1 2018-03-02 Jeff Law <law@redhat.com>
3 * reorg.c (stop_search_p): Handle DEBUG_INSN.
4 (redundant_insn, fill_simple_delay_slots): Likewise.
5 (fill_slots_from_thread): Likewise.
6 * resource.c (mark_referenced_resources): Likewise.
7 (mark_set_resources, find_dead_or_set_registers): Likewise.
9 2018-03-02 Jakub Jelinek <jakub@redhat.com>
11 * substring-locations.h (format_warning_va): Formatting fix for
13 (format_warning_at_substring): Fix up ATTRIBUTE_GCC_DIAG second
15 (format_warning_n_va, format_warning_at_substring_n): New prototypes.
16 * substring-locations.c: Include intl.h.
17 (format_warning_va): Turned into small wrapper around
18 format_warning_n_va, renamed to ...
19 (format_warning_n_va): ... this, add N and PLURAL_GMSGID arguments,
20 rename GMSGID to SINGULAR_GMSGID, if SINGULAR_GMSGID != PLURAL_GMSGID,
22 (format_warning_at_substring_n): New function.
23 * gimple-ssa-sprintf.c: Remove GCC diagnostic ignored pragma.
24 (fmtwarn): Add ATTRIBUTE_GCC_DIAG. Turn into a copy of
25 format_warning_at_substring with just a shorter name instead of
26 const function pointer.
27 (fmtwarn_n): New function.
28 (maybe_warn, format_directive, parse_directive): Use fmtwarn_n where
29 appropriate, get rid of all the fmtstr temporaries, move conditionals
30 with G_() wrapped string literals directly into fmtwarn arguments,
31 cast dir.len to (int), formatting fixes.
33 2018-03-02 Thomas Schwinge <thomas@codesourcery.com>
35 * doc/invoke.texi: Remove "Cilk Plus" references.
37 2018-03-02 Jakub Jelinek <jakub@redhat.com>
38 Richard Biener <rguenther@suse.de>
41 * expr.c (expand_expr_real_1) <case CALL_EXPR>: Don't emit diagnostics
42 for error or warning attributes if CALL_FROM_THUNK_P is set.
45 2018-03-02 Jakub Jelinek <jakub@redhat.com>
48 * config/pa/pa.h (TARGET_CPU_CPP_BUILTINS): Predefine
49 __SIZEOF_128__ macro if HPUX_LONG_DOUBLE_LIBRARY.
52 * config/ia64/ia64.h (TARGET_CPU_CPP_BUILTINS): Predefine
53 __SIZEOF_{FPREG,FLOAT{80,128}}__ macros.
55 * predict.c (test_prediction_value_range): Use PROB_UNINITIALIZED
56 instead of -1U in last predictors element's probability member.
58 2018-03-02 Eric Botcazou <ebotcazou@adacore.com>
61 * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
62 arguments if they are comparable.
64 2018-03-02 Richard Sandiford <richard.sandiford@linaro.org>
66 PR tree-optimization/84634
67 * tree-vect-stmts.c (vectorizable_store, vectorizable_load): Replace
68 masks and masked_loop_p with a single loop_masks, making sure it's
69 null for bb vectorization.
71 2018-03-02 Richard Sandiford <richard.sandiford@linaro.org>
73 * tree-vect-data-refs.c (vect_analyze_data_ref_dependence)
74 (vect_analyze_data_ref_access): Use loop->safe_len rather than
75 loop->force_vectorize to check whether there is no alias.
77 2018-03-02 Jakub Jelinek <jakub@redhat.com>
80 * rtl.h (prev_real_nondebug_insn, next_real_nondebug_insn): New
82 * emit-rtl.c (next_real_insn, prev_real_insn): Fix up function
84 (next_real_nondebug_insn, prev_real_nondebug_insn): New functions.
85 * cfgcleanup.c (try_head_merge_bb): Use prev_real_nondebug_insn
86 instead of a loop around prev_real_insn.
87 * combine.c (move_deaths): Use prev_real_nondebug_insn instead of
91 * config/i386/i386.c (ix86_print_operand): Use conditional
92 output_operand_lossage instead of gcc_assert if CONST_VECTOR is not
95 2018-03-02 Richard Biener <rguenther@suse.de>
97 PR tree-optimization/84427
98 * tree-ssa-pre.c (bitmap_remove_expr_from_set): Remove.
99 (bitmap_set_subtract_values): Rewrite to handle multiple
102 (prune_clobbered_mems): Likewise.
103 (phi_translate): Take edge instead of pred/phiblock.
104 (phi_translate_1): Likewise.
105 (phi_translate_set): Likewise. Insert all translated
106 exprs for a value into the set, keeping possibly multiple
107 expressions per value.
108 (compute_antic_aux): Adjust for phi_translate changes.
109 When intersecting union the expressions and prune those
110 not in the final value set, keeping possibly multiple
111 expressions per value. Do not use value-insertion
112 for unioning ANTIC_OUT U EXP_GEN - TMP_GEN but merge
113 all expressions. Add verification that the value-sets
114 only shrink during iteration.
115 (compute_partial_antic_aux): Adjust for the phi_translate changes.
116 (do_pre_regular_insertion): Likewise.
117 (do_pre_partial_partial_insertion): Likewise.
119 2018-03-02 Richard Biener <rguenther@suse.de>
122 * config/darwin.c (saved_debug_info_level): New static global.
123 (darwin_asm_lto_start): Disable debug info generation for LTO out.
124 (darwin_asm_lto_end): Restore debug info generation settings.
126 2018-03-01 Martin Liska <mliska@suse.cz>
129 * sanopt.c (sanitize_rewrite_addressable_params): Do not handle
132 2018-03-01 Richard Biener <rguenther@suse.de>
135 * dwarf2out.c (gen_variable_die): Properly handle late VLA
136 type annotation with LTO when debug was disabled at compile-time.
138 2018-03-01 Matthew Fortune <mfortune@gmail.com>
140 * config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
142 (mips_final_postscan_insn): Likewise.
144 2018-03-01 Richard Sandiford <richard.sandiford@linaro.org>
146 PR rtl-optimization/84528
147 * alias.c (init_alias_target): Add commentary.
148 (init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
149 a unique base value if the frame pointer is not eliminated
150 to the stack pointer.
152 2018-03-01 Tom de Vries <tom@codesourcery.com>
154 PR rtl-optimization/83327
155 * lra-int.h (hard_regs_spilled_into): Declare.
156 * lra.c (hard_regs_spilled_into): Define.
157 (init_reg_info): Init hard_regs_spilled_into.
158 * lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
159 * lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
160 (process_bb_lives): Handle hard_regs_spilled_into.
161 (lra_create_live_ranges_1): Before doing liveness propagation, clear
162 regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.
164 2018-02-28 David Edelsohn <dje.gcc@gmail.com>
166 * config.gcc (powerpc-ibm-aix7.1.*): New stanza.
167 (powerpc-ibm-aix[789]*): Default to AIX 7.2.
168 * config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
169 * config/rs6000/aix72.h: New file.
171 2018-02-28 Jakub Jelinek <jakub@redhat.com>
173 * gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
174 instead of warning_at with conditional singular and plural messages
178 * stor-layout.c (update_alignment_for_field): For
179 targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
180 && !DECL_PACKED (field), do the alignment update, just use
181 only desired_align instead of MAX (type_align, desired_align)
183 (place_field): Don't do known_align < desired_align handling
184 early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
185 is non-NULL, instead do it after rli->prev_field handling and
186 only if not within a bitfield word. For DECL_PACKED (field)
187 use type_align of BITS_PER_UNIT.
189 2018-02-28 Eric Botcazou <ebotcazou@adacore.com>
191 * config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
192 superfluous parentheses and trailing spaces.
194 2018-02-28 Richard Biener <rguenther@suse.de>
196 PR tree-optimization/84584
197 * graphite-scop-detection.c (scop_detection::add_scop): Discard
198 SCoPs with fake exit edge.
200 2018-02-28 Martin Liska <mliska@suse.cz>
203 * timevar.c (timer::print): Fix format to properly print 100%
206 2018-02-28 Richard Biener <rguenther@suse.de>
209 * genmatch.c (capture_info::walk_match): Do not mark
210 captured expressions without operands as expr_p given
211 they act more like predicates and should be subject to
212 "lost tail" side-effect preserving.
214 2018-02-28 Alexandre Oliva <aoliva@redhat.com>
216 PR rtl-optimization/81611
217 * auto-inc-dec.c (attempt_change): Move dead note from
218 mem_insn if it's the next use of regno
219 (find_address): Take address use of reg holding
220 non-incremented value. Add parm to limit search to the named
222 (merge_in_block): Attempt to use a mem insn that is the next
223 use of the original regno.
225 2018-02-27 Martin Sebor <msebor@redhat.com>
228 * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
229 * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
231 2018-02-27 Martin Sebor <msebor@redhat.com>
234 * diagnostic-core.h (warning_n, error_n, inform_n): Change
235 n argument to unsigned HOST_WIDE_INT.
236 * diagnostic.c (warning_n, error_n, inform_n): Ditto.
237 (diagnostic_n_impl): Ditto. Handle arguments in excess of LONG_MAX.
238 * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
239 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
241 2018-02-27 Richard Biener <rguenther@suse.de>
243 PR tree-optimization/84512
244 * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
245 Do not use the estimate returned from record_stmt_cost for
246 the scalar iteration cost but sum properly using add_stmt_cost.
248 2018-02-27 Richard Biener <rguenther@suse.de>
250 PR tree-optimization/84466
251 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
252 Adjust last change to less strictly validate use operands.
254 2018-02-27 Martin Liska <mliska@suse.cz>
256 PR gcov-profile/84548
257 * gcov.c (process_file): Allow partial overlap and consider it
258 also as group functions.
259 (output_lines): Properly calculate range of lines for a group.
261 2018-02-27 Martin Liska <mliska@suse.cz>
263 * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
264 'ggc' suffixes. Change first column width.
265 (timer::print): Fix formatting of the column.
267 2018-02-27 Alexandre Oliva <aoliva@redhat.com>
269 * tree-ssa-live.c (remove_unused_scope_block_p): Do not
270 preserve inline entry blocks for the sake of debug inline
271 entry point markers alone.
