1 2018-03-02 Richard Sandiford <richard.sandiford@linaro.org>
3 * tree-vect-data-refs.c (vect_analyze_data_ref_dependence)
4 (vect_analyze_data_ref_access): Use loop->safe_len rather than
5 loop->force_vectorize to check whether there is no alias.
7 2018-03-02 Jakub Jelinek <jakub@redhat.com>
10 * rtl.h (prev_real_nondebug_insn, next_real_nondebug_insn): New
12 * emit-rtl.c (next_real_insn, prev_real_insn): Fix up function
14 (next_real_nondebug_insn, prev_real_nondebug_insn): New functions.
15 * cfgcleanup.c (try_head_merge_bb): Use prev_real_nondebug_insn
16 instead of a loop around prev_real_insn.
17 * combine.c (move_deaths): Use prev_real_nondebug_insn instead of
21 * config/i386/i386.c (ix86_print_operand): Use conditional
22 output_operand_lossage instead of gcc_assert if CONST_VECTOR is not
25 2018-03-02 Richard Biener <rguenther@suse.de>
27 PR tree-optimization/84427
28 * tree-ssa-pre.c (bitmap_remove_expr_from_set): Remove.
29 (bitmap_set_subtract_values): Rewrite to handle multiple
32 (prune_clobbered_mems): Likewise.
33 (phi_translate): Take edge instead of pred/phiblock.
34 (phi_translate_1): Likewise.
35 (phi_translate_set): Likewise. Insert all translated
36 exprs for a value into the set, keeping possibly multiple
37 expressions per value.
38 (compute_antic_aux): Adjust for phi_translate changes.
39 When intersecting union the expressions and prune those
40 not in the final value set, keeping possibly multiple
41 expressions per value. Do not use value-insertion
42 for unioning ANTIC_OUT U EXP_GEN - TMP_GEN but merge
43 all expressions. Add verification that the value-sets
44 only shrink during iteration.
45 (compute_partial_antic_aux): Adjust for the phi_translate changes.
46 (do_pre_regular_insertion): Likewise.
47 (do_pre_partial_partial_insertion): Likewise.
49 2018-03-02 Richard Biener <rguenther@suse.de>
52 * config/darwin.c (saved_debug_info_level): New static global.
53 (darwin_asm_lto_start): Disable debug info generation for LTO out.
54 (darwin_asm_lto_end): Restore debug info generation settings.
56 2018-03-01 Martin Liska <mliska@suse.cz>
59 * sanopt.c (sanitize_rewrite_addressable_params): Do not handle
62 2018-03-01 Richard Biener <rguenther@suse.de>
65 * dwarf2out.c (gen_variable_die): Properly handle late VLA
66 type annotation with LTO when debug was disabled at compile-time.
68 2018-03-01 Matthew Fortune <mfortune@gmail.com>
70 * config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
72 (mips_final_postscan_insn): Likewise.
74 2018-03-01 Richard Sandiford <richard.sandiford@linaro.org>
76 PR rtl-optimization/84528
77 * alias.c (init_alias_target): Add commentary.
78 (init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
79 a unique base value if the frame pointer is not eliminated
82 2018-03-01 Tom de Vries <tom@codesourcery.com>
84 PR rtl-optimization/83327
85 * lra-int.h (hard_regs_spilled_into): Declare.
86 * lra.c (hard_regs_spilled_into): Define.
87 (init_reg_info): Init hard_regs_spilled_into.
88 * lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
89 * lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
90 (process_bb_lives): Handle hard_regs_spilled_into.
91 (lra_create_live_ranges_1): Before doing liveness propagation, clear
92 regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.
94 2018-02-28 David Edelsohn <dje.gcc@gmail.com>
96 * config.gcc (powerpc-ibm-aix7.1.*): New stanza.
97 (powerpc-ibm-aix[789]*): Default to AIX 7.2.
98 * config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
99 * config/rs6000/aix72.h: New file.
101 2018-02-28 Jakub Jelinek <jakub@redhat.com>
103 * gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
104 instead of warning_at with conditional singular and plural messages
108 * stor-layout.c (update_alignment_for_field): For
109 targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
110 && !DECL_PACKED (field), do the alignment update, just use
111 only desired_align instead of MAX (type_align, desired_align)
113 (place_field): Don't do known_align < desired_align handling
114 early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
115 is non-NULL, instead do it after rli->prev_field handling and
116 only if not within a bitfield word. For DECL_PACKED (field)
117 use type_align of BITS_PER_UNIT.
119 2018-02-28 Eric Botcazou <ebotcazou@adacore.com>
121 * config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
122 superfluous parentheses and trailing spaces.
124 2018-02-28 Richard Biener <rguenther@suse.de>
126 PR tree-optimization/84584
127 * graphite-scop-detection.c (scop_detection::add_scop): Discard
128 SCoPs with fake exit edge.
130 2018-02-28 Martin Liska <mliska@suse.cz>
133 * timevar.c (timer::print): Fix format to properly print 100%
136 2018-02-28 Richard Biener <rguenther@suse.de>
139 * genmatch.c (capture_info::walk_match): Do not mark
140 captured expressions without operands as expr_p given
141 they act more like predicates and should be subject to
142 "lost tail" side-effect preserving.
144 2018-02-28 Alexandre Oliva <aoliva@redhat.com>
146 PR rtl-optimization/81611
147 * auto-inc-dec.c (attempt_change): Move dead note from
148 mem_insn if it's the next use of regno
149 (find_address): Take address use of reg holding
150 non-incremented value. Add parm to limit search to the named
152 (merge_in_block): Attempt to use a mem insn that is the next
153 use of the original regno.
155 2018-02-27 Martin Sebor <msebor@redhat.com>
158 * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
159 * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
161 2018-02-27 Martin Sebor <msebor@redhat.com>
164 * diagnostic-core.h (warning_n, error_n, inform_n): Change
165 n argument to unsigned HOST_WIDE_INT.
166 * diagnostic.c (warning_n, error_n, inform_n): Ditto.
167 (diagnostic_n_impl): Ditto. Handle arguments in excess of LONG_MAX.
168 * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
169 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
171 2018-02-27 Richard Biener <rguenther@suse.de>
173 PR tree-optimization/84512
174 * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
175 Do not use the estimate returned from record_stmt_cost for
176 the scalar iteration cost but sum properly using add_stmt_cost.
178 2018-02-27 Richard Biener <rguenther@suse.de>
180 PR tree-optimization/84466
181 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
182 Adjust last change to less strictly validate use operands.
184 2018-02-27 Martin Liska <mliska@suse.cz>
186 PR gcov-profile/84548
187 * gcov.c (process_file): Allow partial overlap and consider it
188 also as group functions.
189 (output_lines): Properly calculate range of lines for a group.
191 2018-02-27 Martin Liska <mliska@suse.cz>
193 * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
194 'ggc' suffixes. Change first column width.
195 (timer::print): Fix formatting of the column.
197 2018-02-27 Alexandre Oliva <aoliva@redhat.com>
199 * tree-ssa-live.c (remove_unused_scope_block_p): Do not
200 preserve inline entry blocks for the sake of debug inline
201 entry point markers alone.
202 (remove_unused_locals): Suggest in comments a better place to
203 force the preservation of inline entry blocks that are
204 otherwise unused, but do not preserve them.
206 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
208 * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
210 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
213 * config/i386/constraints.md (Bs): Replace
214 ix86_indirect_branch_register with
215 TARGET_INDIRECT_BRANCH_REGISTER.
217 * config/i386/i386.md (indirect_jump): Likewise.
218 (tablejump): Likewise.
219 (*sibcall_memory): Likewise.
220 (*sibcall_value_memory): Likewise.
221 Peepholes of indirect call and jump via memory: Likewise.
222 (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
223 (*sibcall_value_GOT_32): Likewise.
224 * config/i386/predicates.md (indirect_branch_operand): Likewise.
225 (GOT_memory_operand): Likewise.
226 (call_insn_operand): Likewise.
227 (sibcall_insn_operand): Likewise.
228 (GOT32_symbol_operand): Likewise.
229 * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
231 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
233 PR rtl-optimization/83496
234 * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
235 booleans to RTXes. Call fix_reg_dead_note on every non-null element.
236 (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
237 redundant insn, if any.
238 (relax_delay_slots): Likewise.
239 (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
241 2018-02-26 Richard Sandiford <richard.sandiford@linaro.org>
243 PR tree-optimization/83965
244 * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
245 that grouped statements are part of a reduction chain. Return
246 true if the statement is not marked as a reduction itself but
248 (vect_recog_dot_prod_pattern): Don't check whether the statement
249 is part of a group here.
250 (vect_recog_sad_pattern): Likewise.
251 (vect_recog_widen_sum_pattern): Likewise.
253 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
256 * final.c (rest_of_clean_state): Also look for calls inside sequences.
258 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
261 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
263 (ix86_output_indirect_function_return): New prototype.
264 (ix86_split_simple_return_pop_internal): Likewise.
265 * config/i386/i386.c (indirect_return_via_cx): New.
266 (indirect_return_via_cx_bnd): Likewise.
267 (indirect_thunk_name): Handle return va CX_REG.
268 (output_indirect_thunk_function): Create alias for
269 __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
270 (ix86_output_indirect_jmp): Remove the bool argument.
271 (ix86_output_indirect_function_return): New function.
272 (ix86_split_simple_return_pop_internal): Likewise.
273 * config/i386/i386.md (*indirect_jump): Don't pass false
274 to ix86_output_indirect_jmp.
275 (*tablejump_1): Likewise.
276 (simple_return_pop_internal): Change it to define_insn_and_split.
277 Call ix86_split_simple_return_pop_internal to split it for
279 (simple_return_indirect_internal): Call
280 ix86_output_indirect_function_return instead of
281 ix86_output_indirect_jmp.
283 2018-02-26 Jakub Jelinek <jakub@redhat.com>
286 * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
287 memset and value initialization afterwards.
289 2018-02-26 Christophe Lyon <christophe.lyon@linaro.org>
291 * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
293 2018-02-26 Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
296 * common/config/aarch64/aarch64-common.c
297 (aarch_option_optimization_table[]): Switch
298 off fomit-frame-pointer
300 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
301 Chung-Ju Wu <jasonwucj@gmail.com>
303 * config/nds32/nds32-multiple.md (load_multiple): Disallow
305 (store_multiple): Ditto.
307 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
309 * config.gcc: Add --with-cpu support for nds32 target.
310 * config/nds32/nds32-opts.h (nds32_cpu_type): New.
311 * config/nds32/nds32.opt: Add -mcpu= option.
313 2018-02-25 Segher Boessenkool <segher@kernel.crashing.org>
315 * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
316 isel=yes): Warn for these deprecated options.
318 2018-02-23 David Edelsohn <dje.gcc@gmail.com>
320 * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
321 ISA_2_5_MASKS_EMBEDDED.
323 2018-02-23 Jakub Jelinek <jakub@redhat.com>
325 * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
326 p->max as pointers rather than using iterative_hash_expr.
328 2018-02-23 Carl Love <cel@us.ibm.com>
330 * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
331 macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
333 * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
334 P8V_BUILTIN_VEC_VSIGNED2. Change VSX_BUILTIN_VEC_VUNSIGNED2 to
335 P8V_BUILTIN_VEC_VUNSIGNED2.
337 2018-02-22 Vladimir Makarov <vmakarov@redhat.com>
340 * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
341 * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
343 * lra-constraints.c (curr_insn_transform): Set up
344 LRA_NON_CLOBBERED_ALT for moves processed on the fast path. Use
346 (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
347 * lra-eliminations.c (spill_pseudos): Ditto.
348 (process_insn_for_elimination): Ditto.
349 * lra-lives.c (reg_early_clobber_p): Use the new macros.
350 * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
351 LRA_NON_CLOBBERED_ALT.
353 2018-02-22 Martin Sebor <msebor@redhat.com>
355 PR tree-optimization/84480
356 * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
357 to maybe_diag_stxncpy_trunc. Call it.
358 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
359 from gimple_fold_builtin_strcpy. Print inlining stack.
360 (handle_builtin_stxncpy): Print inlining stack.
361 * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
363 2018-02-22 H.J. Lu <hongjiu.lu@intel.com>
366 * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
367 error when -mindirect-branch=thunk-extern, -fcf-protection=branch
368 and -fcheck-pointer-bounds are used together.
369 (indirect_thunk_prefix): New enum.
370 (indirect_thunk_need_prefix): New function.
371 (indirect_thunk_name): Replace need_bnd_p with need_prefix. Use
372 "_nt" instead of "_bnd" for NOTRACK prefix.
373 (output_indirect_thunk): Replace need_bnd_p with need_prefix.
374 (output_indirect_thunk_function): Likewise.
376 (ix86_code_end): Update output_indirect_thunk_function calls.
377 (ix86_output_indirect_branch_via_reg): Replace
378 ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
379 (ix86_output_indirect_branch_via_push): Likewise.
380 (ix86_output_function_return): Likewise.
381 * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
382 incompatible with -fcf-protection=branch and
383 -fcheck-pointer-bounds.
385 2018-02-22 Steve Ellcey <sellcey@cavium.com>
388 * config/aarch64/aarch64.c (aarch64_print_address_internal):
389 Change gcc_assert call to output_operand_lossage.
391 2018-02-22 Steve Ellcey <sellcey@cavium.com>
393 * doc/extend.texi (__builtin_extend_pointer): Document builtin.
395 2018-02-22 DJ Delorie <dj@redhat.com>
396 Sebastian Perta <sebastian.perta@renesas.com>
397 Oleg Endo <olegendo@gcc.gnu.org>
399 * config/rx/rx.c (rx_rtx_costs): New function.
400 (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
402 2018-02-22 Thomas Preud'homme <thomas.preudhomme@arm.com>
404 * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
406 2018-02-22 Martin Liska <mliska@suse.cz>
409 * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
410 Add "native" as a possible value.
412 2018-02-22 Martin Liska <mliska@suse.cz>
415 * config/i386/i386.c (ix86_option_override_internal):
416 Add "native" as a possible value for -march and -mtune.
418 2018-02-22 Jakub Jelinek <jakub@redhat.com>
421 * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
422 to all type variants.
424 PR tree-optimization/84503
425 * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
426 width as info->bitpos + info->bitsize - start.
427 (merged_store_group::merge_overlapping): Simplify width computation.
428 (check_no_overlap): New function.
429 (imm_store_chain_info::try_coalesce_bswap): Compute expected
430 start + width and last_order of the group, fail if check_no_overlap
432 (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
433 to group if check_no_overlap fails.
435 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
437 * config/rs6000/altivec.md: Delete contraint arguments to
438 define_expand, define_split, and define_peephole2, and in
439 define_insn_and_split if always unused.
440 * config/rs6000/darwin.md: Ditto.
441 * config/rs6000/dfp.md: Ditto.
442 * config/rs6000/rs6000.md: Ditto.
443 * config/rs6000/sync.md: Ditto.
444 * config/rs6000/vector.md: Ditto.
445 * config/rs6000/vsx.md: Ditto.
447 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
449 * config/rs6000/altivec.md: Write output control strings as braced
450 blocks instead of double-quoted strings.
451 * config/rs6000/darwin.md: Ditto.
452 * config/rs6000/rs6000.md: Ditto.
453 * config/rs6000/vector.md: Ditto.
454 * config/rs6000/vsx.md: Ditto.
456 2018-02-21 Jason Merrill <jason@redhat.com>
458 PR c++/84314 - ICE with templates and fastcall attribute.
459 * attribs.c (build_type_attribute_qual_variant): Remove assert.
461 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
463 * ipa-cp.c (determine_versionability): Fix comment typos.
465 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
468 * ipa-cp.c (determine_versionability): Do not version functions caling
471 2018-02-21 Martin Liska <mliska@suse.cz>
474 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
475 Add "native" as a possible value.
476 * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT): Define
477 the macro when native cpu detection is available.
479 2018-02-21 Martin Liska <mliska@suse.cz>
482 * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
483 Add "native" as a possible value.
484 * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
485 when native cpu detection is available.
487 2018-02-21 Jakub Jelinek <jakub@redhat.com>
488 Martin Sebor <msebor@redhat.com>
490 PR tree-optimization/84478
491 * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
493 * gimple-fold.c (get_range_strlen): Make minlen const and assume it
494 can't be NULL. Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
495 support which is conservatively correct, for 2 only stay conservative
496 for maxlen. Formatting and comment capitalization fixes. Add STRICT
497 argument to the 2 argument get_range_strlen, adjust 6 arg
498 get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
500 (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
501 (gimple_fold_builtin_strlen): Pass true as last argument to
504 2018-02-20 Martin Sebor <msebor@redhat.com>
507 * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
508 (builtin_memref::set_base_and_offset): Same. Handle inner references.
509 (builtin_memref::builtin_memref): Factor out parts into
510 set_base_and_offset and call it.
512 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
515 * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
516 is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
517 greater precision. If to_mode is a MODE_PARTIAL_INT, stop the
518 search at the associated MODE_INT.
520 2018-02-20 Jeff Law <law@redhat.com>
523 PR tree-optimization/81592
525 * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
526 for range data rather than using global data.
527 * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
528 range data rather than using global data.
529 * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
530 pass it to children as needed.
531 (struct directive::fmtresult): Similarly.
532 (struct directive::set_width): Similarly.
533 (struct directive::set_precision): Similarly.
534 (format_integer, format_directive, parse_directive): Similarly.
535 (format_none): Accept unnamed vr_values parameter.
536 (format_percent, format_floating, format_character): Similarly.
537 (format_string, format_plain): Similarly.
538 * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
539 the EVRP range analyzer for range data rather than using global data.
540 * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
541 gimple-ssa-evrp-analyze.h
542 (class sprintf_dom_walker): Add after_dom_children member function.
543 Add evrp_range_analyzer member.
544 (sprintf_dom_walker::before_dom_children): Call into the EVRP
545 range analyzer as needed.
546 (sprintf_dom_walker::after_dom_children): New member function.
547 * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
549 (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
550 (evrp_range_analyzer::pop_to_marker): Likewise.
552 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
554 PR tree-optimization/84419
555 * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
556 with the required type if its current type is compatible but
559 2018-02-20 Jakub Jelinek <jakub@redhat.com>
562 * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
565 2018-02-20 Martin Liska <mliska@suse.cz>
568 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
569 possible values if we don't have a hint.
571 2018-02-20 Martin Liska <mliska@suse.cz>
575 * final.c (shorten_branches): Build align_tab array with one
577 * opts.c (finish_options): Add alignment option limit check.
578 (MAX_CODE_ALIGN): Likewise.
579 (MAX_CODE_ALIGN_VALUE): Likewise.
580 * doc/invoke.texi: Document maximum allowed option value for
581 all -falign-* options.
583 2018-02-19 Jakub Jelinek <jakub@redhat.com>
586 * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
587 * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
588 * var-tracking.c (emit_note_insn_var_location): Remove all references
589 to NOTE_INSN_CALL_ARG_LOCATION.
590 (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
591 the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
592 Use copy_rtx_if_shared.
593 * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
594 NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
595 (dwarf2out_var_location): Remove handling of
596 NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
598 * final.c (final_scan_insn): Remove all references to
599 NOTE_INSN_CALL_ARG_LOCATION.
600 (rest_of_clean_state): Likewise. Remove REG_CALL_ARG_LOCATION notes
601 before dumping final insns.
602 * except.c (emit_note_eh_region_end): Remove all references to
603 NOTE_INSN_CALL_ARG_LOCATION.
604 * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
605 * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
606 * config/arc/arc.c (hwloop_optimize): Likewise.
607 * config/arm/arm.c (create_fix_barrier): Likewise.
608 * config/s390/s390.c (s390_chunkify_start): Likewise.
609 * config/sh/sh.c (find_barrier): Likewise.
610 * config/i386/i386.c (rest_of_insert_endbranch,
611 ix86_seh_fixup_eh_fallthru): Likewise.
612 * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
613 * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
614 * config/frv/frv.c (frv_function_prologue): Likewise.
615 * emit-rtl.c (try_split): Likewise. Copy over REG_CALL_ARG_LOCATION
617 (note_outside_basic_block_p): Remove all references to
618 NOTE_INSN_CALL_ARG_LOCATION.
619 * gengtype.c (adjust_field_rtx_def): Likewise.
620 * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
622 * jump.c (cleanup_barriers, delete_related_insns): Likewise.
623 * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
626 * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
629 PR tree-optimization/84452
630 * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
631 expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
634 2018-02-19 Martin Liska <mliska@suse.cz>
637 * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
639 2018-02-19 Martin Liska <mliska@suse.cz>
640 Richard Sandiford <richard.sandiford@linaro.org>
642 PR tree-optimization/82491
643 * gimple-fold.c (get_base_constructor): Make earlier bail out
646 2018-02-19 Carl Love <cel@us.ibm.com>
648 * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
649 BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
651 * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
654 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
656 * config/rl78/rl78.md (movdf): New define expand.
658 2018-02-19 Martin Liska <mliska@suse.cz>
661 * doc/invoke.texi: Fix typo.
662 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
664 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
666 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
667 handle rs6000_single_float and rs6000_double_float specially for
670 2018-02-16 Jeff Law <law@redhat.com>
672 * config/rx/rx.c (add_pop_cfi_notes): New function.;
675 2018-02-16 Jakub Jelinek <jakub@redhat.com>
678 * ipa-inline.c (inline_small_functions): Fix a typo.
680 2018-02-16 Nathan Sidwell <nathan@acm.org>
682 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
684 2018-02-16 Carl Love <cel@us.ibm.com>
686 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
687 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
688 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
689 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
690 expansion to P8V_BUILTIN_VEC_FLOAT2.
692 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
694 PR rtl-optimization/70023
695 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
696 src_regno into account.
698 2018-02-16 Carl Love <cel@us.ibm.com>
700 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
701 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
702 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
703 * config/rs6000/rs6000.c: Remove case statements for
704 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
705 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
706 and P9V_BUILTIN_VEC_VINSERT4B.
707 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
708 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
709 * config/rs6000/vsx.md:
710 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
713 2018-02-16 Carl Love <cel@us.ibm.com>
715 * config/rs6000/altivec.h: Add builtin names vec_extract4b
717 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
719 * config/rs6000/rs6000-c.c: Add the definitions for
720 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
721 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
722 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
723 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
724 definition for insert4b and define insn *insert3b_internal.
725 * doc/extend.texi: Add documentation for vec_extract4b.
727 2018-02-16 Nathan Sidwell <nathan@acm.org>
729 * doc/extend.texi (Backwards Compatibility): Mention friend
730 injection. Note for-scope is deprecated.
731 * doc/invoke.texi (-ffriend-injection): Deprecate.
733 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
735 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
736 that moved to I2, also allow destinations that are a paradoxical
737 subreg (instead of a normal reg).
739 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
742 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
745 2018-02-16 Richard Biener <rguenther@suse.de>
747 PR tree-optimization/84037
748 PR tree-optimization/84016
750 * config/i386/i386.c (ix86_builtin_vectorization_cost):
751 Adjust vec_construct for the fact we need additional higher latency
752 128bit inserts for AVX256 and AVX512 vector builds.
753 (ix86_add_stmt_cost): Scale vector construction cost for
756 2018-02-16 Richard Biener <rguenther@suse.de>
758 PR tree-optimization/84417
759 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
760 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
761 (non_rewritable_lvalue_p): Likewise, use poly-ints.
763 2018-02-16 Martin Liska <mliska@suse.cz>
766 * internal-fn.def (ASAN_CHECK): Set proper flags.
767 (ASAN_MARK): Likewise.
769 2018-02-16 Julia Koval <julia.koval@intel.com>
771 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
774 2018-02-16 Jakub Jelinek <jakub@redhat.com>
777 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
778 Use ++iter rather than iter++ for std::list iterators.
779 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
780 defer deleting them until all nodes in the forest are processed. Do
781 free even leaf nodes. Change to_process into auto_vec.
784 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
785 * vec.h (vec_default_construct): Use memset instead of placement new
786 if BROKEN_VALUE_INITIALIZATION is defined.
787 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
788 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
791 PR rtl-optimization/83723
792 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
793 * lra.c (lra_substitute_pseudo): Likewise. If true, use
794 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
796 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
798 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
800 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
802 PR rtl-optimization/81443
803 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
804 from inner REGs to paradoxical SUBREGs.
806 2018-02-16 Richard Biener <rguenther@suse.de>
808 PR tree-optimization/84399
809 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
810 For operands we can analyze at their definition make sure we can
811 analyze them at each use as well.
813 2018-02-16 Richard Biener <rguenther@suse.de>
815 PR tree-optimization/84190
816 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
817 volatile accesses if the decl isn't volatile.
819 2018-02-15 Jason Merrill <jason@redhat.com>
821 PR c++/84314 - ICE with templates and fastcall attribute.
822 * attribs.c (build_type_attribute_qual_variant): Don't clobber
823 TYPE_CANONICAL on an existing type.
825 2018-02-15 Jakub Jelinek <jakub@redhat.com>
827 PR tree-optimization/84383
828 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
829 dstoff nor call operand_equal_p if dstbase is NULL.
831 PR tree-optimization/84334
832 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
833 also a CONSTANT_CLASS_P, punt.
835 2018-02-14 Jim Wilson <jimw@sifive.com>
837 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
838 first SMALL_OPERAND check. New local min_second_step. Move assert
839 to where locals are set. Add TARGET_RVC support.
840 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
842 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
844 * doc/invoke.texi: Correct -Wformat-overflow code sample.
846 2018-02-14 Martin Sebor <msebor@redhat.com>
848 PR tree-optimization/83698
849 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
850 arrays constrain the offset range to their bounds.
851 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
852 (builtin_access::overlap): Avoid setting the size of overlap if it's
854 (maybe_diag_overlap): Also consider arrays when deciding what values
855 of offsets to include in diagnostics.
857 2018-02-14 Martin Sebor <msebor@redhat.com>
860 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
861 that correspond to the kind of a declaration.
863 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
866 * config/pa/pa.md: Load address of PIC label using the linkage table
867 if the label is nonlocal.
869 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
871 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
872 warning message if user requests -maltivec=be.
873 * doc/invoke.texi: Document deprecation of -maltivec=be.
875 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
878 * config/rs6000/rs6000-c.c: Update definitions for
879 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
880 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
882 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
885 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
886 add _get_ssp intrinsics. Remove argument from
887 __builtin_ia32_rdssp[d|q].
888 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
889 * config/i386/i386-builtin.def: Remove argument from
890 __builtin_ia32_rdssp[d|q].
891 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
892 ix86_expand_special_args_builtin for _rdssp[d|q].
893 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
894 Clear register before usage.
895 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
896 Add documentation for new _get_ssp and _inc_ssp intrinsics.
898 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
900 PR tree-optimization/84357
901 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
902 operand 1 of an ARRAY_REF too.
904 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
907 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
908 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
910 (set_of_reg): New struct.
911 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
912 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
913 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
915 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
916 Split into bitclr, bitset, bitinvert patterns if appropriate.
917 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
918 use rx_fuse_in_memory_bitop.
919 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
920 to named insn, correct maximum insn length.
922 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
925 * machmode.def: Define a complex mode for PARTIAL_INT.
926 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
928 * doc/rtl.texi: Document CSPImode.
929 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
931 (msp430_hard_regno_nregs_with_padding): Likewise.
933 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
936 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
938 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
940 PR rtl-optimization/84169
941 * combine.c (try_combine): New variable split_i2i3. Set it to true if
942 we generated a parallel as new i3 and we split that to new i2 and i3
943 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
944 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
945 those to i2, not i1. Partially rewrite this scan code.
947 2018-02-13 Jakub Jelinek <jakub@redhat.com>
950 * stor-layout.c (place_field): For variable length fields, adjust
951 offset_align afterwards not just based on the field's alignment,
952 but also on the size.
955 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
956 of exps and logs in the use_exp2 case.
958 2018-02-13 Jeff Law <law@redhat.com>
960 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
963 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
966 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
970 * common.opt (gas-loc-support, gas-locview-support): New.
971 (ginline-points, ginternal-reset-location-views): New.
972 * doc/invoke.texi: Document them. Use @itemx where intended.
973 (gvariable-location-views): Adjust.
974 * target.def (reset_location_view): New.
975 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
976 (TARGET_RESET_LOCATION_VIEW): New.
977 * doc/tm.texi: Rebuilt.
978 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
979 (dwarf2out_default_as_locview_support): New.
980 (output_asm_line_debug_info): Use option variables.
981 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
982 (output_loc_list): Likewise.
983 (add_high_low_attributes): Check option variables.
984 Don't output entry view attribute in strict mode.
985 (gen_inlined_subroutine_die): Check option variables.
986 (dwarf2out_inline_entry): Likewise.
987 (init_sections_and_labels): Likewise.
988 (dwarf2out_early_finish): Likewise.
989 (maybe_reset_location_view): New, from...
990 (dwarf2out_var_location): ... here. Call it.
991 * debug.h (dwarf2out_default_as_loc_support): Declare.
992 (dwarf2out_default_as_locview_support): Declare.
993 * hooks.c (hook_int_rtx_insn_0): New.
994 * hooks.h (hook_int_rtx_insn_0): Declare.
995 * toplev.c (process_options): Take -gas-loc-support and
996 -gas-locview-support from dwarf2out. Enable
997 -gvariable-location-views by default only with locview
998 assembler support. Enable -ginternal-reset-location-views by
999 default only if the target defines the corresponding hook.
1000 Enable -ginline-points by default if location views are
1001 enabled; force it disabled if statement frontiers are
1003 * tree-inline.c (expand_call_inline): Check option variables.
