value): Likewise.
6871 * convert.c (convert_to_integer_1): Likewise.
6872 * cse.c (cse_insn): Likewise.
6873 * expr.c (expand_expr_real_1): Likewise.
6874 * lra-constraints.c (simplify_operand_subreg): Likewise.
6875 * optabs-query.c (can_atomic_load_p): Likewise.
6876 * optabs.c (expand_atomic_load): Likewise.
6877 (expand_atomic_store): Likewise.
6878 * ree.c (combine_reaching_defs): Likewise.
6879 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
6880 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
6881 * tree.h (type_has_mode_precision_p): Likewise.
6882 * ubsan.c (instrument_si_overflow): Likewise.
6884 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6885 Alan Hayward <alan.hayward@arm.com>
6886 David Sherwood <david.sherwood@arm.com>
6888 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
6889 polynomial numbers of units.
6890 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
6891 (valid_vector_subparts_p): New function.
6892 (build_vector_type): Remove temporary shim and take the number
6893 of units as a poly_uint64 rather than an int.
6894 (build_opaque_vector_type): Take the number of units as a
6895 poly_uint64 rather than an int.
6896 * tree.c (build_vector_from_ctor): Handle polynomial
6897 TYPE_VECTOR_SUBPARTS.
6898 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
6899 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
6900 (build_vector_from_val): If the number of units is variable,
6901 use build_vec_duplicate_cst for constant operands and
6902 VEC_DUPLICATE_EXPR otherwise.
6903 (make_vector_type): Remove temporary is_constant ().
6904 (build_vector_type, build_opaque_vector_type): Take the number of
6905 units as a poly_uint64 rather than an int.
6906 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
6908 * cfgexpand.c (expand_debug_expr): Likewise.
6909 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
6910 (store_constructor, expand_expr_real_1): Likewise.
6911 (const_scalar_mask_from_tree): Likewise.
6912 * fold-const-call.c (fold_const_reduction): Likewise.
6913 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
6914 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
6915 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
6916 (fold_relational_const): Likewise.
6917 (native_interpret_vector): Likewise. Change the size from an
6918 int to an unsigned int.
6919 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
6920 TYPE_VECTOR_SUBPARTS.
6921 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
6922 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
6923 duplicating a non-constant operand into a variable-length vector.
6924 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
6925 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
6926 * ipa-icf.c (sem_variable::equals): Likewise.
6927 * match.pd: Likewise.
6928 * omp-simd-clone.c (simd_clone_subparts): Likewise.
6929 * print-tree.c (print_node): Likewise.
6930 * stor-layout.c (layout_type): Likewise.
6931 * targhooks.c (default_builtin_vectorization_cost): Likewise.
6932 * tree-cfg.c (verify_gimple_comparison): Likewise.
6933 (verify_gimple_assign_binary): Likewise.
6934 (verify_gimple_assign_ternary): Likewise.
6935 (verify_gimple_assign_single): Likewise.
6936 * tree-pretty-print.c (dump_generic_node): Likewise.
6937 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6938 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
6939 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
6940 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
6941 (vect_shift_permute_load_chain): Likewise.
6942 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
6943 (expand_vector_condition, optimize_vector_constructor): Likewise.
6944 (lower_vec_perm, get_compute_type): Likewise.
6945 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6946 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6947 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6948 (vect_recog_mask_conversion_pattern): Likewise.
6949 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6950 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6951 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6952 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6953 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6954 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6955 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6956 (supportable_widening_operation): Likewise.
6957 (supportable_narrowing_operation): Likewise.
6958 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6960 * varasm.c (output_constant): Likewise.
6962 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6963 Alan Hayward <alan.hayward@arm.com>
6964 David Sherwood <david.sherwood@arm.com>
6966 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6967 so that both the length == 3 and length != 3 cases set up their
6968 own permute vectors. Add comments explaining why we know the
6969 number of elements is constant.
6970 (vect_permute_load_chain): Likewise.
6972 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6973 Alan Hayward <alan.hayward@arm.com>
6974 David Sherwood <david.sherwood@arm.com>
6976 * machmode.h (mode_nunits): Change from unsigned char to
6978 (ONLY_FIXED_SIZE_MODES): New macro.
6979 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6980 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6981 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6983 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6984 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6985 or if measurement_type is not polynomial.
6986 * genmodes.c (ZERO_COEFFS): New macro.
6987 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6989 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6990 Use ZERO_COEFFS when emitting initializers.
6991 * data-streamer.h (bp_pack_poly_value): New function.
6992 (bp_unpack_poly_value): Likewise.
6993 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6994 for GET_MODE_NUNITS.
6995 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6996 for GET_MODE_NUNITS.
6997 * tree.c (make_vector_type): Remove temporary shim and make
6998 the real function take the number of units as a poly_uint64
7000 (build_vector_type_for_mode): Handle polynomial nunits.
7001 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
7002 * emit-rtl.c (const_vec_series_p_1): Likewise.
7003 (gen_rtx_CONST_VECTOR): Likewise.
7004 * fold-const.c (test_vec_duplicate_folding): Likewise.
7005 * genrecog.c (validate_pattern): Likewise.
7006 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
7007 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
7008 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
7009 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
7010 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
7011 * rtlanal.c (subreg_get_info): Likewise.
7012 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7013 (vect_grouped_load_supported): Likewise.
7014 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
7015 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
7016 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
7017 (simplify_const_unary_operation, simplify_binary_operation_1)
7018 (simplify_const_binary_operation, simplify_ternary_operation)
7019 (test_vector_ops_duplicate, test_vector_ops): Likewise.