272 (remove_unused_locals): Suggest in comments a better place to
273 force the preservation of inline entry blocks that are
274 otherwise unused, but do not preserve them.
276 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
278 * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
280 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
283 * config/i386/constraints.md (Bs): Replace
284 ix86_indirect_branch_register with
285 TARGET_INDIRECT_BRANCH_REGISTER.
287 * config/i386/i386.md (indirect_jump): Likewise.
288 (tablejump): Likewise.
289 (*sibcall_memory): Likewise.
290 (*sibcall_value_memory): Likewise.
291 Peepholes of indirect call and jump via memory: Likewise.
292 (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
293 (*sibcall_value_GOT_32): Likewise.
294 * config/i386/predicates.md (indirect_branch_operand): Likewise.
295 (GOT_memory_operand): Likewise.
296 (call_insn_operand): Likewise.
297 (sibcall_insn_operand): Likewise.
298 (GOT32_symbol_operand): Likewise.
299 * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
301 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
303 PR rtl-optimization/83496
304 * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
305 booleans to RTXes. Call fix_reg_dead_note on every non-null element.
306 (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
307 redundant insn, if any.
308 (relax_delay_slots): Likewise.
309 (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
311 2018-02-26 Richard Sandiford <richard.sandiford@linaro.org>
313 PR tree-optimization/83965
314 * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
315 that grouped statements are part of a reduction chain. Return
316 true if the statement is not marked as a reduction itself but
318 (vect_recog_dot_prod_pattern): Don't check whether the statement
319 is part of a group here.
320 (vect_recog_sad_pattern): Likewise.
321 (vect_recog_widen_sum_pattern): Likewise.
323 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
326 * final.c (rest_of_clean_state): Also look for calls inside sequences.
328 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
331 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
333 (ix86_output_indirect_function_return): New prototype.
334 (ix86_split_simple_return_pop_internal): Likewise.
335 * config/i386/i386.c (indirect_return_via_cx): New.
336 (indirect_return_via_cx_bnd): Likewise.
337 (indirect_thunk_name): Handle return va CX_REG.
338 (output_indirect_thunk_function): Create alias for
339 __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
340 (ix86_output_indirect_jmp): Remove the bool argument.
341 (ix86_output_indirect_function_return): New function.
342 (ix86_split_simple_return_pop_internal): Likewise.
343 * config/i386/i386.md (*indirect_jump): Don't pass false
344 to ix86_output_indirect_jmp.
345 (*tablejump_1): Likewise.
346 (simple_return_pop_internal): Change it to define_insn_and_split.
347 Call ix86_split_simple_return_pop_internal to split it for
349 (simple_return_indirect_internal): Call
350 ix86_output_indirect_function_return instead of
351 ix86_output_indirect_jmp.
353 2018-02-26 Jakub Jelinek <jakub@redhat.com>
356 * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
357 memset and value initialization afterwards.
359 2018-02-26 Christophe Lyon <christophe.lyon@linaro.org>
361 * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
363 2018-02-26 Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
366 * common/config/aarch64/aarch64-common.c
367 (aarch_option_optimization_table[]): Switch
368 off fomit-frame-pointer
370 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
371 Chung-Ju Wu <jasonwucj@gmail.com>
373 * config/nds32/nds32-multiple.md (load_multiple): Disallow
375 (store_multiple): Ditto.
377 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
379 * config.gcc: Add --with-cpu support for nds32 target.
380 * config/nds32/nds32-opts.h (nds32_cpu_type): New.
381 * config/nds32/nds32.opt: Add -mcpu= option.
383 2018-02-25 Segher Boessenkool <segher@kernel.crashing.org>
385 * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
386 isel=yes): Warn for these deprecated options.
388 2018-02-23 David Edelsohn <dje.gcc@gmail.com>
390 * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
391 ISA_2_5_MASKS_EMBEDDED.
393 2018-02-23 Jakub Jelinek <jakub@redhat.com>
395 * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
396 p->max as pointers rather than using iterative_hash_expr.
398 2018-02-23 Carl Love <cel@us.ibm.com>
400 * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
401 macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
403 * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
404 P8V_BUILTIN_VEC_VSIGNED2. Change VSX_BUILTIN_VEC_VUNSIGNED2 to
405 P8V_BUILTIN_VEC_VUNSIGNED2.
407 2018-02-22 Vladimir Makarov <vmakarov@redhat.com>
410 * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
411 * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
413 * lra-constraints.c (curr_insn_transform): Set up
414 LRA_NON_CLOBBERED_ALT for moves processed on the fast path. Use
416 (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
417 * lra-eliminations.c (spill_pseudos): Ditto.
418 (process_insn_for_elimination): Ditto.
419 * lra-lives.c (reg_early_clobber_p): Use the new macros.
420 * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
421 LRA_NON_CLOBBERED_ALT.
423 2018-02-22 Martin Sebor <msebor@redhat.com>
425 PR tree-optimization/84480
426 * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
427 to maybe_diag_stxncpy_trunc. Call it.
428 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
429 from gimple_fold_builtin_strcpy. Print inlining stack.
430 (handle_builtin_stxncpy): Print inlining stack.
431 * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
433 2018-02-22 H.J. Lu <hongjiu.lu@intel.com>
436 * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
437 error when -mindirect-branch=thunk-extern, -fcf-protection=branch
438 and -fcheck-pointer-bounds are used together.
439 (indirect_thunk_prefix): New enum.
440 (indirect_thunk_need_prefix): New function.
441 (indirect_thunk_name): Replace need_bnd_p with need_prefix. Use
442 "_nt" instead of "_bnd" for NOTRACK prefix.
443 (output_indirect_thunk): Replace need_bnd_p with need_prefix.
444 (output_indirect_thunk_function): Likewise.
446 (ix86_code_end): Update output_indirect_thunk_function calls.
447 (ix86_output_indirect_branch_via_reg): Replace
448 ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
449 (ix86_output_indirect_branch_via_push): Likewise.
450 (ix86_output_function_return): Likewise.
451 * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
452 incompatible with -fcf-protection=branch and
453 -fcheck-pointer-bounds.
455 2018-02-22 Steve Ellcey <sellcey@cavium.com>
458 * config/aarch64/aarch64.c (aarch64_print_address_internal):
459 Change gcc_assert call to output_operand_lossage.
461 2018-02-22 Steve Ellcey <sellcey@cavium.com>
463 * doc/extend.texi (__builtin_extend_pointer): Document builtin.
465 2018-02-22 DJ Delorie <dj@redhat.com>
466 Sebastian Perta <sebastian.perta@renesas.com>
467 Oleg Endo <olegendo@gcc.gnu.org>
469 * config/rx/rx.c (rx_rtx_costs): New function.
470 (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
472 2018-02-22 Thomas Preud'homme <thomas.preudhomme@arm.com>
474 * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
476 2018-02-22 Martin Liska <mliska@suse.cz>
479 * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
480 Add "native" as a possible value.
482 2018-02-22 Martin Liska <mliska@suse.cz>
485 * config/i386/i386.c (ix86_option_override_internal):
486 Add "native" as a possible value for -march and -mtune.
488 2018-02-22 Jakub Jelinek <jakub@redhat.com>
491 * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
492 to all type variants.
494 PR tree-optimization/84503
495 * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
496 width as info->bitpos + info->bitsize - start.
497 (merged_store_group::merge_overlapping): Simplify width computation.
498 (check_no_overlap): New function.
499 (imm_store_chain_info::try_coalesce_bswap): Compute expected
500 start + width and last_order of the group, fail if check_no_overlap
502 (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
503 to group if check_no_overlap fails.
505 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
507 * config/rs6000/altivec.md: Delete contraint arguments to
508 define_expand, define_split, and define_peephole2, and in
509 define_insn_and_split if always unused.
510 * config/rs6000/darwin.md: Ditto.
511 * config/rs6000/dfp.md: Ditto.
512 * config/rs6000/rs6000.md: Ditto.
513 * config/rs6000/sync.md: Ditto.
514 * config/rs6000/vector.md: Ditto.
515 * config/rs6000/vsx.md: Ditto.
517 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
519 * config/rs6000/altivec.md: Write output control strings as braced
520 blocks instead of double-quoted strings.
521 * config/rs6000/darwin.md: Ditto.
522 * config/rs6000/rs6000.md: Ditto.
523 * config/rs6000/vector.md: Ditto.
524 * config/rs6000/vsx.md: Ditto.
526 2018-02-21 Jason Merrill <jason@redhat.com>
528 PR c++/84314 - ICE with templates and fastcall attribute.
529 * attribs.c (build_type_attribute_qual_variant): Remove assert.
531 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
533 * ipa-cp.c (determine_versionability): Fix comment typos.
535 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
538 * ipa-cp.c (determine_versionability): Do not version functions caling
541 2018-02-21 Martin Liska <mliska@suse.cz>
544 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
545 Add "native" as a possible value.
546 * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT): Define
547 the macro when native cpu detection is available.
549 2018-02-21 Martin Liska <mliska@suse.cz>
552 * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
553 Add "native" as a possible value.
554 * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
555 when native cpu detection is available.
557 2018-02-21 Jakub Jelinek <jakub@redhat.com>
558 Martin Sebor <msebor@redhat.com>
560 PR tree-optimization/84478
561 * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
563 * gimple-fold.c (get_range_strlen): Make minlen const and assume it
564 can't be NULL. Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
565 support which is conservatively correct, for 2 only stay conservative
566 for maxlen. Formatting and comment capitalization fixes. Add STRICT
567 argument to the 2 argument get_range_strlen, adjust 6 arg
568 get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
570 (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
571 (gimple_fold_builtin_strlen): Pass true as last argument to
574 2018-02-20 Martin Sebor <msebor@redhat.com>
577 * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
578 (builtin_memref::set_base_and_offset): Same. Handle inner references.
579 (builtin_memref::builtin_memref): Factor out parts into
580 set_base_and_offset and call it.
582 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
585 * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
586 is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
587 greater precision. If to_mode is a MODE_PARTIAL_INT, stop the
588 search at the associated MODE_INT.
590 2018-02-20 Jeff Law <law@redhat.com>
593 PR tree-optimization/81592
595 * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
596 for range data rather than using global data.
597 * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
598 range data rather than using global data.
599 * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
600 pass it to children as needed.