1004 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
1006 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
1008 PR tree-optimization/84321
1009 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
1010 handling. Also check whether the anti-range contains any values
1011 that satisfy the mask; switch to a VR_RANGE if not.
1013 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
1016 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
1018 2018-02-13 Martin Jambor <mjambor@suse.cz>
1021 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
1022 of call statements, also set location of a load to a temporary.
1024 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
1026 * config/rl78/rl78.c (add_vector_labels): New function.
1027 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
1028 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
1029 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
1030 which checks that no arguments are passed.
1031 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
1032 * doc/extend.texi: Documentation for the new attribute.
1034 2018-02-13 Andreas Schwab <schwab@suse.de>
1036 * config/riscv/linux.h (CPP_SPEC): Define.
1038 2018-02-13 Jakub Jelinek <jakub@redhat.com>
1041 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
1042 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
1043 OPTION_MASK_ISA_AES as first argument to def_builtin_const
1044 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
1045 instead of OPTION_MASK_ISA_PCLMUL as first argument to
1046 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
1047 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
1048 temporarily for AES and PCLMUL builtins.
1050 PR tree-optimization/84339
1051 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
1052 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
1056 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
1057 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
1058 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
1060 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
1062 * omp-simd-clone.h: New file.
1063 * omp-simd-clone.c: Include omp-simd-clone.h.
1064 (expand_simd_clones): No longer static.
1065 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
1066 cgraph.h and omp-simd-clone.h.
1067 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
1068 (vect_recog_widen_shift_pattern): Formatting fix.
1069 (vect_pattern_recog_1): Don't check optab for calls.
1072 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
1073 operands[2] into a REG before using gen_lowpart on it.
1075 2018-02-12 Jeff Law <law@redhat.com>
1078 * config/sh/sh.c (find_barrier): Consider a sibling call
1081 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
1082 successfully back substituting a reg.
1084 2018-02-12 Richard Biener <rguenther@suse.de>
1086 PR tree-optimization/84037
1087 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
1088 parameter, move visited init to caller.
1089 (vect_slp_analyze_operations): Separate cost from validity
1090 check, initialize visited once for all instances.
1091 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
1093 * tree-vect-stmts.c (vect_model_simple_cost): Make early
1095 (vect_model_promotion_demotion_cost): Likewise.
1096 (vectorizable_bswap): Guard cost modeling with !slp_node
1097 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
1099 (vectorizable_call): Likewise.
1100 (vectorizable_conversion): Likewise.
1101 (vectorizable_assignment): Likewise.
1102 (vectorizable_shift): Likewise.
1103 (vectorizable_operation): Likewise.
1104 (vectorizable_store): Likewise.
1105 (vectorizable_load): Likewise.
1106 (vectorizable_condition): Likewise.
1107 (vectorizable_comparison): Likewise.
1109 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
1112 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
1113 (ASAN_MARK): Fix fnspec to account for return value, change pointer
1114 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
1116 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1119 * params.def (inline-min-speedup): Increase from 8 to 15.
1120 (max-inline-insns-auto): Decrease from 40 to 30.
1121 * ipa-split.c (consider_split): Add some buffer for function to
1122 be considered inlining candidate.
1123 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
1126 2018-02-12 Richard Biener <rguenther@suse.de>
1128 PR tree-optimization/84037
1129 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
1130 matched stmts if we cannot swap the non-matched ones.
1132 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
1134 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
1135 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
1136 _mm_maskz_scalef_round_ss): New intrinsics.
1137 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
1138 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
1139 __builtin_ia32_scalefss_round): Remove.
1140 (__builtin_ia32_scalefsd_mask_round,
1141 __builtin_ia32_scalefss_mask_round): New intrinsics.
1142 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
1143 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
1144 ((match_operand:VF_128 2 "<round_nimm_predicate>"
1145 "<round_constraint>")): Changed to ...
1146 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
1147 "<round_scalar_constraint>")): ... this.
1148 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
1149 %0, %1, %2<round_op3>}"): Changed to ...
1150 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
1151 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
1152 %2<round_scalar_mask_op3>}"): ... this.
1153 * config/i386/subst.md (round_scalar_nimm_predicate): New.
1155 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
1157 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
1158 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
1159 (_mm_maskz_sqrt_round_ss): New intrinsics.
1160 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
1161 (__builtin_ia32_sqrtsd_mask_round)
1162 (__builtin_ia32_sqrtss_mask_round): New builtins.
1163 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
1164 (__builtin_ia32_sqrtss_round): Remove.
1165 (__builtin_ia32_sqrtsd_mask_round)
1166 (__builtin_ia32_sqrtss_mask_round): New builtins.
1167 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
1168 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
1169 ((match_operand:VF_128 1 "vector_operand"
1170 "xBm,<round_constraint>")): Changed to ...
1171 ((match_operand:VF_128 1 "vector_operand"
1172 "xBm,<round_scalar_constraint>")): ... this.
1173 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
1174 %0, %2, %<iptr>1<round_op3>}): Changed to ...
1175 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
1176 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
1177 %<iptr>1<round_scalar_mask_op3>}): ... this.
1178 ((set_attr "prefix" "<round_prefix>")): Changed to ...
1179 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
1181 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
1184 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
1185 Cast vec_cmpeq result to correct type.
1186 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
1187 Cast vec_cmpgt result to correct type.
1189 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
1191 * final.c (final_scan_insn_1): Renamed from...
1192 (final_scan_insn): ... this. New wrapper, to recover
1193 seen from the outermost call in recursive ones.
1194 * config/sparc/sparc.c (output_return): Drop seen from call.
1195 (output_sibcall): Likewise.
1196 * config/visium/visium.c (output_branch): Likewise.
1198 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
1200 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
1203 2018-02-10 Alan Modra <amodra@gmail.com>
1206 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
1207 Specify LR as an input.
1209 2018-02-10 Jakub Jelinek <jakub@redhat.com>
1212 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
1213 remove_member_access_dummy_vars): New functions.
1214 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
1215 lower_omp_1, execute_lower_omp): Use them.
1217 PR rtl-optimization/84308
1218 * shrink-wrap.c (spread_components): Release todo vector.
1220 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
1222 PR rtl-optimization/57193
1223 * ira-color.c (struct allocno_color_data): Add member
1224 conflict_allocno_hard_prefs.
1225 (update_conflict_allocno_hard_prefs): New.
1226 (bucket_allocno_compare_func): Add a preference based on
1227 conflict_allocno_hard_prefs.
1228 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
1229 (color_allocnos): Remove a dead code. Initiate
1230 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
1232 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1235 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
1236 constraint from =wa to wa. Avoid a subreg on the output operand,
1237 instead use a pseudo and subreg it in a move.
1238 (p9_xxbrd_<mode>): Changed to ...
1239 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
1240 (p9_xxbrd_v2df): New expander.
1241 (p9_xxbrw_<mode>): Changed to ...
1242 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
1243 (p9_xxbrw_v4sf): New expander.
1245 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1247 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
1249 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
1252 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
1253 multiply in 32-bit mode.
1254 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
1255 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
1258 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1260 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
1261 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
1262 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
1263 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
1265 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
1268 * dwarf2out.c (is_trivial_indirect_ref): New function.
1269 (dwarf2out_late_global_decl): Do not generate a location
1270 attribute for variables that have a non-trivial DECL_VALUE_EXPR
1271 and that are not defined in the current unit.
1273 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1275 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
1276 instead of a libcall for UNORDERED.
1278 2018-02-09 Tamar Christina <tamar.christina@arm.com>
1281 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
1282 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
1284 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1287 * config/s390/s390.c (s390_set_current_function): Invoke
1288 s390_indirect_branch_settings also if fndecl didn't change.
1290 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1292 * config/rs6000/rs6000.md (blockage): Set length to zero.
1294 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1296 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
1298 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1301 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
1302 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
1306 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
1307 PARALLEL incoming that failed vt_get_decl_and_offset check.
1310 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
1311 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
1313 (get_variable_section): For decls in named .bss* sections pass true as
1314 second argument to bss_initializer_p.
1316 2018-02-09 Marek Polacek <polacek@redhat.com>
1317 Jakub Jelinek <jakub@redhat.com>
1320 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
1321 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
1322 Sync some changes from cxx_fold_indirect_ref.
1324 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1326 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
1328 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
1329 (BLOCK_INLINE_ENTRY_LABEL): New.
1330 (dwarf2out_var_location): Disregard inline entry markers.
1331 (inline_entry_data): New struct.
1332 (inline_entry_data_hasher): New hashtable type.
1333 (inline_entry_data_hasher::hash): New.
1334 (inline_entry_data_hasher::equal): New.
1335 (inline_entry_data_table): New variable.
1336 (add_high_low_attributes): Add DW_AT_entry_pc and
1337 DW_AT_GNU_entry_view attributes if a pending entry is found
1338 in inline_entry_data_table. Add old entry_pc attribute only
1339 if debug nonbinding markers are disabled.
1340 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
1341 markers are enabled.
1342 (block_within_block_p, dwarf2out_inline_entry): New.
1343 (dwarf2out_finish): Check that no entries remained in
1344 inline_entry_data_table.
1345 * final.c (reemit_insn_block_notes): Handle inline entry notes.
1346 (final_scan_insn, notice_source_line): Likewise.
1347 (rest_of_clean_state): Skip inline entry markers.
1348 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
1350 * gimple.c (gimple_build_debug_inline_entry): New.
1351 * gimple.h (enum gimple_debug_subcode): Add
1352 GIMPLE_DEBUG_INLINE_ENTRY.
1353 (gimple_build_debug_inline_entry): Declare.
1354 (gimple_debug_inline_entry_p): New.
1355 (gimple_debug_nonbind_marker_p): Adjust.
1356 * insn-notes.def (INLINE_ENTRY): New.
1357 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
1358 inline entry marker notes.
1359 (print_insn): Likewise.
1360 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
1361 (INSN_DEBUG_MARKER_KIND): Likewise.
1362 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
1363 * tree-inline.c (expand_call_inline): Build and insert
1364 debug_inline_entry stmt.
1365 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
1366 inline entry blocks early, if nonbind markers are enabled.
1367 (dump_scope_block): Dump fragment info.
1368 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
1369 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
1370 (gimple_build_debug_inline_entry): New.
1371 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
1372 Enable/disable inline entry points too.
1373 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
1374 (DEBUG_INSN): Describe inline entry markers.
1376 * common.opt (gvariable-location-views): New.
1377 (gvariable-location-views=incompat5): New.
1378 * config.in: Rebuilt.
1379 * configure: Rebuilt.
1380 * configure.ac: Test assembler for view support.
1381 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
1382 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
1383 * dwarf2out.c (var_loc_view): New typedef.
1384 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
1385 (dwarf2out_locviews_in_attribute): New.
1386 (dwarf2out_locviews_in_loclist): New.
1387 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
1388 (enum dw_line_info_opcode): Add LI_adv_address.
1389 (struct dw_line_info_table): Add view.
1390 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
1391 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
1392 (zero_view_p): New variable.
1393 (ZERO_VIEW_P): New macro.
1394 (output_asm_line_debug_info): New.
1395 (struct var_loc_node): Add view.
1396 (add_AT_view_list, AT_loc_list): New.
1397 (add_var_loc_to_decl): Add view param. Test it against last.
1398 (new_loc_list): Add view params. Record them.
1399 (AT_loc_list_ptr): Handle loc and view lists.
1400 (view_list_to_loc_list_val_node): New.
1401 (print_dw_val): Handle dw_val_class_view_list.
1402 (size_of_die): Likewise.
1403 (value_format): Likewise.
1404 (loc_list_has_views): New.
1405 (gen_llsym): Set vl_symbol too.
1406 (maybe_gen_llsym, skip_loc_list_entry): New.
1407 (dwarf2out_maybe_output_loclist_view_pair): New.
1408 (output_loc_list): Output view list or entries too.
1409 (output_view_list_offset): New.
1410 (output_die): Handle dw_val_class_view_list.
1411 (output_dwarf_version): New.
1412 (output_compilation_unit_header): Use it.
1413 (output_skeleton_debug_sections): Likewise.
1414 (output_rnglists, output_line_info): Likewise.
1415 (output_pubnames, output_aranges): Update version comments.
1416 (output_one_line_info_table): Output view numbers in asm comments.
1417 (dw_loc_list): Determine current endview, pass it to new_loc_list.
1418 Call maybe_gen_llsym.
1419 (loc_list_from_tree_1): Adjust.
1420 (add_AT_location_description): Create view list attribute if
1421 needed, check it's absent otherwise.
1422 (convert_cfa_to_fb_loc_list): Adjust.
1423 (maybe_emit_file): Call output_asm_line_debug_info for test.
1424 (dwarf2out_var_location): Reset views as needed. Precompute
1425 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
1426 attribute. Set view.
1427 (new_line_info_table): Reset next view.
1428 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
1429 (dwarf2out_source_line): Likewise. Output view resets and labels to
1430 the assembler, or select appropriate line info opcodes.
1431 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
1432 (optimize_string_length): Catch it. Adjust.
1433 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
1434 dw_val_class_view_list, and remove it if no longer needed.
1435 (hash_loc_list): Hash view numbers.
1436 (loc_list_hasher::equal): Compare them.
1437 (optimize_location_lists): Check whether a view list symbol is
1438 needed, and whether the locview attribute is present, and
1439 whether they match. Remove the locview attribute if no longer
1441 (index_location_lists): Call skip_loc_list_entry for test.
1442 (dwarf2out_finish): Call output_asm_line_debug_info for test.
1443 Use output_dwarf_version.
1444 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
1445 (struct dw_val_node): Add val_view_list.
1446 * final.c (SEEN_NEXT_VIEW): New.
1447 (set_next_view_needed): New.
1448 (clear_next_view_needed): New.
1449 (maybe_output_next_view): New.
1450 (final_start_function): Rename to...
1451 (final_start_function_1): ... this. Take pointer to FIRST,
1452 add SEEN parameter. Emit param bindings in the initial view.
1453 (final_start_function): Reintroduce SEEN-less interface.
1454 (final): Rename to...
1455 (final_1): ... this. Take SEEN parameter. Output final pending
1456 next view at the end.
1457 (final): Reintroduce seen-less interface.
1458 (final_scan_insn): Output pending next view before switching
1459 sections or ending a block. Mark the next view as needed when
1460 outputting variable locations. Notify debug backend of section
1461 changes, and of location view changes.
1462 (rest_of_handle_final): Adjust.
1463 * toplev.c (process_options): Autodetect value for debug variable
1464 location views option. Warn on incompat5 without -gdwarf-5.
1465 * doc/invoke.texi (gvariable-location-views): New.
1466 (gvariable-location-views=incompat5): New.
1467 (gno-variable-location-views): New.
1469 2018-02-08 David Malcolm <dmalcolm@redhat.com>
1471 PR tree-optimization/84136
1472 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
1473 that the result of find_edge is non-NULL.
1475 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1478 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
1479 storing integer register in SImode. Fix cost of 256 and 512
1480 byte aligned SSE register store.
1482 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1484 * config/i386/i386.c (ix86_multiplication_cost): Fix
1485 multiplication cost for TARGET_AVX512DQ.
1487 2018-02-08 Marek Polacek <polacek@redhat.com>
1489 PR tree-optimization/84238
1490 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
1493 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1495 PR tree-optimization/84265
1496 * tree-vect-stmts.c (vectorizable_store): Don't treat
1497 VMAT_CONTIGUOUS accesses as grouped.
1498 (vectorizable_load): Likewise.
1500 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1502 PR tree-optimization/81635
1503 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
1504 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
1505 (test_round_for_mask): New functions.
1506 (wide_int_cc_tests): Call test_round_for_mask.
1507 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
1508 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
1509 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
1510 range returned by get_range_info.
1512 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1515 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
1516 * symtab.c: Include builtins.h
1517 (symtab_node::output_to_lto_symbol_table_p): Move here
1518 from lto-streamer-out.c:output_symbol_p.
1519 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
1520 (output_symbol_p): Move all logic to symtab.c
1521 (produce_symtab): Update.
1523 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1525 * config/s390/s390-opts.h (enum indirect_branch): Define.
1526 * config/s390/s390-protos.h (s390_return_addr_from_memory)
1527 (s390_indirect_branch_via_thunk)
1528 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
1529 (enum s390_indirect_branch_type): Define.
1530 * config/s390/s390.c (struct s390_frame_layout, struct
1531 machine_function): Remove.
1532 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
1533 (indirect_branch_table_label_no, indirect_branch_table_name):
1535 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
1536 (enum s390_indirect_branch_option): Define.
1537 (s390_return_addr_from_memory): New function.
1538 (s390_handle_string_attribute): New function.
1539 (s390_attribute_table): Add new attribute handler.
1540 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
1541 (s390_indirect_branch_via_thunk): New function.
1542 (s390_indirect_branch_via_inline_thunk): New function.
1543 (s390_function_ok_for_sibcall): When jumping via thunk disallow
1544 sibling call optimization for non z10 compiles.
1545 (s390_emit_call): Force indirect branch target to be a single
1546 register. Add r1 clobber for non-z10 compiles.
1547 (s390_emit_epilogue): Emit return jump via return_use expander.
1548 (s390_reorg): Handle JUMP_INSNs as execute targets.
1549 (s390_option_override_internal): Perform validity checks for the
1550 new command line options.
1551 (s390_indirect_branch_attrvalue): New function.
1552 (s390_indirect_branch_settings): New function.
1553 (s390_set_current_function): Invoke s390_indirect_branch_settings.
1554 (s390_output_indirect_thunk_function): New function.
1555 (s390_code_end): Implement target hook.
1556 (s390_case_values_threshold): Implement target hook.
1557 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
1559 * config/s390/s390.h (struct s390_frame_layout)
1560 (struct machine_function): Move here from s390.c.
1561 (TARGET_INDIRECT_BRANCH_NOBP_RET)
1562 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
1563 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
1564 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
1565 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
1566 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
1567 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
1568 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
1569 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
1570 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
1571 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
1572 (mnemonic attribute): Add values which aren't recognized
1574 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
1575 pattern for branch conversion. Fix mnemonic attribute.
1576 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
1577 indirect branch via thunk if requested.
1578 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
1579 ("*indirect_jump"): Disable for branch conversion using out of
1581 ("indirect_jump_via_thunk<mode>_z10")
1582 ("indirect_jump_via_thunk<mode>")
1583 ("indirect_jump_via_inlinethunk<mode>_z10")
1584 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
1585 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
1586 ("casesi_jump_via_inlinethunk<mode>_z10")
1587 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
1588 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
1589 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
1590 ("*indirect2_jump"): Disable for branch conversion.
1591 ("casesi_jump"): Turn into expander and expand patterns for branch
1593 ("return_use"): New expander.
1594 ("*return"): Emit return via thunk and rename it to ...
1595 ("*return<mode>"): ... this one.
1596 * config/s390/s390.opt: Add new options and and enum for the
1599 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1601 * lra-constraints.c (match_reload): Unconditionally use
1602 gen_lowpart_SUBREG, rather than selecting between that
1603 and equivalent gen_rtx_SUBREG code.
1605 2018-02-08 Richard Biener <rguenther@suse.de>
1607 PR tree-optimization/84233
1608 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
1609 changed flag instead of boguously re-using phi_inserted.
1611 2018-02-08 Martin Jambor <mjambor@suse.cz>
1613 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
1614 static local variables.
1616 2018-02-08 Richard Biener <rguenther@suse.de>
1618 PR tree-optimization/84278
1619 * tree-vect-stmts.c (vectorizable_store): When looking for
1620 smaller vector types to perform grouped strided loads/stores
1621 make sure the mode is supported by the target.
1622 (vectorizable_load): Likewise.
1624 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1626 * config/aarch64/aarch64.c (aarch64_components_for_bb):
1627 Increase LDP/STP opportunities by adding adjacent callee-saves.
1629 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1631 PR rtl-optimization/84068
1632 PR rtl-optimization/83459
1633 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
1635 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
1637 PR tree-optimization/84224
1638 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
1639 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
1642 2018-02-07 Iain Sandoe <iain@codesourcery.com>
1645 * config/rs6000/altivec.md (*restore_world): Remove LR use.
1646 * config/rs6000/predicates.md (restore_world_operation): Adjust op
1647 count, remove one USE.
1649 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
1651 * doc/install.texi (Configuration): Document the
1652 --with-long-double-format={ibm,ieee} PowerPC configuration
1656 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
1657 Convert from define_expand to be define_insn_and_split. Rework
1658 float/double/_Float128 conversions to QI/HI/SImode to work with
1659 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
1660 conversions to QI/HImode types did a store and then a load to
1661 truncate the value. For conversions to VSX registers, don't split
1662 the insn, instead emit the code directly. Use the code iterator
1663 any_fix to combine signed and unsigned conversions.
1664 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
1665 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
1666 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1667 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
1668 (fix_<mode>di2_hw): Likewise.
1669 (fixuns_<mode>di2_hw): Likewise.
1670 (fix_<mode>si2_hw): Likewise.
1671 (fixuns_<mode>si2_hw): Likewise.
1672 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
1673 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1674 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1675 fix<uns>_trunc<SFDF:mode>si2_p8.
1676 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1678 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1679 (fix<uns>_<mode>_mem): Likewise.
1680 (fctiw<u>z_<mode>_mem): Likewise.
1681 (fix<uns>_<mode>_mem): Likewise.
1682 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1683 the register allocator from doing a direct move to the GPRs to do
1684 a store, and instead use the ISA 3.0 store byte/half-word from
1685 vector register instruction. For IEEE 128-bit floating point,
1686 also optimize stores of 32-bit ints.
1687 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1689 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1691 * genextract.c (push_pathstr_operand): New function to support
1693 (walk_rtx): Call push_pathstr_operand.
1694 (print_path): Support [a-zA-Z].
1696 2018-02-07 Richard Biener <rguenther@suse.de>
1698 PR tree-optimization/84037
1699 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1700 (cse_and_gimplify_to_preheader): Declare.
1701 (vect_get_place_in_interleaving_chain): Likewise.
1702 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1704 (_loop_vec_info::~_loop_vec_info): Delete it.
1705 (cse_and_gimplify_to_preheader): New function.
1706 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1707 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1708 (vectorizable_load): Likewise. For grouped stores always base
1709 the IV on the first element.
1710 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1711 condition before gimplifying.
1713 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1715 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1716 *DIV_EXPR and *MOD_EXPR.
1718 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1721 * config/i386/i386.c (ix86_option_override_internal): Mask out
1722 the CF_SET bit when checking -fcf-protection.
1724 2018-02-07 Tom de Vries <tom@codesourcery.com>
1727 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1730 2018-02-07 Richard Biener <rguenther@suse.de>
1732 PR tree-optimization/84204
1733 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1736 PR tree-optimization/84205
1737 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1738 special-case isl_ast_op_zdiv_r.
1740 PR tree-optimization/84223
1741 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1742 Only add conditions from within the region.
1743 (gather_bbs::after_dom_children): Adjust.
1745 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1748 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1749 * config/avr/avr.md: Only post-reload split REG-REG moves if
1750 either register is REGERAL_REG_P.
1752 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1754 PR tree-optimization/84235
1755 * tree-ssa-scopedtables.c
1756 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1757 if the subtraction is performed in floating point type where NaNs are
1758 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1759 build 1. Formatting fix.
1761 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1764 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1765 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1766 and skip it regardless of bb boundaries. Use CALL_P macro,
1767 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1768 unnecessarily, formatting fix.
1770 2018-02-06 Michael Collison <michael.collison@arm.com>
1772 * config/arm/thumb2.md:
1773 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1774 (*thumb_mov_notscc): Ditto.
1776 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1779 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1780 unsigned_fix, not "s".
1782 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1784 * configure.ac (gcc_fn_eh_frame_ro): New function.
1785 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1786 correct .eh_frame permissions.
1787 * configure: Regenerate.
1789 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1791 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1794 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1796 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1797 Display warning message for -mno-speculate-indirect-jumps.
1799 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1801 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1803 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1805 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1807 PR tree-optimization/84225
1808 * tree-eh.c (find_trapping_overflow): Only call
1809 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1811 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1814 * config/i386/i386.c: Reimplement the check of possible options
1815 -mibt/-mshstk conbination. Change error messages.
1816 * doc/invoke.texi: Fix a typo: remove extra '='.
1818 2018-02-06 Marek Polacek <polacek@redhat.com>
1820 PR tree-optimization/84228
1821 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1823 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1826 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1827 emitted arch directives.
1828 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1829 __ARM_FEATURE_COPROC before changing architectures.
1831 2018-02-06 Richard Biener <rguenther@suse.de>
1833 * config/i386/i386.c (print_reg): Fix typo.
1834 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1836 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1838 * configure: Regenerate.
1840 2018-02-05 Martin Sebor <msebor@redhat.com>
1842 PR tree-optimization/83369
1843 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1846 2018-02-05 Martin Liska <mliska@suse.cz>
1848 * doc/invoke.texi: Cherry-pick upstream r323995.
1850 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1852 * ira.c (ira_init_register_move_cost): Adjust comment.
1854 2018-02-05 Martin Liska <mliska@suse.cz>
1856 PR gcov-profile/84137
1857 * doc/gcov.texi: Fix typo in documentation.
1859 2018-02-05 Martin Liska <mliska@suse.cz>
1861 PR gcov-profile/83879
1862 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1863 using dlopen functionality.
1865 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1867 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1868 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1869 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1870 _mm_maskz_range_round_ss): New intrinsics.
1871 (__builtin_ia32_rangesd128_round)
1872 (__builtin_ia32_rangess128_round): Remove.
1873 (__builtin_ia32_rangesd128_mask_round,
1874 __builtin_ia32_rangess128_mask_round): New builtins.
1875 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1876 __builtin_ia32_rangess128_round): Remove.
1877 (__builtin_ia32_rangesd128_mask_round,
1878 __builtin_ia32_rangess128_mask_round): New builtins.
1879 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1880 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1881 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1882 "<round_saeonly_constraint>")): Changed to ...
1883 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1884 "<round_saeonly_scalar_constraint>")): ... this.
1885 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1886 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1887 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1888 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1889 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1891 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1893 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1895 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1896 Remove all values except native, 8540 and 8548.
1898 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1900 * config/i386/i386.c (ix86_output_function_return): Pass
1901 INVALID_REGNUM, instead of -1, as invalid register number to
1902 indirect_thunk_name and output_indirect_thunk.
1904 2018-02-02 Julia Koval <julia.koval@intel.com>
1906 * config.gcc: Add -march=icelake.
1907 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1908 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1909 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1910 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1911 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1912 (processor_target_table): Add icelake.
1913 (ix86_option_override_internal): Handle new PTAs.
1914 (get_builtin_code_for_version): Handle icelake.
1915 (M_INTEL_COREI7_ICELAKE): New.
1916 (fold_builtin_cpu): Handle icelake.
1917 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1918 * doc/invoke.texi: Add -march=icelake.
1920 2018-02-02 Julia Koval <julia.koval@intel.com>
1922 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1923 to wide_int_bitmask.
1924 * wide-int-bitmask.h: New.
1926 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1929 * config/i386/i386.md: Replace Pmode with word_mode in
1930 builtin_setjmp_setup and builtin_longjmp to support x32.
1932 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1936 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1938 (rs6000_supported_cpu_names): New static variable.
1939 (linux_cpu_translation_table): Likewise.
1940 (elf_platform) <cpu>: Define new static variable and use it.
1941 Translate kernel AT_PLATFORM name to canonical name if needed.
1942 Error if platform name is unknown.
1944 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1947 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1949 2018-02-01 Jeff Law <law@redhat.com>
1952 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1953 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1954 the scratch if RELEASE_VIA_POP is false.
1955 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1956 If we have to save a temporary register, decrement SIZE appropriately.
1957 Pass new arguments to release_scratch_register_on_entry.
1958 (ix86_adjust_stack_and_probe): Likewise.
1959 (ix86_emit_probe_stack_range): Pass new arguments to
1960 release_scratch_register_on_entry.
1962 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1964 PR rtl-optimization/84157
1965 * combine.c (change_zero_ext): Use REG_P predicate in
1966 front of HARD_REGISTER_P predicate.
1968 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1970 * config/avr/avr.c (avr_option_override): Move disabling of
1971 -fdelete-null-pointer-checks to...
1972 * common/config/avr/avr-common.c (avr_option_optimization_table):
1975 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1977 PR tree-optimization/81635
1978 * tree-data-ref.c (split_constant_offset_1): For types that
1979 wrap on overflow, try to use range info to prove that wrapping
1982 2018-02-01 Renlin Li <renlin.li@arm.com>
1985 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1987 (aarch64_register_move_cost): Likewise.
1988 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1990 (REG_CLASS_NAMES): Likewise.
1991 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1992 TAILCALL_ADDR_REGS. Remove IP registers.
1993 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1995 2018-02-01 Richard Biener <rguenther@suse.de>
1997 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1998 for specifying RPO order and allow NULL for that.
1999 * domwalk.c (dom_walker::dom_walker): Likewise.
2000 (dom_walker::walk): Handle NULL RPO order.
2001 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
2003 (rewrite_update_dom_walker): Likewise.
2004 (mark_def_dom_walker): Likewise.
2006 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2008 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
2009 (aarch64_maybe_expand_sve_subreg_move): Declare.
2010 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
2011 * config/aarch64/predicates.md (aarch64_any_register_operand): New
2013 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
2014 that are semantically a reverse operation.
2015 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
2016 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
2017 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
2019 (aarch64_can_change_mode_class): For big-endian, forbid changes
2020 between two SVE modes if they have different element sizes.
2022 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2024 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
2025 the TImode handling for big-endian targets.