7020 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
7021 instead of CONST_VECTOR_NUNITS.
7022 * varasm.c (output_constant_pool_2): Likewise.
7023 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
7024 explicit-encoded elements in the XVEC for variable-length vectors.
7026 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7028 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
7030 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7031 Alan Hayward <alan.hayward@arm.com>
7032 David Sherwood <david.sherwood@arm.com>
7034 * coretypes.h (fixed_size_mode): Declare.
7035 (fixed_size_mode_pod): New typedef.
7036 * builtins.h (target_builtins::x_apply_args_mode)
7037 (target_builtins::x_apply_result_mode): Change type to
7038 fixed_size_mode_pod.
7039 * builtins.c (apply_args_size, apply_result_size, result_vector)
7040 (expand_builtin_apply_args_1, expand_builtin_apply)
7041 (expand_builtin_return): Update accordingly.
7043 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7045 * cse.c (hash_rtx_cb): Hash only the encoded elements.
7046 * cselib.c (cselib_hash_rtx): Likewise.
7047 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
7048 CONST_VECTOR encoding.
7050 2018-01-03 Jakub Jelinek <jakub@redhat.com>
7051 Jeff Law <law@redhat.com>
7054 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
7055 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
7056 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
7057 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
7060 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
7061 explicitly probe *sp in a noreturn function if there were any callee
7062 register saves or frame pointer is needed.
7064 2018-01-03 Jakub Jelinek <jakub@redhat.com>
7067 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
7068 BLKmode for ternary, binary or unary expressions.
7071 * var-tracking.c (delete_vta_debug_insn): New inline function.
7072 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
7073 insns from get_insns () to NULL instead of each bb separately.
7074 Use delete_vta_debug_insn. No longer static.
7075 (vt_debug_insns_local, variable_tracking_main_1): Adjust
7076 delete_vta_debug_insns callers.
7077 * rtl.h (delete_vta_debug_insns): Declare.
7078 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
7079 instead of variable_tracking_main.
7081 2018-01-03 Martin Sebor <msebor@redhat.com>
7083 PR tree-optimization/83603
7084 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
7085 arguments past the endof the argument list in functions declared
7086 without a prototype.
7087 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
7088 Avoid checking when arguments are null.
7090 2018-01-03 Martin Sebor <msebor@redhat.com>
7093 * doc/extend.texi (attribute const): Fix a typo.
7094 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
7095 issuing -Wsuggest-attribute for void functions.
7097 2018-01-03 Martin Sebor <msebor@redhat.com>
7099 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
7100 offset_int::from instead of wide_int::to_shwi.
7101 (maybe_diag_overlap): Remove assertion.
7102 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
7103 * gimple-ssa-sprintf.c (format_directive): Same.
7104 (parse_directive): Same.
7105 (sprintf_dom_walker::compute_format_length): Same.
7106 (try_substitute_return_value): Same.
7108 2018-01-03 Jeff Law <law@redhat.com>
7111 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
7112 non-constant residual for zero at runtime and avoid probing in
7113 that case. Reorganize code for trailing problem to mirror handling
7116 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
7118 PR tree-optimization/83501
7119 * tree-ssa-strlen.c (get_string_cst): New.
7120 (handle_char_store): Call get_string_cst.
7122 2018-01-03 Martin Liska <mliska@suse.cz>
7124 PR tree-optimization/83593
7125 * tree-ssa-strlen.c: Include tree-cfg.h.
7126 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
7127 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
7128 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
7130 (strlen_dom_walker::before_dom_children): Call
7131 gimple_purge_dead_eh_edges. Dump tranformation with details
7133 (strlen_dom_walker::before_dom_children): Update call by adding
7134 new argument cleanup_eh.
7135 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
7137 2018-01-03 Martin Liska <mliska@suse.cz>
7140 * cif-code.def (VARIADIC_THUNK): New enum value.
7141 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
7144 2018-01-03 Jan Beulich <jbeulich@suse.com>
7146 * sse.md (mov<mode>_internal): Tighten condition for when to use
7147 vmovdqu<ssescalarsize> for TI and OI modes.
7149 2018-01-03 Jakub Jelinek <jakub@redhat.com>
7151 Update copyright years.
7153 2018-01-03 Martin Liska <mliska@suse.cz>
7156 * ipa-visibility.c (function_and_variable_visibility): Skip
7157 functions with noipa attribure.
7159 2018-01-03 Jakub Jelinek <jakub@redhat.com>
7161 * gcc.c (process_command): Update copyright notice dates.
7162 * gcov-dump.c (print_version): Ditto.
7163 * gcov.c (print_version): Ditto.
7164 * gcov-tool.c (print_version): Ditto.
7165 * gengtype.c (create_file): Ditto.
7166 * doc/cpp.texi: Bump @copying's copyright year.
7167 * doc/cppinternals.texi: Ditto.
7168 * doc/gcc.texi: Ditto.
7169 * doc/gccint.texi: Ditto.
7170 * doc/gcov.texi: Ditto.
7171 * doc/install.texi: Ditto.
7172 * doc/invoke.texi: Ditto.
7174 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7176 * vector-builder.h (vector_builder::m_full_nelts): Change from
7177 unsigned int to poly_uint64.
7178 (vector_builder::full_nelts): Update prototype accordingly.
7179 (vector_builder::new_vector): Likewise.
7180 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
7181 (vector_builder::operator ==): Likewise.
7182 (vector_builder::finalize): Likewise.
7183 * int-vector-builder.h (int_vector_builder::int_vector_builder):
7184 Take the number of elements as a poly_uint64 rather than an
7186 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
7187 from unsigned int to poly_uint64.