601 (struct directive::fmtresult): Similarly.
602 (struct directive::set_width): Similarly.
603 (struct directive::set_precision): Similarly.
604 (format_integer, format_directive, parse_directive): Similarly.
605 (format_none): Accept unnamed vr_values parameter.
606 (format_percent, format_floating, format_character): Similarly.
607 (format_string, format_plain): Similarly.
608 * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
609 the EVRP range analyzer for range data rather than using global data.
610 * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
611 gimple-ssa-evrp-analyze.h
612 (class sprintf_dom_walker): Add after_dom_children member function.
613 Add evrp_range_analyzer member.
614 (sprintf_dom_walker::before_dom_children): Call into the EVRP
615 range analyzer as needed.
616 (sprintf_dom_walker::after_dom_children): New member function.
617 * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
619 (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
620 (evrp_range_analyzer::pop_to_marker): Likewise.
622 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
624 PR tree-optimization/84419
625 * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
626 with the required type if its current type is compatible but
629 2018-02-20 Jakub Jelinek <jakub@redhat.com>
632 * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
635 2018-02-20 Martin Liska <mliska@suse.cz>
638 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
639 possible values if we don't have a hint.
641 2018-02-20 Martin Liska <mliska@suse.cz>
645 * final.c (shorten_branches): Build align_tab array with one
647 * opts.c (finish_options): Add alignment option limit check.
648 (MAX_CODE_ALIGN): Likewise.
649 (MAX_CODE_ALIGN_VALUE): Likewise.
650 * doc/invoke.texi: Document maximum allowed option value for
651 all -falign-* options.
653 2018-02-19 Jakub Jelinek <jakub@redhat.com>
656 * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
657 * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
658 * var-tracking.c (emit_note_insn_var_location): Remove all references
659 to NOTE_INSN_CALL_ARG_LOCATION.
660 (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
661 the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
662 Use copy_rtx_if_shared.
663 * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
664 NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
665 (dwarf2out_var_location): Remove handling of
666 NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
668 * final.c (final_scan_insn): Remove all references to
669 NOTE_INSN_CALL_ARG_LOCATION.
670 (rest_of_clean_state): Likewise. Remove REG_CALL_ARG_LOCATION notes
671 before dumping final insns.
672 * except.c (emit_note_eh_region_end): Remove all references to
673 NOTE_INSN_CALL_ARG_LOCATION.
674 * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
675 * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
676 * config/arc/arc.c (hwloop_optimize): Likewise.
677 * config/arm/arm.c (create_fix_barrier): Likewise.
678 * config/s390/s390.c (s390_chunkify_start): Likewise.
679 * config/sh/sh.c (find_barrier): Likewise.
680 * config/i386/i386.c (rest_of_insert_endbranch,
681 ix86_seh_fixup_eh_fallthru): Likewise.
682 * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
683 * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
684 * config/frv/frv.c (frv_function_prologue): Likewise.
685 * emit-rtl.c (try_split): Likewise. Copy over REG_CALL_ARG_LOCATION
687 (note_outside_basic_block_p): Remove all references to
688 NOTE_INSN_CALL_ARG_LOCATION.
689 * gengtype.c (adjust_field_rtx_def): Likewise.
690 * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
692 * jump.c (cleanup_barriers, delete_related_insns): Likewise.
693 * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
696 * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
699 PR tree-optimization/84452
700 * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
701 expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
704 2018-02-19 Martin Liska <mliska@suse.cz>
707 * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
709 2018-02-19 Martin Liska <mliska@suse.cz>
710 Richard Sandiford <richard.sandiford@linaro.org>
712 PR tree-optimization/82491
713 * gimple-fold.c (get_base_constructor): Make earlier bail out
716 2018-02-19 Carl Love <cel@us.ibm.com>
718 * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
719 BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
721 * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
724 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
726 * config/rl78/rl78.md (movdf): New define expand.
728 2018-02-19 Martin Liska <mliska@suse.cz>
731 * doc/invoke.texi: Fix typo.
732 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
734 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
736 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
737 handle rs6000_single_float and rs6000_double_float specially for
740 2018-02-16 Jeff Law <law@redhat.com>
742 * config/rx/rx.c (add_pop_cfi_notes): New function.;
745 2018-02-16 Jakub Jelinek <jakub@redhat.com>
748 * ipa-inline.c (inline_small_functions): Fix a typo.
750 2018-02-16 Nathan Sidwell <nathan@acm.org>
752 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
754 2018-02-16 Carl Love <cel@us.ibm.com>
756 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
757 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
758 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
759 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
760 expansion to P8V_BUILTIN_VEC_FLOAT2.
762 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
764 PR rtl-optimization/70023
765 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
766 src_regno into account.
768 2018-02-16 Carl Love <cel@us.ibm.com>
770 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
771 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
772 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
773 * config/rs6000/rs6000.c: Remove case statements for
774 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
775 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
776 and P9V_BUILTIN_VEC_VINSERT4B.
777 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
778 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
779 * config/rs6000/vsx.md:
780 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
783 2018-02-16 Carl Love <cel@us.ibm.com>
785 * config/rs6000/altivec.h: Add builtin names vec_extract4b
787 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
789 * config/rs6000/rs6000-c.c: Add the definitions for
790 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
791 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
792 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
793 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
794 definition for insert4b and define insn *insert3b_internal.
795 * doc/extend.texi: Add documentation for vec_extract4b.
797 2018-02-16 Nathan Sidwell <nathan@acm.org>
799 * doc/extend.texi (Backwards Compatibility): Mention friend
800 injection. Note for-scope is deprecated.
801 * doc/invoke.texi (-ffriend-injection): Deprecate.
803 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
805 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
806 that moved to I2, also allow destinations that are a paradoxical
807 subreg (instead of a normal reg).
809 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
812 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
815 2018-02-16 Richard Biener <rguenther@suse.de>
817 PR tree-optimization/84037
818 PR tree-optimization/84016
820 * config/i386/i386.c (ix86_builtin_vectorization_cost):
821 Adjust vec_construct for the fact we need additional higher latency
822 128bit inserts for AVX256 and AVX512 vector builds.
823 (ix86_add_stmt_cost): Scale vector construction cost for
826 2018-02-16 Richard Biener <rguenther@suse.de>
828 PR tree-optimization/84417
829 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
830 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
831 (non_rewritable_lvalue_p): Likewise, use poly-ints.
833 2018-02-16 Martin Liska <mliska@suse.cz>
836 * internal-fn.def (ASAN_CHECK): Set proper flags.
837 (ASAN_MARK): Likewise.
839 2018-02-16 Julia Koval <julia.koval@intel.com>
841 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
844 2018-02-16 Jakub Jelinek <jakub@redhat.com>
847 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
848 Use ++iter rather than iter++ for std::list iterators.
849 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
850 defer deleting them until all nodes in the forest are processed. Do
851 free even leaf nodes. Change to_process into auto_vec.
854 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
855 * vec.h (vec_default_construct): Use memset instead of placement new
856 if BROKEN_VALUE_INITIALIZATION is defined.
857 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
858 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
861 PR rtl-optimization/83723
862 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
863 * lra.c (lra_substitute_pseudo): Likewise. If true, use
864 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
866 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
868 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
870 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
872 PR rtl-optimization/81443
873 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
874 from inner REGs to paradoxical SUBREGs.
876 2018-02-16 Richard Biener <rguenther@suse.de>
878 PR tree-optimization/84399
879 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
880 For operands we can analyze at their definition make sure we can
881 analyze them at each use as well.
883 2018-02-16 Richard Biener <rguenther@suse.de>
885 PR tree-optimization/84190
886 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
887 volatile accesses if the decl isn't volatile.
889 2018-02-15 Jason Merrill <jason@redhat.com>
891 PR c++/84314 - ICE with templates and fastcall attribute.
892 * attribs.c (build_type_attribute_qual_variant): Don't clobber
893 TYPE_CANONICAL on an existing type.
895 2018-02-15 Jakub Jelinek <jakub@redhat.com>
897 PR tree-optimization/84383
898 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
899 dstoff nor call operand_equal_p if dstbase is NULL.
901 PR tree-optimization/84334
902 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
903 also a CONSTANT_CLASS_P, punt.
905 2018-02-14 Jim Wilson <jimw@sifive.com>
907 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
908 first SMALL_OPERAND check. New local min_second_step. Move assert
909 to where locals are set. Add TARGET_RVC support.
910 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
912 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
914 * doc/invoke.texi: Correct -Wformat-overflow code sample.
916 2018-02-14 Martin Sebor <msebor@redhat.com>
918 PR tree-optimization/83698
919 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
920 arrays constrain the offset range to their bounds.
921 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
922 (builtin_access::overlap): Avoid setting the size of overlap if it's
924 (maybe_diag_overlap): Also consider arrays when deciding what values
925 of offsets to include in diagnostics.
927 2018-02-14 Martin Sebor <msebor@redhat.com>
930 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
931 that correspond to the kind of a declaration.
933 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
936 * config/pa/pa.md: Load address of PIC label using the linkage table
937 if the label is nonlocal.
939 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
941 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
942 warning message if user requests -maltivec=be.
943 * doc/invoke.texi: Document deprecation of -maltivec=be.
945 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
948 * config/rs6000/rs6000-c.c: Update definitions for
949 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
950 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
952 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
955 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
956 add _get_ssp intrinsics. Remove argument from
957 __builtin_ia32_rdssp[d|q].
958 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
959 * config/i386/i386-builtin.def: Remove argument from
960 __builtin_ia32_rdssp[d|q].
961 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
962 ix86_expand_special_args_builtin for _rdssp[d|q].
963 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
964 Clear register before usage.
965 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
966 Add documentation for new _get_ssp and _inc_ssp intrinsics.
968 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
970 PR tree-optimization/84357
971 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
972 operand 1 of an ARRAY_REF too.
974 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
977 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
978 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
980 (set_of_reg): New struct.
981 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
982 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
983 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
985 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
986 Split into bitclr, bitset, bitinvert patterns if appropriate.
987 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
988 use rx_fuse_in_memory_bitop.
989 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
990 to named insn, correct maximum insn length.
992 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
995 * machmode.def: Define a complex mode for PARTIAL_INT.
996 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
998 * doc/rtl.texi: Document CSPImode.