2027 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2029 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
2030 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
2032 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
2033 Remove BSWAP handing for big-endian targets and use the form of
2034 LD1RQ appropariate for the mode.
2036 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2038 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
2039 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
2042 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2045 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
2046 check for operands that need to go through aarch64_sve_reload_be.
2048 2018-02-01 Jakub Jelinek <jakub@redhat.com>
2050 PR tree-optimization/81661
2051 PR tree-optimization/84117
2052 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
2053 * tree-eh.c: Include gimplify.h.
2054 (find_trapping_overflow, replace_trapping_overflow,
2055 rewrite_to_non_trapping_overflow): New functions.
2056 * tree-vect-loop.c: Include tree-eh.h.
2057 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
2058 * tree-data-ref.c: Include tree-eh.h.
2059 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
2061 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
2063 PR rtl-optimization/84123
2064 * combine.c (change_zero_ext): Check if hard register satisfies
2065 can_change_dest_mode before calling gen_lowpart_SUBREG.
2067 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
2070 * ira.c (ira_init_register_move_cost): Remove assert.
2072 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2074 PR rtl-optimization/84071
2075 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
2076 * doc/tm.texi: Regenerate.
2078 2018-01-31 Richard Biener <rguenther@suse.de>
2080 PR tree-optimization/84132
2081 * tree-data-ref.c (analyze_miv_subscript): Properly
2082 check whether evolution_function_is_affine_multivariate_p
2083 before calling gcd_of_steps_may_divide_p.
2085 2018-01-31 Julia Koval <julia.koval@intel.com>
2088 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
2089 * config/i386/i386.md (rdpid_rex64) New.
2090 (rdpid): Make 32bit only.
2092 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
2095 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
2096 an IDENTIFIER_NODE for FUNCTION_TYPE's.
2098 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2101 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2103 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2105 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2107 PR rtl-optimization/84071
2108 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
2109 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
2111 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
2113 * config/arc/arc.c (arc_handle_aux_attribute): New function.
2114 (arc_attribute_table): Add 'aux' attribute.
2115 (arc_in_small_data_p): Consider aux like variables.
2116 (arc_is_aux_reg_p): New function.
2117 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
2118 (arc_get_aux_arg): New function.
2119 (prepare_move_operands): Handle aux-register access.
2120 (arc_handle_aux_attribute): New function.
2121 * doc/extend.texi (ARC Variable attributes): Add subsection.
2123 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
2125 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
2126 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
2127 (arc_attribute_table): Add 'uncached' attribute.
2128 (arc_print_operand): Print '.di' flag for uncached memory
2130 (arc_in_small_data_p): Do not consider for small data the uncached
2132 (arc_is_uncached_mem_p): New function.
2133 * config/arc/predicates.md (compact_store_memory_operand): Check
2134 for uncached memory accesses.
2135 (nonvol_nonimm_operand): Likewise.
2136 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
2138 2018-01-31 Jakub Jelinek <jakub@redhat.com>
2141 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
2142 falign-loops=): Add Optimization flag.
2144 2018-01-30 Jeff Law <law@redhat.com>
2147 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
2148 INT_REGISTERS_SAVED. Check it prior to calling
2149 get_scratch_register_on_entry.
2150 (ix86_adjust_stack_and_probe): Similarly.
2151 (ix86_emit_probe_stack_range): Similarly.
2152 (ix86_expand_prologue): Corresponding changes.
2154 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2157 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
2158 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
2160 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
2163 * lra-constraints.c (curr_insn_transform): Process AND in the
2166 2018-01-30 Jakub Jelinek <jakub@redhat.com>
2168 PR rtl-optimization/83986
2169 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
2170 dependence against last_pending_memory_flush in addition to
2173 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
2175 PR tree-optimization/81611
2176 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
2179 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2182 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
2185 2018-01-30 Richard Biener <rguenther@suse.de>
2186 Jakub Jelinek <jakub@redhat.com>
2188 PR tree-optimization/84111
2189 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
2190 inner loops added during recursion, as they don't have up-to-date
2193 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2196 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
2197 (can_inline_edge_by_limits_p): ... here.
2198 (can_early_inline_edge_p, check_callers,
2199 update_caller_keys, update_callee_keys, recursive_inlining,
2200 add_new_edges_to_heap, speculation_useful_p,
2201 inline_small_functions,
2202 inline_small_functions, flatten_function,
2203 inline_to_all_callers_1): Update.
2205 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2207 * profile-count.c (profile_count::combine_with_ipa_count): Handle
2210 2018-01-30 Richard Biener <rguenther@suse.de>
2212 PR tree-optimization/83008
2213 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
2214 invariant and constant vector uses in stmts when they need
2217 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2220 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
2221 * configure: Regenerate.
2223 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2225 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
2227 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
2228 Use gen_rtx_REG rather than gen_lowpart.
2230 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2232 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
2233 rather than 0 when creating partial subregs.
2235 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2237 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
2240 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
2243 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
2244 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
2245 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
2246 flags. This restores the settings used before the 2017-07-24.
2247 Turning off pre increment/decrement/modify allows IVOPTS to
2248 optimize DF/SF loops where the index is an int.
2250 2018-01-29 Richard Biener <rguenther@suse.de>
2251 Kelvin Nilsen <kelvin@gcc.gnu.org>
2254 * tree-vect-stmts.c (vectorizable_call): Don't call
2255 targetm.vectorize_builtin_md_vectorized_function if callee is
2258 2018-01-22 Carl Love <cel@us.ibm.com>
2260 * doc/extend.tex: Fix typo in second arg in
2261 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
2263 2018-01-29 Richard Biener <rguenther@suse.de>
2265 PR tree-optimization/84086
2266 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
2267 (flush_ssaname_freelist): When SSA names were released reset
2268 the SCEV hash table.
2270 2018-01-29 Richard Biener <rguenther@suse.de>
2272 PR tree-optimization/84057
2273 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
2274 removed paths when removing edges.
2276 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
2278 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
2279 -mfunction-return=@var{choice}.
2281 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
2284 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
2286 (layout::print_source_line): Likewise.
2287 (test_get_line_width_without_trailing_whitespace): Add test cases.
2289 2018-01-27 Jakub Jelinek <jakub@redhat.com>
2292 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
2295 2018-01-26 Jim Wilson <jimw@sifive.com>
2297 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
2299 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
2302 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2304 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
2305 and CMP + SUB-immediate -> SUBS.
2307 2018-01-26 Martin Sebor <msebor@redhat.com>
2309 PR tree-optimization/83896
2310 * tree-ssa-strlen.c (get_string_len): Rename...
2311 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
2312 Avoid assuming length is constant.
2313 (handle_char_store): Use HOST_WIDE_INT for string length.
2315 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
2318 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
2319 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
2321 2018-01-26 Richard Biener <rguenther@suse.de>
2323 PR rtl-optimization/84003
2324 * dse.c (record_store): Only record redundant stores when
2325 the earlier store aliases at least all accesses the later one does.
2327 2018-01-26 Jakub Jelinek <jakub@redhat.com>
2329 PR rtl-optimization/83985
2330 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
2331 REG_CFA_RESTORE insns.
2332 (delete_unmarked_insns): Don't ignore separate shrink wrapping
2333 REG_CFA_RESTORE insns here.
2336 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
2337 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
2339 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2341 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
2342 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
2343 (arc_init): Likewise.
2344 (arc_override_options): Likewise.
2345 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
2347 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
2349 * config/arc/arc.h (TARGET_DBNZ): Define.
2350 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
2351 properly set the tune attribute.
2352 (dbnz): Use TARGET_DBNZ guard.
2353 * config/arc/arc.opt (mtune): Add core3 option.
2355 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2357 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
2358 recognize new pic like addresses.
2359 (arc_delegitimize_address): Clean up.
2361 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2363 * config/arc/arc-arches.def: Option mrf16 valid for all
2365 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
2366 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
2367 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
2368 * config/arc/arc-tables.opt: Regenerate.
2369 * config/arc/arc.c (arc_conditional_register_usage): Handle
2370 reduced register file case.
2371 (arc_file_start): Set must have build attributes.
2372 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
2374 * config/arc/arc.opt (mrf16): Add new option.
2375 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
2376 * config/arc/genmultilib.awk: Handle new mrf16 option.
2377 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
2378 * config/arc/t-multilib: Regenerate.
2379 * doc/invoke.texi (ARC Options): Document mrf16 option.
2381 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2383 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
2384 * config/arc/arc.c (arc_handle_secure_attribute): New function.
2385 (arc_attribute_table): Add 'secure_call' attribute.
2386 (arc_print_operand): Print secure call operand.
2387 (arc_function_ok_for_sibcall): Don't optimize tail calls when
2389 (arc_is_secure_call_p): New function. * config/arc/arc.md
2390 (call_i): Add support for sjli instruction.
2391 (call_value_i): Likewise.
2392 * config/arc/constraints.md (Csc): New constraint.
2394 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2395 John Eric Martin <John.Martin@emmicro-us.com>
2397 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
2398 * config/arc/arc.c (_arc_jli_section): New struct.
2399 (arc_jli_section): New type.
2400 (rc_jli_sections): New static variable.
2401 (arc_handle_jli_attribute): New function.
2402 (arc_attribute_table): Add jli_always and jli_fixed attribute.
2403 (arc_file_end): New function.
2404 (TARGET_ASM_FILE_END): Define.
2405 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
2406 (arc_add_jli_section): New function.
2407 (jli_call_scan): Likewise.
2408 (arc_reorg): Call jli_call_scan.
2409 (arc_output_addsi): Remove 'S' from printing asm operand.
2410 (arc_is_jli_call_p): New function.
2411 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
2413 (movhi_insn): Likewise.
2414 (movsi_insn): Likewise.
2415 (movsi_set_cc_insn): Likewise.
2416 (loadqi_update): Likewise.
2417 (load_zeroextendqisi_update): Likewise.
2418 (load_signextendqisi_update): Likewise.
2419 (loadhi_update): Likewise.
2420 (load_zeroextendhisi_update): Likewise.
2421 (load_signextendhisi_update): Likewise.
2422 (loadsi_update): Likewise.
2423 (loadsf_update): Likewise.
2424 (movsicc_insn): Likewise.
2425 (bset_insn): Likewise.
2426 (bxor_insn): Likewise.
2427 (bclr_insn): Likewise.
2428 (bmsk_insn): Likewise.
2429 (bicsi3_insn): Likewise.
2430 (cmpsi_cc_c_insn): Likewise.
2431 (movsi_ne): Likewise.
2432 (movsi_cond_exec): Likewise.
2433 (clrsbsi2): Likewise.
2444 (call_i): Remove 'S' asm letter, add jli instruction.
2445 (call_value_i): Likewise.
2446 * config/arc/arc.op (mjli-always): New option.
2447 * config/arc/constraints.md (Cji): New constraint.
2448 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
2450 (subsf3_fpx): Likewise.
2451 (mulsf3_fpx): Likewise.
2452 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
2454 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
2456 * doc/invoke.texi (ARC): Document mjli-always option.
2458 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
2460 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
2461 avoid addition with 0 and use incw and decw where possible.
2463 2018-01-26 Richard Biener <rguenther@suse.de>
2465 PR tree-optimization/81082
2466 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
2467 association if it requires casting to unsigned.
2468 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
2469 from fold_plusminus_mult_expr to catch important cases late when
2470 range info is available.
2472 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2474 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
2475 * configure.ac (hidden_linkonce): New test.
2476 * configure: Regenerate.
2477 * config.in: Regenerate.
2479 2018-01-26 Julia Koval <julia.koval@intel.com>
2481 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
2482 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
2483 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
2484 _mm_mask_bitshuffle_epi64_mask): Fix type.
2485 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
2486 USI_FTYPE_V4DI_V4DI_USI): Remove.
2487 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
2488 __builtin_ia32_vpshufbitqmb256_mask,
2489 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
2490 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
2491 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
2493 2018-01-26 Alan Modra <amodra@gmail.com>
2496 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
2497 UNSPEC_VBPERMQ. Sort other unspecs.
2499 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
2501 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
2503 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
2506 * predict.c (drop_profile): Do not push/pop cfun; update also
2508 (handle_missing_profiles): Fix logic looking for zero profiles.
2510 2018-01-25 Jakub Jelinek <jakub@redhat.com>
2513 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
2514 on functions with #pragma omp declare simd or functions with simd
2516 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
2517 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
2518 Remove trailing \n from warning_at calls.
2520 2018-01-25 Tom de Vries <tom@codesourcery.com>
2523 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2524 for neutered workers.
2526 2018-01-24 Joseph Myers <joseph@codesourcery.com>
2529 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
2530 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
2532 2018-01-24 Jeff Law <law@redhat.com>
2535 * i386.c (get_probe_interval): Move to earlier point.
2536 (ix86_compute_frame_layout): If -fstack-clash-protection and
2537 the frame is larger than the probe interval, then use pushes
2538 to save registers rather than reg->mem moves.
2539 (ix86_expand_prologue): Remove conditional for int_registers_saved
2542 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
2545 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
2546 min/max for never referenced object.
2548 2018-01-24 Jakub Jelinek <jakub@redhat.com>
2551 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
2553 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
2554 attributes from DECL_ATTRIBUTES (decl) without affecting
2555 DECL_ATTRIBUTES (current_function_decl).
2556 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
2557 functions with non-NULL DECL_ABSTRACT_ORIGIN.
2559 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
2561 PR tree-optimization/83979
2562 * fold-const.c (fold_comparison): Use constant_boolean_node
2563 instead of boolean_{true,false}_node.
2565 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
2567 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
2570 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2572 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2573 Simplify the clause that sets the length attribute.
2574 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2575 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
2576 clause that sets the length attribute.
2577 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2579 2018-01-24 Tom de Vries <tom@codesourcery.com>
2582 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
2583 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
2584 Add strict parameter.
2585 (prevent_branch_around_nothing): Insert dummy insn between branch to
2586 label and label with no ptx insn inbetween.
2587 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
2589 2018-01-24 Tom de Vries <tom@codesourcery.com>
2592 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2593 for neutered threads in warp.
2594 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
2596 2018-01-24 Richard Biener <rguenther@suse.de>
2598 PR tree-optimization/83176
2599 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
2602 2018-01-24 Richard Biener <rguenther@suse.de>
2604 PR tree-optimization/82819
2605 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
2606 code generating pluses that are no-ops in the target precision.
2608 2018-01-24 Richard Biener <rguenther@suse.de>
2611 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
2613 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2615 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
2616 to merge probabilities.
2617 * predict.c (probably_never_executed): Also mark as cold functions
2618 with global 0 profile and guessed local profile.
2619 * profile-count.c (profile_probability::combine_with_count): New
2621 * profile-count.h (profile_probability::operator*,
2622 profile_probability::operator*=, profile_probability::operator/,
2623 profile_probability::operator/=): Reduce precision to adjusted
2624 and set value to guessed on contradictory divisions.
2625 (profile_probability::combine_with_freq): Remove.
2626 (profile_probability::combine_wiht_count): Declare.
2627 (profile_count::force_nonzero):: Set to adjusted.
2628 (profile_count::probability_in):: Set quality to adjusted.
2629 * tree-ssa-tail-merge.c (replace_block_by): Use
2632 2018-01-23 Andrew Waterman <andrew@sifive.com>
2633 Jim Wilson <jimw@sifive.com>
2635 * config/riscv/riscv.c (riscv_stack_boundary): New.
2636 (riscv_option_override): Set riscv_stack_boundary. Handle
2637 riscv_preferred_stack_boundary_arg.
2638 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
2639 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
2640 (STACK_BOUNDARY): Set to riscv_stack_boundary.
2641 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
2642 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
2643 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
2645 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
2648 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
2649 of struct ix86_frame.
2650 (ix86_expand_epilogue): Likewise. Add a local variable for
2651 the reg_save_offset field in struct ix86_frame.
2653 2018-01-23 Bin Cheng <bin.cheng@arm.com>
2655 PR tree-optimization/82604
2656 * tree-loop-distribution.c (enum partition_kind): New enum item
2657 PKIND_PARTIAL_MEMSET.
2658 (partition_builtin_p): Support above new enum item.
2659 (generate_code_for_partition): Ditto.
2660 (compute_access_range): Differentiate cases that equality can be
2661 proven at all loops, the innermost loops or no loops.
2662 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
2663 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
2664 (finalize_partitions, distribute_loop): Don't fuse partition of
2665 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
2666 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
2669 2018-01-23 Martin Liska <mliska@suse.cz>
2671 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
2672 order to ignore the predictor.
2673 (PRED_POLYMORPHIC_CALL): Likewise.
2674 (PRED_RECURSIVE_CALL): Likewise.
2676 2018-01-23 Martin Liska <mliska@suse.cz>
2678 * tree-profile.c (tree_profiling): Print function header to
2679 aware reader which function we are working on.
2680 * value-prof.c (gimple_find_values_to_profile): Do not print
2681 not interesting value histograms.
2683 2018-01-23 Martin Liska <mliska@suse.cz>
2685 * profile-count.h (enum profile_quality): Add
2686 profile_uninitialized as the first value. Do not number values
2687 as they are zero based.
2688 (profile_count::verify): Update sanity check.
2689 (profile_probability::verify): Likewise.
2691 2018-01-23 Nathan Sidwell <nathan@acm.org>
2693 * doc/invoke.texi (ffor-scope): Deprecate.
2695 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2697 PR tree-optimization/83510
2698 * domwalk.c (set_all_edges_as_executable): New function.
2699 (dom_walker::dom_walker): Convert bool param
2700 "skip_unreachable_blocks" to enum reachability. Move setup of
2701 edge flags to set_all_edges_as_executable and only do it when
2702 reachability is REACHABLE_BLOCKS.
2703 * domwalk.h (enum dom_walker::reachability): New enum.
2704 (dom_walker::dom_walker): Convert bool param
2705 "skip_unreachable_blocks" to enum reachability.
2706 (set_all_edges_as_executable): New decl.
2707 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2708 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2710 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2711 but converting true to REACHABLE_BLOCKS.
2712 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2714 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2715 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2716 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2718 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2719 if check_all_array_refs will be called.
2721 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2723 * tree.c (selftest::test_location_wrappers): Add more test
2726 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2728 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2729 (selftest::test_bit_in_range): Likewise.
2731 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2734 * doc/sourcebuild.texi (vect_float): Say that the selector
2735 only describes the situation when -funsafe-math-optimizations is on.
2736 (vect_float_strict): Document.
2738 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2740 PR tree-optimization/83965
2741 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2742 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2743 instead of checking only for a reduction.
2744 (vect_recog_widen_sum_pattern): Likewise.
2746 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2748 * predict.c (probably_never_executed): Only use precise profile info.
2749 (compute_function_frequency): Skip after inlining hack since we now
2750 have quality checking.
2752 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2754 * profile-count.h (profile_probability::very_unlikely,
2755 profile_probability::unlikely, profile_probability::even): Set
2756 precision to guessed.
2758 2018-01-23 Richard Biener <rguenther@suse.de>
2760 PR tree-optimization/83963
2761 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2762 Properly terminate dominator walk when crossing the exit edge not
2763 when visiting its source block.
2765 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2768 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2769 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2771 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2773 PR tree-optimization/83957
2774 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2775 semicolon after for body surrounded by braces.
2777 PR tree-optimization/83081
2778 * profile-count.h (profile_probability::split): New method.
2779 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2780 Use profile_probability::split.
2781 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2782 when splitting a single conditional jump into 2.
2784 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2786 PR tree-optimization/69452
2787 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2790 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2792 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2793 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2794 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2796 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2798 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2799 * config/rl78/rl78.md (movdi): New define_expand.
2800 * config/rl78/rl78.c (rl78_split_movdi): New function.
2802 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2805 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2807 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2808 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2809 128-bit to produce an UNSPEC move to get the double word with the
2810 signbit and then a shift directly to do signbit.
2811 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2812 implementation with a new version that just does either a direct
2813 move or a regular move. Move memory interface to separate insns.
2814 Move insns so they are next to the expander.
2815 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2816 with signbit move. Split big and little endian case.
2817 (signbit<mode>2_dm_mem_le): Likewise.
2818 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2819 (signbit<mode>2_dm2): Likewise.
2821 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2823 * config/rl78/rl78.md (anddi3): New define_expand.
2825 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2827 * config/rl78/rl78.md (umindi3): New define_expand.
2829 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2831 * config/rl78/rl78.md (smindi3): New define_expand.
2833 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2835 * config/rl78/rl78.md (smaxdi3): New define_expand.
2837 2018-01-22 Carl Love <cel@us.ibm.com>
2839 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2840 LVX_V1TI): Add macro expansion.
2841 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2842 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2843 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2844 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2845 Change check to determine if the instruction is a byte reversing
2846 entry. Fix typo in comment.
2847 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2848 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2849 Add def_builtin calls for new builtins.
2850 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2851 Add define_insn expansion.
2853 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2855 * config/rl78/rl78.md (umaxdi3): New define_expand.
2857 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2859 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2860 for non-QImode registers.
2862 2018-01-22 Richard Biener <rguenther@suse.de>
2864 PR tree-optimization/83963
2865 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2866 including the loop exit block.
2867 (scop_detection::merge_sese): Likewise.
2868 (scop_detection::add_scop): Do it here instead.
2870 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2872 * doc/sourcebuild.texi (arm_softfloat): Document.
2874 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2877 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2878 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2879 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2881 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2882 David Edelsohn <dje.gcc@gmail.com>
2885 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2886 Change "crset eq" to "crset 2".
2887 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2888 (*call_indirect_aix<mode>_nospec): Likewise.
2889 (*call_value_indirect_aix<mode>_nospec): Likewise.
2890 (*call_indirect_elfv2<mode>_nospec): Likewise.
2891 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2892 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2893 change assembly output from . to $.
2894 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2895 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2896 (*tablejump<mode>_internal1_nospec): Likewise.
2898 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2901 * config/sh/sh_optimize_sett_clrt.cc:
2902 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2904 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2906 PR tree-optimization/83940
2907 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2908 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2909 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2910 use it to pass back the definition type.
2911 (vect_check_store_rhs): Likewise rhs_dt_out.
2912 (vect_build_gather_load_calls): Add a mask_dt argument and use
2913 it instead of a call to vect_is_simple_use.
2914 (vectorizable_store): Update calls to vect_check_load_store_mask
2915 and vect_check_store_rhs. Use the dt returned by the latter instead
2916 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2917 instead of calls to vect_is_simple_use. Pass the scalar rather
2918 than the vector operand to vect_is_simple_use when handling
2919 second and subsequent copies of an rhs value.
2920 (vectorizable_load): Update calls to vect_check_load_store_mask
2921 and vect_build_gather_load_calls. Use the cached mask_dt and
2922 gs_info.offset_dt instead of calls to vect_is_simple_use.
2924 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2927 * tree-emutls.c: Include gimplify.h.
2928 (lower_emutls_2): New function.
2929 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2930 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2931 it before further processing.
2934 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2935 UINTVAL (trueop1) instead of INTVAL (op1).
2937 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2941 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2942 INCOMING_FRAME_SP_OFFSET if not defined.
2943 (scan_trace): Add ENTRY argument. If true and
2944 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2945 emit a note to adjust the CFA offset.
2946 (create_cfi_notes): Adjust scan_trace callers.
2947 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2948 INCOMING_FRAME_SP_OFFSET in the CIE.
2949 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2950 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2952 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2953 * doc/tm.texi: Regenerated.
2955 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2957 PR rtl-optimization/83147
2958 * lra-constraints.c (remove_inheritance_pseudos): Use
2959 lra_substitute_pseudo_within_insn.
2961 2018-01-19 Tom de Vries <tom@codesourcery.com>
2962 Cesar Philippidis <cesar@codesourcery.com>
2965 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2967 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2970 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2971 spaces for function labels.
2973 2018-01-19 Martin Liska <mliska@suse.cz>
2975 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2976 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2977 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2978 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2979 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2980 (PRED_CONST_RETURN): Change from 69 to 65.
2981 (PRED_NULL_RETURN): Change from 91 to 71.
2982 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2983 (PRED_LOOP_GUARD): Change from 66 to 73.
2985 2018-01-19 Martin Liska <mliska@suse.cz>
2987 * predict.c (predict_insn_def): Add new assert.
2988 (struct branch_predictor): Change type to signed integer.
2989 (test_prediction_value_range): Amend test to cover
2991 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2992 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2993 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2994 (PRED_LOOP_IV_COMPARE): Likewise.
2995 * predict.h (PROB_UNINITIALIZED): Define new constant.
2997 2018-01-19 Martin Liska <mliska@suse.cz>
2999 * predict.c (dump_prediction): Add new format for
3000 analyze_brprob.py script which is enabled with -details
3002 * profile-count.h (precise_p): New function.
3004 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
3006 PR tree-optimization/83922
3007 * tree-vect-loop.c (vect_verify_full_masking): Return false if
3008 there are no statements that need masking.
3009 (vect_active_double_reduction_p): New function.
3010 (vect_analyze_loop_operations): Use it when handling phis that
3011 are not in the loop header.
3013 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
3015 PR tree-optimization/83914
3016 * tree-vect-loop.c (vectorizable_induction): Don't convert
3017 init_expr or apply the peeling adjustment for inductions
3018 that are nested within the vectorized loop.
3020 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3022 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
3025 2018-01-18 Jakub Jelinek <jakub@redhat.com>
3029 * function.h (gimplify_parameters): Add gimple_seq * argument.
3030 * function.c: Include gimple.h and options.h.
3031 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
3032 for the added local temporaries if needed.
3033 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
3034 if there are any parameter cleanups, wrap whole body into a
3035 try/finally with the cleanups.
3037 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
3040 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
3041 Use GET_MODE_CLASS for scalar floating point.
3043 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
3047 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
3048 Fix call of call_cgraph_insertion_hooks.
3050 2018-01-18 Martin Sebor <msebor@redhat.com>
3052 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
3054 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
3057 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
3060 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
3063 * common.opt: (-ffile-prefix-map): New option.
3064 * opts.c (common_handle_option): Defer it.
3065 * opts-global.c (handle_common_deferred_options): Handle it.
3066 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
3067 * file-prefix-map.h: New file.
3068 (remap_debug_filename, add_debug_prefix_map): ...here.
3069 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
3070 * final.c (debug_prefix_map, add_debug_prefix_map
3071 remap_debug_filename): Move to...
3072 * file-prefix-map.c: New file.
3073 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
3074 generalize, get rid of alloca(), use strrchr() instead of strchr().
3075 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
3076 Implement in terms of add_prefix_map().
3077 (remap_macro_filename, remap_debug_filename): Implement in term of
3079 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
3080 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
3081 * dbxout.c: Include file-prefix-map.h.
3082 * varasm.c: Likewise.
3083 * vmsdbgout.c: Likewise.
3084 * xcoffout.c: Likewise.
3085 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
3086 * doc/cppopts.texi (-fmacro-prefix-map): Document.
3087 * doc/invoke.texi (-ffile-prefix-map): Document.
3088 (-fdebug-prefix-map): Update description.
3090 2018-01-18 Martin Liska <mliska@suse.cz>
3092 * config/i386/i386.c (indirect_thunk_name): Document that also
3094 (output_indirect_thunk): Document why both instructions
3095 (pause and lfence) are generated.
3097 2018-01-18 Richard Biener <rguenther@suse.de>
3099 PR tree-optimization/83887
3100 * graphite-scop-detection.c
3101 (scop_detection::get_nearest_dom_with_single_entry): Remove.
3102 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
3103 (scop_detection::merge_sese): Re-implement with a flood-fill
3104 algorithm that properly finds a SESE region if it exists.
3106 2018-01-18 Jakub Jelinek <jakub@redhat.com>
3109 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
3110 pointer_diff optimizations use view_convert instead of convert.
3112 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3114 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
3115 Generate different code for -mno-speculate-indirect-jumps.
3116 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
3117 (*call_indirect_aix<mode>): Disable for
3118 -mno-speculate-indirect-jumps.
3119 (*call_indirect_aix<mode>_nospec): New define_insn.
3120 (*call_value_indirect_aix<mode>): Disable for
3121 -mno-speculate-indirect-jumps.
3122 (*call_value_indirect_aix<mode>_nospec): New define_insn.
3123 (*sibcall_nonlocal_sysv<mode>): Generate different code for
3124 -mno-speculate-indirect-jumps.
3125 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
3127 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
3129 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
3130 long double type, set the flags for noting the default long double
3131 type, even if we don't pass or return a long double type.
3133 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
3136 * ipa-inline.c (flatten_function): Do not overwrite final inlining
3139 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
3141 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
3142 support for merge[hl].
3143 (fold_mergehl_helper): New helper function.
3144 (tree-vector-builder.h): New #include for tree_vector_builder usage.
3145 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
3146 (altivec_vmrglw_direct): Add xxmrglw insn.
3148 2018-01-17 Andrew Waterman <andrew@sifive.com>
3150 * config/riscv/riscv.c (riscv_conditional_register_usage): If
3151 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
3153 2018-01-17 David Malcolm <dmalcolm@redhat.com>
3156 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
3157 call the lto_location_cache before reading the
3158 DECL_SOURCE_LOCATION of the types.
3160 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
3161 Richard Sandiford <richard.sandiford@linaro.org>
3163 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
3164 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
3165 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
3166 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
3167 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
3169 * config/aarch64/constraints.md (aarch64_movti_operand):
3171 * config/aarch64/predicates.md (Uti): Add new constraint.
3173 2018-01-17 Carl Love <cel@us.ibm.com>
3175 * config/rs6000/vsx.md (define_expand xl_len_r,
3176 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
3177 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
3179 (define_expand, define_insn): Move the shift left from the
3180 define_insn to the define_expand for lxvl and stxvl instructions.
3181 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
3182 and XL_LEN_R definitions to PURE.