7188 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
7189 (vec_perm_indices::new_vector): Likewise.
7190 (vec_perm_indices::length): Likewise.
7191 (vec_perm_indices::nelts_per_input): Likewise.
7192 (vec_perm_indices::input_nelts): Likewise.
7193 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
7194 number of elements per input as a poly_uint64 rather than an
7195 unsigned int. Use the original encoding for variable-length
7196 vectors, rather than clamping each individual element.
7197 For the second and subsequent elements in each pattern,
7198 clamp the step and base before clamping their sum.
7199 (vec_perm_indices::series_p): Handle polynomial element counts.
7200 (vec_perm_indices::all_in_range_p): Likewise.
7201 (vec_perm_indices_to_tree): Likewise.
7202 (vec_perm_indices_to_rtx): Likewise.
7203 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
7204 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
7205 (tree_vector_builder::new_binary_operation): Handle polynomial
7206 element counts. Return false if we need to know the number
7207 of elements at compile time.
7208 * fold-const.c (fold_vec_perm): Punt if the number of elements
7209 isn't known at compile time.
7211 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7213 * vec-perm-indices.h (vec_perm_builder): Change element type
7214 from HOST_WIDE_INT to poly_int64.
7215 (vec_perm_indices::element_type): Update accordingly.
7216 (vec_perm_indices::clamp): Handle polynomial element_types.
7217 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7218 (vec_perm_indices::all_in_range_p): Likewise.
7219 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
7221 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
7222 polynomial vec_perm_indices element types.
7223 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
7224 * fold-const.c (fold_vec_perm): Likewise.
7225 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
7226 * tree-vect-generic.c (lower_vec_perm): Likewise.
7227 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7228 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
7229 element type to HOST_WIDE_INT.
7231 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7232 Alan Hayward <alan.hayward@arm.com>
7233 David Sherwood <david.sherwood@arm.com>
7235 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
7236 rather than an int. Use plus_constant.
7237 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
7238 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
7240 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7241 Alan Hayward <alan.hayward@arm.com>
7242 David Sherwood <david.sherwood@arm.com>
7244 * calls.c (emit_call_1, expand_call): Change struct_value_size from
7245 a HOST_WIDE_INT to a poly_int64.
7247 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7248 Alan Hayward <alan.hayward@arm.com>
7249 David Sherwood <david.sherwood@arm.com>
7251 * calls.c (load_register_parameters): Cope with polynomial
7252 mode sizes. Require a constant size for BLKmode parameters
7253 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
7254 forces a parameter to be padded at the lsb end in order to
7255 fill a complete number of words, require the parameter size
7256 to be ordered wrt UNITS_PER_WORD.
7258 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7259 Alan Hayward <alan.hayward@arm.com>
7260 David Sherwood <david.sherwood@arm.com>
7262 * reload1.c (spill_stack_slot_width): Change element type
7263 from unsigned int to poly_uint64_pod.
7264 (alter_reg): Treat mode sizes as polynomial.
7266 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7267 Alan Hayward <alan.hayward@arm.com>
7268 David Sherwood <david.sherwood@arm.com>
7270 * reload.c (complex_word_subreg_p): New function.
7271 (reload_inner_reg_of_subreg, push_reload): Use it.
7273 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7274 Alan Hayward <alan.hayward@arm.com>
7275 David Sherwood <david.sherwood@arm.com>
7277 * lra-constraints.c (process_alt_operands): Reject matched
7278 operands whose sizes aren't ordered.
7279 (match_reload): Refer to this check here.
7281 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7282 Alan Hayward <alan.hayward@arm.com>
7283 David Sherwood <david.sherwood@arm.com>
7285 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
7286 that the mode size is in the set {1, 2, 4, 8, 16}.
7288 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7289 Alan Hayward <alan.hayward@arm.com>
7290 David Sherwood <david.sherwood@arm.com>
7292 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
7293 Use plus_constant instead of gen_rtx_PLUS.
7295 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7296 Alan Hayward <alan.hayward@arm.com>
7297 David Sherwood <david.sherwood@arm.com>
7299 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
7300 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
7301 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
7302 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
7303 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
7304 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
7305 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
7306 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
7307 * config/i386/i386.c (ix86_push_rounding): ...this new function.
7308 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
7310 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
7311 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
7312 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
7313 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
7314 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
7315 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
7316 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
7317 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
7318 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
7319 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
7321 * expr.c (emit_move_resolve_push): Treat the input and result
7322 of PUSH_ROUNDING as a poly_int64.
7323 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
7324 (emit_push_insn): Likewise.
7325 * lra-eliminations.c (mark_not_eliminable): Likewise.
7326 * recog.c (push_operand): Likewise.
7327 * reload1.c (elimination_effects): Likewise.
7328 * rtlanal.c (nonzero_bits1): Likewise.
7329 * calls.c (store_one_arg): Likewise. Require the padding to be
7330 known at compile time.
7332 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7333 Alan Hayward <alan.hayward@arm.com>
7334 David Sherwood <david.sherwood@arm.com>
7336 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
7337 Use plus_constant instead of gen_rtx_PLUS.
7339 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7340 Alan Hayward <alan.hayward@arm.com>
7341 David Sherwood <david.sherwood@arm.com>
7343 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
7346 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7347 Alan Hayward <alan.hayward@arm.com>
7348 David Sherwood <david.sherwood@arm.com>
7350 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
7351 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
7352 via stack temporaries. Treat the mode size as polynomial too.