999 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
1001 (msp430_hard_regno_nregs_with_padding): Likewise.
1003 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
1006 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
1008 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
1010 PR rtl-optimization/84169
1011 * combine.c (try_combine): New variable split_i2i3. Set it to true if
1012 we generated a parallel as new i3 and we split that to new i2 and i3
1013 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
1014 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
1015 those to i2, not i1. Partially rewrite this scan code.
1017 2018-02-13 Jakub Jelinek <jakub@redhat.com>
1020 * stor-layout.c (place_field): For variable length fields, adjust
1021 offset_align afterwards not just based on the field's alignment,
1022 but also on the size.
1025 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
1026 of exps and logs in the use_exp2 case.
1028 2018-02-13 Jeff Law <law@redhat.com>
1030 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
1033 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
1036 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
1040 * common.opt (gas-loc-support, gas-locview-support): New.
1041 (ginline-points, ginternal-reset-location-views): New.
1042 * doc/invoke.texi: Document them. Use @itemx where intended.
1043 (gvariable-location-views): Adjust.
1044 * target.def (reset_location_view): New.
1045 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
1046 (TARGET_RESET_LOCATION_VIEW): New.
1047 * doc/tm.texi: Rebuilt.
1048 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
1049 (dwarf2out_default_as_locview_support): New.
1050 (output_asm_line_debug_info): Use option variables.
1051 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
1052 (output_loc_list): Likewise.
1053 (add_high_low_attributes): Check option variables.
1054 Don't output entry view attribute in strict mode.
1055 (gen_inlined_subroutine_die): Check option variables.
1056 (dwarf2out_inline_entry): Likewise.
1057 (init_sections_and_labels): Likewise.
1058 (dwarf2out_early_finish): Likewise.
1059 (maybe_reset_location_view): New, from...
1060 (dwarf2out_var_location): ... here. Call it.
1061 * debug.h (dwarf2out_default_as_loc_support): Declare.
1062 (dwarf2out_default_as_locview_support): Declare.
1063 * hooks.c (hook_int_rtx_insn_0): New.
1064 * hooks.h (hook_int_rtx_insn_0): Declare.
1065 * toplev.c (process_options): Take -gas-loc-support and
1066 -gas-locview-support from dwarf2out. Enable
1067 -gvariable-location-views by default only with locview
1068 assembler support. Enable -ginternal-reset-location-views by
1069 default only if the target defines the corresponding hook.
1070 Enable -ginline-points by default if location views are
1071 enabled; force it disabled if statement frontiers are
1073 * tree-inline.c (expand_call_inline): Check option variables.
1074 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
1076 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
1078 PR tree-optimization/84321
1079 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
1080 handling. Also check whether the anti-range contains any values
1081 that satisfy the mask; switch to a VR_RANGE if not.
1083 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
1086 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
1088 2018-02-13 Martin Jambor <mjambor@suse.cz>
1091 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
1092 of call statements, also set location of a load to a temporary.
1094 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
1096 * config/rl78/rl78.c (add_vector_labels): New function.
1097 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
1098 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
1099 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
1100 which checks that no arguments are passed.
1101 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
1102 * doc/extend.texi: Documentation for the new attribute.
1104 2018-02-13 Andreas Schwab <schwab@suse.de>
1106 * config/riscv/linux.h (CPP_SPEC): Define.
1108 2018-02-13 Jakub Jelinek <jakub@redhat.com>
1111 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
1112 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
1113 OPTION_MASK_ISA_AES as first argument to def_builtin_const
1114 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
1115 instead of OPTION_MASK_ISA_PCLMUL as first argument to
1116 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
1117 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
1118 temporarily for AES and PCLMUL builtins.
1120 PR tree-optimization/84339
1121 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
1122 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
1126 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
1127 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
1128 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
1130 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
1132 * omp-simd-clone.h: New file.
1133 * omp-simd-clone.c: Include omp-simd-clone.h.
1134 (expand_simd_clones): No longer static.
1135 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
1136 cgraph.h and omp-simd-clone.h.
1137 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
1138 (vect_recog_widen_shift_pattern): Formatting fix.
1139 (vect_pattern_recog_1): Don't check optab for calls.
1142 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
1143 operands[2] into a REG before using gen_lowpart on it.
1145 2018-02-12 Jeff Law <law@redhat.com>
1148 * config/sh/sh.c (find_barrier): Consider a sibling call
1151 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
1152 successfully back substituting a reg.
1154 2018-02-12 Richard Biener <rguenther@suse.de>
1156 PR tree-optimization/84037
1157 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
1158 parameter, move visited init to caller.
1159 (vect_slp_analyze_operations): Separate cost from validity
1160 check, initialize visited once for all instances.
1161 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
1163 * tree-vect-stmts.c (vect_model_simple_cost): Make early
1165 (vect_model_promotion_demotion_cost): Likewise.
1166 (vectorizable_bswap): Guard cost modeling with !slp_node
1167 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
1169 (vectorizable_call): Likewise.
1170 (vectorizable_conversion): Likewise.
1171 (vectorizable_assignment): Likewise.
1172 (vectorizable_shift): Likewise.
1173 (vectorizable_operation): Likewise.
1174 (vectorizable_store): Likewise.
1175 (vectorizable_load): Likewise.
1176 (vectorizable_condition): Likewise.
1177 (vectorizable_comparison): Likewise.
1179 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
1182 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
1183 (ASAN_MARK): Fix fnspec to account for return value, change pointer
1184 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
1186 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1189 * params.def (inline-min-speedup): Increase from 8 to 15.
1190 (max-inline-insns-auto): Decrease from 40 to 30.
1191 * ipa-split.c (consider_split): Add some buffer for function to
1192 be considered inlining candidate.
1193 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
1196 2018-02-12 Richard Biener <rguenther@suse.de>
1198 PR tree-optimization/84037
1199 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
1200 matched stmts if we cannot swap the non-matched ones.
1202 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
1204 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
1205 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
1206 _mm_maskz_scalef_round_ss): New intrinsics.
1207 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
1208 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
1209 __builtin_ia32_scalefss_round): Remove.
1210 (__builtin_ia32_scalefsd_mask_round,
1211 __builtin_ia32_scalefss_mask_round): New intrinsics.
1212 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
1213 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
1214 ((match_operand:VF_128 2 "<round_nimm_predicate>"
1215 "<round_constraint>")): Changed to ...
1216 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
1217 "<round_scalar_constraint>")): ... this.
1218 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
1219 %0, %1, %2<round_op3>}"): Changed to ...
1220 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
1221 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
1222 %2<round_scalar_mask_op3>}"): ... this.
1223 * config/i386/subst.md (round_scalar_nimm_predicate): New.
1225 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
1227 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
1228 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
1229 (_mm_maskz_sqrt_round_ss): New intrinsics.
1230 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
1231 (__builtin_ia32_sqrtsd_mask_round)
1232 (__builtin_ia32_sqrtss_mask_round): New builtins.
1233 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
1234 (__builtin_ia32_sqrtss_round): Remove.
1235 (__builtin_ia32_sqrtsd_mask_round)
1236 (__builtin_ia32_sqrtss_mask_round): New builtins.
1237 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
1238 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
1239 ((match_operand:VF_128 1 "vector_operand"
1240 "xBm,<round_constraint>")): Changed to ...
1241 ((match_operand:VF_128 1 "vector_operand"
1242 "xBm,<round_scalar_constraint>")): ... this.
1243 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
1244 %0, %2, %<iptr>1<round_op3>}): Changed to ...
1245 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
1246 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
1247 %<iptr>1<round_scalar_mask_op3>}): ... this.
1248 ((set_attr "prefix" "<round_prefix>")): Changed to ...
1249 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
1251 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
1254 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
1255 Cast vec_cmpeq result to correct type.
1256 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
1257 Cast vec_cmpgt result to correct type.
1259 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
1261 * final.c (final_scan_insn_1): Renamed from...
1262 (final_scan_insn): ... this. New wrapper, to recover
1263 seen from the outermost call in recursive ones.
1264 * config/sparc/sparc.c (output_return): Drop seen from call.
1265 (output_sibcall): Likewise.
1266 * config/visium/visium.c (output_branch): Likewise.
1268 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
1270 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
1273 2018-02-10 Alan Modra <amodra@gmail.com>
1276 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
1277 Specify LR as an input.
1279 2018-02-10 Jakub Jelinek <jakub@redhat.com>
1282 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
1283 remove_member_access_dummy_vars): New functions.
1284 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
1285 lower_omp_1, execute_lower_omp): Use them.
1287 PR rtl-optimization/84308
1288 * shrink-wrap.c (spread_components): Release todo vector.
1290 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
1292 PR rtl-optimization/57193
1293 * ira-color.c (struct allocno_color_data): Add member
1294 conflict_allocno_hard_prefs.
1295 (update_conflict_allocno_hard_prefs): New.
1296 (bucket_allocno_compare_func): Add a preference based on
1297 conflict_allocno_hard_prefs.
1298 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
1299 (color_allocnos): Remove a dead code. Initiate
1300 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
1302 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1305 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
1306 constraint from =wa to wa. Avoid a subreg on the output operand,
1307 instead use a pseudo and subreg it in a move.
1308 (p9_xxbrd_<mode>): Changed to ...
1309 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
1310 (p9_xxbrd_v2df): New expander.
1311 (p9_xxbrw_<mode>): Changed to ...
1312 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
1313 (p9_xxbrw_v4sf): New expander.
1315 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1317 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
1319 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
1322 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
1323 multiply in 32-bit mode.
1324 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
1325 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
1328 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1330 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
1331 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
1332 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
1333 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
1335 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
1338 * dwarf2out.c (is_trivial_indirect_ref): New function.
1339 (dwarf2out_late_global_decl): Do not generate a location
1340 attribute for variables that have a non-trivial DECL_VALUE_EXPR
1341 and that are not defined in the current unit.
1343 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1345 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
1346 instead of a libcall for UNORDERED.
1348 2018-02-09 Tamar Christina <tamar.christina@arm.com>
1351 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
1352 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
1354 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1357 * config/s390/s390.c (s390_set_current_function): Invoke
1358 s390_indirect_branch_settings also if fndecl didn't change.
1360 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1362 * config/rs6000/rs6000.md (blockage): Set length to zero.