3184 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
3186 * config/i386/i386.c (indirect_thunk_name): Declare regno
3187 as unsigned int. Compare regno with INVALID_REGNUM.
3188 (output_indirect_thunk): Ditto.
3189 (output_indirect_thunk_function): Ditto.
3190 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
3191 in the call to output_indirect_thunk_function.
3193 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
3196 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
3197 rather than the size of inner_type to determine the stack slot size
3198 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
3200 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
3203 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
3206 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
3208 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
3209 endian Linux systems to optionally enable multilibs for selecting
3210 the long double type if the user configured an explicit type.
3211 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
3212 have no long double multilibs if not defined.
3213 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
3214 warn if the user used -mabi={ieee,ibm}longdouble and we built
3215 multilibs for long double.
3216 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
3217 appropriate multilib option.
3218 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
3220 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
3221 for building long double multilibs.
3222 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
3224 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
3226 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
3229 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
3231 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
3234 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
3237 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
3240 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3242 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
3243 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
3245 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
3247 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
3248 different rtl trees depending on TARGET_64BIT.
3249 (rs6000_gen_lvx): Likewise.
3251 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3253 * config/visium/visium.md (nop): Tweak comment.
3254 (hazard_nop): Likewise.
3256 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3258 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
3259 -mspeculate-indirect-jumps.
3260 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
3261 for -mno-speculate-indirect-jumps.
3262 (*call_indirect_elfv2<mode>_nospec): New define_insn.
3263 (*call_value_indirect_elfv2<mode>): Disable for
3264 -mno-speculate-indirect-jumps.
3265 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
3266 (indirect_jump): Emit different RTL for
3267 -mno-speculate-indirect-jumps.
3268 (*indirect_jump<mode>): Disable for
3269 -mno-speculate-indirect-jumps.
3270 (*indirect_jump<mode>_nospec): New define_insn.
3271 (tablejump): Emit different RTL for
3272 -mno-speculate-indirect-jumps.
3273 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
3274 (tablejumpsi_nospec): New define_expand.
3275 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
3276 (tablejumpdi_nospec): New define_expand.
3277 (*tablejump<mode>_internal1): Disable for
3278 -mno-speculate-indirect-jumps.
3279 (*tablejump<mode>_internal1_nospec): New define_insn.
3280 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
3283 2018-01-16 Artyom Skrobov tyomitch@gmail.com
3285 * caller-save.c (insert_save): Drop unnecessary parameter. All
3288 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3289 Richard Biener <rguenth@suse.de>
3292 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
3293 return early, inline manually is_gimple_sizepos. Make sure if we
3294 call gimplify_expr we don't end up with a gimple constant.
3295 * tree.c (variably_modified_type_p): Don't return true for
3296 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
3297 * gimplify.h (is_gimple_sizepos): Remove.
3299 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3301 PR tree-optimization/83857
3302 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
3303 vectorizable_live_operation for pure SLP statements.
3304 (vectorizable_live_operation): Handle PHIs.
3306 2018-01-16 Richard Biener <rguenther@suse.de>
3308 PR tree-optimization/83867
3309 * tree-vect-stmts.c (vect_transform_stmt): Precompute
3310 nested_in_vect_loop_p since the scalar stmt may get invalidated.
3312 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3315 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
3316 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
3317 If off is not INTEGER_CST, issue a may not be aligned warning
3318 rather than isn't aligned. Use isn%'t rather than isn't.
3319 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
3321 <case MULT_EXPR>: Improve the case when bottom and one of the
3322 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
3323 operand, in that case check if the other operand is multiple of
3324 bottom divided by the INTEGER_CST operand.
3326 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3329 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
3330 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
3331 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
3332 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
3333 * config/pa/pa.c (pa_function_arg_advance): Likewise.
3334 (pa_function_arg, pa_arg_partial_bytes): Likewise.
3335 (pa_function_arg_size): New function.
3337 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3339 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
3340 in a separate statement.
3342 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3344 PR tree-optimization/83847
3345 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
3346 group gathers and scatters.
3348 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3350 PR rtl-optimization/86620
3351 * params.def (max-sched-ready-insns): Bump minimum value to 1.
3353 PR rtl-optimization/83213
3354 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
3355 to last if both are JUMP_INSNs.
3357 PR tree-optimization/83843
3358 * gimple-ssa-store-merging.c
3359 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
3360 store_immediate_info for bswap/nop orig_stores.
3362 2018-01-15 Andrew Waterman <andrew@sifive.com>
3364 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
3366 <UDIV>: Increase cost if !TARGET_DIV.
3368 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
3370 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
3371 (define_attr "cr_logical_3op"): New.
3372 (cceq_ior_compare): Adjust.
3373 (cceq_ior_compare_complement): Adjust.
3374 (*cceq_rev_compare): Adjust.
3375 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
3376 (is_cracked_insn): Adjust.
3377 (insn_must_be_first_in_group): Adjust.
3378 * config/rs6000/40x.md: Adjust.
3379 * config/rs6000/440.md: Adjust.
3380 * config/rs6000/476.md: Adjust.
3381 * config/rs6000/601.md: Adjust.
3382 * config/rs6000/603.md: Adjust.
3383 * config/rs6000/6xx.md: Adjust.
3384 * config/rs6000/7450.md: Adjust.
3385 * config/rs6000/7xx.md: Adjust.
3386 * config/rs6000/8540.md: Adjust.
3387 * config/rs6000/cell.md: Adjust.
3388 * config/rs6000/e300c2c3.md: Adjust.
3389 * config/rs6000/e500mc.md: Adjust.
3390 * config/rs6000/e500mc64.md: Adjust.
3391 * config/rs6000/e5500.md: Adjust.
3392 * config/rs6000/e6500.md: Adjust.
3393 * config/rs6000/mpc.md: Adjust.
3394 * config/rs6000/power4.md: Adjust.
3395 * config/rs6000/power5.md: Adjust.
3396 * config/rs6000/power6.md: Adjust.
3397 * config/rs6000/power7.md: Adjust.
3398 * config/rs6000/power8.md: Adjust.
3399 * config/rs6000/power9.md: Adjust.
3400 * config/rs6000/rs64.md: Adjust.
3401 * config/rs6000/titan.md: Adjust.
3403 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3405 * config/i386/predicates.md (indirect_branch_operand): Rewrite
3406 ix86_indirect_branch_register logic.
3408 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3410 * config/i386/constraints.md (Bs): Update
3411 ix86_indirect_branch_register check. Don't check
3412 ix86_indirect_branch_register with GOT_memory_operand.
3414 * config/i386/predicates.md (GOT_memory_operand): Don't check
3415 ix86_indirect_branch_register here.
3416 (GOT32_symbol_operand): Likewise.
3418 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3420 * config/i386/predicates.md (constant_call_address_operand):
3421 Rewrite ix86_indirect_branch_register logic.
3422 (sibcall_insn_operand): Likewise.
3424 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3426 * config/i386/constraints.md (Bs): Replace
3427 ix86_indirect_branch_thunk_register with
3428 ix86_indirect_branch_register.
3430 * config/i386/i386.md (indirect_jump): Likewise.
3431 (tablejump): Likewise.
3432 (*sibcall_memory): Likewise.
3433 (*sibcall_value_memory): Likewise.
3434 Peepholes of indirect call and jump via memory: Likewise.
3435 * config/i386/i386.opt: Likewise.
3436 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3437 (GOT_memory_operand): Likewise.
3438 (call_insn_operand): Likewise.
3439 (sibcall_insn_operand): Likewise.
3440 (GOT32_symbol_operand): Likewise.
3442 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3445 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
3446 type rather than type addr's type points to.
3447 (expand_omp_atomic_mutex): Likewise.
3448 (expand_omp_atomic): Likewise.
3450 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3453 * config/i386/i386.c (output_indirect_thunk_function): Use
3454 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
3455 for __x86_return_thunk.
3457 2018-01-15 Richard Biener <rguenther@suse.de>
3460 * expmed.c (extract_bit_field_1): Fix typo.
3462 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3465 * config/arm/iterators.md (VF): New mode iterator.
3466 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
3467 Remove integer-related logic from pattern.
3468 (neon_vabd<mode>_3): Likewise.
3470 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3473 * common.opt (fstrict-overflow): No longer an alias.
3474 (fwrapv-pointer): New option.
3475 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
3476 also for pointer types based on flag_wrapv_pointer.
3477 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
3478 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
3479 opts->x_flag_wrapv got set.
3480 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
3481 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
3482 POINTER_TYPE_OVERFLOW_UNDEFINED.
3483 * match.pd: Likewise in address comparison pattern.
3484 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
3486 2018-01-15 Richard Biener <rguenther@suse.de>
3489 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
3490 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
3491 Reset type names to their identifier if their TYPE_DECL doesn't
3492 have linkage (and thus is used for ODR and devirt).
3493 (save_debug_info_for_decl): Remove.
3494 (save_debug_info_for_type): Likewise.
3495 (add_tree_to_fld_list): Adjust.
3496 * tree-pretty-print.c (dump_generic_node): Make dumping of
3497 type names more robust.
3499 2018-01-15 Richard Biener <rguenther@suse.de>
3501 * BASE-VER: Bump to 8.0.1.
3503 2018-01-14 Martin Sebor <msebor@redhat.com>
3506 * builtins.c (check_access): Avoid warning when the no-warning bit
3509 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
3511 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
3512 * ira-color (allocno_hard_regs_compare): Likewise.
3514 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
3517 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
3518 Use .pushsection/.popsection.
3520 2018-01-14 Martin Sebor <msebor@redhat.com>
3523 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
3525 2018-01-14 Jakub Jelinek <jakub@redhat.com>
3527 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
3528 entry from extra_headers.
3529 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
3530 extra_headers, make the list bitwise identical to the i?86-*-* one.
3532 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3534 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
3535 -mcmodel=large with -mindirect-branch=thunk,
3536 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
3537 -mfunction-return=thunk-extern.
3538 * doc/invoke.texi: Document -mcmodel=large is incompatible with
3539 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
3540 -mfunction-return=thunk and -mfunction-return=thunk-extern.
3542 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3544 * config/i386/i386.c (print_reg): Print the name of the full
3545 integer register without '%'.
3546 (ix86_print_operand): Handle 'V'.
3547 * doc/extend.texi: Document 'V' modifier.
3549 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3551 * config/i386/constraints.md (Bs): Disallow memory operand for
3552 -mindirect-branch-register.
3554 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3555 (GOT_memory_operand): Likewise.
3556 (call_insn_operand): Likewise.
3557 (sibcall_insn_operand): Likewise.
3558 (GOT32_symbol_operand): Likewise.
3559 * config/i386/i386.md (indirect_jump): Call convert_memory_address
3560 for -mindirect-branch-register.
3561 (tablejump): Likewise.
3562 (*sibcall_memory): Likewise.
3563 (*sibcall_value_memory): Likewise.
3564 Disallow peepholes of indirect call and jump via memory for
3565 -mindirect-branch-register.
3566 (*call_pop): Replace m with Bw.
3567 (*call_value_pop): Likewise.
3568 (*sibcall_pop_memory): Replace m with Bs.
3569 * config/i386/i386.opt (mindirect-branch-register): New option.
3570 * doc/invoke.texi: Document -mindirect-branch-register option.
3572 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3574 * config/i386/i386-protos.h (ix86_output_function_return): New.
3575 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
3576 set function_return_type.
3577 (indirect_thunk_name): Add ret_p to indicate thunk for function
3579 (output_indirect_thunk_function): Pass false to
3580 indirect_thunk_name.
3581 (ix86_output_indirect_branch_via_reg): Likewise.
3582 (ix86_output_indirect_branch_via_push): Likewise.
3583 (output_indirect_thunk_function): Create alias for function
3584 return thunk if regno < 0.
3585 (ix86_output_function_return): New function.
3586 (ix86_handle_fndecl_attribute): Handle function_return.
3587 (ix86_attribute_table): Add function_return.
3588 * config/i386/i386.h (machine_function): Add
3589 function_return_type.
3590 * config/i386/i386.md (simple_return_internal): Use
3591 ix86_output_function_return.
3592 (simple_return_internal_long): Likewise.
3593 * config/i386/i386.opt (mfunction-return=): New option.
3594 (indirect_branch): Mention -mfunction-return=.
3595 * doc/extend.texi: Document function_return function attribute.
3596 * doc/invoke.texi: Document -mfunction-return= option.
3598 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3600 * config/i386/i386-opts.h (indirect_branch): New.
3601 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
3602 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
3603 with local indirect jump when converting indirect call and jump.
3604 (ix86_set_indirect_branch_type): New.
3605 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
3606 (indirectlabelno): New.
3607 (indirect_thunk_needed): Likewise.
3608 (indirect_thunk_bnd_needed): Likewise.
3609 (indirect_thunks_used): Likewise.
3610 (indirect_thunks_bnd_used): Likewise.
3611 (INDIRECT_LABEL): Likewise.
3612 (indirect_thunk_name): Likewise.
3613 (output_indirect_thunk): Likewise.
3614 (output_indirect_thunk_function): Likewise.
3615 (ix86_output_indirect_branch_via_reg): Likewise.
3616 (ix86_output_indirect_branch_via_push): Likewise.
3617 (ix86_output_indirect_branch): Likewise.
3618 (ix86_output_indirect_jmp): Likewise.
3619 (ix86_code_end): Call output_indirect_thunk_function if needed.
3620 (ix86_output_call_insn): Call ix86_output_indirect_branch if
3622 (ix86_handle_fndecl_attribute): Handle indirect_branch.
3623 (ix86_attribute_table): Add indirect_branch.
3624 * config/i386/i386.h (machine_function): Add indirect_branch_type
3625 and has_local_indirect_jump.
3626 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
3628 (tablejump): Likewise.
3629 (*indirect_jump): Use ix86_output_indirect_jmp.
3630 (*tablejump_1): Likewise.
3631 (simple_return_indirect_internal): Likewise.
3632 * config/i386/i386.opt (mindirect-branch=): New option.
3633 (indirect_branch): New.
3636 (thunk-inline): Likewise.
3637 (thunk-extern): Likewise.
3638 * doc/extend.texi: Document indirect_branch function attribute.
3639 * doc/invoke.texi: Document -mindirect-branch= option.
3641 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
3644 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
3646 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
3648 * ipa-inline.c (want_inline_small_function_p): Return false if
3649 inlining has already failed with CIF_FINAL_ERROR.
3650 (update_caller_keys): Call want_inline_small_function_p before
3652 (update_callee_keys): Likewise.
3654 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
3656 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
3658 (rs6000_quadword_masked_address_p): Likewise.
3659 (quad_aligned_load_p): Likewise.
3660 (quad_aligned_store_p): Likewise.
3661 (const_load_sequence_p): Add comment to describe the outer-most loop.
3662 (mimic_memory_attributes_and_flags): New function.
3663 (rs6000_gen_stvx): Likewise.
3664 (replace_swapped_aligned_store): Likewise.
3665 (rs6000_gen_lvx): Likewise.
3666 (replace_swapped_aligned_load): Likewise.
3667 (replace_swapped_load_constant): Capitalize argument name in
3668 comment describing this function.
3669 (rs6000_analyze_swaps): Add a third pass to search for vector loads
3670 and stores that access quad-word aligned addresses and replace
3671 with stvx or lvx instructions when appropriate.
3672 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
3673 New function prototype.
3674 (rs6000_quadword_masked_address_p): Likewise.
3675 (rs6000_gen_lvx): Likewise.
3676 (rs6000_gen_stvx): Likewise.
3677 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3678 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3679 when memory address is aligned.
3680 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3681 this split to select lvx instruction when memory address is aligned.
3682 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3683 instruction when memory address is aligned.
3684 (*vsx_le_perm_load_v16qi): Likewise.
3685 (four unnamed splitters): Modify to select the stvx instruction
3686 when memory is aligned.
3688 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3690 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3691 which appears in the queue multiple times.
3693 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3694 Alan Hayward <alan.hayward@arm.com>
3695 David Sherwood <david.sherwood@arm.com>
3697 * tree-vectorizer.h (vec_lower_bound): New structure.
3698 (_loop_vec_info): Add check_nonzero and lower_bounds.
3699 (LOOP_VINFO_CHECK_NONZERO): New macro.
3700 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3701 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3702 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3703 fields. Make seg_len the distance travelled, not including the
3705 (dr_direction_indicator): Declare.
3706 (dr_zero_step_indicator): Likewise.
3707 (dr_known_forward_stride_p): Likewise.
3708 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3710 (runtime_alias_check_p): Allow runtime alias checks with
3712 (operator ==): Compare access_size and align.
3713 (prune_runtime_alias_test_list): Rework for new distinction between
3714 the access_size and seg_len.
3715 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3717 (get_segment_min_max): New function.
3718 (create_intersect_range_checks): Use it.
3719 (dr_step_indicator): New function.
3720 (dr_direction_indicator): Likewise.
3721 (dr_zero_step_indicator): Likewise.
3722 (dr_known_forward_stride_p): Likewise.
3723 * tree-loop-distribution.c (data_ref_segment_size): Return
3724 DR_STEP * (niters - 1).
3725 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3727 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3728 (vect_preserves_scalar_order_p): New function, split out from...
3729 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3730 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3731 (vect_vfa_access_size): New function.
3732 (vect_vfa_align): Likewise.
3733 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3734 (dump_lower_bound): New function.
3735 (vect_check_lower_bound): Likewise.
3736 (vect_small_gap_p): Likewise.
3737 (vectorizable_with_step_bound_p): Likewise.
3738 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3739 depencies if the vectorization factor is 1. Convert the checks
3740 for nonzero steps into checks on the bounds of DR_STEP. Try using
3741 a bunds check for variable steps if the minimum required step is
3742 relatively small. Update calls to the dr_with_seg_len
3743 constructor and to vect_compile_time_alias.
3744 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3746 (vect_loop_versioning): Call it.
3747 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3749 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3751 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3752 Alan Hayward <alan.hayward@arm.com>
3753 David Sherwood <david.sherwood@arm.com>
3755 * doc/sourcebuild.texi (vect_scatter_store): Document.
3756 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3758 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3760 * genopinit.c (main): Add supports_vec_scatter_store and
3761 supports_vec_scatter_store_cached to target_optabs.
3762 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3763 IFN_MASK_SCATTER_STORE.
3764 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3766 * internal-fn.h (internal_store_fn_p): Declare.
3767 (internal_fn_stored_value_index): Likewise.
3768 * internal-fn.c (scatter_store_direct): New macro.
3769 (expand_scatter_store_optab_fn): New function.
3770 (direct_scatter_store_optab_supported_p): New macro.
3771 (internal_store_fn_p): New function.
3772 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3773 IFN_MASK_SCATTER_STORE.
3774 (internal_fn_mask_index): Likewise.
3775 (internal_fn_stored_value_index): New function.
3776 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3778 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3779 * optabs-query.c (supports_vec_scatter_store_p): New function.
3780 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3781 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3782 true for scatter stores.
3783 (vect_gather_scatter_fn_p): Handle scatter stores too.
3784 (vect_check_gather_scatter): Consider using scatter stores if
3785 supports_vec_scatter_store_p.
3786 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3788 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3789 internal_fn_stored_value_index.
3790 (check_load_store_masking): Handle scatter stores too.
3791 (vect_get_store_rhs): Make public.
3792 (vectorizable_call): Use internal_store_fn_p.
3793 (vectorizable_store): Handle scatter store internal functions.
3794 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3795 when deciding whether the end of the group has been reached.
3796 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3797 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3798 (mask_scatter_store<mode>): New insns.
3800 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3801 Alan Hayward <alan.hayward@arm.com>
3802 David Sherwood <david.sherwood@arm.com>
3804 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3805 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3806 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3808 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3809 Use vect_truncate_gather_scatter_offset if we can't treat the
3810 operation as a normal gather load or scatter store.
3811 (get_group_load_store_type): Take the gather_scatter_info
3812 as argument. Try using a gather load or scatter store for
3813 single-element groups.
3814 (get_load_store_type): Update calls to get_group_load_store_type
3815 and vect_use_strided_gather_scatters_p.
3817 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3818 Alan Hayward <alan.hayward@arm.com>
3819 David Sherwood <david.sherwood@arm.com>
3821 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3822 optional tree argument.
3823 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3825 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3826 but continue to use the current value as a fallback.
3827 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3828 to compare the updates.
3829 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3830 (get_load_store_type): Use it when handling a strided access.
3831 (vect_get_strided_load_store_ops): New function.
3832 (vect_get_data_ptr_increment): Likewise.
3833 (vectorizable_load): Handle strided gather loads. Always pass
3834 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3836 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3837 Alan Hayward <alan.hayward@arm.com>
3838 David Sherwood <david.sherwood@arm.com>
3840 * doc/md.texi (gather_load@var{m}): Document.
3841 (mask_gather_load@var{m}): Likewise.
3842 * genopinit.c (main): Add supports_vec_gather_load and
3843 supports_vec_gather_load_cached to target_optabs.
3844 * optabs-tree.c (init_tree_optimization_optabs): Use
3845 ggc_cleared_alloc to allocate target_optabs.
3846 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3847 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3849 * internal-fn.h (internal_load_fn_p): Declare.
3850 (internal_gather_scatter_fn_p): Likewise.
3851 (internal_fn_mask_index): Likewise.
3852 (internal_gather_scatter_fn_supported_p): Likewise.
3853 * internal-fn.c (gather_load_direct): New macro.
3854 (expand_gather_load_optab_fn): New function.
3855 (direct_gather_load_optab_supported_p): New macro.
3856 (direct_internal_fn_optab): New function.
3857 (internal_load_fn_p): Likewise.
3858 (internal_gather_scatter_fn_p): Likewise.
3859 (internal_fn_mask_index): Likewise.
3860 (internal_gather_scatter_fn_supported_p): Likewise.
3861 * optabs-query.c (supports_at_least_one_mode_p): New function.
3862 (supports_vec_gather_load_p): Likewise.
3863 * optabs-query.h (supports_vec_gather_load_p): Declare.
3864 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3865 and memory_type field.
3866 (NUM_PATTERNS): Bump to 15.
3867 * tree-vect-data-refs.c: Include internal-fn.h.
3868 (vect_gather_scatter_fn_p): New function.
3869 (vect_describe_gather_scatter_call): Likewise.
3870 (vect_check_gather_scatter): Try using internal functions for
3871 gather loads. Recognize existing calls to a gather load function.
3872 (vect_analyze_data_refs): Consider using gather loads if
3873 supports_vec_gather_load_p.
3874 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3875 (vect_get_gather_scatter_offset_type): Likewise.
3876 (vect_convert_mask_for_vectype): Likewise.
3877 (vect_add_conversion_to_patterm): Likewise.
3878 (vect_try_gather_scatter_pattern): Likewise.
3879 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3880 (vect_vect_recog_func_ptrs): Add it.
3881 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3882 internal_fn_mask_index and internal_gather_scatter_fn_p.
3883 (check_load_store_masking): Take the gather_scatter_info as an
3884 argument and handle gather loads.
3885 (vect_get_gather_scatter_ops): New function.
3886 (vectorizable_call): Check internal_load_fn_p.
3887 (vectorizable_load): Likewise. Handle gather load internal
3889 (vectorizable_store): Update call to check_load_store_masking.
3890 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3891 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3892 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3893 (aarch64_gather_scale_operand_d): New predicates.
3894 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3895 (mask_gather_load<mode>): New insns.
3897 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3898 Alan Hayward <alan.hayward@arm.com>
3899 David Sherwood <david.sherwood@arm.com>
3901 * optabs.def (fold_left_plus_optab): New optab.
3902 * doc/md.texi (fold_left_plus_@var{m}): Document.
3903 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3904 * internal-fn.c (fold_left_direct): Define.
3905 (expand_fold_left_optab_fn): Likewise.
3906 (direct_fold_left_optab_supported_p): Likewise.
3907 * fold-const-call.c (fold_const_fold_left): New function.
3908 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3909 * tree-parloops.c (valid_reduction_p): New function.
3910 (gather_scalar_reductions): Use it.
3911 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3912 (vect_finish_replace_stmt): Declare.
3913 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3914 (needs_fold_left_reduction_p): New function, split out from...
3915 (vect_is_simple_reduction): ...here. Accept reductions that
3916 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3917 (vect_force_simple_reduction): Also store the reduction type in
3918 the assignment's STMT_VINFO_REDUC_TYPE.
3919 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3920 (merge_with_identity): New function.
3921 (vect_expand_fold_left): Likewise.
3922 (vectorize_fold_left_reduction): Likewise.
3923 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3924 scalar phi in place for it. Check for target support and reject
3925 cases that would reassociate the operation. Defer the transform
3926 phase to vectorize_fold_left_reduction.
3927 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3928 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3929 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3931 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3933 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3934 call to ifc_temp_var.
3936 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3937 Alan Hayward <alan.hayward@arm.com>
3938 David Sherwood <david.sherwood@arm.com>
3940 * target.def (legitimize_address_displacement): Take the original
3941 offset as a poly_int.
3942 * targhooks.h (default_legitimize_address_displacement): Update
3944 * targhooks.c (default_legitimize_address_displacement): Likewise.
3945 * doc/tm.texi: Regenerate.
3946 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3947 as an argument, moving assert of ad->disp == ad->disp_term to...
3948 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3949 Try calling targetm.legitimize_address_displacement before expanding
3950 the address rather than afterwards, and adjust for the new interface.
3951 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3952 Match the new hook interface. Handle SVE addresses.
3953 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3956 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3958 * Makefile.in (OBJS): Add early-remat.o.
3959 * target.def (select_early_remat_modes): New hook.
3960 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3961 * doc/tm.texi: Regenerate.
3962 * targhooks.h (default_select_early_remat_modes): Declare.
3963 * targhooks.c (default_select_early_remat_modes): New function.
3964 * timevar.def (TV_EARLY_REMAT): New timevar.
3965 * passes.def (pass_early_remat): New pass.
3966 * tree-pass.h (make_pass_early_remat): Declare.
3967 * early-remat.c: New file.
3968 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3970 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3972 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3973 Alan Hayward <alan.hayward@arm.com>
3974 David Sherwood <david.sherwood@arm.com>
3976 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3977 vfm1 with a bound_epilog parameter.
3978 (vect_do_peeling): Update calls accordingly, and move the prologue
3979 call earlier in the function. Treat the base bound_epilog as 0 for
3980 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3981 this base when peeling for gaps.
3982 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3983 with fully-masked loops.
3984 (vect_estimate_min_profitable_iters): Handle the single peeled
3985 iteration in that case.
3987 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3988 Alan Hayward <alan.hayward@arm.com>
3989 David Sherwood <david.sherwood@arm.com>
3991 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3992 single-element interleaving even if the size is not a power of 2.
3993 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3994 accesses for single-element interleaving if the group size is
3997 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3998 Alan Hayward <alan.hayward@arm.com>
3999 David Sherwood <david.sherwood@arm.com>
4001 * doc/md.texi (fold_extract_last_@var{m}): Document.
4002 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
4003 * optabs.def (fold_extract_last_optab): New optab.
4004 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
4005 * internal-fn.c (fold_extract_direct): New macro.
4006 (expand_fold_extract_optab_fn): Likewise.
4007 (direct_fold_extract_optab_supported_p): Likewise.
4008 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
4009 * tree-vect-loop.c (vect_model_reduction_cost): Handle
4010 EXTRACT_LAST_REDUCTION.
4011 (get_initial_def_for_reduction): Do not create an initial vector
4012 for EXTRACT_LAST_REDUCTION reductions.
4013 (vectorizable_reduction): Leave the scalar phi in place for
4014 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
4015 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
4016 epilogue code for EXTRACT_LAST_REDUCTION and defer the
4017 transform phase to vectorizable_condition.
4018 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
4020 (vect_finish_stmt_generation): ...here.
4021 (vect_finish_replace_stmt): New function.
4022 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
4023 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
4025 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
4027 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4028 Alan Hayward <alan.hayward@arm.com>
4029 David Sherwood <david.sherwood@arm.com>
4031 * doc/md.texi (extract_last_@var{m}): Document.
4032 * optabs.def (extract_last_optab): New optab.
4033 * internal-fn.def (EXTRACT_LAST): New internal function.
4034 * internal-fn.c (cond_unary_direct): New macro.
4035 (expand_cond_unary_optab_fn): Likewise.
4036 (direct_cond_unary_optab_supported_p): Likewise.
4037 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
4038 loops using EXTRACT_LAST.
4039 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
4040 (extract_last_<mode>): ...this optab.
4041 (vec_extract<mode><Vel>): Update accordingly.
4043 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4044 Alan Hayward <alan.hayward@arm.com>
4045 David Sherwood <david.sherwood@arm.com>
4047 * target.def (empty_mask_is_expensive): New hook.
4048 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
4049 * doc/tm.texi: Regenerate.
4050 * targhooks.h (default_empty_mask_is_expensive): Declare.
4051 * targhooks.c (default_empty_mask_is_expensive): New function.
4052 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
4053 if the target says that empty masks are expensive.
4054 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
4056 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
4058 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4059 Alan Hayward <alan.hayward@arm.com>
4060 David Sherwood <david.sherwood@arm.com>
4062 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
4063 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
4064 (vect_use_loop_mask_for_alignment_p): New function.
4065 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
4066 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
4067 niters_skip argument. Make sure that the first niters_skip elements
4068 of the first iteration are inactive.
4069 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
4070 Update call to vect_set_loop_masks_directly.
4071 (get_misalign_in_elems): New function, split out from...
4072 (vect_gen_prolog_loop_niters): ...here.
4073 (vect_update_init_of_dr): Take a code argument that specifies whether
4074 the adjustment should be added or subtracted.