7354 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7355 Alan Hayward <alan.hayward@arm.com>
7356 David Sherwood <david.sherwood@arm.com>
7358 * expr.c (expand_expr_real_2): When handling conversions involving
7359 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
7360 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
7361 as a poly_uint64 too.
7363 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7364 Alan Hayward <alan.hayward@arm.com>
7365 David Sherwood <david.sherwood@arm.com>
7367 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
7369 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7370 Alan Hayward <alan.hayward@arm.com>
7371 David Sherwood <david.sherwood@arm.com>
7373 * combine.c (can_change_dest_mode): Handle polynomial
7374 REGMODE_NATURAL_SIZE.
7375 * expmed.c (store_bit_field_1): Likewise.
7376 * expr.c (store_constructor): Likewise.
7377 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
7378 and polynomial REGMODE_NATURAL_SIZE.
7379 (gen_lowpart_common): Likewise.
7380 * reginfo.c (record_subregs_of_mode): Likewise.
7381 * rtlanal.c (read_modify_subreg_p): Likewise.
7383 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7384 Alan Hayward <alan.hayward@arm.com>
7385 David Sherwood <david.sherwood@arm.com>
7387 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
7388 numbers of elements.
7390 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7391 Alan Hayward <alan.hayward@arm.com>
7392 David Sherwood <david.sherwood@arm.com>
7394 * match.pd: Cope with polynomial numbers of vector elements.
7396 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7397 Alan Hayward <alan.hayward@arm.com>
7398 David Sherwood <david.sherwood@arm.com>
7400 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
7401 in a POINTER_PLUS_EXPR.
7403 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7404 Alan Hayward <alan.hayward@arm.com>
7405 David Sherwood <david.sherwood@arm.com>
7407 * omp-simd-clone.c (simd_clone_subparts): New function.
7408 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
7409 (ipa_simd_modify_function_body): Likewise.
7411 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7412 Alan Hayward <alan.hayward@arm.com>
7413 David Sherwood <david.sherwood@arm.com>
7415 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
7416 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
7417 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
7418 (expand_vector_condition, vector_element): Likewise.
7419 (subparts_gt): New function.
7420 (get_compute_type): Use subparts_gt.
7421 (count_type_subparts): Delete.
7422 (expand_vector_operations_1): Use subparts_gt instead of
7423 count_type_subparts.
7425 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7426 Alan Hayward <alan.hayward@arm.com>
7427 David Sherwood <david.sherwood@arm.com>
7429 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
7430 (vect_compile_time_alias): ...this new function. Do the calculation
7431 on poly_ints rather than trees.
7432 (vect_prune_runtime_alias_test_list): Update call accordingly.
7434 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7435 Alan Hayward <alan.hayward@arm.com>
7436 David Sherwood <david.sherwood@arm.com>
7438 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
7440 (vect_schedule_slp_instance): Likewise.
7442 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7443 Alan Hayward <alan.hayward@arm.com>
7444 David Sherwood <david.sherwood@arm.com>
7446 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
7447 constant and extern definitions for variable-length vectors.
7448 (vect_get_constant_vectors): Note that the number of units
7449 is known to be constant.
7451 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7452 Alan Hayward <alan.hayward@arm.com>
7453 David Sherwood <david.sherwood@arm.com>
7455 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
7456 of units as polynomial. Choose between WIDE and NARROW based
7459 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7460 Alan Hayward <alan.hayward@arm.com>
7461 David Sherwood <david.sherwood@arm.com>
7463 * tree-vect-stmts.c (simd_clone_subparts): New function.
7464 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
7466 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7467 Alan Hayward <alan.hayward@arm.com>
7468 David Sherwood <david.sherwood@arm.com>
7470 * tree-vect-stmts.c (vectorizable_call): Treat the number of
7471 vectors as polynomial. Use build_index_vector for
7474 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7475 Alan Hayward <alan.hayward@arm.com>
7476 David Sherwood <david.sherwood@arm.com>
7478 * tree-vect-stmts.c (get_load_store_type): Treat the number of
7479 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
7480 for variable-length vectors.
7481 (vectorizable_mask_load_store): Treat the number of units as
7482 polynomial, asserting that it is constant if the condition has
7483 already been enforced.
7484 (vectorizable_store, vectorizable_load): Likewise.
7486 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7487 Alan Hayward <alan.hayward@arm.com>
7488 David Sherwood <david.sherwood@arm.com>
7490 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
7491 of units as polynomial. Punt if we can't tell at compile time
7492 which vector contains the final result.
7494 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7495 Alan Hayward <alan.hayward@arm.com>
7496 David Sherwood <david.sherwood@arm.com>
7498 * tree-vect-loop.c (vectorizable_induction): Treat the number
7499 of units as polynomial. Punt on SLP inductions. Use an integer
7500 VEC_SERIES_EXPR for variable-length integer reductions. Use a
7501 cast of such a series for variable-length floating-point
7504 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7505 Alan Hayward <alan.hayward@arm.com>
7506 David Sherwood <david.sherwood@arm.com>
7508 * tree.h (build_index_vector): Declare.
7509 * tree.c (build_index_vector): New function.
7510 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
7511 of units as polynomial, forcibly converting it to a constant if
7512 vectorizable_reduction has already enforced the condition.
7513 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
7514 to create a {1,2,3,...} vector.
7515 (vectorizable_reduction): Treat the number of units as polynomial.
7516 Choose vectype_in based on the largest scalar element size rather
7517 than the smallest number of units. Enforce the restrictions
7520 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7521 Alan Hayward <alan.hayward@arm.com>
7522 David Sherwood <david.sherwood@arm.com>
7524 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
7525 number of units as polynomial.