1364 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1366 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
1368 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1371 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
1372 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
1376 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
1377 PARALLEL incoming that failed vt_get_decl_and_offset check.
1380 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
1381 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
1383 (get_variable_section): For decls in named .bss* sections pass true as
1384 second argument to bss_initializer_p.
1386 2018-02-09 Marek Polacek <polacek@redhat.com>
1387 Jakub Jelinek <jakub@redhat.com>
1390 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
1391 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
1392 Sync some changes from cxx_fold_indirect_ref.
1394 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1396 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
1398 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
1399 (BLOCK_INLINE_ENTRY_LABEL): New.
1400 (dwarf2out_var_location): Disregard inline entry markers.
1401 (inline_entry_data): New struct.
1402 (inline_entry_data_hasher): New hashtable type.
1403 (inline_entry_data_hasher::hash): New.
1404 (inline_entry_data_hasher::equal): New.
1405 (inline_entry_data_table): New variable.
1406 (add_high_low_attributes): Add DW_AT_entry_pc and
1407 DW_AT_GNU_entry_view attributes if a pending entry is found
1408 in inline_entry_data_table. Add old entry_pc attribute only
1409 if debug nonbinding markers are disabled.
1410 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
1411 markers are enabled.
1412 (block_within_block_p, dwarf2out_inline_entry): New.
1413 (dwarf2out_finish): Check that no entries remained in
1414 inline_entry_data_table.
1415 * final.c (reemit_insn_block_notes): Handle inline entry notes.
1416 (final_scan_insn, notice_source_line): Likewise.
1417 (rest_of_clean_state): Skip inline entry markers.
1418 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
1420 * gimple.c (gimple_build_debug_inline_entry): New.
1421 * gimple.h (enum gimple_debug_subcode): Add
1422 GIMPLE_DEBUG_INLINE_ENTRY.
1423 (gimple_build_debug_inline_entry): Declare.
1424 (gimple_debug_inline_entry_p): New.
1425 (gimple_debug_nonbind_marker_p): Adjust.
1426 * insn-notes.def (INLINE_ENTRY): New.
1427 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
1428 inline entry marker notes.
1429 (print_insn): Likewise.
1430 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
1431 (INSN_DEBUG_MARKER_KIND): Likewise.
1432 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
1433 * tree-inline.c (expand_call_inline): Build and insert
1434 debug_inline_entry stmt.
1435 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
1436 inline entry blocks early, if nonbind markers are enabled.
1437 (dump_scope_block): Dump fragment info.
1438 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
1439 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
1440 (gimple_build_debug_inline_entry): New.
1441 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
1442 Enable/disable inline entry points too.
1443 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
1444 (DEBUG_INSN): Describe inline entry markers.
1446 * common.opt (gvariable-location-views): New.
1447 (gvariable-location-views=incompat5): New.
1448 * config.in: Rebuilt.
1449 * configure: Rebuilt.
1450 * configure.ac: Test assembler for view support.
1451 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
1452 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
1453 * dwarf2out.c (var_loc_view): New typedef.
1454 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
1455 (dwarf2out_locviews_in_attribute): New.
1456 (dwarf2out_locviews_in_loclist): New.
1457 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
1458 (enum dw_line_info_opcode): Add LI_adv_address.
1459 (struct dw_line_info_table): Add view.
1460 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
1461 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
1462 (zero_view_p): New variable.
1463 (ZERO_VIEW_P): New macro.
1464 (output_asm_line_debug_info): New.
1465 (struct var_loc_node): Add view.
1466 (add_AT_view_list, AT_loc_list): New.
1467 (add_var_loc_to_decl): Add view param. Test it against last.
1468 (new_loc_list): Add view params. Record them.
1469 (AT_loc_list_ptr): Handle loc and view lists.
1470 (view_list_to_loc_list_val_node): New.
1471 (print_dw_val): Handle dw_val_class_view_list.
1472 (size_of_die): Likewise.
1473 (value_format): Likewise.
1474 (loc_list_has_views): New.
1475 (gen_llsym): Set vl_symbol too.
1476 (maybe_gen_llsym, skip_loc_list_entry): New.
1477 (dwarf2out_maybe_output_loclist_view_pair): New.
1478 (output_loc_list): Output view list or entries too.
1479 (output_view_list_offset): New.
1480 (output_die): Handle dw_val_class_view_list.
1481 (output_dwarf_version): New.
1482 (output_compilation_unit_header): Use it.
1483 (output_skeleton_debug_sections): Likewise.
1484 (output_rnglists, output_line_info): Likewise.
1485 (output_pubnames, output_aranges): Update version comments.
1486 (output_one_line_info_table): Output view numbers in asm comments.
1487 (dw_loc_list): Determine current endview, pass it to new_loc_list.
1488 Call maybe_gen_llsym.
1489 (loc_list_from_tree_1): Adjust.
1490 (add_AT_location_description): Create view list attribute if
1491 needed, check it's absent otherwise.
1492 (convert_cfa_to_fb_loc_list): Adjust.
1493 (maybe_emit_file): Call output_asm_line_debug_info for test.
1494 (dwarf2out_var_location): Reset views as needed. Precompute
1495 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
1496 attribute. Set view.
1497 (new_line_info_table): Reset next view.
1498 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
1499 (dwarf2out_source_line): Likewise. Output view resets and labels to
1500 the assembler, or select appropriate line info opcodes.
1501 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
1502 (optimize_string_length): Catch it. Adjust.
1503 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
1504 dw_val_class_view_list, and remove it if no longer needed.
1505 (hash_loc_list): Hash view numbers.
1506 (loc_list_hasher::equal): Compare them.
1507 (optimize_location_lists): Check whether a view list symbol is
1508 needed, and whether the locview attribute is present, and
1509 whether they match. Remove the locview attribute if no longer
1511 (index_location_lists): Call skip_loc_list_entry for test.
1512 (dwarf2out_finish): Call output_asm_line_debug_info for test.
1513 Use output_dwarf_version.
1514 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
1515 (struct dw_val_node): Add val_view_list.
1516 * final.c (SEEN_NEXT_VIEW): New.
1517 (set_next_view_needed): New.
1518 (clear_next_view_needed): New.
1519 (maybe_output_next_view): New.
1520 (final_start_function): Rename to...
1521 (final_start_function_1): ... this. Take pointer to FIRST,
1522 add SEEN parameter. Emit param bindings in the initial view.
1523 (final_start_function): Reintroduce SEEN-less interface.
1524 (final): Rename to...
1525 (final_1): ... this. Take SEEN parameter. Output final pending
1526 next view at the end.
1527 (final): Reintroduce seen-less interface.
1528 (final_scan_insn): Output pending next view before switching
1529 sections or ending a block. Mark the next view as needed when
1530 outputting variable locations. Notify debug backend of section
1531 changes, and of location view changes.
1532 (rest_of_handle_final): Adjust.
1533 * toplev.c (process_options): Autodetect value for debug variable
1534 location views option. Warn on incompat5 without -gdwarf-5.
1535 * doc/invoke.texi (gvariable-location-views): New.
1536 (gvariable-location-views=incompat5): New.
1537 (gno-variable-location-views): New.
1539 2018-02-08 David Malcolm <dmalcolm@redhat.com>
1541 PR tree-optimization/84136
1542 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
1543 that the result of find_edge is non-NULL.
1545 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1548 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
1549 storing integer register in SImode. Fix cost of 256 and 512
1550 byte aligned SSE register store.
1552 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1554 * config/i386/i386.c (ix86_multiplication_cost): Fix
1555 multiplication cost for TARGET_AVX512DQ.
1557 2018-02-08 Marek Polacek <polacek@redhat.com>
1559 PR tree-optimization/84238
1560 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
1563 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1565 PR tree-optimization/84265
1566 * tree-vect-stmts.c (vectorizable_store): Don't treat
1567 VMAT_CONTIGUOUS accesses as grouped.
1568 (vectorizable_load): Likewise.
1570 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1572 PR tree-optimization/81635
1573 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
1574 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
1575 (test_round_for_mask): New functions.
1576 (wide_int_cc_tests): Call test_round_for_mask.
1577 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
1578 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
1579 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
1580 range returned by get_range_info.
1582 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1585 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
1586 * symtab.c: Include builtins.h
1587 (symtab_node::output_to_lto_symbol_table_p): Move here
1588 from lto-streamer-out.c:output_symbol_p.
1589 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
1590 (output_symbol_p): Move all logic to symtab.c
1591 (produce_symtab): Update.
1593 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1595 * config/s390/s390-opts.h (enum indirect_branch): Define.
1596 * config/s390/s390-protos.h (s390_return_addr_from_memory)
1597 (s390_indirect_branch_via_thunk)
1598 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
1599 (enum s390_indirect_branch_type): Define.
1600 * config/s390/s390.c (struct s390_frame_layout, struct
1601 machine_function): Remove.
1602 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
1603 (indirect_branch_table_label_no, indirect_branch_table_name):
1605 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
1606 (enum s390_indirect_branch_option): Define.
1607 (s390_return_addr_from_memory): New function.
1608 (s390_handle_string_attribute): New function.
1609 (s390_attribute_table): Add new attribute handler.
1610 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
1611 (s390_indirect_branch_via_thunk): New function.
1612 (s390_indirect_branch_via_inline_thunk): New function.
1613 (s390_function_ok_for_sibcall): When jumping via thunk disallow
1614 sibling call optimization for non z10 compiles.
1615 (s390_emit_call): Force indirect branch target to be a single
1616 register. Add r1 clobber for non-z10 compiles.
1617 (s390_emit_epilogue): Emit return jump via return_use expander.
1618 (s390_reorg): Handle JUMP_INSNs as execute targets.
1619 (s390_option_override_internal): Perform validity checks for the
1620 new command line options.
1621 (s390_indirect_branch_attrvalue): New function.
1622 (s390_indirect_branch_settings): New function.
1623 (s390_set_current_function): Invoke s390_indirect_branch_settings.
1624 (s390_output_indirect_thunk_function): New function.
1625 (s390_code_end): Implement target hook.
1626 (s390_case_values_threshold): Implement target hook.
1627 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
1629 * config/s390/s390.h (struct s390_frame_layout)
1630 (struct machine_function): Move here from s390.c.