4075 (vect_update_init_of_drs): Likewise.
4076 (vect_prepare_for_masked_peels): New function.
4077 (vect_do_peeling): Skip prologue peeling if we're using a mask
4078 instead. Update call to vect_update_inits_of_drs.
4079 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4081 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
4082 alignment. Do not include the number of peeled iterations in
4083 the minimum threshold in that case.
4084 (vectorizable_induction): Adjust the start value down by
4085 LOOP_VINFO_MASK_SKIP_NITERS iterations.
4086 (vect_transform_loop): Call vect_prepare_for_masked_peels.
4087 Take the number of skipped iterations into account when calculating
4089 * tree-vect-stmts.c (vect_gen_while_not): New function.
4091 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4092 Alan Hayward <alan.hayward@arm.com>
4093 David Sherwood <david.sherwood@arm.com>
4095 * doc/sourcebuild.texi (vect_fully_masked): Document.
4096 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
4098 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
4100 (vect_analyze_loop_2): ...here. Don't check the vectorization
4101 factor against the number of loop iterations if the loop is
4104 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4105 Alan Hayward <alan.hayward@arm.com>
4106 David Sherwood <david.sherwood@arm.com>
4108 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
4109 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
4110 (dump_groups): Update accordingly.
4111 (iv_use::mem_type): New member variable.
4112 (address_p): New function.
4113 (record_use): Add a mem_type argument and initialize the new
4115 (record_group_use): Add a mem_type argument. Use address_p.
4116 Remove obsolete null checks of base_object. Update call to record_use.
4117 (find_interesting_uses_op): Update call to record_group_use.
4118 (find_interesting_uses_cond): Likewise.
4119 (find_interesting_uses_address): Likewise.
4120 (get_mem_type_for_internal_fn): New function.
4121 (find_address_like_use): Likewise.
4122 (find_interesting_uses_stmt): Try find_address_like_use before
4123 calling find_interesting_uses_op.
4124 (addr_offset_valid_p): Use the iv mem_type field as the type
4125 of the addressed memory.
4126 (add_autoinc_candidates): Likewise.
4127 (get_address_cost): Likewise.
4128 (split_small_address_groups_p): Use address_p.
4129 (split_address_groups): Likewise.
4130 (add_iv_candidate_for_use): Likewise.
4131 (autoinc_possible_for_pair): Likewise.
4132 (rewrite_groups): Likewise.
4133 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
4134 (determine_group_iv_cost): Update after split of USE_ADDRESS.
4135 (get_alias_ptr_type_for_ptr_address): New function.
4136 (rewrite_use_address): Rewrite address uses in calls that were
4137 identified by find_address_like_use.
4139 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4140 Alan Hayward <alan.hayward@arm.com>
4141 David Sherwood <david.sherwood@arm.com>
4143 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
4145 * gimple-expr.h (is_gimple_addressable: Likewise.
4146 * gimple-expr.c (is_gimple_address): Likewise.
4147 * internal-fn.c (expand_call_mem_ref): New function.
4148 (expand_mask_load_optab_fn): Use it.
4149 (expand_mask_store_optab_fn): Likewise.
4151 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4152 Alan Hayward <alan.hayward@arm.com>
4153 David Sherwood <david.sherwood@arm.com>
4155 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
4156 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
4157 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
4158 (cond_umax@var{mode}): Document.
4159 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
4160 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
4161 (cond_umin_optab, cond_umax_optab): New optabs.
4162 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
4163 (COND_IOR, COND_XOR): New internal functions.
4164 * internal-fn.h (get_conditional_internal_fn): Declare.
4165 * internal-fn.c (cond_binary_direct): New macro.
4166 (expand_cond_binary_optab_fn): Likewise.
4167 (direct_cond_binary_optab_supported_p): Likewise.
4168 (get_conditional_internal_fn): New function.
4169 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
4170 Cope with reduction statements that are vectorized as calls rather
4172 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
4173 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
4174 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
4175 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
4176 (UNSPEC_COND_EOR): New unspecs.
4177 (optab): Add mappings for them.
4178 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
4179 (sve_int_op, sve_fp_op): New int attributes.
4181 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4182 Alan Hayward <alan.hayward@arm.com>
4183 David Sherwood <david.sherwood@arm.com>
4185 * optabs.def (while_ult_optab): New optab.
4186 * doc/md.texi (while_ult@var{m}@var{n}): Document.
4187 * internal-fn.def (WHILE_ULT): New internal function.
4188 * internal-fn.h (direct_internal_fn_supported_p): New override
4189 that takes two types as argument.
4190 * internal-fn.c (while_direct): New macro.
4191 (expand_while_optab_fn): New function.
4192 (convert_optab_supported_p): Likewise.
4193 (direct_while_optab_supported_p): New macro.
4194 * wide-int.h (wi::udiv_ceil): New function.
4195 * tree-vectorizer.h (rgroup_masks): New structure.
4196 (vec_loop_masks): New typedef.
4197 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
4199 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
4200 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
4201 (vect_max_vf): New function.
4202 (slpeel_make_loop_iterate_ntimes): Delete.
4203 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
4204 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
4205 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
4206 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
4207 internal-fn.h, stor-layout.h and optabs-query.h.
4208 (vect_set_loop_mask): New function.
4209 (add_preheader_seq): Likewise.
4210 (add_header_seq): Likewise.
4211 (interleave_supported_p): Likewise.
4212 (vect_maybe_permute_loop_masks): Likewise.
4213 (vect_set_loop_masks_directly): Likewise.
4214 (vect_set_loop_condition_masked): Likewise.
4215 (vect_set_loop_condition_unmasked): New function, split out from
4216 slpeel_make_loop_iterate_ntimes.
4217 (slpeel_make_loop_iterate_ntimes): Rename to..
4218 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
4219 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
4220 (vect_do_peeling): Update call accordingly.
4221 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
4223 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4224 mask_compare_type, can_fully_mask_p and fully_masked_p.
4225 (release_vec_loop_masks): New function.
4226 (_loop_vec_info): Use it to free the loop masks.
4227 (can_produce_all_loop_masks_p): New function.
4228 (vect_get_max_nscalars_per_iter): Likewise.
4229 (vect_verify_full_masking): Likewise.
4230 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
4231 retries, and free the mask rgroups before retrying. Check loop-wide
4232 reasons for disallowing fully-masked loops. Make the final decision
4233 about whether use a fully-masked loop or not.
4234 (vect_estimate_min_profitable_iters): Do not assume that peeling
4235 for the number of iterations will be needed for fully-masked loops.
4236 (vectorizable_reduction): Disable fully-masked loops.
4237 (vectorizable_live_operation): Likewise.
4238 (vect_halve_mask_nunits): New function.
4239 (vect_double_mask_nunits): Likewise.
4240 (vect_record_loop_mask): Likewise.
4241 (vect_get_loop_mask): Likewise.
4242 (vect_transform_loop): Handle the case in which the final loop
4243 iteration might handle a partial vector. Call vect_set_loop_condition
4244 instead of slpeel_make_loop_iterate_ntimes.
4245 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
4246 (check_load_store_masking): New function.
4247 (prepare_load_store_mask): Likewise.
4248 (vectorizable_store): Handle fully-masked loops.
4249 (vectorizable_load): Likewise.
4250 (supportable_widening_operation): Use vect_halve_mask_nunits for
4252 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
4253 (vect_gen_while): New function.
4254 * config/aarch64/aarch64.md (umax<mode>3): New expander.
4255 (aarch64_uqdec<mode>): New insn.
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 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
4262 (reduc_xor_scal_optab): New optabs.
4263 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
4264 (reduc_xor_scal_@var{m}): Document.
4265 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
4266 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
4268 * fold-const-call.c (fold_const_call): Handle them.
4269 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
4270 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
4271 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
4272 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
4273 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
4274 (UNSPEC_XORV): New unspecs.
4275 (optab): Add entries for them.
4276 (BITWISEV): New int iterator.
4277 (bit_reduc_op): New int attributes.
4279 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4280 Alan Hayward <alan.hayward@arm.com>
4281 David Sherwood <david.sherwood@arm.com>
4283 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
4284 * internal-fn.def (VEC_SHL_INSERT): New internal function.
4285 * optabs.def (vec_shl_insert_optab): New optab.
4286 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
4287 (duplicate_and_interleave): Likewise.
4288 * tree-vect-loop.c: Include internal-fn.h.
4289 (neutral_op_for_slp_reduction): New function, split out from
4290 get_initial_defs_for_reduction.
4291 (get_initial_def_for_reduction): Handle option 2 for variable-length
4292 vectors by loading the neutral value into a vector and then shifting
4293 the initial value into element 0.
4294 (get_initial_defs_for_reduction): Replace the code argument with
4295 the neutral value calculated by neutral_op_for_slp_reduction.
4296 Use gimple_build_vector for constant-length vectors.
4297 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
4298 but the first group_size elements have a neutral value.
4299 Use duplicate_and_interleave otherwise.
4300 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
4301 Update call to get_initial_defs_for_reduction. Handle SLP
4302 reductions for variable-length vectors by creating one vector
4303 result for each scalar result, with the elements associated
4304 with other scalar results stubbed out with the neutral value.
4305 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
4306 Require IFN_VEC_SHL_INSERT for double reductions on
4307 variable-length vectors, or SLP reductions that have
4308 a neutral value. Require can_duplicate_and_interleave_p
4309 support for variable-length unchained SLP reductions if there
4310 is no neutral value, such as for MIN/MAX reductions. Also require
4311 the number of vector elements to be a multiple of the number of
4312 SLP statements when doing variable-length unchained SLP reductions.
4313 Update call to vect_create_epilog_for_reduction.
4314 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
4315 and remove initial values.
4316 (duplicate_and_interleave): Make public.
4317 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
4318 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
4320 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4321 Alan Hayward <alan.hayward@arm.com>
4322 David Sherwood <david.sherwood@arm.com>
4324 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
4325 (can_duplicate_and_interleave_p): New function.
4326 (vect_get_and_check_slp_defs): Take the vector of statements
4327 rather than just the current one. Remove excess parentheses.
4328 Restriction rejectinon of vect_constant_def and vect_external_def
4329 for variable-length vectors to boolean types, or types for which
4330 can_duplicate_and_interleave_p is false.
4331 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
4332 (duplicate_and_interleave): New function.
4333 (vect_get_constant_vectors): Use gimple_build_vector for
4334 constant-length vectors and suitable variable-length constant
4335 vectors. Use duplicate_and_interleave for other variable-length
4336 vectors. Don't defer the update when inserting new statements.
4338 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4339 Alan Hayward <alan.hayward@arm.com>
4340 David Sherwood <david.sherwood@arm.com>
4342 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
4343 min_profitable_iters doesn't go negative.
4345 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4346 Alan Hayward <alan.hayward@arm.com>
4347 David Sherwood <david.sherwood@arm.com>
4349 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
4350 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
4351 * optabs.def (vec_mask_load_lanes_optab): New optab.
4352 (vec_mask_store_lanes_optab): Likewise.
4353 * internal-fn.def (MASK_LOAD_LANES): New internal function.
4354 (MASK_STORE_LANES): Likewise.
4355 * internal-fn.c (mask_load_lanes_direct): New macro.
4356 (mask_store_lanes_direct): Likewise.
4357 (expand_mask_load_optab_fn): Handle masked operations.
4358 (expand_mask_load_lanes_optab_fn): New macro.
4359 (expand_mask_store_optab_fn): Handle masked operations.
4360 (expand_mask_store_lanes_optab_fn): New macro.
4361 (direct_mask_load_lanes_optab_supported_p): Likewise.
4362 (direct_mask_store_lanes_optab_supported_p): Likewise.
4363 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
4365 (vect_load_lanes_supported): Likewise.
4366 * tree-vect-data-refs.c (strip_conversion): New function.
4367 (can_group_stmts_p): Likewise.
4368 (vect_analyze_data_ref_accesses): Use it instead of checking
4369 for a pair of assignments.
4370 (vect_store_lanes_supported): Take a masked_p parameter.
4371 (vect_load_lanes_supported): Likewise.
4372 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
4373 vect_store_lanes_supported and vect_load_lanes_supported.
4374 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
4375 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
4376 parameter. Don't allow gaps for masked accesses.
4377 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
4378 and vect_load_lanes_supported.
4379 (get_load_store_type): Take a masked_p parameter and update
4380 call to get_group_load_store_type.
4381 (vectorizable_store): Update call to get_load_store_type.
4382 Handle IFN_MASK_STORE_LANES.
4383 (vectorizable_load): Update call to get_load_store_type.
4384 Handle IFN_MASK_LOAD_LANES.
4386 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4387 Alan Hayward <alan.hayward@arm.com>
4388 David Sherwood <david.sherwood@arm.com>
4390 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
4392 * config/aarch64/aarch64-protos.h
4393 (aarch64_sve_struct_memory_operand_p): Declare.
4394 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
4395 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
4396 (VPRED, vpred): Handle SVE structure modes.
4397 * config/aarch64/constraints.md (Utx): New constraint.
4398 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
4399 (aarch64_sve_struct_nonimmediate_operand): New predicates.
4400 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
4401 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
4402 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
4403 structure modes. Split into pieces after RA.
4404 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
4405 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
4407 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
4408 SVE structure modes.
4409 (aarch64_classify_address): Likewise.
4410 (sizetochar): Move earlier in file.
4411 (aarch64_print_operand): Handle SVE register lists.
4412 (aarch64_array_mode): New function.
4413 (aarch64_sve_struct_memory_operand_p): Likewise.
4414 (TARGET_ARRAY_MODE): Redefine.
4416 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4417 Alan Hayward <alan.hayward@arm.com>
4418 David Sherwood <david.sherwood@arm.com>
4420 * target.def (array_mode): New target hook.
4421 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
4422 * doc/tm.texi: Regenerate.
4423 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
4424 * hooks.c (hook_optmode_mode_uhwi_none): New function.
4425 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
4427 * stor-layout.c (mode_for_array): Likewise. Support polynomial
4430 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4431 Alan Hayward <alan.hayward@arm.com>
4432 David Sherwood <david.sherwood@arm.com>
4434 * fold-const.c (fold_binary_loc): Check the argument types
4435 rather than the result type when testing for a vector operation.
4437 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4439 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
4440 * doc/tm.texi: Regenerate.
4442 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4443 Alan Hayward <alan.hayward@arm.com>
4444 David Sherwood <david.sherwood@arm.com>
4446 * doc/invoke.texi (-msve-vector-bits=): Document new option.
4447 (sve): Document new AArch64 extension.
4448 * doc/md.texi (w): Extend the description of the AArch64
4449 constraint to include SVE vectors.
4450 (Upl, Upa): Document new AArch64 predicate constraints.
4451 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
4453 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
4454 (msve-vector-bits=): New option.
4455 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
4456 SVE when these are disabled.
4457 (sve): New extension.
4458 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
4459 modes. Adjust their number of units based on aarch64_sve_vg.
4460 (MAX_BITSIZE_MODE_ANY_MODE): Define.
4461 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
4462 aarch64_addr_query_type.
4463 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
4464 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
4465 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
4466 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
4467 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
4468 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
4469 (aarch64_simd_imm_zero_p): Delete.
4470 (aarch64_check_zero_based_sve_index_immediate): Declare.
4471 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4472 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4473 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4474 (aarch64_sve_float_mul_immediate_p): Likewise.
4475 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4477 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
4478 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
4479 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
4480 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
4481 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
4482 (aarch64_regmode_natural_size): Likewise.
4483 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
4484 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
4486 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
4487 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
4488 for VG and the SVE predicate registers.
4489 (V_ALIASES): Add a "z"-prefixed alias.
4490 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
4491 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
4492 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
4493 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
4494 (REG_CLASS_NAMES): Add entries for them.
4495 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
4496 and the predicate registers.
4497 (aarch64_sve_vg): Declare.
4498 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
4499 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
4500 (REGMODE_NATURAL_SIZE): Define.
4501 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
4503 * config/aarch64/aarch64.c: Include cfgrtl.h.
4504 (simd_immediate_info): Add a constructor for series vectors,
4505 and an associated step field.
4506 (aarch64_sve_vg): New variable.
4507 (aarch64_dbx_register_number): Handle VG and the predicate registers.
4508 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
4509 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
4510 (VEC_ANY_DATA, VEC_STRUCT): New constants.
4511 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
4512 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
4513 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
4514 (aarch64_get_mask_mode): New functions.
4515 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
4516 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
4517 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
4518 predicate modes and predicate registers. Explicitly restrict
4519 GPRs to modes of 16 bytes or smaller. Only allow FP registers
4520 to store a vector mode if it is recognized by
4521 aarch64_classify_vector_mode.
4522 (aarch64_regmode_natural_size): New function.
4523 (aarch64_hard_regno_caller_save_mode): Return the original mode
4525 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
4526 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
4527 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
4528 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
4530 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
4531 does not overlap dest if the function is frame-related. Handle
4533 (aarch64_split_add_offset): New function.
4534 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
4535 them aarch64_add_offset.
4536 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
4537 and update call to aarch64_sub_sp.
4538 (aarch64_add_cfa_expression): New function.
4539 (aarch64_expand_prologue): Pass extra temporary registers to the
4540 functions above. Handle the case in which we need to emit new
4541 DW_CFA_expressions for registers that were originally saved
4542 relative to the stack pointer, but now have to be expressed
4543 relative to the frame pointer.
4544 (aarch64_output_mi_thunk): Pass extra temporary registers to the
4546 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
4547 IP0 and IP1 values for SVE frames.
4548 (aarch64_expand_vec_series): New function.
4549 (aarch64_expand_sve_widened_duplicate): Likewise.
4550 (aarch64_expand_sve_const_vector): Likewise.
4551 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
4552 Handle SVE constants. Use emit_move_insn to move a force_const_mem
4553 into the register, rather than emitting a SET directly.
4554 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
4555 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
4556 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
4557 (offset_9bit_signed_scaled_p): New functions.
4558 (aarch64_replicate_bitmask_imm): New function.
4559 (aarch64_bitmask_imm): Use it.
4560 (aarch64_cannot_force_const_mem): Reject expressions involving
4561 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
4562 (aarch64_classify_index): Handle SVE indices, by requiring
4563 a plain register index with a scale that matches the element size.
4564 (aarch64_classify_address): Handle SVE addresses. Assert that
4565 the mode of the address is VOIDmode or an integer mode.
4566 Update call to aarch64_classify_symbol.
4567 (aarch64_classify_symbolic_expression): Update call to
4568 aarch64_classify_symbol.
4569 (aarch64_const_vec_all_in_range_p): New function.
4570 (aarch64_print_vector_float_operand): Likewise.
4571 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
4572 "vN" for FP registers with SVE modes. Handle (const ...) vectors
4573 and the FP immediates 1.0 and 0.5.
4574 (aarch64_print_address_internal): Handle SVE addresses.
4575 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
4576 (aarch64_regno_regclass): Handle predicate registers.
4577 (aarch64_secondary_reload): Handle big-endian reloads of SVE
4579 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
4580 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
4581 (aarch64_convert_sve_vector_bits): New function.
4582 (aarch64_override_options): Use it to handle -msve-vector-bits=.
4583 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4585 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
4586 Handle SVE vector and predicate modes. Accept VL-based constants
4587 that need only one temporary register, and VL offsets that require
4588 no temporary registers.
4589 (aarch64_conditional_register_usage): Mark the predicate registers
4590 as fixed if SVE isn't available.
4591 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
4592 Return true for SVE vector and predicate modes.
4593 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
4594 rather than an unsigned int. Handle SVE modes.
4595 (aarch64_preferred_simd_mode): Update call accordingly. Handle
4597 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
4599 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4600 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4601 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4602 (aarch64_sve_float_mul_immediate_p): New functions.
4603 (aarch64_sve_valid_immediate): New function.
4604 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
4605 Explicitly reject structure modes. Check for INDEX constants.
4606 Handle PTRUE and PFALSE constants.
4607 (aarch64_check_zero_based_sve_index_immediate): New function.
4608 (aarch64_simd_imm_zero_p): Delete.
4609 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
4610 vector modes. Accept constants in the range of CNT[BHWD].
4611 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
4612 ask for an Advanced SIMD mode.
4613 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
4614 (aarch64_simd_vector_alignment): Handle SVE predicates.
4615 (aarch64_vectorize_preferred_vector_alignment): New function.
4616 (aarch64_simd_vector_alignment_reachable): Use it instead of
4618 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
4619 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
4621 (MAX_VECT_LEN): Delete.
4622 (expand_vec_perm_d): Add a vec_flags field.
4623 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
4624 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
4625 (aarch64_evpc_ext): Don't apply a big-endian lane correction
4627 (aarch64_evpc_rev): Rename to...
4628 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
4629 (aarch64_evpc_rev_global): New function.
4630 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
4631 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
4633 (aarch64_evpc_sve_tbl): New function.
4634 (aarch64_expand_vec_perm_const_1): Update after rename of
4635 aarch64_evpc_rev. Handle SVE permutes too, trying
4636 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
4637 than aarch64_evpc_tbl.
4638 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
4639 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
4640 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
4641 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
4642 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
4643 (aarch64_expand_sve_vcond): New functions.
4644 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
4645 of aarch64_vector_mode_p.
4646 (aarch64_dwarf_poly_indeterminate_value): New function.
4647 (aarch64_compute_pressure_classes): Likewise.
4648 (aarch64_can_change_mode_class): Likewise.
4649 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
4650 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
4651 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
4652 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
4653 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
4654 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
4655 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
4656 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
4658 (Dn, Dl, Dr): Accept const as well as const_vector.
4659 (Dz): Likewise. Compare against CONST0_RTX.
4660 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
4661 of "vector" where appropriate.
4662 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
4663 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
4664 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
4665 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
4666 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
4667 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
4668 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
4669 (v_int_equiv): Extend to SVE modes.
4670 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
4672 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
4673 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4674 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4675 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4676 (SVE_COND_FP_CMP): New int iterators.
4677 (perm_hilo): Handle the new unpack unspecs.
4678 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4680 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4681 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4682 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4683 (aarch64_equality_operator, aarch64_constant_vector_operand)
4684 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4685 (aarch64_sve_nonimmediate_operand): Likewise.
4686 (aarch64_sve_general_operand): Likewise.
4687 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4688 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4689 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4690 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4691 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4692 (aarch64_sve_float_arith_immediate): Likewise.
4693 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4694 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4695 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4696 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4697 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4698 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4699 (aarch64_sve_float_arith_operand): Likewise.
4700 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4701 (aarch64_sve_float_mul_operand): Likewise.
4702 (aarch64_sve_vec_perm_operand): Likewise.
4703 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4704 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4705 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4706 as well as const_vector.
4707 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4708 in file. Use CONST0_RTX and CONSTM1_RTX.
4709 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4710 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4711 Use aarch64_simd_imm_zero.
4712 * config/aarch64/aarch64-sve.md: New file.
4713 * config/aarch64/aarch64.md: Include it.
4714 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4715 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4716 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4717 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4718 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4719 (sve): New attribute.
4720 (enabled): Disable instructions with the sve attribute unless
4722 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4723 aarch64_expand_mov_immediate.
4724 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4725 CNT[BHSD] immediates.
4726 (movti): Split CONST_POLY_INT moves into two halves.
4727 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4728 Split additions that need a temporary here if the destination
4729 is the stack pointer.
4730 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4731 (*add<mode>3_poly_1): New instruction.
4732 (set_clobber_cc): New expander.
4734 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4736 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4737 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4738 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4739 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4740 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4741 Change innermode from fixed_mode_size to machine_mode.
4742 (simplify_subreg): Update call accordingly. Handle a constant-sized
4743 subreg of a variable-length CONST_VECTOR.
4745 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4746 Alan Hayward <alan.hayward@arm.com>
4747 David Sherwood <david.sherwood@arm.com>
4749 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4750 (add_offset_to_base): New function, split out from...
4751 (create_mem_ref): ...here. When handling a scale other than 1,
4752 check first whether the address is valid without the offset.
4753 Add it into the base if so, leaving the index and scale as-is.
4755 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4758 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4759 fold_for_warn before checking if arg2 is INTEGER_CST.
4761 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4763 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4764 (store_multiple_operation): Delete.
4765 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4766 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4767 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4768 guarded by TARGET_STRING.
4769 (rs6000_output_load_multiple): Delete.
4770 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4771 OPTION_MASK_STRING / TARGET_STRING handling.
4772 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4773 (const rs6000_opt_masks) <"string">: Change mask to 0.
4774 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4775 (MASK_STRING): Delete.
4776 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4778 (load_multiple): Delete.
4785 (store_multiple): Delete.
4792 (movmemsi_8reg): Delete.
4793 (corresponding unnamed define_insn): Delete.
4794 (movmemsi_6reg): Delete.
4795 (corresponding unnamed define_insn): Delete.
4796 (movmemsi_4reg): Delete.
4797 (corresponding unnamed define_insn): Delete.
4798 (movmemsi_2reg): Delete.
4799 (corresponding unnamed define_insn): Delete.
4800 (movmemsi_1reg): Delete.
4801 (corresponding unnamed define_insn): Delete.
4802 * config/rs6000/rs6000.opt (mno-string): New.
4803 (mstring): Replace by deprecation warning stub.
4804 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4806 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4808 * regrename.c (regrename_do_replace): If replacing the same
4809 reg multiple times, try to reuse last created gen_raw_REG.
4812 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4813 main to workaround a bug in GDB.
4815 2018-01-12 Tom de Vries <tom@codesourcery.com>
4818 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4820 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4822 PR rtl-optimization/80481
4823 * ira-color.c (get_cap_member): New function.
4824 (allocnos_conflict_by_live_ranges_p): Use it.
4825 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4826 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4828 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4831 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4832 (*saddl_se_1): Ditto.
4834 (*ssubl_se_1): Ditto.
4836 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4838 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4839 rather than wi::to_widest for DR_INITs.
4840 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4841 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4842 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4844 (vect_analyze_group_access_1): Note that here.
4846 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4848 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4849 polynomial type sizes.
4851 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4853 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4854 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4855 (gimple_add_tmp_var): Likewise.
4857 2018-01-12 Martin Liska <mliska@suse.cz>
4859 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4860 (gimple_alloc_sizes): Likewise.
4861 (dump_gimple_statistics): Use PRIu64 in printf format.
4862 * gimple.h: Change uint64_t to int.
4864 2018-01-12 Martin Liska <mliska@suse.cz>
4866 * tree-core.h: Use uint64_t instead of int.
4867 * tree.c (tree_node_counts): Likewise.
4868 (tree_node_sizes): Likewise.
4869 (dump_tree_statistics): Use PRIu64 in printf format.
4871 2018-01-12 Martin Liska <mliska@suse.cz>
4873 * Makefile.in: As qsort_chk is implemented in vec.c, add
4874 vec.o to linkage of gencfn-macros.
4875 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4876 passing the info to record_node_allocation_statistics.
4877 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4879 * ggc-common.c (struct ggc_usage): Add operator== and use
4880 it in operator< and compare function.
4881 * mem-stats.h (struct mem_usage): Likewise.
4882 * vec.c (struct vec_usage): Remove operator< and compare
4883 function. Can be simply inherited.
4885 2018-01-12 Martin Jambor <mjambor@suse.cz>
4888 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4889 * tree-ssa-math-opts.c: Include domwalk.h.
4890 (convert_mult_to_fma_1): New function.
4891 (fma_transformation_info): New type.
4892 (fma_deferring_state): Likewise.
4893 (cancel_fma_deferring): New function.
4894 (result_of_phi): Likewise.
4895 (last_fma_candidate_feeds_initial_phi): Likewise.
4896 (convert_mult_to_fma): Added deferring logic, split actual
4897 transformation to convert_mult_to_fma_1.
4898 (math_opts_dom_walker): New type.
4899 (math_opts_dom_walker::after_dom_children): New method, body moved
4900 here from pass_optimize_widening_mul::execute, added deferring logic
4902 (pass_optimize_widening_mul::execute): Moved most of code to
4903 math_opts_dom_walker::after_dom_children.
4904 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4905 * config/i386/i386.c (ix86_option_override_internal): Added
4906 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4908 2018-01-12 Richard Biener <rguenther@suse.de>
4911 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4912 inline instance vars.
4914 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4917 * config/rx/rx.c (rx_is_restricted_memory_address):
4920 2018-01-12 Richard Biener <rguenther@suse.de>
4922 PR tree-optimization/80846
4923 * target.def (split_reduction): New target hook.
4924 * targhooks.c (default_split_reduction): New function.
4925 * targhooks.h (default_split_reduction): Declare.
4926 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4927 target requests first reduce vectors by combining low and high
4929 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4930 (get_vectype_for_scalar_type_and_size): Export.
4931 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4932 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4933 * doc/tm.texi: Regenerate.
4934 * config/i386/i386.c (ix86_split_reduction): Implement
4935 TARGET_VECTORIZE_SPLIT_REDUCTION.
4937 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4940 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4941 in PIC mode except for TARGET_VXWORKS_RTP.
4942 * config/sparc/sparc.c: Include cfgrtl.h.
4943 (TARGET_INIT_PIC_REG): Define.
4944 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4945 (sparc_pic_register_p): New predicate.
4946 (sparc_legitimate_address_p): Use it.
4947 (sparc_legitimize_pic_address): Likewise.
4948 (sparc_delegitimize_address): Likewise.
4949 (sparc_mode_dependent_address_p): Likewise.
4950 (gen_load_pcrel_sym): Remove 4th parameter.
4951 (load_got_register): Adjust call to above. Remove obsolete stuff.
4952 (sparc_expand_prologue): Do not call load_got_register here.
4953 (sparc_flat_expand_prologue): Likewise.