7527 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7528 Alan Hayward <alan.hayward@arm.com>
7529 David Sherwood <david.sherwood@arm.com>
7531 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
7532 * target.def (autovectorize_vector_sizes): Return the vector sizes
7533 by pointer, using vector_sizes rather than a bitmask.
7534 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
7535 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
7536 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
7538 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
7539 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
7540 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
7541 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
7542 * omp-general.c (omp_max_vf): Likewise.
7543 * omp-low.c (omp_clause_aligned_alignment): Likewise.
7544 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
7545 * tree-vect-loop.c (vect_analyze_loop): Likewise.
7546 * tree-vect-slp.c (vect_slp_bb): Likewise.
7547 * doc/tm.texi: Regenerate.
7548 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
7550 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
7551 the vector size as a poly_uint64 rather than an unsigned int.
7552 (current_vector_size): Change from an unsigned int to a poly_uint64.
7553 (get_vectype_for_scalar_type): Update accordingly.
7554 * tree.h (build_truth_vector_type): Take the size and number of
7555 units as a poly_uint64 rather than an unsigned int.
7556 (build_vector_type): Add a temporary overload that takes
7557 the number of units as a poly_uint64 rather than an unsigned int.
7558 * tree.c (make_vector_type): Likewise.
7559 (build_truth_vector_type): Take the number of units as a poly_uint64
7560 rather than an unsigned int.
7562 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7563 Alan Hayward <alan.hayward@arm.com>
7564 David Sherwood <david.sherwood@arm.com>
7566 * target.def (get_mask_mode): Take the number of units and length
7567 as poly_uint64s rather than unsigned ints.
7568 * targhooks.h (default_get_mask_mode): Update accordingly.
7569 * targhooks.c (default_get_mask_mode): Likewise.
7570 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
7571 * doc/tm.texi: Regenerate.
7573 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7574 Alan Hayward <alan.hayward@arm.com>
7575 David Sherwood <david.sherwood@arm.com>
7577 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
7578 * omp-general.c (omp_max_vf): Likewise.
7579 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
7580 (expand_omp_simd): Handle polynomial safelen.
7581 * omp-low.c (omplow_simd_context): Add a default constructor.
7582 (omplow_simd_context::max_vf): Change from int to poly_uint64.
7583 (lower_rec_simd_input_clauses): Update accordingly.
7584 (lower_rec_input_clauses): Likewise.
7586 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7587 Alan Hayward <alan.hayward@arm.com>
7588 David Sherwood <david.sherwood@arm.com>
7590 * tree-vectorizer.h (vect_nunits_for_cost): New function.
7591 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
7592 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
7593 (vect_analyze_slp_cost): Likewise.
7594 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
7595 (vect_model_load_cost): Likewise.
7597 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7598 Alan Hayward <alan.hayward@arm.com>
7599 David Sherwood <david.sherwood@arm.com>
7601 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
7602 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
7603 from an unsigned int * to a poly_uint64_pod *.
7604 (calculate_unrolling_factor): New function.
7605 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
7607 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7608 Alan Hayward <alan.hayward@arm.com>
7609 David Sherwood <david.sherwood@arm.com>
7611 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
7612 from an unsigned int to a poly_uint64.
7613 (_loop_vec_info::slp_unrolling_factor): Likewise.
7614 (_loop_vec_info::vectorization_factor): Change from an int
7616 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
7617 (vect_get_num_vectors): New function.
7618 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
7619 (vect_get_num_copies): Use vect_get_num_vectors.
7620 (vect_analyze_data_ref_dependences): Change max_vf from an int *
7621 to an unsigned int *.
7622 (vect_analyze_data_refs): Change min_vf from an int * to a
7624 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7625 than an unsigned HOST_WIDE_INT.
7626 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
7627 (vect_analyze_data_ref_dependence): Change max_vf from an int *
7628 to an unsigned int *.
7629 (vect_analyze_data_ref_dependences): Likewise.
7630 (vect_compute_data_ref_alignment): Handle polynomial vf.
7631 (vect_enhance_data_refs_alignment): Likewise.
7632 (vect_prune_runtime_alias_test_list): Likewise.
7633 (vect_shift_permute_load_chain): Likewise.
7634 (vect_supportable_dr_alignment): Likewise.
7635 (dependence_distance_ge_vf): Take the vectorization factor as a
7636 poly_uint64 rather than an unsigned HOST_WIDE_INT.
7637 (vect_analyze_data_refs): Change min_vf from an int * to a
7639 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
7640 vfm1 as a poly_uint64 rather than an int. Make the same change
7641 for the returned bound_scalar.
7642 (vect_gen_vector_loop_niters): Handle polynomial vf.
7643 (vect_do_peeling): Likewise. Update call to
7644 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
7645 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
7647 * tree-vect-loop.c (vect_determine_vectorization_factor)
7648 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
7649 (vect_get_known_peeling_cost): Likewise.
7650 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
7651 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
7652 (vect_transform_loop): Likewise. Use the lowest possible VF when
7653 updating the upper bounds of the loop.
7654 (vect_min_worthwhile_factor): Make static. Return an unsigned int
7656 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
7657 polynomial unroll factors.
7658 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
7659 (vect_make_slp_decision): Likewise.
7660 (vect_supported_load_permutation_p): Likewise, and polynomial
7662 (vect_analyze_slp_cost): Handle polynomial vf.
7663 (vect_slp_analyze_node_operations): Likewise.
7664 (vect_slp_analyze_bb_1): Likewise.
7665 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7666 than an unsigned HOST_WIDE_INT.