1631 (TARGET_INDIRECT_BRANCH_NOBP_RET)
1632 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
1633 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
1634 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
1635 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
1636 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
1637 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
1638 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
1639 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
1640 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
1641 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
1642 (mnemonic attribute): Add values which aren't recognized
1644 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
1645 pattern for branch conversion. Fix mnemonic attribute.
1646 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
1647 indirect branch via thunk if requested.
1648 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
1649 ("*indirect_jump"): Disable for branch conversion using out of
1651 ("indirect_jump_via_thunk<mode>_z10")
1652 ("indirect_jump_via_thunk<mode>")
1653 ("indirect_jump_via_inlinethunk<mode>_z10")
1654 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
1655 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
1656 ("casesi_jump_via_inlinethunk<mode>_z10")
1657 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
1658 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
1659 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
1660 ("*indirect2_jump"): Disable for branch conversion.
1661 ("casesi_jump"): Turn into expander and expand patterns for branch
1663 ("return_use"): New expander.
1664 ("*return"): Emit return via thunk and rename it to ...
1665 ("*return<mode>"): ... this one.
1666 * config/s390/s390.opt: Add new options and and enum for the
1669 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1671 * lra-constraints.c (match_reload): Unconditionally use
1672 gen_lowpart_SUBREG, rather than selecting between that
1673 and equivalent gen_rtx_SUBREG code.
1675 2018-02-08 Richard Biener <rguenther@suse.de>
1677 PR tree-optimization/84233
1678 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
1679 changed flag instead of boguously re-using phi_inserted.
1681 2018-02-08 Martin Jambor <mjambor@suse.cz>
1683 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
1684 static local variables.
1686 2018-02-08 Richard Biener <rguenther@suse.de>
1688 PR tree-optimization/84278
1689 * tree-vect-stmts.c (vectorizable_store): When looking for
1690 smaller vector types to perform grouped strided loads/stores
1691 make sure the mode is supported by the target.
1692 (vectorizable_load): Likewise.
1694 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1696 * config/aarch64/aarch64.c (aarch64_components_for_bb):
1697 Increase LDP/STP opportunities by adding adjacent callee-saves.
1699 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1701 PR rtl-optimization/84068
1702 PR rtl-optimization/83459
1703 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
1705 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
1707 PR tree-optimization/84224
1708 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
1709 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
1712 2018-02-07 Iain Sandoe <iain@codesourcery.com>
1715 * config/rs6000/altivec.md (*restore_world): Remove LR use.
1716 * config/rs6000/predicates.md (restore_world_operation): Adjust op
1717 count, remove one USE.
1719 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
1721 * doc/install.texi (Configuration): Document the
1722 --with-long-double-format={ibm,ieee} PowerPC configuration
1726 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
1727 Convert from define_expand to be define_insn_and_split. Rework
1728 float/double/_Float128 conversions to QI/HI/SImode to work with
1729 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
1730 conversions to QI/HImode types did a store and then a load to
1731 truncate the value. For conversions to VSX registers, don't split
1732 the insn, instead emit the code directly. Use the code iterator
1733 any_fix to combine signed and unsigned conversions.
1734 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
1735 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
1736 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1737 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
1738 (fix_<mode>di2_hw): Likewise.
1739 (fixuns_<mode>di2_hw): Likewise.
1740 (fix_<mode>si2_hw): Likewise.
1741 (fixuns_<mode>si2_hw): Likewise.
1742 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
1743 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1744 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1745 fix<uns>_trunc<SFDF:mode>si2_p8.
1746 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1748 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1749 (fix<uns>_<mode>_mem): Likewise.
1750 (fctiw<u>z_<mode>_mem): Likewise.
1751 (fix<uns>_<mode>_mem): Likewise.
1752 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1753 the register allocator from doing a direct move to the GPRs to do
1754 a store, and instead use the ISA 3.0 store byte/half-word from
1755 vector register instruction. For IEEE 128-bit floating point,
1756 also optimize stores of 32-bit ints.
1757 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1759 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1761 * genextract.c (push_pathstr_operand): New function to support
1763 (walk_rtx): Call push_pathstr_operand.
1764 (print_path): Support [a-zA-Z].
1766 2018-02-07 Richard Biener <rguenther@suse.de>
1768 PR tree-optimization/84037
1769 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1770 (cse_and_gimplify_to_preheader): Declare.
1771 (vect_get_place_in_interleaving_chain): Likewise.
1772 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1774 (_loop_vec_info::~_loop_vec_info): Delete it.
1775 (cse_and_gimplify_to_preheader): New function.
1776 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1777 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1778 (vectorizable_load): Likewise. For grouped stores always base
1779 the IV on the first element.
1780 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1781 condition before gimplifying.
1783 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1785 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1786 *DIV_EXPR and *MOD_EXPR.
1788 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1791 * config/i386/i386.c (ix86_option_override_internal): Mask out
1792 the CF_SET bit when checking -fcf-protection.
1794 2018-02-07 Tom de Vries <tom@codesourcery.com>
1797 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1800 2018-02-07 Richard Biener <rguenther@suse.de>
1802 PR tree-optimization/84204
1803 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1806 PR tree-optimization/84205
1807 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1808 special-case isl_ast_op_zdiv_r.
1810 PR tree-optimization/84223
1811 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1812 Only add conditions from within the region.
1813 (gather_bbs::after_dom_children): Adjust.
1815 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1818 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1819 * config/avr/avr.md: Only post-reload split REG-REG moves if
1820 either register is REGERAL_REG_P.
1822 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1824 PR tree-optimization/84235
1825 * tree-ssa-scopedtables.c
1826 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1827 if the subtraction is performed in floating point type where NaNs are
1828 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1829 build 1. Formatting fix.
1831 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1834 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1835 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1836 and skip it regardless of bb boundaries. Use CALL_P macro,
1837 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1838 unnecessarily, formatting fix.
1840 2018-02-06 Michael Collison <michael.collison@arm.com>
1842 * config/arm/thumb2.md:
1843 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1844 (*thumb_mov_notscc): Ditto.
1846 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1849 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1850 unsigned_fix, not "s".
1852 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1854 * configure.ac (gcc_fn_eh_frame_ro): New function.
1855 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1856 correct .eh_frame permissions.
1857 * configure: Regenerate.
1859 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1861 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1864 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1866 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1867 Display warning message for -mno-speculate-indirect-jumps.
1869 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1871 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1873 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1875 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1877 PR tree-optimization/84225
1878 * tree-eh.c (find_trapping_overflow): Only call
1879 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1881 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1884 * config/i386/i386.c: Reimplement the check of possible options
1885 -mibt/-mshstk conbination. Change error messages.
1886 * doc/invoke.texi: Fix a typo: remove extra '='.
1888 2018-02-06 Marek Polacek <polacek@redhat.com>
1890 PR tree-optimization/84228
1891 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1893 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1896 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1897 emitted arch directives.
1898 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1899 __ARM_FEATURE_COPROC before changing architectures.
1901 2018-02-06 Richard Biener <rguenther@suse.de>
1903 * config/i386/i386.c (print_reg): Fix typo.
1904 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1906 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1908 * configure: Regenerate.
1910 2018-02-05 Martin Sebor <msebor@redhat.com>
1912 PR tree-optimization/83369
1913 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1916 2018-02-05 Martin Liska <mliska@suse.cz>
1918 * doc/invoke.texi: Cherry-pick upstream r323995.
1920 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1922 * ira.c (ira_init_register_move_cost): Adjust comment.
1924 2018-02-05 Martin Liska <mliska@suse.cz>
1926 PR gcov-profile/84137
1927 * doc/gcov.texi: Fix typo in documentation.
1929 2018-02-05 Martin Liska <mliska@suse.cz>
1931 PR gcov-profile/83879
1932 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1933 using dlopen functionality.
1935 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1937 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1938 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1939 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1940 _mm_maskz_range_round_ss): New intrinsics.
1941 (__builtin_ia32_rangesd128_round)
1942 (__builtin_ia32_rangess128_round): Remove.
1943 (__builtin_ia32_rangesd128_mask_round,
1944 __builtin_ia32_rangess128_mask_round): New builtins.
1945 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1946 __builtin_ia32_rangess128_round): Remove.
1947 (__builtin_ia32_rangesd128_mask_round,
1948 __builtin_ia32_rangess128_mask_round): New builtins.
1949 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1950 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1951 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1952 "<round_saeonly_constraint>")): Changed to ...
1953 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1954 "<round_saeonly_scalar_constraint>")): ... this.
1955 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1956 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1957 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1958 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1959 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1961 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1963 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1965 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1966 Remove all values except native, 8540 and 8548.
1968 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1970 * config/i386/i386.c (ix86_output_function_return): Pass
1971 INVALID_REGNUM, instead of -1, as invalid register number to
1972 indirect_thunk_name and output_indirect_thunk.
1974 2018-02-02 Julia Koval <julia.koval@intel.com>
1976 * config.gcc: Add -march=icelake.
1977 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1978 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1979 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1980 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1981 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1982 (processor_target_table): Add icelake.
1983 (ix86_option_override_internal): Handle new PTAs.
1984 (get_builtin_code_for_version): Handle icelake.
1985 (M_INTEL_COREI7_ICELAKE): New.
1986 (fold_builtin_cpu): Handle icelake.
1987 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1988 * doc/invoke.texi: Add -march=icelake.
1990 2018-02-02 Julia Koval <julia.koval@intel.com>
1992 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1993 to wide_int_bitmask.
1994 * wide-int-bitmask.h: New.
1996 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1999 * config/i386/i386.md: Replace Pmode with word_mode in
2000 builtin_setjmp_setup and builtin_longjmp to support x32.
2002 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
2006 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
2008 (rs6000_supported_cpu_names): New static variable.
2009 (linux_cpu_translation_table): Likewise.
2010 (elf_platform) <cpu>: Define new static variable and use it.
2011 Translate kernel AT_PLATFORM name to canonical name if needed.
2012 Error if platform name is unknown.
2014 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
2017 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
2019 2018-02-01 Jeff Law <law@redhat.com>
2022 * config/i386/i386.c (release_scratch_register_on_entry): Add new
2023 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
2024 the scratch if RELEASE_VIA_POP is false.
2025 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
2026 If we have to save a temporary register, decrement SIZE appropriately.
2027 Pass new arguments to release_scratch_register_on_entry.
2028 (ix86_adjust_stack_and_probe): Likewise.