4954 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4955 (sparc_use_pseudo_pic_reg): New function.
4956 (sparc_init_pic_reg): Likewise.
4957 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4958 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4960 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4962 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4963 Add item for branch_cost.
4965 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4967 PR rtl-optimization/83565
4968 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4969 not extend the result to a larger mode for rotate operations.
4970 (num_sign_bit_copies1): Likewise.
4972 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4975 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4977 Use values-Xc.o for -pedantic.
4978 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4980 2018-01-12 Martin Liska <mliska@suse.cz>
4983 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4985 (possible_polymorphic_call_targets): Use it.
4986 (ipa_devirt): Likewise.
4988 2018-01-12 Martin Liska <mliska@suse.cz>
4990 * profile-count.h (enum profile_quality): Use 0 as invalid
4991 enum value of profile_quality.
4993 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4995 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4996 -mext-string options.
4998 2018-01-12 Richard Biener <rguenther@suse.de>
5000 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
5001 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
5002 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
5004 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
5006 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
5008 * configure.ac (--with-long-double-format): Add support for the
5009 configuration option to change the default long double format on
5011 * config.gcc (powerpc*-linux*-*): Likewise.
5012 * configure: Regenerate.
5013 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
5014 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
5015 used without modification.
5017 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5019 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
5020 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
5021 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
5022 MISC_BUILTIN_SPEC_BARRIER.
5023 (rs6000_init_builtins): Likewise.
5024 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
5026 (speculation_barrier): New define_insn.
5027 * doc/extend.texi: Document __builtin_speculation_barrier.
5029 2018-01-11 Jakub Jelinek <jakub@redhat.com>
5032 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
5033 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
5034 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
5036 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
5037 integral modes instead of "ss" and "sd".
5038 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
5039 vectors with 32-bit and 64-bit elements.
5040 (vecdupssescalarmodesuffix): New mode attribute.
5041 (vec_dup<mode>): Use it.
5043 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
5046 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
5047 frame if argument is passed on stack.
5049 2018-01-11 Jakub Jelinek <jakub@redhat.com>
5052 * ree.c (combine_reaching_defs): Optimize also
5053 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
5054 reg2=any_extend(exp); reg1=reg2;, formatting fix.
5056 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
5059 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
5061 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
5064 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
5065 after they are computed.
5067 2018-01-11 Bin Cheng <bin.cheng@arm.com>
5069 PR tree-optimization/83695
5070 * gimple-loop-linterchange.cc
5071 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
5072 reset cached scev information after interchange.
5073 (pass_linterchange::execute): Remove call to scev_reset_htab.
5075 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5077 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
5078 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
5079 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
5080 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
5081 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
5082 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
5083 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
5084 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
5085 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
5086 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
5087 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
5088 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
5089 (V_lane_reg): Likewise.
5090 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
5092 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
5093 (vfmal_lane_low<mode>_intrinsic,
5094 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
5095 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
5096 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
5097 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
5098 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
5099 vfmsl_lane_high<mode>_intrinsic): New define_insns.
5101 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5103 * config/arm/arm-cpus.in (fp16fml): New feature.
5104 (ALL_SIMD): Add fp16fml.
5105 (armv8.2-a): Add fp16fml as an option.
5106 (armv8.3-a): Likewise.
5107 (armv8.4-a): Add fp16fml as part of fp16.
5108 * config/arm/arm.h (TARGET_FP16FML): Define.
5109 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
5111 * config/arm/arm-modes.def (V2HF): Define.
5112 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
5113 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
5114 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
5115 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
5116 vfmsl_low, vfmsl_high): New set of builtins.
5117 * config/arm/iterators.md (PLUSMINUS): New code iterator.
5118 (vfml_op): New code attribute.
5119 (VFMLHALVES): New int iterator.
5120 (VFML, VFMLSEL): New mode attributes.
5121 (V_reg): Define mapping for V2HF.
5122 (V_hi, V_lo): New mode attributes.
5123 (VF_constraint): Likewise.
5124 (vfml_half, vfml_half_selector): New int attributes.
5125 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
5127 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
5128 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
5130 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
5131 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
5132 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
5133 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
5135 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
5136 Document new effective target and option set.
5138 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5140 * config/arm/arm-cpus.in (armv8_4): New feature.
5141 (ARMv8_4a): New fgroup.
5142 (armv8.4-a): New arch.
5143 * config/arm/arm-tables.opt: Regenerate.
5144 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
5145 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
5146 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
5147 Add matching rules for -march=armv8.4-a and extensions.
5148 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
5150 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
5153 * config/rx/rx.md (BW): New mode attribute.
5154 (sync_lock_test_and_setsi): Add mode suffix to insn output.
5156 2018-01-11 Richard Biener <rguenther@suse.de>
5158 PR tree-optimization/83435
5159 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
5160 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
5161 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
5163 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5164 Alan Hayward <alan.hayward@arm.com>
5165 David Sherwood <david.sherwood@arm.com>
5167 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
5169 (aarch64_classify_address): Initialize it. Track polynomial offsets.
5170 (aarch64_print_address_internal): Use it to check for a zero offset.
5172 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5173 Alan Hayward <alan.hayward@arm.com>
5174 David Sherwood <david.sherwood@arm.com>
5176 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
5177 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
5178 Return a poly_int64 rather than a HOST_WIDE_INT.
5179 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
5180 rather than a HOST_WIDE_INT.
5181 * config/aarch64/aarch64.h (aarch64_frame): Protect with
5182 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
5183 hard_fp_offset, frame_size, initial_adjust, callee_offset and
5184 final_offset from HOST_WIDE_INT to poly_int64.
5185 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
5186 to_constant when getting the number of units in an Advanced SIMD
5188 (aarch64_builtin_vectorized_function): Check for a constant number
5190 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
5192 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
5193 attribute instead of GET_MODE_NUNITS.
5194 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
5195 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
5196 GET_MODE_SIZE for fixed-size registers.
5197 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
5198 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
5199 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
5200 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
5201 (aarch64_print_operand, aarch64_print_address_internal)
5202 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
5203 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
5204 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
5205 Handle polynomial GET_MODE_SIZE.
5206 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
5207 wider than SImode without modification.
5208 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
5209 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
5210 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
5211 passing and returning SVE modes.
5212 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
5213 rather than GEN_INT.
5214 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
5215 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
5216 (aarch64_allocate_and_probe_stack_space): Likewise.
5217 (aarch64_layout_frame): Cope with polynomial offsets.
5218 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
5219 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
5221 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
5222 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
5223 poly_int64 rather than a HOST_WIDE_INT.
5224 (aarch64_get_separate_components, aarch64_process_components)
5225 (aarch64_expand_prologue, aarch64_expand_epilogue)
5226 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
5227 (aarch64_anchor_offset): New function, split out from...
5228 (aarch64_legitimize_address): ...here.
5229 (aarch64_builtin_vectorization_cost): Handle polynomial
5230 TYPE_VECTOR_SUBPARTS.
5231 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
5233 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
5234 number of elements from the PARALLEL rather than the mode.
5235 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
5236 rather than GET_MODE_BITSIZE.
5237 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
5238 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
5239 (aarch64_expand_vec_perm_const_1): Handle polynomial
5240 d->perm.length () and d->perm elements.
5241 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
5242 Apply to_constant to d->perm elements.
5243 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
5244 polynomial CONST_VECTOR_NUNITS.
5245 (aarch64_move_pointer): Take amount as a poly_int64 rather
5247 (aarch64_progress_pointer): Avoid temporary variable.
5248 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
5249 the mode attribute instead of GET_MODE.
5251 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5252 Alan Hayward <alan.hayward@arm.com>
5253 David Sherwood <david.sherwood@arm.com>
5255 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
5256 x exists before using it.
5257 (aarch64_add_constant_internal): Rename to...
5258 (aarch64_add_offset_1): ...this. Replace regnum with separate
5259 src and dest rtxes. Handle the case in which they're different,
5260 including when the offset is zero. Replace scratchreg with an rtx.
5261 Use 2 additions if there is no spare register into which we can
5262 move a 16-bit constant.
5263 (aarch64_add_constant): Delete.
5264 (aarch64_add_offset): Replace reg with separate src and dest
5265 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
5266 Use aarch64_add_offset_1.
5267 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
5268 an rtx rather than an int. Take the delta as a poly_int64
5269 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
5270 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
5271 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
5272 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
5273 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
5275 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
5276 aarch64_add_constant.
5278 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5280 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
5281 Use scalar_float_mode.
5283 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5285 * config/aarch64/aarch64-simd.md
5286 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
5287 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
5288 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
5289 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
5290 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
5291 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
5292 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
5293 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
5294 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
5295 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
5297 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5300 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
5301 targ_options->x_arm_arch_string is non NULL.
5303 2018-01-11 Tamar Christina <tamar.christina@arm.com>
5305 * config/aarch64/aarch64.h
5306 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
5308 2018-01-11 Sudakshina Das <sudi.das@arm.com>
5311 * expmed.c (emit_store_flag_force): Swap if const op0
5312 and change VOIDmode to mode of op0.
5314 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5316 PR rtl-optimization/83761
5317 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
5318 than bytes to mode_for_size.
5320 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5323 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
5324 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
5327 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5330 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
5331 when in layout mode.
5332 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
5333 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
5336 2018-01-10 Michael Collison <michael.collison@arm.com>
5338 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
5339 * config/aarch64/aarch64-option-extension.def: Add
5340 AARCH64_OPT_EXTENSION of 'fp16fml'.
5341 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5342 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
5343 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
5344 * config/aarch64/constraints.md (Ui7): New constraint.
5345 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
5346 (VFMLA_SEL_W): Ditto.
5349 (VFMLA16_LOW): New int iterator.
5350 (VFMLA16_HIGH): Ditto.
5351 (UNSPEC_FMLAL): New unspec.
5352 (UNSPEC_FMLSL): Ditto.
5353 (UNSPEC_FMLAL2): Ditto.
5354 (UNSPEC_FMLSL2): Ditto.
5355 (f16mac): New code attribute.
5356 * config/aarch64/aarch64-simd-builtins.def
5357 (aarch64_fmlal_lowv2sf): Ditto.
5358 (aarch64_fmlsl_lowv2sf): Ditto.
5359 (aarch64_fmlalq_lowv4sf): Ditto.
5360 (aarch64_fmlslq_lowv4sf): Ditto.
5361 (aarch64_fmlal_highv2sf): Ditto.
5362 (aarch64_fmlsl_highv2sf): Ditto.
5363 (aarch64_fmlalq_highv4sf): Ditto.
5364 (aarch64_fmlslq_highv4sf): Ditto.
5365 (aarch64_fmlal_lane_lowv2sf): Ditto.
5366 (aarch64_fmlsl_lane_lowv2sf): Ditto.
5367 (aarch64_fmlal_laneq_lowv2sf): Ditto.
5368 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
5369 (aarch64_fmlalq_lane_lowv4sf): Ditto.
5370 (aarch64_fmlsl_lane_lowv4sf): Ditto.
5371 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
5372 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
5373 (aarch64_fmlal_lane_highv2sf): Ditto.
5374 (aarch64_fmlsl_lane_highv2sf): Ditto.
5375 (aarch64_fmlal_laneq_highv2sf): Ditto.
5376 (aarch64_fmlsl_laneq_highv2sf): Ditto.
5377 (aarch64_fmlalq_lane_highv4sf): Ditto.
5378 (aarch64_fmlsl_lane_highv4sf): Ditto.
5379 (aarch64_fmlalq_laneq_highv4sf): Ditto.
5380 (aarch64_fmlsl_laneq_highv4sf): Ditto.
5381 * config/aarch64/aarch64-simd.md:
5382 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
5383 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5384 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
5385 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5386 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
5387 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
5388 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
5389 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
5390 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
5391 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
5392 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
5393 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
5394 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
5395 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
5396 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
5397 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
5398 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
5399 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
5400 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
5401 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
5402 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
5403 (vfmlsl_low_u32): Ditto.
5404 (vfmlalq_low_u32): Ditto.
5405 (vfmlslq_low_u32): Ditto.
5406 (vfmlal_high_u32): Ditto.
5407 (vfmlsl_high_u32): Ditto.
5408 (vfmlalq_high_u32): Ditto.
5409 (vfmlslq_high_u32): Ditto.
5410 (vfmlal_lane_low_u32): Ditto.
5411 (vfmlsl_lane_low_u32): Ditto.
5412 (vfmlal_laneq_low_u32): Ditto.
5413 (vfmlsl_laneq_low_u32): Ditto.
5414 (vfmlalq_lane_low_u32): Ditto.
5415 (vfmlslq_lane_low_u32): Ditto.
5416 (vfmlalq_laneq_low_u32): Ditto.
5417 (vfmlslq_laneq_low_u32): Ditto.
5418 (vfmlal_lane_high_u32): Ditto.
5419 (vfmlsl_lane_high_u32): Ditto.
5420 (vfmlal_laneq_high_u32): Ditto.
5421 (vfmlsl_laneq_high_u32): Ditto.
5422 (vfmlalq_lane_high_u32): Ditto.
5423 (vfmlslq_lane_high_u32): Ditto.
5424 (vfmlalq_laneq_high_u32): Ditto.
5425 (vfmlslq_laneq_high_u32): Ditto.
5426 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
5427 (AARCH64_FL_FOR_ARCH8_4): New.
5428 (AARCH64_ISA_F16FML): New ISA flag.
5429 (TARGET_F16FML): New feature flag for fp16fml.
5430 (doc/invoke.texi): Document new fp16fml option.
5432 2018-01-10 Michael Collison <michael.collison@arm.com>
5434 * config/aarch64/aarch64-builtins.c:
5435 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
5436 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5437 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
5438 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
5439 (AARCH64_ISA_SHA3): New ISA flag.
5440 (TARGET_SHA3): New feature flag for sha3.
5441 * config/aarch64/iterators.md (sha512_op): New int attribute.
5442 (CRYPTO_SHA512): New int iterator.
5443 (UNSPEC_SHA512H): New unspec.
5444 (UNSPEC_SHA512H2): Ditto.
5445 (UNSPEC_SHA512SU0): Ditto.
5446 (UNSPEC_SHA512SU1): Ditto.
5447 * config/aarch64/aarch64-simd-builtins.def
5448 (aarch64_crypto_sha512hqv2di): New builtin.
5449 (aarch64_crypto_sha512h2qv2di): Ditto.
5450 (aarch64_crypto_sha512su0qv2di): Ditto.
5451 (aarch64_crypto_sha512su1qv2di): Ditto.
5452 (aarch64_eor3qv8hi): Ditto.
5453 (aarch64_rax1qv2di): Ditto.
5454 (aarch64_xarqv2di): Ditto.
5455 (aarch64_bcaxqv8hi): Ditto.
5456 * config/aarch64/aarch64-simd.md:
5457 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
5458 (aarch64_crypto_sha512su0qv2di): Ditto.
5459 (aarch64_crypto_sha512su1qv2di): Ditto.
5460 (aarch64_eor3qv8hi): Ditto.
5461 (aarch64_rax1qv2di): Ditto.
5462 (aarch64_xarqv2di): Ditto.
5463 (aarch64_bcaxqv8hi): Ditto.
5464 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
5465 (vsha512h2q_u64): Ditto.
5466 (vsha512su0q_u64): Ditto.
5467 (vsha512su1q_u64): Ditto.
5468 (veor3q_u16): Ditto.
5469 (vrax1q_u64): Ditto.
5471 (vbcaxq_u16): Ditto.
5472 * config/arm/types.md (crypto_sha512): New type attribute.
5473 (crypto_sha3): Ditto.
5474 (doc/invoke.texi): Document new sha3 option.
5476 2018-01-10 Michael Collison <michael.collison@arm.com>
5478 * config/aarch64/aarch64-builtins.c:
5479 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
5480 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5481 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
5482 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
5483 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
5484 (AARCH64_ISA_SM4): New ISA flag.
5485 (TARGET_SM4): New feature flag for sm4.
5486 * config/aarch64/aarch64-simd-builtins.def
5487 (aarch64_sm3ss1qv4si): Ditto.
5488 (aarch64_sm3tt1aq4si): Ditto.
5489 (aarch64_sm3tt1bq4si): Ditto.
5490 (aarch64_sm3tt2aq4si): Ditto.
5491 (aarch64_sm3tt2bq4si): Ditto.
5492 (aarch64_sm3partw1qv4si): Ditto.
5493 (aarch64_sm3partw2qv4si): Ditto.
5494 (aarch64_sm4eqv4si): Ditto.
5495 (aarch64_sm4ekeyqv4si): Ditto.
5496 * config/aarch64/aarch64-simd.md:
5497 (aarch64_sm3ss1qv4si): Ditto.
5498 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
5499 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
5500 (aarch64_sm4eqv4si): Ditto.
5501 (aarch64_sm4ekeyqv4si): Ditto.
5502 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
5503 (sm3part_op): Ditto.
5504 (CRYPTO_SM3TT): Ditto.
5505 (CRYPTO_SM3PART): Ditto.
5506 (UNSPEC_SM3SS1): New unspec.
5507 (UNSPEC_SM3TT1A): Ditto.
5508 (UNSPEC_SM3TT1B): Ditto.
5509 (UNSPEC_SM3TT2A): Ditto.
5510 (UNSPEC_SM3TT2B): Ditto.
5511 (UNSPEC_SM3PARTW1): Ditto.
5512 (UNSPEC_SM3PARTW2): Ditto.
5513 (UNSPEC_SM4E): Ditto.
5514 (UNSPEC_SM4EKEY): Ditto.
5515 * config/aarch64/constraints.md (Ui2): New constraint.
5516 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
5517 * config/arm/types.md (crypto_sm3): New type attribute.
5518 (crypto_sm4): Ditto.
5519 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
5520 (vsm3tt1aq_u32): Ditto.
5521 (vsm3tt1bq_u32): Ditto.
5522 (vsm3tt2aq_u32): Ditto.
5523 (vsm3tt2bq_u32): Ditto.
5524 (vsm3partw1q_u32): Ditto.
5525 (vsm3partw2q_u32): Ditto.
5526 (vsm4eq_u32): Ditto.
5527 (vsm4ekeyq_u32): Ditto.
5528 (doc/invoke.texi): Document new sm4 option.
5530 2018-01-10 Michael Collison <michael.collison@arm.com>
5532 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
5533 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
5534 (AARCH64_FL_FOR_ARCH8_4): New.
5535 (AARCH64_FL_V8_4): New flag.
5536 (doc/invoke.texi): Document new armv8.4-a option.
5538 2018-01-10 Michael Collison <michael.collison@arm.com>
5540 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5541 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
5542 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
5543 * config/aarch64/aarch64-option-extension.def: Add
5544 AARCH64_OPT_EXTENSION of 'sha2'.
5545 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
5546 (crypto): Disable sha2 and aes if crypto disabled.
5547 (crypto): Enable aes and sha2 if enabled.
5548 (simd): Disable sha2 and aes if simd disabled.
5549 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
5551 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
5552 (TARGET_SHA2): New feature flag for sha2.
5553 (TARGET_AES): New feature flag for aes.
5554 * config/aarch64/aarch64-simd.md:
5555 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
5556 conditional on TARGET_AES.
5557 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
5558 (aarch64_crypto_sha1hsi): Make pattern conditional
5560 (aarch64_crypto_sha1hv4si): Ditto.
5561 (aarch64_be_crypto_sha1hv4si): Ditto.
5562 (aarch64_crypto_sha1su1v4si): Ditto.
5563 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
5564 (aarch64_crypto_sha1su0v4si): Ditto.
5565 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
5566 (aarch64_crypto_sha256su0v4si): Ditto.
5567 (aarch64_crypto_sha256su1v4si): Ditto.
5568 (doc/invoke.texi): Document new aes and sha2 options.
5570 2018-01-10 Martin Sebor <msebor@redhat.com>
5572 PR tree-optimization/83781
5573 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
5576 2018-01-11 Martin Sebor <msebor@gmail.com>
5577 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5579 PR tree-optimization/83501
5580 PR tree-optimization/81703
5582 * tree-ssa-strlen.c (get_string_cst): Rename...
5583 (get_string_len): ...to this. Handle global constants.
5584 (handle_char_store): Adjust.
5586 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
5587 Jim Wilson <jimw@sifive.com>
5589 * config/riscv/riscv-protos.h (riscv_output_return): New.
5590 * config/riscv/riscv.c (struct machine_function): New naked_p field.
5591 (riscv_attribute_table, riscv_output_return),
5592 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
5593 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
5594 (riscv_compute_frame_info): Only compute frame->mask if not a naked
5596 (riscv_expand_prologue): Add early return for naked function.
5597 (riscv_expand_epilogue): Likewise.
5598 (riscv_function_ok_for_sibcall): Return false for naked function.
5599 (riscv_set_current_function): New.
5600 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
5601 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
5602 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
5603 * doc/extend.texi (RISC-V Function Attributes): New.
5605 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
5607 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
5608 check for 128-bit long double before checking TCmode.
5609 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
5610 128-bit long doubles before checking TFmode or TCmode.
5611 (FLOAT128_IBM_P): Likewise.
5613 2018-01-10 Martin Sebor <msebor@redhat.com>
5615 PR tree-optimization/83671
5616 * builtins.c (c_strlen): Unconditionally return zero for the empty
5618 Use -Warray-bounds for warnings.
5619 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
5620 for non-constant array indices with COMPONENT_REF, arrays of
5621 arrays, and pointers to arrays.
5622 (gimple_fold_builtin_strlen): Determine and set length range for
5623 non-constant character arrays.
5625 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
5628 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
5631 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
5633 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
5635 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5638 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
5639 VECTOR_MEM_ALTIVEC_OR_VSX_P.
5640 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
5641 indexed_or_indirect_operand predicate.
5642 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
5643 (*vsx_le_perm_load_v8hi): Likewise.
5644 (*vsx_le_perm_load_v16qi): Likewise.
5645 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
5646 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
5647 (*vsx_le_perm_store_v8hi): Likewise.
5648 (*vsx_le_perm_store_v16qi): Likewise.
5649 (eight unnamed splitters): Likewise.
5651 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5653 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
5654 * config/rs6000/emmintrin.h: Likewise.
5655 * config/rs6000/mmintrin.h: Likewise.
5656 * config/rs6000/xmmintrin.h: Likewise.
5658 2018-01-10 David Malcolm <dmalcolm@redhat.com>
5661 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
5663 * tree.c (tree_nop_conversion): Return true for location wrapper
5665 (maybe_wrap_with_location): New function.
5666 (selftest::check_strip_nops): New function.
5667 (selftest::test_location_wrappers): New function.
5668 (selftest::tree_c_tests): Call it.
5669 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
5670 (maybe_wrap_with_location): New decl.
5671 (EXPR_LOCATION_WRAPPER_P): New macro.
5672 (location_wrapper_p): New inline function.
5673 (tree_strip_any_location_wrapper): New inline function.
5675 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5678 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5679 stack_realign_offset for the largest alignment of stack slot
5681 (ix86_find_max_used_stack_alignment): New function.
5682 (ix86_finalize_stack_frame_flags): Use it. Set
5683 max_used_stack_alignment if we don't realign stack.
5684 * config/i386/i386.h (machine_function): Add
5685 max_used_stack_alignment.
5687 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5689 * config/arm/arm.opt (-mbranch-cost): New option.
5690 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5693 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5696 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5697 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5699 2018-01-10 Richard Biener <rguenther@suse.de>
5702 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5703 early out so it also covers the case where we have a non-NULL
5706 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5708 PR tree-optimization/83753
5709 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5710 for non-strided grouped accesses if the number of elements is 1.
5712 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5715 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5716 * i386.h (TARGET_USE_GATHER): Define.
5717 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5719 2018-01-10 Martin Liska <mliska@suse.cz>
5722 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5723 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5725 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5726 CLEANUP_NO_PARTITIONING is not set.
5728 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5730 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5731 for vectors, as a partial revert of r254296.
5732 * rtl.h (const_vec_p): Delete.
5733 (const_vec_duplicate_p): Don't test for vector CONSTs.
5734 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5735 * expmed.c (make_tree): Likewise.
5738 * common.md (E, F): Use CONSTANT_P instead of checking for
5740 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5741 checking for CONST_VECTOR.
5743 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5746 * predict.c (force_edge_cold): Handle in more sane way edges
5749 2018-01-09 Carl Love <cel@us.ibm.com>
5751 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5753 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5754 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5755 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5756 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5757 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5758 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5759 * config/rs6000/rs6000-protos.h: Add extern defition for
5760 rs6000_generate_float2_double_code.
5761 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5763 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5764 (float2_v2df): Add define_expand.
5766 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5769 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5770 op_mode in the force_to_mode call.
5772 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5774 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5775 instead of checking each element individually.
5776 (aarch64_evpc_uzp): Likewise.
5777 (aarch64_evpc_zip): Likewise.
5778 (aarch64_evpc_ext): Likewise.
5779 (aarch64_evpc_rev): Likewise.
5780 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5781 instead of checking each element individually. Return true without
5783 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5784 whether all selected elements come from the same input, instead of
5785 checking each element individually. Remove calls to gen_rtx_REG,
5786 start_sequence and end_sequence and instead assert that no rtl is
5789 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5791 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5792 order of HIGH and CONST checks.
5794 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5796 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5797 if the destination isn't an SSA_NAME.
5799 2018-01-09 Richard Biener <rguenther@suse.de>
5801 PR tree-optimization/83668
5802 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5804 (canonicalize_loop_form): ... here, renamed from ...
5805 (canonicalize_loop_closed_ssa_form): ... this and amended to
5806 swap successor edges for loop exit blocks to make us use
5807 the RPO order we need for initial schedule generation.
5809 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5811 PR tree-optimization/64811
5812 * match.pd: When optimizing comparisons with Inf, avoid
5813 introducing or losing exceptions from comparisons with NaN.
5815 2018-01-09 Martin Liska <mliska@suse.cz>
5818 * asan.c (shadow_mem_size): Add gcc_assert.
5820 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5822 Don't save registers in main().
5825 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5826 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5827 * config/avr/avr.c (avr_set_current_function): Don't error if
5828 naked, OS_task or OS_main are specified at the same time.
5829 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5831 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5833 * common/config/avr/avr-common.c (avr_option_optimization_table):
5834 Switch on -mmain-is-OS_task for optimizing compilations.
5836 2018-01-09 Richard Biener <rguenther@suse.de>
5838 PR tree-optimization/83572
5839 * graphite.c: Include cfganal.h.
5840 (graphite_transform_loops): Connect infinite loops to exit
5841 and remove fake edges at the end.
5843 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5845 * ipa-inline.c (edge_badness): Revert accidental checkin.
5847 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5850 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5851 symbols; not inline clones.
5853 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5856 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5857 hard registers. Formatting fixes.
5859 PR preprocessor/83722
5860 * gcc.c (try_generate_repro): Pass
5861 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5862 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5865 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5866 Kito Cheng <kito.cheng@gmail.com>
5868 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5869 (riscv_leaf_function_p): Delete.
5870 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5872 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5874 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5876 (do_ifelse): New function.
5877 (do_isel): New function.
5878 (do_sub3): New function.
5879 (do_add3): New function.
5880 (do_load_mask_compare): New function.
5881 (do_overlap_load_compare): New function.
5882 (expand_compare_loop): New function.
5883 (expand_block_compare): Call expand_compare_loop() when appropriate.
5884 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5886 (-mblock-compare-inline-loop-limit): New option.
5888 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5891 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5892 Reverse order of second and third operands in first alternative.
5893 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5894 of first and second elements in UNSPEC_VPERMR vector.
5895 (altivec_expand_vec_perm_le): Likewise.
5897 2018-01-08 Jeff Law <law@redhat.com>
5899 PR rtl-optimizatin/81308
5900 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5901 (process_switch): If group_case_labels makes a change, then set
5903 (pass_convert_switch::execute): If a switch is converted, then
5904 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5906 PR rtl-optimization/81308
5907 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5910 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5912 PR target/83663 - Revert r255946
5913 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5914 generation for cases where splatting a value is not useful.
5915 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5916 across a vec_duplicate and a paradoxical subreg forming a vector
5917 mode to a vec_concat.
5919 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5921 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5922 -march=armv8.3-a variants.
5923 * config/arm/t-multilib: Likewise.
5924 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5926 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5928 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5930 (cceq_ior_compare_complement): Give it a name so I can use it, and
5931 change boolean_or_operator predicate to boolean_operator so it can
5932 be used to generate a crand.
5933 (eqne): New code iterator.
5934 (bd/bd_neg): New code_attrs.
5935 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5936 a single define_insn.
5937 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5938 decrement (bdnzt/bdnzf/bdzt/bdzf).
5939 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5940 with the new names of the branch decrement patterns, and added the
5941 names of the branch decrement conditional patterns.
5943 2018-01-08 Richard Biener <rguenther@suse.de>
5945 PR tree-optimization/83563
5946 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5949 2018-01-08 Richard Biener <rguenther@suse.de>
5952 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5954 2018-01-08 Richard Biener <rguenther@suse.de>
5956 PR tree-optimization/83685
5957 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5958 references to abnormals.
5960 2018-01-08 Richard Biener <rguenther@suse.de>
5963 * dwarf2out.c (output_indirect_strings): Handle empty
5964 skeleton_debug_str_hash.
5965 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5967 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5969 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5970 (emit_store_direct): Likewise.
5971 (arc_trampoline_adjust_address): Likewise.
5972 (arc_asm_trampoline_template): New function.
5973 (arc_initialize_trampoline): Use asm_trampoline_template.
5974 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5975 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5976 * config/arc/arc.md (flush_icache): Delete pattern.