7667 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
7668 (vectorizable_load): Handle polynomial vf.
7669 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
7671 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
7673 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7674 Alan Hayward <alan.hayward@arm.com>
7675 David Sherwood <david.sherwood@arm.com>
7677 * match.pd: Handle bit operations involving three constants
7678 and try to fold one pair.
7680 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7682 * tree-vect-loop-manip.c: Include gimple-fold.h.
7683 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
7684 niters_maybe_zero parameters. Handle other cases besides a step of 1.
7685 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
7686 Add a path that uses a step of VF instead of 1, but disable it
7688 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
7689 and niters_no_overflow parameters. Update calls to
7690 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
7691 Create a new SSA name if the latter choses to use a ste other
7692 than zero, and return it via niters_vector_mult_vf_var.
7693 * tree-vect-loop.c (vect_transform_loop): Update calls to
7694 vect_do_peeling, vect_gen_vector_loop_niters and
7695 slpeel_make_loop_iterate_ntimes.
7696 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
7697 (vect_gen_vector_loop_niters): Update declarations after above changes.
7699 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
7701 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
7702 128-bit round to integer instructions.
7703 (ceil<mode>2): Likewise.
7704 (btrunc<mode>2): Likewise.
7705 (round<mode>2): Likewise.
7707 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7709 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
7710 unaligned VSX load/store on P8/P9.
7711 (expand_block_clear): Allow the use of unaligned VSX
7712 load/store on P8/P9.
7714 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
7716 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
7718 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
7719 swap associated with both a load and a store.
7721 2018-01-02 Andrew Waterman <andrew@sifive.com>
7723 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
7724 * config/riscv/riscv.md (clear_cache): Use it.
7726 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
7728 * web.c: Remove out-of-date comment.
7730 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7732 * expr.c (fixup_args_size_notes): Check that any existing
7733 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
7734 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
7735 (emit_single_push_insn): ...here.
7737 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7739 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
7740 (const_vector_encoded_nelts): New function.
7741 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
7742 (const_vector_int_elt, const_vector_elt): Declare.
7743 * emit-rtl.c (const_vector_int_elt_1): New function.
7744 (const_vector_elt): Likewise.
7745 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
7746 of CONST_VECTOR_ELT.
7748 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7750 * expr.c: Include rtx-vector-builder.h.
7751 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
7752 directly on the tree encoding.
7753 (const_vector_from_tree): Likewise.
7754 * optabs.c: Include rtx-vector-builder.h.
7755 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
7756 sequence of "u" values.
7757 * vec-perm-indices.c: Include rtx-vector-builder.h.
7758 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
7759 directly on the vec_perm_indices encoding.
7761 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7763 * doc/rtl.texi (const_vector): Describe new encoding scheme.
7764 * Makefile.in (OBJS): Add rtx-vector-builder.o.
7765 * rtx-vector-builder.h: New file.
7766 * rtx-vector-builder.c: Likewise.
7767 * rtl.h (rtx_def::u2): Add a const_vector field.
7768 (CONST_VECTOR_NPATTERNS): New macro.
7769 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
7770 (CONST_VECTOR_DUPLICATE_P): Likewise.
7771 (CONST_VECTOR_STEPPED_P): Likewise.
7772 (CONST_VECTOR_ENCODED_ELT): Likewise.
7773 (const_vec_duplicate_p): Check for a duplicated vector encoding.
7774 (unwrap_const_vec_duplicate): Likewise.
7775 (const_vec_series_p): Check for a non-duplicated vector encoding.
7776 Say that the function only returns true for integer vectors.
7777 * emit-rtl.c: Include rtx-vector-builder.h.
7778 (gen_const_vec_duplicate_1): Delete.
7779 (gen_const_vector): Call gen_const_vec_duplicate instead of
7780 gen_const_vec_duplicate_1.
7781 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
7782 (gen_const_vec_duplicate): Use rtx_vector_builder.
7783 (gen_const_vec_series): Likewise.
7784 (gen_rtx_CONST_VECTOR): Likewise.
7785 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
7786 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7787 Build a new vector rather than modifying a CONST_VECTOR in-place.
7788 (handle_special_swappables): Update call accordingly.
7789 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
7790 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7791 Build a new vector rather than modifying a CONST_VECTOR in-place.
7792 (handle_special_swappables): Update call accordingly.
7794 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7796 * simplify-rtx.c (simplify_const_binary_operation): Use
7797 CONST_VECTOR_ELT instead of XVECEXP.
7799 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7801 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
7802 the selector elements to be different from the data elements
7803 if the selector is a VECTOR_CST.
7804 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
7805 ssizetype for the selector.
7807 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7809 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
7810 before testing each element individually.
7811 * tree-vect-generic.c (lower_vec_perm): Likewise.
7813 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7815 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
7816 * selftest-run-tests.c (selftest::run_tests): Call it.
7817 * vector-builder.h (vector_builder::operator ==): New function.
7818 (vector_builder::operator !=): Likewise.
7819 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
7820 (vec_perm_indices::all_from_input_p): New function.
7821 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7822 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
7823 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
7824 instead of reading the VECTOR_CST directly. Detect whether both
7825 vector inputs are the same before constructing the vec_perm_indices,
7826 and update the number of inputs argument accordingly. Use the
7827 utility functions added above. Only construct sel2 if we need to.
7829 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7831 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
7832 the broadcast of the low byte.
7833 (expand_mult_highpart): Use an explicit encoding for the permutes.
7834 * optabs-query.c (can_mult_highpart_p): Likewise.