2029 (ix86_emit_probe_stack_range): Pass new arguments to
2030 release_scratch_register_on_entry.
2032 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
2034 PR rtl-optimization/84157
2035 * combine.c (change_zero_ext): Use REG_P predicate in
2036 front of HARD_REGISTER_P predicate.
2038 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
2040 * config/avr/avr.c (avr_option_override): Move disabling of
2041 -fdelete-null-pointer-checks to...
2042 * common/config/avr/avr-common.c (avr_option_optimization_table):
2045 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2047 PR tree-optimization/81635
2048 * tree-data-ref.c (split_constant_offset_1): For types that
2049 wrap on overflow, try to use range info to prove that wrapping
2052 2018-02-01 Renlin Li <renlin.li@arm.com>
2055 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
2057 (aarch64_register_move_cost): Likewise.
2058 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
2060 (REG_CLASS_NAMES): Likewise.
2061 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
2062 TAILCALL_ADDR_REGS. Remove IP registers.
2063 * config/aarch64/aarch64.md (Ucs): Update register constraint.
2065 2018-02-01 Richard Biener <rguenther@suse.de>
2067 * domwalk.h (dom_walker::dom_walker): Add additional constructor
2068 for specifying RPO order and allow NULL for that.
2069 * domwalk.c (dom_walker::dom_walker): Likewise.
2070 (dom_walker::walk): Handle NULL RPO order.
2071 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
2073 (rewrite_update_dom_walker): Likewise.
2074 (mark_def_dom_walker): Likewise.
2076 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2078 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
2079 (aarch64_maybe_expand_sve_subreg_move): Declare.
2080 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
2081 * config/aarch64/predicates.md (aarch64_any_register_operand): New
2083 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
2084 that are semantically a reverse operation.
2085 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
2086 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
2087 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
2089 (aarch64_can_change_mode_class): For big-endian, forbid changes
2090 between two SVE modes if they have different element sizes.
2092 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2094 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
2095 the TImode handling for big-endian targets.
2097 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2099 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
2100 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
2102 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
2103 Remove BSWAP handing for big-endian targets and use the form of
2104 LD1RQ appropariate for the mode.
2106 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2108 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
2109 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
2112 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2115 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
2116 check for operands that need to go through aarch64_sve_reload_be.
2118 2018-02-01 Jakub Jelinek <jakub@redhat.com>
2120 PR tree-optimization/81661
2121 PR tree-optimization/84117
2122 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
2123 * tree-eh.c: Include gimplify.h.
2124 (find_trapping_overflow, replace_trapping_overflow,
2125 rewrite_to_non_trapping_overflow): New functions.
2126 * tree-vect-loop.c: Include tree-eh.h.
2127 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
2128 * tree-data-ref.c: Include tree-eh.h.
2129 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
2131 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
2133 PR rtl-optimization/84123
2134 * combine.c (change_zero_ext): Check if hard register satisfies
2135 can_change_dest_mode before calling gen_lowpart_SUBREG.
2137 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
2140 * ira.c (ira_init_register_move_cost): Remove assert.
2142 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2144 PR rtl-optimization/84071
2145 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
2146 * doc/tm.texi: Regenerate.
2148 2018-01-31 Richard Biener <rguenther@suse.de>
2150 PR tree-optimization/84132
2151 * tree-data-ref.c (analyze_miv_subscript): Properly
2152 check whether evolution_function_is_affine_multivariate_p
2153 before calling gcd_of_steps_may_divide_p.
2155 2018-01-31 Julia Koval <julia.koval@intel.com>
2158 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
2159 * config/i386/i386.md (rdpid_rex64) New.
2160 (rdpid): Make 32bit only.
2162 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
2165 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
2166 an IDENTIFIER_NODE for FUNCTION_TYPE's.
2168 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2171 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2173 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2175 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2177 PR rtl-optimization/84071
2178 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
2179 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
2181 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
2183 * config/arc/arc.c (arc_handle_aux_attribute): New function.
2184 (arc_attribute_table): Add 'aux' attribute.
2185 (arc_in_small_data_p): Consider aux like variables.
2186 (arc_is_aux_reg_p): New function.
2187 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
2188 (arc_get_aux_arg): New function.
2189 (prepare_move_operands): Handle aux-register access.
2190 (arc_handle_aux_attribute): New function.
2191 * doc/extend.texi (ARC Variable attributes): Add subsection.
2193 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
2195 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
2196 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
2197 (arc_attribute_table): Add 'uncached' attribute.
2198 (arc_print_operand): Print '.di' flag for uncached memory
2200 (arc_in_small_data_p): Do not consider for small data the uncached
2202 (arc_is_uncached_mem_p): New function.
2203 * config/arc/predicates.md (compact_store_memory_operand): Check
2204 for uncached memory accesses.
2205 (nonvol_nonimm_operand): Likewise.
2206 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
2208 2018-01-31 Jakub Jelinek <jakub@redhat.com>
2211 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
2212 falign-loops=): Add Optimization flag.
2214 2018-01-30 Jeff Law <law@redhat.com>
2217 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
2218 INT_REGISTERS_SAVED. Check it prior to calling
2219 get_scratch_register_on_entry.
2220 (ix86_adjust_stack_and_probe): Similarly.
2221 (ix86_emit_probe_stack_range): Similarly.
2222 (ix86_expand_prologue): Corresponding changes.
2224 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2227 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
2228 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
2230 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
2233 * lra-constraints.c (curr_insn_transform): Process AND in the
2236 2018-01-30 Jakub Jelinek <jakub@redhat.com>
2238 PR rtl-optimization/83986
2239 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
2240 dependence against last_pending_memory_flush in addition to
2243 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
2245 PR tree-optimization/81611
2246 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
2249 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2252 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
2255 2018-01-30 Richard Biener <rguenther@suse.de>
2256 Jakub Jelinek <jakub@redhat.com>
2258 PR tree-optimization/84111
2259 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
2260 inner loops added during recursion, as they don't have up-to-date
2263 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2266 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
2267 (can_inline_edge_by_limits_p): ... here.
2268 (can_early_inline_edge_p, check_callers,
2269 update_caller_keys, update_callee_keys, recursive_inlining,
2270 add_new_edges_to_heap, speculation_useful_p,
2271 inline_small_functions,
2272 inline_small_functions, flatten_function,
2273 inline_to_all_callers_1): Update.
2275 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2277 * profile-count.c (profile_count::combine_with_ipa_count): Handle
2280 2018-01-30 Richard Biener <rguenther@suse.de>
2282 PR tree-optimization/83008
2283 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
2284 invariant and constant vector uses in stmts when they need
2287 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2290 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
2291 * configure: Regenerate.
2293 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2295 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
2297 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
2298 Use gen_rtx_REG rather than gen_lowpart.
2300 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2302 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
2303 rather than 0 when creating partial subregs.
2305 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2307 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
2310 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
2313 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
2314 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
2315 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
2316 flags. This restores the settings used before the 2017-07-24.
2317 Turning off pre increment/decrement/modify allows IVOPTS to
2318 optimize DF/SF loops where the index is an int.
2320 2018-01-29 Richard Biener <rguenther@suse.de>
2321 Kelvin Nilsen <kelvin@gcc.gnu.org>
2324 * tree-vect-stmts.c (vectorizable_call): Don't call
2325 targetm.vectorize_builtin_md_vectorized_function if callee is
2328 2018-01-22 Carl Love <cel@us.ibm.com>
2330 * doc/extend.tex: Fix typo in second arg in
2331 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
2333 2018-01-29 Richard Biener <rguenther@suse.de>
2335 PR tree-optimization/84086
2336 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
2337 (flush_ssaname_freelist): When SSA names were released reset
2338 the SCEV hash table.
2340 2018-01-29 Richard Biener <rguenther@suse.de>
2342 PR tree-optimization/84057
2343 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
2344 removed paths when removing edges.
2346 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
2348 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
2349 -mfunction-return=@var{choice}.
2351 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
2354 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
2356 (layout::print_source_line): Likewise.
2357 (test_get_line_width_without_trailing_whitespace): Add test cases.
2359 2018-01-27 Jakub Jelinek <jakub@redhat.com>
2362 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
2365 2018-01-26 Jim Wilson <jimw@sifive.com>
2367 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
2369 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
2372 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2374 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
2375 and CMP + SUB-immediate -> SUBS.
2377 2018-01-26 Martin Sebor <msebor@redhat.com>
2379 PR tree-optimization/83896
2380 * tree-ssa-strlen.c (get_string_len): Rename...
2381 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
2382 Avoid assuming length is constant.
2383 (handle_char_store): Use HOST_WIDE_INT for string length.
2385 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
2388 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
2389 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
2391 2018-01-26 Richard Biener <rguenther@suse.de>
2393 PR rtl-optimization/84003
2394 * dse.c (record_store): Only record redundant stores when
2395 the earlier store aliases at least all accesses the later one does.
2397 2018-01-26 Jakub Jelinek <jakub@redhat.com>
2399 PR rtl-optimization/83985
2400 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
2401 REG_CFA_RESTORE insns.
2402 (delete_unmarked_insns): Don't ignore separate shrink wrapping
2403 REG_CFA_RESTORE insns here.
2406 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
2407 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
2409 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2411 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
2412 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
2413 (arc_init): Likewise.
2414 (arc_override_options): Likewise.
2415 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
2417 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
2419 * config/arc/arc.h (TARGET_DBNZ): Define.
2420 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
2421 properly set the tune attribute.
2422 (dbnz): Use TARGET_DBNZ guard.
2423 * config/arc/arc.opt (mtune): Add core3 option.
2425 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2427 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
2428 recognize new pic like addresses.
2429 (arc_delegitimize_address): Clean up.
2431 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2433 * config/arc/arc-arches.def: Option mrf16 valid for all
2435 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
2436 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
2437 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
2438 * config/arc/arc-tables.opt: Regenerate.
2439 * config/arc/arc.c (arc_conditional_register_usage): Handle
2440 reduced register file case.
2441 (arc_file_start): Set must have build attributes.
2442 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
2444 * config/arc/arc.opt (mrf16): Add new option.
2445 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
2446 * config/arc/genmultilib.awk: Handle new mrf16 option.
2447 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
2448 * config/arc/t-multilib: Regenerate.