5978 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5980 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5981 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5984 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5987 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5988 by not USED_FOR_TARGET.
5989 (make_pass_resolve_sw_modes): Likewise.
5991 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5993 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5996 2018-01-08 Richard Biener <rguenther@suse.de>
5999 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
6001 2018-01-08 Richard Biener <rguenther@suse.de>
6004 * match.pd ((t * 2) / 2) -> t): Add missing :c.
6006 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
6009 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
6010 basic blocks with a small number of successors.
6011 (convert_control_dep_chain_into_preds): Improve handling of
6013 (dump_predicates): Split apart into...
6014 (dump_pred_chain): ...here...
6015 (dump_pred_info): ...and here.
6016 (can_one_predicate_be_invalidated_p): Add debugging printfs.
6017 (can_chain_union_be_invalidated_p): Improve check for invalidation
6019 (uninit_uses_cannot_happen): Avoid unnecessary if
6020 convert_control_dep_chain_into_preds yielded nothing.
6022 2018-01-06 Martin Sebor <msebor@redhat.com>
6024 PR tree-optimization/83640
6025 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
6026 subtracting negative offset from size.
6027 (builtin_access::overlap): Adjust offset bounds of the access to fall
6028 within the size of the object if possible.
6030 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
6032 PR rtl-optimization/83699
6033 * expmed.c (extract_bit_field_1): Restrict the vector usage of
6034 extract_bit_field_as_subreg to cases in which the extracted
6035 value is also a vector.
6037 * lra-constraints.c (process_alt_operands): Test for the equivalence
6038 substitutions when detecting a possible reload cycle.
6040 2018-01-06 Jakub Jelinek <jakub@redhat.com>
6043 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
6044 by default if flag_selective_schedling{,2}. Formatting fixes.
6046 PR rtl-optimization/83682
6047 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
6048 if it has non-VECTOR_MODE element mode.
6049 (vec_duplicate_p): Likewise.
6052 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
6053 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
6055 2018-01-05 Jakub Jelinek <jakub@redhat.com>
6058 * config/i386/i386-builtin.def
6059 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
6060 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
6061 Require also OPTION_MASK_ISA_AVX512F in addition to
6062 OPTION_MASK_ISA_GFNI.
6063 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
6064 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
6065 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
6066 to OPTION_MASK_ISA_GFNI.
6067 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
6068 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
6069 OPTION_MASK_ISA_AVX512BW.
6070 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
6071 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
6072 addition to OPTION_MASK_ISA_GFNI.
6073 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
6074 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
6075 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
6076 to OPTION_MASK_ISA_GFNI.
6077 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
6078 a requirement for all ISAs rather than any of them with a few
6080 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
6082 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
6083 bitmasks to be enabled with 3 exceptions, instead of requiring any
6084 enabled ISA with lots of exceptions.
6085 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
6086 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
6087 Change avx512bw in isa attribute to avx512f.
6088 * config/i386/sgxintrin.h: Add license boilerplate.
6089 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
6090 to __AVX512F__ and __AVX512VL to __AVX512VL__.
6091 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
6092 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
6094 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
6095 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
6096 temporarily sse2 rather than sse if not enabled already.
6099 * config/i386/sse.md (VI248_VLBW): Rename to ...
6100 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
6101 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
6102 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
6103 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
6104 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
6105 mode iterator instead of VI248_VLBW.
6107 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
6109 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
6110 (record_modified): Skip clobbers; add debug output.
6111 (param_change_prob): Use sreal frequencies.
6113 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
6115 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
6116 punt for user-aligned variables.
6118 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
6120 * tree-chrec.c (chrec_contains_symbols): Return true for
6123 2018-01-05 Sudakshina Das <sudi.das@arm.com>
6126 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
6127 of (x|y) == x for BICS pattern.
6129 2018-01-05 Jakub Jelinek <jakub@redhat.com>
6131 PR tree-optimization/83605
6132 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
6133 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
6136 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
6138 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
6139 * config/epiphany/rtems.h: New file.
6141 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6142 Uros Bizjak <ubizjak@gmail.com>
6145 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
6146 QIreg_operand instead of register_operand predicate.
6147 * config/i386/i386.c (ix86_rop_should_change_byte_p,
6148 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
6149 comments instead of -fmitigate[-_]rop.
6151 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
6154 * cgraphunit.c (symbol_table::compile): Switch to text_section
6155 before calling assembly_start debug hook.
6156 * run-rtl-passes.c (run_rtl_passes): Likewise.
6159 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6161 * tree-vrp.c (extract_range_from_binary_expr_1): Check
6162 range_int_cst_p rather than !symbolic_range_p before calling
6163 extract_range_from_multiplicative_op_1.
6165 2018-01-04 Jeff Law <law@redhat.com>
6167 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
6168 redundant test in assertion.
6170 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6172 * doc/rtl.texi: Document machine_mode wrapper classes.
6174 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6176 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
6179 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6181 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
6182 the VEC_PERM_EXPR fold to fail.
6184 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6187 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
6188 to switched_sections.
6190 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6193 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
6196 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
6199 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
6200 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
6202 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6205 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
6206 is BLKmode and bitpos not zero or mode change is needed.
6208 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6211 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
6214 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
6217 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
6218 instead of MULT rtx. Update all corresponding splitters.
6220 (*ssub<modesuffix>): Ditto.
6222 (*cmp_sadd_di): Update split patterns.
6223 (*cmp_sadd_si): Ditto.
6224 (*cmp_sadd_sidi): Ditto.
6225 (*cmp_ssub_di): Ditto.
6226 (*cmp_ssub_si): Ditto.
6227 (*cmp_ssub_sidi): Ditto.
6228 * config/alpha/predicates.md (const23_operand): New predicate.
6229 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
6230 Look for ASHIFT, not MULT inner operand.
6231 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
6233 2018-01-04 Martin Liska <mliska@suse.cz>
6235 PR gcov-profile/83669
6236 * gcov.c (output_intermediate_file): Add version to intermediate
6238 * doc/gcov.texi: Document new field 'version' in intermediate
6239 file format. Fix location of '-k' option of gcov command.
6241 2018-01-04 Martin Liska <mliska@suse.cz>
6244 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
6246 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6248 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
6250 2018-01-03 Martin Sebor <msebor@redhat.com>
6252 PR tree-optimization/83655
6253 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
6254 checking calls with invalid arguments.
6256 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6258 * tree-vect-stmts.c (vect_get_store_rhs): New function.
6259 (vectorizable_mask_load_store): Delete.
6260 (vectorizable_call): Return false for masked loads and stores.
6261 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
6262 instead of gimple_assign_rhs1.
6263 (vectorizable_load): Handle IFN_MASK_LOAD.
6264 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
6266 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6268 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
6270 (vectorizable_mask_load_store): ...here.
6271 (vectorizable_load): ...and here.
6273 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6275 * tree-vect-stmts.c (vect_build_all_ones_mask)
6276 (vect_build_zero_merge_argument): New functions, split out from...
6277 (vectorizable_load): ...here.
6279 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6281 * tree-vect-stmts.c (vect_check_store_rhs): New function,
6283 (vectorizable_mask_load_store): ...here.
6284 (vectorizable_store): ...and here.
6286 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6288 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
6290 (vectorizable_mask_load_store): ...here.
6292 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6294 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
6295 (vect_model_store_cost): Take a vec_load_store_type instead of a
6297 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
6298 (vect_model_store_cost): Take a vec_load_store_type instead of a
6300 (vectorizable_mask_load_store): Update accordingly.
6301 (vectorizable_store): Likewise.
6302 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
6304 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6306 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
6307 IFN_MASK_LOAD calls here rather than...
6308 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
6310 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6311 Alan Hayward <alan.hayward@arm.com>
6312 David Sherwood <david.sherwood@arm.com>
6314 * expmed.c (extract_bit_field_1): For vector extracts,
6315 fall back to extract_bit_field_as_subreg if vec_extract
6318 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6319 Alan Hayward <alan.hayward@arm.com>
6320 David Sherwood <david.sherwood@arm.com>
6322 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
6323 they are variable or constant sized.
6324 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
6325 slots for constant-sized data.
6327 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6328 Alan Hayward <alan.hayward@arm.com>
6329 David Sherwood <david.sherwood@arm.com>
6331 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
6332 handling COND_EXPRs with boolean comparisons, try to find a better
6333 basis for the mask type than the boolean itself.
6335 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6337 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
6338 is calculated and how it can be overridden.
6339 * genmodes.c (max_bitsize_mode_any_mode): New variable.
6340 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
6342 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
6345 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6346 Alan Hayward <alan.hayward@arm.com>
6347 David Sherwood <david.sherwood@arm.com>
6349 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
6350 Remove the mode argument.
6351 (aarch64_simd_valid_immediate): Remove the mode and inverse
6353 * config/aarch64/iterators.md (bitsize): New iterator.
6354 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
6355 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
6356 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
6357 aarch64_simd_valid_immediate.
6358 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
6359 (aarch64_reg_or_bic_imm): Likewise.
6360 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
6361 with an insn_type enum and msl with a modifier_type enum.
6362 Replace element_width with a scalar_mode. Change the shift
6363 to unsigned int. Add constructors for scalar_float_mode and
6364 scalar_int_mode elements.
6365 (aarch64_vect_float_const_representable_p): Delete.
6366 (aarch64_can_const_movi_rtx_p)
6367 (aarch64_simd_scalar_immediate_valid_for_move)
6368 (aarch64_simd_make_constant): Update call to
6369 aarch64_simd_valid_immediate.
6370 (aarch64_advsimd_valid_immediate_hs): New function.
6371 (aarch64_advsimd_valid_immediate): Likewise.
6372 (aarch64_simd_valid_immediate): Remove mode and inverse
6373 arguments. Rewrite to use the above. Use const_vec_duplicate_p
6374 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
6375 and aarch64_float_const_representable_p on the result.
6376 (aarch64_output_simd_mov_immediate): Remove mode argument.
6377 Update call to aarch64_simd_valid_immediate and use of
6378 simd_immediate_info.
6379 (aarch64_output_scalar_simd_mov_immediate): Update call
6382 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6383 Alan Hayward <alan.hayward@arm.com>
6384 David Sherwood <david.sherwood@arm.com>
6386 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
6387 (mode_nunits): Likewise CONST_MODE_NUNITS.
6388 * machmode.def (ADJUST_NUNITS): Document.
6389 * genmodes.c (mode_data::need_nunits_adj): New field.
6390 (blank_mode): Update accordingly.
6391 (adj_nunits): New variable.
6392 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
6394 (emit_mode_size_inline): Set need_bytesize_adj for all modes
6395 listed in adj_nunits.
6396 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
6397 listed in adj_nunits. Don't emit case statements for such modes.
6398 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
6399 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
6400 nothing if adj_nunits is nonnull.
6401 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
6402 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
6403 (emit_mode_fbit): Update use of print_maybe_const_decl.
6404 (emit_move_size): Likewise. Treat the array as non-const
6406 (emit_mode_adjustments): Handle adj_nunits.
6408 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6410 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
6411 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
6412 (VECTOR_MODES): Use it.
6413 (make_vector_modes): Take the prefix as an argument.
6415 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6416 Alan Hayward <alan.hayward@arm.com>
6417 David Sherwood <david.sherwood@arm.com>
6419 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
6420 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
6421 for MODE_VECTOR_BOOL.
6422 * machmode.def (VECTOR_BOOL_MODE): Document.
6423 * genmodes.c (VECTOR_BOOL_MODE): New macro.
6424 (make_vector_bool_mode): New function.
6425 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
6427 * lto-streamer-in.c (lto_input_mode_table): Likewise.
6428 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
6430 * stor-layout.c (int_mode_for_mode): Likewise.
6431 * tree.c (build_vector_type_for_mode): Likewise.
6432 * varasm.c (output_constant_pool_2): Likewise.
6433 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
6434 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
6435 for MODE_VECTOR_BOOL.
6436 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
6437 of mode class checks.
6438 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
6439 instead of a list of mode class checks.
6440 (expand_vector_scalar_condition): Likewise.
6441 (type_for_widest_vector_mode): Handle BImode as an inner mode.
6443 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6444 Alan Hayward <alan.hayward@arm.com>
6445 David Sherwood <david.sherwood@arm.com>
6447 * machmode.h (mode_size): Change from unsigned short to
6449 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
6450 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6451 or if measurement_type is not polynomial.
6452 (fixed_size_mode::includes_p): Check for constant-sized modes.
6453 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
6454 return a poly_uint16 rather than an unsigned short.
6455 (emit_mode_size): Change the type of mode_size from unsigned short
6456 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
6457 (emit_mode_adjustments): Cope with polynomial vector sizes.
6458 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6460 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6462 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
6463 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
6464 * caller-save.c (setup_save_areas): Likewise.
6465 (replace_reg_with_saved_mem): Likewise.
6466 * calls.c (emit_library_call_value_1): Likewise.
6467 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
6468 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
6469 (gen_lowpart_for_combine): Likewise.
6470 * convert.c (convert_to_integer_1): Likewise.
6471 * cse.c (equiv_constant, cse_insn): Likewise.
6472 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
6473 (cselib_subst_to_values): Likewise.
6474 * dce.c (word_dce_process_block): Likewise.
6475 * df-problems.c (df_word_lr_mark_ref): Likewise.
6476 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
6477 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
6478 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
6479 (rtl_for_decl_location): Likewise.
6480 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
6481 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
6482 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
6483 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
6484 (expand_expr_real_1): Likewise.
6485 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
6486 (pad_below): Likewise.
6487 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6488 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
6489 * ira.c (get_subreg_tracking_sizes): Likewise.
6490 * ira-build.c (ira_create_allocno_objects): Likewise.
6491 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
6492 (ira_sort_regnos_for_alter_reg): Likewise.
6493 * ira-costs.c (record_operand_costs): Likewise.
6494 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
6495 (resolve_simple_move): Likewise.
6496 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
6497 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
6498 (lra_constraints): Likewise.
6499 (CONST_POOL_OK_P): Reject variable-sized modes.
6500 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
6501 (add_pseudo_to_slot, lra_spill): Likewise.
6502 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6503 * optabs-query.c (get_best_extraction_insn): Likewise.
6504 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6505 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
6506 (expand_mult_highpart, valid_multiword_target_p): Likewise.
6507 * recog.c (offsettable_address_addr_space_p): Likewise.
6508 * regcprop.c (maybe_mode_change): Likewise.
6509 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
6510 * regrename.c (build_def_use): Likewise.
6511 * regstat.c (dump_reg_info): Likewise.
6512 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
6513 (find_reloads, find_reloads_subreg_address): Likewise.
6514 * reload1.c (eliminate_regs_1): Likewise.
6515 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
6516 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
6517 (simplify_binary_operation_1, simplify_subreg): Likewise.
6518 * targhooks.c (default_function_arg_padding): Likewise.
6519 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
6520 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
6521 (verify_gimple_assign_ternary): Likewise.
6522 * tree-inline.c (estimate_move_cost): Likewise.
6523 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6524 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
6525 (get_address_cost_ainc): Likewise.
6526 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
6527 (vect_supportable_dr_alignment): Likewise.
6528 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6529 (vectorizable_reduction): Likewise.
6530 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
6531 (vectorizable_operation, vectorizable_load): Likewise.
6532 * tree.c (build_same_sized_truth_vector_type): Likewise.
6533 * valtrack.c (cleanup_auto_inc_dec): Likewise.
6534 * var-tracking.c (emit_note_insn_var_location): Likewise.
6535 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
6536 (ADDR_VEC_ALIGN): Likewise.
6538 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6539 Alan Hayward <alan.hayward@arm.com>
6540 David Sherwood <david.sherwood@arm.com>
6542 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
6544 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6545 or if measurement_type is polynomial.
6546 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
6547 * combine.c (make_extraction): Likewise.
6548 * dse.c (find_shift_sequence): Likewise.
6549 * dwarf2out.c (mem_loc_descriptor): Likewise.
6550 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
6551 (extract_bit_field, extract_low_bits): Likewise.
6552 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
6553 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
6554 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
6555 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
6556 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6557 * reload.c (find_reloads): Likewise.
6558 * reload1.c (alter_reg): Likewise.
6559 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
6560 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
6561 * tree-if-conv.c (predicate_mem_writes): Likewise.
6562 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
6563 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
6564 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
6565 * valtrack.c (dead_debug_insert_temp): Likewise.
6566 * varasm.c (mergeable_constant_section): Likewise.
6567 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
6569 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6570 Alan Hayward <alan.hayward@arm.com>
6571 David Sherwood <david.sherwood@arm.com>
6573 * expr.c (expand_assignment): Cope with polynomial mode sizes
6574 when assigning to a CONCAT.
6576 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6577 Alan Hayward <alan.hayward@arm.com>
6578 David Sherwood <david.sherwood@arm.com>
6580 * machmode.h (mode_precision): Change from unsigned short to
6582 (mode_to_precision): Return a poly_uint16 rather than an unsigned
6584 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
6585 or if measurement_type is not polynomial.
6586 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
6587 in which the mode is already known to be a scalar_int_mode.
6588 * genmodes.c (emit_mode_precision): Change the type of mode_precision
6589 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
6591 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6592 for GET_MODE_PRECISION.
6593 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6594 for GET_MODE_PRECISION.
6595 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
6597 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
6598 (expand_field_assignment, make_extraction): Likewise.
6599 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
6600 (get_last_value): Likewise.
6601 * convert.c (convert_to_integer_1): Likewise.
6602 * cse.c (cse_insn): Likewise.
6603 * expr.c (expand_expr_real_1): Likewise.
6604 * lra-constraints.c (simplify_operand_subreg): Likewise.
6605 * optabs-query.c (can_atomic_load_p): Likewise.
6606 * optabs.c (expand_atomic_load): Likewise.
6607 (expand_atomic_store): Likewise.
6608 * ree.c (combine_reaching_defs): Likewise.
6609 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
6610 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
6611 * tree.h (type_has_mode_precision_p): Likewise.
6612 * ubsan.c (instrument_si_overflow): Likewise.
6614 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6615 Alan Hayward <alan.hayward@arm.com>
6616 David Sherwood <david.sherwood@arm.com>
6618 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
6619 polynomial numbers of units.
6620 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
6621 (valid_vector_subparts_p): New function.
6622 (build_vector_type): Remove temporary shim and take the number
6623 of units as a poly_uint64 rather than an int.
6624 (build_opaque_vector_type): Take the number of units as a
6625 poly_uint64 rather than an int.
6626 * tree.c (build_vector_from_ctor): Handle polynomial
6627 TYPE_VECTOR_SUBPARTS.
6628 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
6629 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
6630 (build_vector_from_val): If the number of units is variable,
6631 use build_vec_duplicate_cst for constant operands and
6632 VEC_DUPLICATE_EXPR otherwise.
6633 (make_vector_type): Remove temporary is_constant ().
6634 (build_vector_type, build_opaque_vector_type): Take the number of
6635 units as a poly_uint64 rather than an int.
6636 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
6638 * cfgexpand.c (expand_debug_expr): Likewise.
6639 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
6640 (store_constructor, expand_expr_real_1): Likewise.
6641 (const_scalar_mask_from_tree): Likewise.
6642 * fold-const-call.c (fold_const_reduction): Likewise.
6643 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
6644 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
6645 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
6646 (fold_relational_const): Likewise.
6647 (native_interpret_vector): Likewise. Change the size from an
6648 int to an unsigned int.
6649 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
6650 TYPE_VECTOR_SUBPARTS.
6651 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
6652 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
6653 duplicating a non-constant operand into a variable-length vector.
6654 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
6655 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
6656 * ipa-icf.c (sem_variable::equals): Likewise.
6657 * match.pd: Likewise.
6658 * omp-simd-clone.c (simd_clone_subparts): Likewise.
6659 * print-tree.c (print_node): Likewise.
6660 * stor-layout.c (layout_type): Likewise.
6661 * targhooks.c (default_builtin_vectorization_cost): Likewise.
6662 * tree-cfg.c (verify_gimple_comparison): Likewise.
6663 (verify_gimple_assign_binary): Likewise.
6664 (verify_gimple_assign_ternary): Likewise.
6665 (verify_gimple_assign_single): Likewise.
6666 * tree-pretty-print.c (dump_generic_node): Likewise.
6667 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6668 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
6669 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
6670 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
6671 (vect_shift_permute_load_chain): Likewise.
6672 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
6673 (expand_vector_condition, optimize_vector_constructor): Likewise.
6674 (lower_vec_perm, get_compute_type): Likewise.
6675 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6676 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6677 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6678 (vect_recog_mask_conversion_pattern): Likewise.
6679 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6680 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6681 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6682 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6683 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6684 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6685 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6686 (supportable_widening_operation): Likewise.
6687 (supportable_narrowing_operation): Likewise.
6688 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6690 * varasm.c (output_constant): Likewise.
6692 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6693 Alan Hayward <alan.hayward@arm.com>
6694 David Sherwood <david.sherwood@arm.com>
6696 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6697 so that both the length == 3 and length != 3 cases set up their
6698 own permute vectors. Add comments explaining why we know the
6699 number of elements is constant.
6700 (vect_permute_load_chain): Likewise.
6702 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6703 Alan Hayward <alan.hayward@arm.com>
6704 David Sherwood <david.sherwood@arm.com>
6706 * machmode.h (mode_nunits): Change from unsigned char to
6708 (ONLY_FIXED_SIZE_MODES): New macro.
6709 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6710 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6711 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6713 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6714 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6715 or if measurement_type is not polynomial.
6716 * genmodes.c (ZERO_COEFFS): New macro.
6717 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6719 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6720 Use ZERO_COEFFS when emitting initializers.
6721 * data-streamer.h (bp_pack_poly_value): New function.
6722 (bp_unpack_poly_value): Likewise.
6723 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6724 for GET_MODE_NUNITS.
6725 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6726 for GET_MODE_NUNITS.
6727 * tree.c (make_vector_type): Remove temporary shim and make
6728 the real function take the number of units as a poly_uint64
6730 (build_vector_type_for_mode): Handle polynomial nunits.
6731 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6732 * emit-rtl.c (const_vec_series_p_1): Likewise.
6733 (gen_rtx_CONST_VECTOR): Likewise.
6734 * fold-const.c (test_vec_duplicate_folding): Likewise.
6735 * genrecog.c (validate_pattern): Likewise.
6736 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6737 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6738 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6739 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6740 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6741 * rtlanal.c (subreg_get_info): Likewise.
6742 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6743 (vect_grouped_load_supported): Likewise.
6744 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6745 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6746 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6747 (simplify_const_unary_operation, simplify_binary_operation_1)
6748 (simplify_const_binary_operation, simplify_ternary_operation)
6749 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6750 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6751 instead of CONST_VECTOR_NUNITS.
6752 * varasm.c (output_constant_pool_2): Likewise.
6753 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6754 explicit-encoded elements in the XVEC for variable-length vectors.
6756 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6758 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6760 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6761 Alan Hayward <alan.hayward@arm.com>
6762 David Sherwood <david.sherwood@arm.com>
6764 * coretypes.h (fixed_size_mode): Declare.
6765 (fixed_size_mode_pod): New typedef.
6766 * builtins.h (target_builtins::x_apply_args_mode)
6767 (target_builtins::x_apply_result_mode): Change type to
6768 fixed_size_mode_pod.
6769 * builtins.c (apply_args_size, apply_result_size, result_vector)
6770 (expand_builtin_apply_args_1, expand_builtin_apply)
6771 (expand_builtin_return): Update accordingly.
6773 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6775 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6776 * cselib.c (cselib_hash_rtx): Likewise.
6777 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6778 CONST_VECTOR encoding.
6780 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6781 Jeff Law <law@redhat.com>
6784 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6785 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6786 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6787 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6790 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6791 explicitly probe *sp in a noreturn function if there were any callee
6792 register saves or frame pointer is needed.
6794 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6797 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6798 BLKmode for ternary, binary or unary expressions.
6801 * var-tracking.c (delete_vta_debug_insn): New inline function.
6802 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6803 insns from get_insns () to NULL instead of each bb separately.
6804 Use delete_vta_debug_insn. No longer static.
6805 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6806 delete_vta_debug_insns callers.
6807 * rtl.h (delete_vta_debug_insns): Declare.
6808 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6809 instead of variable_tracking_main.
6811 2018-01-03 Martin Sebor <msebor@redhat.com>
6813 PR tree-optimization/83603
6814 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6815 arguments past the endof the argument list in functions declared
6816 without a prototype.
6817 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6818 Avoid checking when arguments are null.
6820 2018-01-03 Martin Sebor <msebor@redhat.com>
6823 * doc/extend.texi (attribute const): Fix a typo.
6824 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6825 issuing -Wsuggest-attribute for void functions.
6827 2018-01-03 Martin Sebor <msebor@redhat.com>
6829 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6830 offset_int::from instead of wide_int::to_shwi.
6831 (maybe_diag_overlap): Remove assertion.
6832 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6833 * gimple-ssa-sprintf.c (format_directive): Same.
6834 (parse_directive): Same.
6835 (sprintf_dom_walker::compute_format_length): Same.
6836 (try_substitute_return_value): Same.
6838 2018-01-03 Jeff Law <law@redhat.com>
6841 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6842 non-constant residual for zero at runtime and avoid probing in
6843 that case. Reorganize code for trailing problem to mirror handling
6846 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6848 PR tree-optimization/83501
6849 * tree-ssa-strlen.c (get_string_cst): New.
6850 (handle_char_store): Call get_string_cst.
6852 2018-01-03 Martin Liska <mliska@suse.cz>
6854 PR tree-optimization/83593
6855 * tree-ssa-strlen.c: Include tree-cfg.h.
6856 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6857 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6858 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6860 (strlen_dom_walker::before_dom_children): Call
6861 gimple_purge_dead_eh_edges. Dump tranformation with details
6863 (strlen_dom_walker::before_dom_children): Update call by adding
6864 new argument cleanup_eh.
6865 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6867 2018-01-03 Martin Liska <mliska@suse.cz>
6870 * cif-code.def (VARIADIC_THUNK): New enum value.
6871 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6874 2018-01-03 Jan Beulich <jbeulich@suse.com>
6876 * sse.md (mov<mode>_internal): Tighten condition for when to use
6877 vmovdqu<ssescalarsize> for TI and OI modes.
6879 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6881 Update copyright years.
6883 2018-01-03 Martin Liska <mliska@suse.cz>
6886 * ipa-visibility.c (function_and_variable_visibility): Skip
6887 functions with noipa attribure.
6889 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6891 * gcc.c (process_command): Update copyright notice dates.
6892 * gcov-dump.c (print_version): Ditto.
6893 * gcov.c (print_version): Ditto.
6894 * gcov-tool.c (print_version): Ditto.
6895 * gengtype.c (create_file): Ditto.
6896 * doc/cpp.texi: Bump @copying's copyright year.
6897 * doc/cppinternals.texi: Ditto.
6898 * doc/gcc.texi: Ditto.
6899 * doc/gccint.texi: Ditto.
6900 * doc/gcov.texi: Ditto.
6901 * doc/install.texi: Ditto.
6902 * doc/invoke.texi: Ditto.
6904 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6906 * vector-builder.h (vector_builder::m_full_nelts): Change from
6907 unsigned int to poly_uint64.
6908 (vector_builder::full_nelts): Update prototype accordingly.
6909 (vector_builder::new_vector): Likewise.
6910 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6911 (vector_builder::operator ==): Likewise.
6912 (vector_builder::finalize): Likewise.
6913 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6914 Take the number of elements as a poly_uint64 rather than an
6916 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6917 from unsigned int to poly_uint64.
6918 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6919 (vec_perm_indices::new_vector): Likewise.
6920 (vec_perm_indices::length): Likewise.
6921 (vec_perm_indices::nelts_per_input): Likewise.
6922 (vec_perm_indices::input_nelts): Likewise.
6923 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6924 number of elements per input as a poly_uint64 rather than an
6925 unsigned int. Use the original encoding for variable-length
6926 vectors, rather than clamping each individual element.
6927 For the second and subsequent elements in each pattern,
6928 clamp the step and base before clamping their sum.
6929 (vec_perm_indices::series_p): Handle polynomial element counts.
6930 (vec_perm_indices::all_in_range_p): Likewise.
6931 (vec_perm_indices_to_tree): Likewise.
6932 (vec_perm_indices_to_rtx): Likewise.
6933 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6934 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6935 (tree_vector_builder::new_binary_operation): Handle polynomial
6936 element counts. Return false if we need to know the number
6937 of elements at compile time.
6938 * fold-const.c (fold_vec_perm): Punt if the number of elements
6939 isn't known at compile time.
6941 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6943 * vec-perm-indices.h (vec_perm_builder): Change element type
6944 from HOST_WIDE_INT to poly_int64.
6945 (vec_perm_indices::element_type): Update accordingly.
6946 (vec_perm_indices::clamp): Handle polynomial element_types.
6947 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6948 (vec_perm_indices::all_in_range_p): Likewise.
6949 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6951 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6952 polynomial vec_perm_indices element types.
6953 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6954 * fold-const.c (fold_vec_perm): Likewise.
6955 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6956 * tree-vect-generic.c (lower_vec_perm): Likewise.
6957 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6958 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6959 element type to HOST_WIDE_INT.
6961 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6962 Alan Hayward <alan.hayward@arm.com>
6963 David Sherwood <david.sherwood@arm.com>
6965 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6966 rather than an int. Use plus_constant.
6967 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6968 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6970 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6971 Alan Hayward <alan.hayward@arm.com>
6972 David Sherwood <david.sherwood@arm.com>
6974 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6975 a HOST_WIDE_INT to a poly_int64.