7835 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
7836 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7837 (vectorizable_bswap): Likewise.
7838 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
7839 explicit encoding for the power-of-2 permutes.
7840 (vect_permute_store_chain): Likewise.
7841 (vect_grouped_load_supported): Likewise.
7842 (vect_permute_load_chain): Likewise.
7844 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7846 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
7847 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
7848 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
7849 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7850 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
7851 (vect_gen_perm_mask_any): Likewise.
7853 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7855 * int-vector-builder.h: New file.
7856 * vec-perm-indices.h: Include int-vector-builder.h.
7857 (vec_perm_indices): Redefine as an int_vector_builder.
7858 (auto_vec_perm_indices): Delete.
7859 (vec_perm_builder): Redefine as a stand-alone class.
7860 (vec_perm_indices::vec_perm_indices): New function.
7861 (vec_perm_indices::clamp): Likewise.
7862 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
7863 (vec_perm_indices::new_vector): New function.
7864 (vec_perm_indices::new_expanded_vector): Update for new
7865 vec_perm_indices class.
7866 (vec_perm_indices::rotate_inputs): New function.
7867 (vec_perm_indices::all_in_range_p): Operate directly on the
7868 encoded form, without computing elided elements.
7869 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
7870 encoding. Update for new vec_perm_indices class.
7871 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
7872 the given vec_perm_builder.
7873 (expand_vec_perm_var): Update vec_perm_builder constructor.
7874 (expand_mult_highpart): Use vec_perm_builder instead of
7875 auto_vec_perm_indices.
7876 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
7877 vec_perm_indices instead of auto_vec_perm_indices. Use a single
7878 or double series encoding as appropriate.
7879 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
7880 vec_perm_indices instead of auto_vec_perm_indices.
7881 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7882 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7883 (vect_permute_store_chain): Likewise.
7884 (vect_grouped_load_supported): Likewise.
7885 (vect_permute_load_chain): Likewise.
7886 (vect_shift_permute_load_chain): Likewise.
7887 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7888 (vect_transform_slp_perm_load): Likewise.
7889 (vect_schedule_slp_instance): Likewise.
7890 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7891 (vectorizable_mask_load_store): Likewise.
7892 (vectorizable_bswap): Likewise.
7893 (vectorizable_store): Likewise.
7894 (vectorizable_load): Likewise.
7895 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
7896 vec_perm_indices instead of auto_vec_perm_indices. Use
7897 tree_to_vec_perm_builder to read the vector from a tree.
7898 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
7899 vec_perm_builder instead of a vec_perm_indices.
7900 (have_whole_vector_shift): Use vec_perm_builder and
7901 vec_perm_indices instead of auto_vec_perm_indices. Leave the
7902 truncation to calc_vec_perm_mask_for_shift.
7903 (vect_create_epilog_for_reduction): Likewise.
7904 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
7905 from auto_vec_perm_indices to vec_perm_indices.
7906 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7907 instead of changing individual elements.
7908 (aarch64_vectorize_vec_perm_const): Use new_vector to install
7909 the vector in d.perm.
7910 * config/arm/arm.c (expand_vec_perm_d::perm): Change
7911 from auto_vec_perm_indices to vec_perm_indices.
7912 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7913 instead of changing individual elements.
7914 (arm_vectorize_vec_perm_const): Use new_vector to install
7915 the vector in d.perm.
7916 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
7917 Update vec_perm_builder constructor.
7918 (rs6000_expand_interleave): Likewise.
7919 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
7920 (rs6000_expand_interleave): Likewise.
7922 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7924 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
7925 to qimode could truncate the indices.
7926 * optabs.c (expand_vec_perm_var): Likewise.
7928 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7930 * Makefile.in (OBJS): Add vec-perm-indices.o.
7931 * vec-perm-indices.h: New file.
7932 * vec-perm-indices.c: Likewise.
7933 * target.h (vec_perm_indices): Replace with a forward class
7935 (auto_vec_perm_indices): Move to vec-perm-indices.h.
7936 * optabs.h: Include vec-perm-indices.h.
7937 (expand_vec_perm): Delete.
7938 (selector_fits_mode_p, expand_vec_perm_var): Declare.
7939 (expand_vec_perm_const): Declare.
7940 * target.def (vec_perm_const_ok): Replace with...
7941 (vec_perm_const): ...this new hook.
7942 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
7943 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7944 * doc/tm.texi: Regenerate.
7945 * optabs.def (vec_perm_const): Delete.
7946 * doc/md.texi (vec_perm_const): Likewise.
7947 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7948 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7949 expand_vec_perm for constant permutation vectors. Assert that
7950 the mode of variable permutation vectors is the integer equivalent
7951 of the mode that is being permuted.
7952 * optabs-query.h (selector_fits_mode_p): Declare.
7953 * optabs-query.c: Include vec-perm-indices.h.
7954 (selector_fits_mode_p): New function.
7955 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7956 is defined, instead of checking whether the vec_perm_const_optab
7957 exists. Use targetm.vectorize.vec_perm_const instead of
7958 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7959 fit in the vector mode before using a variable permute.
7960 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7961 vec_perm_indices instead of an rtx.
7962 (expand_vec_perm): Replace with...
7963 (expand_vec_perm_const): ...this new function. Take the selector
7964 as a vec_perm_indices rather than an rtx. Also take the mode of
7965 the selector. Update call to shift_amt_for_vec_perm_mask.
7966 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7967 Use vec_perm_indices::new_expanded_vector to expand the original
7968 selector into bytes. Check whether the indices fit in the vector
7969 mode before using a variable permute.