6977 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6978 Alan Hayward <alan.hayward@arm.com>
6979 David Sherwood <david.sherwood@arm.com>
6981 * calls.c (load_register_parameters): Cope with polynomial
6982 mode sizes. Require a constant size for BLKmode parameters
6983 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6984 forces a parameter to be padded at the lsb end in order to
6985 fill a complete number of words, require the parameter size
6986 to be ordered wrt UNITS_PER_WORD.
6988 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6989 Alan Hayward <alan.hayward@arm.com>
6990 David Sherwood <david.sherwood@arm.com>
6992 * reload1.c (spill_stack_slot_width): Change element type
6993 from unsigned int to poly_uint64_pod.
6994 (alter_reg): Treat mode sizes as polynomial.
6996 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6997 Alan Hayward <alan.hayward@arm.com>
6998 David Sherwood <david.sherwood@arm.com>
7000 * reload.c (complex_word_subreg_p): New function.
7001 (reload_inner_reg_of_subreg, push_reload): Use it.
7003 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7004 Alan Hayward <alan.hayward@arm.com>
7005 David Sherwood <david.sherwood@arm.com>
7007 * lra-constraints.c (process_alt_operands): Reject matched
7008 operands whose sizes aren't ordered.
7009 (match_reload): Refer to this check here.
7011 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7012 Alan Hayward <alan.hayward@arm.com>
7013 David Sherwood <david.sherwood@arm.com>
7015 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
7016 that the mode size is in the set {1, 2, 4, 8, 16}.
7018 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7019 Alan Hayward <alan.hayward@arm.com>
7020 David Sherwood <david.sherwood@arm.com>
7022 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
7023 Use plus_constant instead of gen_rtx_PLUS.
7025 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7026 Alan Hayward <alan.hayward@arm.com>
7027 David Sherwood <david.sherwood@arm.com>
7029 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
7030 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
7031 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
7032 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
7033 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
7034 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
7035 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
7036 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
7037 * config/i386/i386.c (ix86_push_rounding): ...this new function.
7038 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
7040 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
7041 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
7042 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
7043 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
7044 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
7045 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
7046 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
7047 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
7048 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
7049 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
7051 * expr.c (emit_move_resolve_push): Treat the input and result
7052 of PUSH_ROUNDING as a poly_int64.
7053 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
7054 (emit_push_insn): Likewise.
7055 * lra-eliminations.c (mark_not_eliminable): Likewise.
7056 * recog.c (push_operand): Likewise.
7057 * reload1.c (elimination_effects): Likewise.
7058 * rtlanal.c (nonzero_bits1): Likewise.
7059 * calls.c (store_one_arg): Likewise. Require the padding to be
7060 known at compile time.
7062 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7063 Alan Hayward <alan.hayward@arm.com>
7064 David Sherwood <david.sherwood@arm.com>
7066 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
7067 Use plus_constant instead of gen_rtx_PLUS.
7069 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7070 Alan Hayward <alan.hayward@arm.com>
7071 David Sherwood <david.sherwood@arm.com>
7073 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
7076 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7077 Alan Hayward <alan.hayward@arm.com>
7078 David Sherwood <david.sherwood@arm.com>
7080 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
7081 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
7082 via stack temporaries. Treat the mode size as polynomial too.
7084 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7085 Alan Hayward <alan.hayward@arm.com>
7086 David Sherwood <david.sherwood@arm.com>
7088 * expr.c (expand_expr_real_2): When handling conversions involving
7089 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
7090 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
7091 as a poly_uint64 too.
7093 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7094 Alan Hayward <alan.hayward@arm.com>
7095 David Sherwood <david.sherwood@arm.com>
7097 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
7099 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7100 Alan Hayward <alan.hayward@arm.com>
7101 David Sherwood <david.sherwood@arm.com>
7103 * combine.c (can_change_dest_mode): Handle polynomial
7104 REGMODE_NATURAL_SIZE.
7105 * expmed.c (store_bit_field_1): Likewise.
7106 * expr.c (store_constructor): Likewise.
7107 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
7108 and polynomial REGMODE_NATURAL_SIZE.
7109 (gen_lowpart_common): Likewise.
7110 * reginfo.c (record_subregs_of_mode): Likewise.
7111 * rtlanal.c (read_modify_subreg_p): Likewise.
7113 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7114 Alan Hayward <alan.hayward@arm.com>
7115 David Sherwood <david.sherwood@arm.com>
7117 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
7118 numbers of elements.
7120 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7121 Alan Hayward <alan.hayward@arm.com>
7122 David Sherwood <david.sherwood@arm.com>
7124 * match.pd: Cope with polynomial numbers of vector elements.
7126 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7127 Alan Hayward <alan.hayward@arm.com>
7128 David Sherwood <david.sherwood@arm.com>
7130 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
7131 in a POINTER_PLUS_EXPR.
7133 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7134 Alan Hayward <alan.hayward@arm.com>
7135 David Sherwood <david.sherwood@arm.com>
7137 * omp-simd-clone.c (simd_clone_subparts): New function.
7138 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
7139 (ipa_simd_modify_function_body): Likewise.
7141 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7142 Alan Hayward <alan.hayward@arm.com>
7143 David Sherwood <david.sherwood@arm.com>
7145 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
7146 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
7147 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
7148 (expand_vector_condition, vector_element): Likewise.
7149 (subparts_gt): New function.
7150 (get_compute_type): Use subparts_gt.
7151 (count_type_subparts): Delete.
7152 (expand_vector_operations_1): Use subparts_gt instead of
7153 count_type_subparts.
7155 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7156 Alan Hayward <alan.hayward@arm.com>
7157 David Sherwood <david.sherwood@arm.com>
7159 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
7160 (vect_compile_time_alias): ...this new function. Do the calculation
7161 on poly_ints rather than trees.
7162 (vect_prune_runtime_alias_test_list): Update call accordingly.
7164 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7165 Alan Hayward <alan.hayward@arm.com>
7166 David Sherwood <david.sherwood@arm.com>
7168 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
7170 (vect_schedule_slp_instance): Likewise.
7172 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7173 Alan Hayward <alan.hayward@arm.com>
7174 David Sherwood <david.sherwood@arm.com>
7176 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
7177 constant and extern definitions for variable-length vectors.
7178 (vect_get_constant_vectors): Note that the number of units
7179 is known to be constant.
7181 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7182 Alan Hayward <alan.hayward@arm.com>
7183 David Sherwood <david.sherwood@arm.com>
7185 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
7186 of units as polynomial. Choose between WIDE and NARROW based
7189 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7190 Alan Hayward <alan.hayward@arm.com>
7191 David Sherwood <david.sherwood@arm.com>
7193 * tree-vect-stmts.c (simd_clone_subparts): New function.
7194 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
7196 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7197 Alan Hayward <alan.hayward@arm.com>
7198 David Sherwood <david.sherwood@arm.com>
7200 * tree-vect-stmts.c (vectorizable_call): Treat the number of
7201 vectors as polynomial. Use build_index_vector for
7204 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7205 Alan Hayward <alan.hayward@arm.com>
7206 David Sherwood <david.sherwood@arm.com>
7208 * tree-vect-stmts.c (get_load_store_type): Treat the number of
7209 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
7210 for variable-length vectors.
7211 (vectorizable_mask_load_store): Treat the number of units as
7212 polynomial, asserting that it is constant if the condition has
7213 already been enforced.
7214 (vectorizable_store, vectorizable_load): Likewise.
7216 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7217 Alan Hayward <alan.hayward@arm.com>
7218 David Sherwood <david.sherwood@arm.com>
7220 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
7221 of units as polynomial. Punt if we can't tell at compile time
7222 which vector contains the final result.
7224 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7225 Alan Hayward <alan.hayward@arm.com>
7226 David Sherwood <david.sherwood@arm.com>
7228 * tree-vect-loop.c (vectorizable_induction): Treat the number
7229 of units as polynomial. Punt on SLP inductions. Use an integer
7230 VEC_SERIES_EXPR for variable-length integer reductions. Use a
7231 cast of such a series for variable-length floating-point
7234 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7235 Alan Hayward <alan.hayward@arm.com>
7236 David Sherwood <david.sherwood@arm.com>
7238 * tree.h (build_index_vector): Declare.
7239 * tree.c (build_index_vector): New function.
7240 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
7241 of units as polynomial, forcibly converting it to a constant if
7242 vectorizable_reduction has already enforced the condition.
7243 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
7244 to create a {1,2,3,...} vector.
7245 (vectorizable_reduction): Treat the number of units as polynomial.
7246 Choose vectype_in based on the largest scalar element size rather
7247 than the smallest number of units. Enforce the restrictions
7250 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7251 Alan Hayward <alan.hayward@arm.com>
7252 David Sherwood <david.sherwood@arm.com>
7254 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
7255 number of units as polynomial.
7257 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7258 Alan Hayward <alan.hayward@arm.com>
7259 David Sherwood <david.sherwood@arm.com>
7261 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
7262 * target.def (autovectorize_vector_sizes): Return the vector sizes
7263 by pointer, using vector_sizes rather than a bitmask.
7264 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
7265 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
7266 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
7268 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
7269 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
7270 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
7271 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
7272 * omp-general.c (omp_max_vf): Likewise.
7273 * omp-low.c (omp_clause_aligned_alignment): Likewise.
7274 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
7275 * tree-vect-loop.c (vect_analyze_loop): Likewise.
7276 * tree-vect-slp.c (vect_slp_bb): Likewise.
7277 * doc/tm.texi: Regenerate.
7278 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
7280 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
7281 the vector size as a poly_uint64 rather than an unsigned int.
7282 (current_vector_size): Change from an unsigned int to a poly_uint64.
7283 (get_vectype_for_scalar_type): Update accordingly.
7284 * tree.h (build_truth_vector_type): Take the size and number of
7285 units as a poly_uint64 rather than an unsigned int.
7286 (build_vector_type): Add a temporary overload that takes
7287 the number of units as a poly_uint64 rather than an unsigned int.
7288 * tree.c (make_vector_type): Likewise.
7289 (build_truth_vector_type): Take the number of units as a poly_uint64
7290 rather than an unsigned int.
7292 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7293 Alan Hayward <alan.hayward@arm.com>
7294 David Sherwood <david.sherwood@arm.com>
7296 * target.def (get_mask_mode): Take the number of units and length
7297 as poly_uint64s rather than unsigned ints.
7298 * targhooks.h (default_get_mask_mode): Update accordingly.
7299 * targhooks.c (default_get_mask_mode): Likewise.
7300 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
7301 * doc/tm.texi: Regenerate.
7303 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7304 Alan Hayward <alan.hayward@arm.com>
7305 David Sherwood <david.sherwood@arm.com>
7307 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
7308 * omp-general.c (omp_max_vf): Likewise.
7309 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
7310 (expand_omp_simd): Handle polynomial safelen.
7311 * omp-low.c (omplow_simd_context): Add a default constructor.
7312 (omplow_simd_context::max_vf): Change from int to poly_uint64.
7313 (lower_rec_simd_input_clauses): Update accordingly.
7314 (lower_rec_input_clauses): Likewise.
7316 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7317 Alan Hayward <alan.hayward@arm.com>
7318 David Sherwood <david.sherwood@arm.com>
7320 * tree-vectorizer.h (vect_nunits_for_cost): New function.
7321 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
7322 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
7323 (vect_analyze_slp_cost): Likewise.
7324 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
7325 (vect_model_load_cost): Likewise.
7327 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7328 Alan Hayward <alan.hayward@arm.com>
7329 David Sherwood <david.sherwood@arm.com>
7331 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
7332 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
7333 from an unsigned int * to a poly_uint64_pod *.
7334 (calculate_unrolling_factor): New function.
7335 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
7337 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7338 Alan Hayward <alan.hayward@arm.com>
7339 David Sherwood <david.sherwood@arm.com>
7341 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
7342 from an unsigned int to a poly_uint64.
7343 (_loop_vec_info::slp_unrolling_factor): Likewise.
7344 (_loop_vec_info::vectorization_factor): Change from an int
7346 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
7347 (vect_get_num_vectors): New function.
7348 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
7349 (vect_get_num_copies): Use vect_get_num_vectors.
7350 (vect_analyze_data_ref_dependences): Change max_vf from an int *
7351 to an unsigned int *.
7352 (vect_analyze_data_refs): Change min_vf from an int * to a
7354 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7355 than an unsigned HOST_WIDE_INT.
7356 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
7357 (vect_analyze_data_ref_dependence): Change max_vf from an int *
7358 to an unsigned int *.
7359 (vect_analyze_data_ref_dependences): Likewise.
7360 (vect_compute_data_ref_alignment): Handle polynomial vf.
7361 (vect_enhance_data_refs_alignment): Likewise.
7362 (vect_prune_runtime_alias_test_list): Likewise.
7363 (vect_shift_permute_load_chain): Likewise.
7364 (vect_supportable_dr_alignment): Likewise.
7365 (dependence_distance_ge_vf): Take the vectorization factor as a
7366 poly_uint64 rather than an unsigned HOST_WIDE_INT.
7367 (vect_analyze_data_refs): Change min_vf from an int * to a
7369 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
7370 vfm1 as a poly_uint64 rather than an int. Make the same change
7371 for the returned bound_scalar.
7372 (vect_gen_vector_loop_niters): Handle polynomial vf.
7373 (vect_do_peeling): Likewise. Update call to
7374 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
7375 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
7377 * tree-vect-loop.c (vect_determine_vectorization_factor)
7378 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
7379 (vect_get_known_peeling_cost): Likewise.
7380 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
7381 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
7382 (vect_transform_loop): Likewise. Use the lowest possible VF when
7383 updating the upper bounds of the loop.
7384 (vect_min_worthwhile_factor): Make static. Return an unsigned int
7386 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
7387 polynomial unroll factors.
7388 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
7389 (vect_make_slp_decision): Likewise.
7390 (vect_supported_load_permutation_p): Likewise, and polynomial
7392 (vect_analyze_slp_cost): Handle polynomial vf.
7393 (vect_slp_analyze_node_operations): Likewise.
7394 (vect_slp_analyze_bb_1): Likewise.
7395 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7396 than an unsigned HOST_WIDE_INT.
7397 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
7398 (vectorizable_load): Handle polynomial vf.
7399 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
7401 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
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 * match.pd: Handle bit operations involving three constants
7408 and try to fold one pair.
7410 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7412 * tree-vect-loop-manip.c: Include gimple-fold.h.
7413 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
7414 niters_maybe_zero parameters. Handle other cases besides a step of 1.
7415 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
7416 Add a path that uses a step of VF instead of 1, but disable it
7418 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
7419 and niters_no_overflow parameters. Update calls to
7420 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
7421 Create a new SSA name if the latter choses to use a ste other
7422 than zero, and return it via niters_vector_mult_vf_var.
7423 * tree-vect-loop.c (vect_transform_loop): Update calls to
7424 vect_do_peeling, vect_gen_vector_loop_niters and
7425 slpeel_make_loop_iterate_ntimes.
7426 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
7427 (vect_gen_vector_loop_niters): Update declarations after above changes.
7429 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
7431 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
7432 128-bit round to integer instructions.
7433 (ceil<mode>2): Likewise.
7434 (btrunc<mode>2): Likewise.
7435 (round<mode>2): Likewise.
7437 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7439 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
7440 unaligned VSX load/store on P8/P9.
7441 (expand_block_clear): Allow the use of unaligned VSX
7442 load/store on P8/P9.
7444 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
7446 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
7448 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
7449 swap associated with both a load and a store.
7451 2018-01-02 Andrew Waterman <andrew@sifive.com>
7453 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
7454 * config/riscv/riscv.md (clear_cache): Use it.
7456 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
7458 * web.c: Remove out-of-date comment.
7460 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7462 * expr.c (fixup_args_size_notes): Check that any existing
7463 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
7464 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
7465 (emit_single_push_insn): ...here.
7467 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7469 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
7470 (const_vector_encoded_nelts): New function.
7471 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
7472 (const_vector_int_elt, const_vector_elt): Declare.
7473 * emit-rtl.c (const_vector_int_elt_1): New function.
7474 (const_vector_elt): Likewise.
7475 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
7476 of CONST_VECTOR_ELT.
7478 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7480 * expr.c: Include rtx-vector-builder.h.
7481 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
7482 directly on the tree encoding.
7483 (const_vector_from_tree): Likewise.
7484 * optabs.c: Include rtx-vector-builder.h.
7485 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
7486 sequence of "u" values.
7487 * vec-perm-indices.c: Include rtx-vector-builder.h.
7488 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
7489 directly on the vec_perm_indices encoding.
7491 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7493 * doc/rtl.texi (const_vector): Describe new encoding scheme.
7494 * Makefile.in (OBJS): Add rtx-vector-builder.o.
7495 * rtx-vector-builder.h: New file.
7496 * rtx-vector-builder.c: Likewise.
7497 * rtl.h (rtx_def::u2): Add a const_vector field.
7498 (CONST_VECTOR_NPATTERNS): New macro.
7499 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
7500 (CONST_VECTOR_DUPLICATE_P): Likewise.
7501 (CONST_VECTOR_STEPPED_P): Likewise.
7502 (CONST_VECTOR_ENCODED_ELT): Likewise.
7503 (const_vec_duplicate_p): Check for a duplicated vector encoding.
7504 (unwrap_const_vec_duplicate): Likewise.
7505 (const_vec_series_p): Check for a non-duplicated vector encoding.
7506 Say that the function only returns true for integer vectors.
7507 * emit-rtl.c: Include rtx-vector-builder.h.
7508 (gen_const_vec_duplicate_1): Delete.
7509 (gen_const_vector): Call gen_const_vec_duplicate instead of
7510 gen_const_vec_duplicate_1.
7511 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
7512 (gen_const_vec_duplicate): Use rtx_vector_builder.
7513 (gen_const_vec_series): Likewise.
7514 (gen_rtx_CONST_VECTOR): Likewise.
7515 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
7516 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7517 Build a new vector rather than modifying a CONST_VECTOR in-place.
7518 (handle_special_swappables): Update call accordingly.
7519 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
7520 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7521 Build a new vector rather than modifying a CONST_VECTOR in-place.
7522 (handle_special_swappables): Update call accordingly.
7524 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7526 * simplify-rtx.c (simplify_const_binary_operation): Use
7527 CONST_VECTOR_ELT instead of XVECEXP.
7529 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7531 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
7532 the selector elements to be different from the data elements
7533 if the selector is a VECTOR_CST.
7534 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
7535 ssizetype for the selector.
7537 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7539 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
7540 before testing each element individually.
7541 * tree-vect-generic.c (lower_vec_perm): Likewise.
7543 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7545 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
7546 * selftest-run-tests.c (selftest::run_tests): Call it.
7547 * vector-builder.h (vector_builder::operator ==): New function.
7548 (vector_builder::operator !=): Likewise.
7549 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
7550 (vec_perm_indices::all_from_input_p): New function.
7551 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7552 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
7553 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
7554 instead of reading the VECTOR_CST directly. Detect whether both
7555 vector inputs are the same before constructing the vec_perm_indices,
7556 and update the number of inputs argument accordingly. Use the
7557 utility functions added above. Only construct sel2 if we need to.
7559 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7561 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
7562 the broadcast of the low byte.
7563 (expand_mult_highpart): Use an explicit encoding for the permutes.
7564 * optabs-query.c (can_mult_highpart_p): Likewise.
7565 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
7566 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7567 (vectorizable_bswap): Likewise.
7568 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
7569 explicit encoding for the power-of-2 permutes.
7570 (vect_permute_store_chain): Likewise.
7571 (vect_grouped_load_supported): Likewise.
7572 (vect_permute_load_chain): Likewise.
7574 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7576 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
7577 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
7578 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
7579 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7580 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
7581 (vect_gen_perm_mask_any): Likewise.
7583 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7585 * int-vector-builder.h: New file.
7586 * vec-perm-indices.h: Include int-vector-builder.h.
7587 (vec_perm_indices): Redefine as an int_vector_builder.
7588 (auto_vec_perm_indices): Delete.
7589 (vec_perm_builder): Redefine as a stand-alone class.
7590 (vec_perm_indices::vec_perm_indices): New function.
7591 (vec_perm_indices::clamp): Likewise.
7592 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
7593 (vec_perm_indices::new_vector): New function.
7594 (vec_perm_indices::new_expanded_vector): Update for new
7595 vec_perm_indices class.
7596 (vec_perm_indices::rotate_inputs): New function.
7597 (vec_perm_indices::all_in_range_p): Operate directly on the
7598 encoded form, without computing elided elements.
7599 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
7600 encoding. Update for new vec_perm_indices class.
7601 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
7602 the given vec_perm_builder.
7603 (expand_vec_perm_var): Update vec_perm_builder constructor.
7604 (expand_mult_highpart): Use vec_perm_builder instead of
7605 auto_vec_perm_indices.
7606 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
7607 vec_perm_indices instead of auto_vec_perm_indices. Use a single
7608 or double series encoding as appropriate.
7609 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
7610 vec_perm_indices instead of auto_vec_perm_indices.
7611 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7612 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7613 (vect_permute_store_chain): Likewise.
7614 (vect_grouped_load_supported): Likewise.
7615 (vect_permute_load_chain): Likewise.
7616 (vect_shift_permute_load_chain): Likewise.
7617 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7618 (vect_transform_slp_perm_load): Likewise.
7619 (vect_schedule_slp_instance): Likewise.
7620 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7621 (vectorizable_mask_load_store): Likewise.
7622 (vectorizable_bswap): Likewise.
7623 (vectorizable_store): Likewise.
7624 (vectorizable_load): Likewise.
7625 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
7626 vec_perm_indices instead of auto_vec_perm_indices. Use
7627 tree_to_vec_perm_builder to read the vector from a tree.
7628 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
7629 vec_perm_builder instead of a vec_perm_indices.
7630 (have_whole_vector_shift): Use vec_perm_builder and
7631 vec_perm_indices instead of auto_vec_perm_indices. Leave the
7632 truncation to calc_vec_perm_mask_for_shift.
7633 (vect_create_epilog_for_reduction): Likewise.
7634 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
7635 from auto_vec_perm_indices to vec_perm_indices.
7636 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7637 instead of changing individual elements.
7638 (aarch64_vectorize_vec_perm_const): Use new_vector to install
7639 the vector in d.perm.
7640 * config/arm/arm.c (expand_vec_perm_d::perm): Change
7641 from auto_vec_perm_indices to vec_perm_indices.
7642 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7643 instead of changing individual elements.
7644 (arm_vectorize_vec_perm_const): Use new_vector to install
7645 the vector in d.perm.
7646 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
7647 Update vec_perm_builder constructor.
7648 (rs6000_expand_interleave): Likewise.
7649 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
7650 (rs6000_expand_interleave): Likewise.
7652 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7654 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
7655 to qimode could truncate the indices.
7656 * optabs.c (expand_vec_perm_var): Likewise.
7658 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7660 * Makefile.in (OBJS): Add vec-perm-indices.o.
7661 * vec-perm-indices.h: New file.
7662 * vec-perm-indices.c: Likewise.
7663 * target.h (vec_perm_indices): Replace with a forward class
7665 (auto_vec_perm_indices): Move to vec-perm-indices.h.
7666 * optabs.h: Include vec-perm-indices.h.
7667 (expand_vec_perm): Delete.
7668 (selector_fits_mode_p, expand_vec_perm_var): Declare.
7669 (expand_vec_perm_const): Declare.
7670 * target.def (vec_perm_const_ok): Replace with...
7671 (vec_perm_const): ...this new hook.
7672 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
7673 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7674 * doc/tm.texi: Regenerate.
7675 * optabs.def (vec_perm_const): Delete.
7676 * doc/md.texi (vec_perm_const): Likewise.
7677 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7678 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7679 expand_vec_perm for constant permutation vectors. Assert that
7680 the mode of variable permutation vectors is the integer equivalent
7681 of the mode that is being permuted.
7682 * optabs-query.h (selector_fits_mode_p): Declare.
7683 * optabs-query.c: Include vec-perm-indices.h.
7684 (selector_fits_mode_p): New function.
7685 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7686 is defined, instead of checking whether the vec_perm_const_optab
7687 exists. Use targetm.vectorize.vec_perm_const instead of
7688 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7689 fit in the vector mode before using a variable permute.
7690 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7691 vec_perm_indices instead of an rtx.
7692 (expand_vec_perm): Replace with...
7693 (expand_vec_perm_const): ...this new function. Take the selector
7694 as a vec_perm_indices rather than an rtx. Also take the mode of
7695 the selector. Update call to shift_amt_for_vec_perm_mask.
7696 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7697 Use vec_perm_indices::new_expanded_vector to expand the original
7698 selector into bytes. Check whether the indices fit in the vector
7699 mode before using a variable permute.
7700 (expand_vec_perm_var): Make global.
7701 (expand_mult_highpart): Use expand_vec_perm_const.
7702 * fold-const.c: Includes vec-perm-indices.h.
7703 * tree-ssa-forwprop.c: Likewise.
7704 * tree-vect-data-refs.c: Likewise.
7705 * tree-vect-generic.c: Likewise.
7706 * tree-vect-loop.c: Likewise.
7707 * tree-vect-slp.c: Likewise.
7708 * tree-vect-stmts.c: Likewise.
7709 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7711 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7712 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7713 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7714 (aarch64_vectorize_vec_perm_const): ...this new function.
7715 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7716 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7717 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7718 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7719 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7720 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7721 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7723 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7724 check for NEON modes.
7725 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7726 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7727 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7728 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7730 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7731 the old VEC_PERM_CONST conditions.
7732 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7733 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7734 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7735 (ia64_vectorize_vec_perm_const_ok): Merge into...
7736 (ia64_vectorize_vec_perm_const): ...this new function.
7737 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7738 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7739 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7740 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7741 * config/mips/mips.c (mips_expand_vec_perm_const)
7742 (mips_vectorize_vec_perm_const_ok): Merge into...
7743 (mips_vectorize_vec_perm_const): ...this new function.
7744 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7745 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7746 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7747 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7748 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7749 (rs6000_expand_vec_perm_const): Delete.
7750 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7752 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7753 (altivec_expand_vec_perm_const_le): Take each operand individually.
7754 Operate on constant selectors rather than rtxes.
7755 (altivec_expand_vec_perm_const): Likewise. Update call to
7756 altivec_expand_vec_perm_const_le.
7757 (rs6000_expand_vec_perm_const): Delete.
7758 (rs6000_vectorize_vec_perm_const_ok): Delete.
7759 (rs6000_vectorize_vec_perm_const): New function.
7760 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7761 an element count and rtx array.
7762 (rs6000_expand_extract_even): Update call accordingly.
7763 (rs6000_expand_interleave): Likewise.
7764 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7765 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7766 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7767 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7768 (rs6000_expand_vec_perm_const): Delete.
7769 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7770 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7771 (altivec_expand_vec_perm_const_le): Take each operand individually.
7772 Operate on constant selectors rather than rtxes.
7773 (altivec_expand_vec_perm_const): Likewise. Update call to
7774 altivec_expand_vec_perm_const_le.
7775 (rs6000_expand_vec_perm_const): Delete.
7776 (rs6000_vectorize_vec_perm_const_ok): Delete.
7777 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7778 reference to the SPE evmerge intructions.
7779 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7780 an element count and rtx array.
7781 (rs6000_expand_extract_even): Update call accordingly.
7782 (rs6000_expand_interleave): Likewise.
7783 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7784 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7786 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7788 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7790 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7791 vector mode and that that mode matches the mode of the data
7793 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7794 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7795 directly using expand_vec_perm_1 when forcing selectors into
7797 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7799 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7801 * optabs-query.h (can_vec_perm_p): Delete.
7802 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7803 * optabs-query.c (can_vec_perm_p): Split into...
7804 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7805 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7806 particular selector is valid.
7807 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7808 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7809 (vect_grouped_load_supported): Likewise.
7810 (vect_shift_permute_load_chain): Likewise.
7811 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7812 (vect_transform_slp_perm_load): Likewise.
7813 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7814 (vectorizable_bswap): Likewise.
7815 (vect_gen_perm_mask_checked): Likewise.
7816 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7817 implementations of variable permutation vectors into account
7818 when deciding which selector to use.
7819 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7820 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7821 with a false third argument.
7822 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7823 to test whether the constant selector is valid and can_vec_perm_var_p
7824 to test whether a variable selector is valid.
7826 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7828 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7829 * optabs-query.c (can_vec_perm_p): Likewise.
7830 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7831 instead of vec_perm_indices.
7832 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7833 (vect_gen_perm_mask_checked): Likewise,
7834 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7835 (vect_gen_perm_mask_checked): Likewise,
7837 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7839 * optabs-query.h (qimode_for_vec_perm): Declare.
7840 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7841 (qimode_for_vec_perm): ...this new function.
7842 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7844 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7846 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7847 does not have a conditional at the top.
7849 2018-01-02 Richard Biener <rguenther@suse.de>
7851 * ipa-inline.c (big_speedup_p): Fix expression.
7853 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7856 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7859 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7863 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7864 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7865 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7866 cond_taken_branch_cost 3->4.
7868 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7870 PR tree-optimization/83581
7871 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7872 TODO_cleanup_cfg if any changes have been made.
7875 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7876 convert_modes if target mode has the right side, but different mode
7880 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7881 last argument when extracting from CONCAT. If either from_real or
7882 from_imag is NULL, use expansion through memory. If result is not
7883 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7884 the parts directly to inner mode, if even that fails, use expansion
7888 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7889 check for bswap in mode rather than HImode and use that in expand_unop
7892 Copyright (C) 2018 Free Software Foundation, Inc.
7894 Copying and distribution of this file, with or without modification,
7895 are permitted in any medium without royalty provided the copyright
7896 notice and this notice are preserved.