7970 (expand_vec_perm_var): Make global.
7971 (expand_mult_highpart): Use expand_vec_perm_const.
7972 * fold-const.c: Includes vec-perm-indices.h.
7973 * tree-ssa-forwprop.c: Likewise.
7974 * tree-vect-data-refs.c: Likewise.
7975 * tree-vect-generic.c: Likewise.
7976 * tree-vect-loop.c: Likewise.
7977 * tree-vect-slp.c: Likewise.
7978 * tree-vect-stmts.c: Likewise.
7979 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7981 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7982 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7983 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7984 (aarch64_vectorize_vec_perm_const): ...this new function.
7985 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7986 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7987 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7988 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7989 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7990 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7991 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7993 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7994 check for NEON modes.
7995 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7996 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7997 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7998 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
8000 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
8001 the old VEC_PERM_CONST conditions.
8002 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
8003 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
8004 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
8005 (ia64_vectorize_vec_perm_const_ok): Merge into...
8006 (ia64_vectorize_vec_perm_const): ...this new function.
8007 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
8008 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
8009 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
8010 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
8011 * config/mips/mips.c (mips_expand_vec_perm_const)
8012 (mips_vectorize_vec_perm_const_ok): Merge into...
8013 (mips_vectorize_vec_perm_const): ...this new function.
8014 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
8015 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
8016 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
8017 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
8018 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
8019 (rs6000_expand_vec_perm_const): Delete.
8020 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
8022 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
8023 (altivec_expand_vec_perm_const_le): Take each operand individually.
8024 Operate on constant selectors rather than rtxes.
8025 (altivec_expand_vec_perm_const): Likewise. Update call to
8026 altivec_expand_vec_perm_const_le.
8027 (rs6000_expand_vec_perm_const): Delete.
8028 (rs6000_vectorize_vec_perm_const_ok): Delete.
8029 (rs6000_vectorize_vec_perm_const): New function.
8030 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
8031 an element count and rtx array.
8032 (rs6000_expand_extract_even): Update call accordingly.
8033 (rs6000_expand_interleave): Likewise.
8034 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
8035 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
8036 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
8037 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
8038 (rs6000_expand_vec_perm_const): Delete.
8039 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
8040 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
8041 (altivec_expand_vec_perm_const_le): Take each operand individually.
8042 Operate on constant selectors rather than rtxes.
8043 (altivec_expand_vec_perm_const): Likewise. Update call to
8044 altivec_expand_vec_perm_const_le.
8045 (rs6000_expand_vec_perm_const): Delete.
8046 (rs6000_vectorize_vec_perm_const_ok): Delete.
8047 (rs6000_vectorize_vec_perm_const): New function. Remove stray
8048 reference to the SPE evmerge intructions.
8049 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
8050 an element count and rtx array.
8051 (rs6000_expand_extract_even): Update call accordingly.
8052 (rs6000_expand_interleave): Likewise.
8053 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
8054 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
8056 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
8058 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8060 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
8061 vector mode and that that mode matches the mode of the data
8063 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
8064 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
8065 directly using expand_vec_perm_1 when forcing selectors into
8067 (expand_vec_perm_var): New function, split out from expand_vec_perm.
8069 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8071 * optabs-query.h (can_vec_perm_p): Delete.
8072 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
8073 * optabs-query.c (can_vec_perm_p): Split into...
8074 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
8075 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
8076 particular selector is valid.
8077 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
8078 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
8079 (vect_grouped_load_supported): Likewise.
8080 (vect_shift_permute_load_chain): Likewise.
8081 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
8082 (vect_transform_slp_perm_load): Likewise.
8083 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
8084 (vectorizable_bswap): Likewise.
8085 (vect_gen_perm_mask_checked): Likewise.
8086 * fold-const.c (fold_ternary_loc): Likewise. Don't take
8087 implementations of variable permutation vectors into account
8088 when deciding which selector to use.
8089 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
8090 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
8091 with a false third argument.
8092 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
8093 to test whether the constant selector is valid and can_vec_perm_var_p
8094 to test whether a variable selector is valid.
8096 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8098 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
8099 * optabs-query.c (can_vec_perm_p): Likewise.
8100 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
8101 instead of vec_perm_indices.
8102 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
8103 (vect_gen_perm_mask_checked): Likewise,
8104 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
8105 (vect_gen_perm_mask_checked): Likewise,
8107 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
8109 * optabs-query.h (qimode_for_vec_perm): Declare.
8110 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
8111 (qimode_for_vec_perm): ...this new function.
8112 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
8114 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
8116 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
8117 does not have a conditional at the top.
8119 2018-01-02 Richard Biener <rguenther@suse.de>
8121 * ipa-inline.c (big_speedup_p): Fix expression.
8123 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
8126 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
8129 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
8133 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
8134 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
8135 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
8136 cond_taken_branch_cost 3->4.
8138 2018-01-01 Jakub Jelinek <jakub@redhat.com>
8140 PR tree-optimization/83581
8141 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
8142 TODO_cleanup_cfg if any changes have been made.
8145 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
8146 convert_modes if target mode has the right side, but different mode
8150 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
8151 last argument when extracting from CONCAT. If either from_real or
8152 from_imag is NULL, use expansion through memory. If result is not
8153 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
8154 the parts directly to inner mode, if even that fails, use expansion
8158 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
8159 check for bswap in mode rather than HImode and use that in expand_unop
8162 Copyright (C) 2018 Free Software Foundation, Inc.
8164 Copying and distribution of this file, with or without modification,
8165 are permitted in any medium without royalty provided the copyright
8166 notice and this notice are preserved.