2449 * doc/invoke.texi (ARC Options): Document mrf16 option.
2451 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2453 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
2454 * config/arc/arc.c (arc_handle_secure_attribute): New function.
2455 (arc_attribute_table): Add 'secure_call' attribute.
2456 (arc_print_operand): Print secure call operand.
2457 (arc_function_ok_for_sibcall): Don't optimize tail calls when
2459 (arc_is_secure_call_p): New function. * config/arc/arc.md
2460 (call_i): Add support for sjli instruction.
2461 (call_value_i): Likewise.
2462 * config/arc/constraints.md (Csc): New constraint.
2464 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2465 John Eric Martin <John.Martin@emmicro-us.com>
2467 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
2468 * config/arc/arc.c (_arc_jli_section): New struct.
2469 (arc_jli_section): New type.
2470 (rc_jli_sections): New static variable.
2471 (arc_handle_jli_attribute): New function.
2472 (arc_attribute_table): Add jli_always and jli_fixed attribute.
2473 (arc_file_end): New function.
2474 (TARGET_ASM_FILE_END): Define.
2475 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
2476 (arc_add_jli_section): New function.
2477 (jli_call_scan): Likewise.
2478 (arc_reorg): Call jli_call_scan.
2479 (arc_output_addsi): Remove 'S' from printing asm operand.
2480 (arc_is_jli_call_p): New function.
2481 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
2483 (movhi_insn): Likewise.
2484 (movsi_insn): Likewise.
2485 (movsi_set_cc_insn): Likewise.
2486 (loadqi_update): Likewise.
2487 (load_zeroextendqisi_update): Likewise.
2488 (load_signextendqisi_update): Likewise.
2489 (loadhi_update): Likewise.
2490 (load_zeroextendhisi_update): Likewise.
2491 (load_signextendhisi_update): Likewise.
2492 (loadsi_update): Likewise.
2493 (loadsf_update): Likewise.
2494 (movsicc_insn): Likewise.
2495 (bset_insn): Likewise.
2496 (bxor_insn): Likewise.
2497 (bclr_insn): Likewise.
2498 (bmsk_insn): Likewise.
2499 (bicsi3_insn): Likewise.
2500 (cmpsi_cc_c_insn): Likewise.
2501 (movsi_ne): Likewise.
2502 (movsi_cond_exec): Likewise.
2503 (clrsbsi2): Likewise.
2514 (call_i): Remove 'S' asm letter, add jli instruction.
2515 (call_value_i): Likewise.
2516 * config/arc/arc.op (mjli-always): New option.
2517 * config/arc/constraints.md (Cji): New constraint.
2518 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
2520 (subsf3_fpx): Likewise.
2521 (mulsf3_fpx): Likewise.
2522 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
2524 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
2526 * doc/invoke.texi (ARC): Document mjli-always option.
2528 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
2530 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
2531 avoid addition with 0 and use incw and decw where possible.
2533 2018-01-26 Richard Biener <rguenther@suse.de>
2535 PR tree-optimization/81082
2536 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
2537 association if it requires casting to unsigned.
2538 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
2539 from fold_plusminus_mult_expr to catch important cases late when
2540 range info is available.
2542 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2544 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
2545 * configure.ac (hidden_linkonce): New test.
2546 * configure: Regenerate.
2547 * config.in: Regenerate.
2549 2018-01-26 Julia Koval <julia.koval@intel.com>
2551 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
2552 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
2553 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
2554 _mm_mask_bitshuffle_epi64_mask): Fix type.
2555 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
2556 USI_FTYPE_V4DI_V4DI_USI): Remove.
2557 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
2558 __builtin_ia32_vpshufbitqmb256_mask,
2559 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
2560 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
2561 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
2563 2018-01-26 Alan Modra <amodra@gmail.com>
2566 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
2567 UNSPEC_VBPERMQ. Sort other unspecs.
2569 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
2571 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
2573 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
2576 * predict.c (drop_profile): Do not push/pop cfun; update also
2578 (handle_missing_profiles): Fix logic looking for zero profiles.
2580 2018-01-25 Jakub Jelinek <jakub@redhat.com>
2583 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
2584 on functions with #pragma omp declare simd or functions with simd
2586 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
2587 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
2588 Remove trailing \n from warning_at calls.
2590 2018-01-25 Tom de Vries <tom@codesourcery.com>
2593 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2594 for neutered workers.
2596 2018-01-24 Joseph Myers <joseph@codesourcery.com>
2599 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
2600 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
2602 2018-01-24 Jeff Law <law@redhat.com>
2605 * i386.c (get_probe_interval): Move to earlier point.
2606 (ix86_compute_frame_layout): If -fstack-clash-protection and
2607 the frame is larger than the probe interval, then use pushes
2608 to save registers rather than reg->mem moves.
2609 (ix86_expand_prologue): Remove conditional for int_registers_saved
2612 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
2615 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
2616 min/max for never referenced object.
2618 2018-01-24 Jakub Jelinek <jakub@redhat.com>
2621 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
2623 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
2624 attributes from DECL_ATTRIBUTES (decl) without affecting
2625 DECL_ATTRIBUTES (current_function_decl).
2626 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
2627 functions with non-NULL DECL_ABSTRACT_ORIGIN.
2629 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
2631 PR tree-optimization/83979
2632 * fold-const.c (fold_comparison): Use constant_boolean_node
2633 instead of boolean_{true,false}_node.
2635 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
2637 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
2640 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2642 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2643 Simplify the clause that sets the length attribute.
2644 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2645 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
2646 clause that sets the length attribute.
2647 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2649 2018-01-24 Tom de Vries <tom@codesourcery.com>
2652 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
2653 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
2654 Add strict parameter.
2655 (prevent_branch_around_nothing): Insert dummy insn between branch to
2656 label and label with no ptx insn inbetween.
2657 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
2659 2018-01-24 Tom de Vries <tom@codesourcery.com>
2662 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2663 for neutered threads in warp.
2664 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
2666 2018-01-24 Richard Biener <rguenther@suse.de>
2668 PR tree-optimization/83176
2669 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
2672 2018-01-24 Richard Biener <rguenther@suse.de>
2674 PR tree-optimization/82819
2675 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
2676 code generating pluses that are no-ops in the target precision.
2678 2018-01-24 Richard Biener <rguenther@suse.de>
2681 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
2683 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2685 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
2686 to merge probabilities.
2687 * predict.c (probably_never_executed): Also mark as cold functions
2688 with global 0 profile and guessed local profile.
2689 * profile-count.c (profile_probability::combine_with_count): New
2691 * profile-count.h (profile_probability::operator*,
2692 profile_probability::operator*=, profile_probability::operator/,
2693 profile_probability::operator/=): Reduce precision to adjusted
2694 and set value to guessed on contradictory divisions.
2695 (profile_probability::combine_with_freq): Remove.
2696 (profile_probability::combine_wiht_count): Declare.
2697 (profile_count::force_nonzero):: Set to adjusted.
2698 (profile_count::probability_in):: Set quality to adjusted.
2699 * tree-ssa-tail-merge.c (replace_block_by): Use
2702 2018-01-23 Andrew Waterman <andrew@sifive.com>
2703 Jim Wilson <jimw@sifive.com>
2705 * config/riscv/riscv.c (riscv_stack_boundary): New.
2706 (riscv_option_override): Set riscv_stack_boundary. Handle
2707 riscv_preferred_stack_boundary_arg.
2708 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
2709 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
2710 (STACK_BOUNDARY): Set to riscv_stack_boundary.
2711 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
2712 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
2713 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
2715 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
2718 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
2719 of struct ix86_frame.
2720 (ix86_expand_epilogue): Likewise. Add a local variable for
2721 the reg_save_offset field in struct ix86_frame.
2723 2018-01-23 Bin Cheng <bin.cheng@arm.com>
2725 PR tree-optimization/82604
2726 * tree-loop-distribution.c (enum partition_kind): New enum item
2727 PKIND_PARTIAL_MEMSET.
2728 (partition_builtin_p): Support above new enum item.
2729 (generate_code_for_partition): Ditto.
2730 (compute_access_range): Differentiate cases that equality can be
2731 proven at all loops, the innermost loops or no loops.
2732 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
2733 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
2734 (finalize_partitions, distribute_loop): Don't fuse partition of
2735 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
2736 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
2739 2018-01-23 Martin Liska <mliska@suse.cz>
2741 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
2742 order to ignore the predictor.
2743 (PRED_POLYMORPHIC_CALL): Likewise.
2744 (PRED_RECURSIVE_CALL): Likewise.
2746 2018-01-23 Martin Liska <mliska@suse.cz>
2748 * tree-profile.c (tree_profiling): Print function header to
2749 aware reader which function we are working on.
2750 * value-prof.c (gimple_find_values_to_profile): Do not print
2751 not interesting value histograms.
2753 2018-01-23 Martin Liska <mliska@suse.cz>
2755 * profile-count.h (enum profile_quality): Add
2756 profile_uninitialized as the first value. Do not number values
2757 as they are zero based.
2758 (profile_count::verify): Update sanity check.
2759 (profile_probability::verify): Likewise.
2761 2018-01-23 Nathan Sidwell <nathan@acm.org>
2763 * doc/invoke.texi (ffor-scope): Deprecate.
2765 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2767 PR tree-optimization/83510
2768 * domwalk.c (set_all_edges_as_executable): New function.
2769 (dom_walker::dom_walker): Convert bool param
2770 "skip_unreachable_blocks" to enum reachability. Move setup of
2771 edge flags to set_all_edges_as_executable and only do it when
2772 reachability is REACHABLE_BLOCKS.
2773 * domwalk.h (enum dom_walker::reachability): New enum.
2774 (dom_walker::dom_walker): Convert bool param
2775 "skip_unreachable_blocks" to enum reachability.
2776 (set_all_edges_as_executable): New decl.
2777 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2778 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2780 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2781 but converting true to REACHABLE_BLOCKS.
2782 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2784 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2785 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2786 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2788 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2789 if check_all_array_refs will be called.
2791 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2793 * tree.c (selftest::test_location_wrappers): Add more test
2796 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2798 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2799 (selftest::test_bit_in_range): Likewise.
2801 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2804 * doc/sourcebuild.texi (vect_float): Say that the selector
2805 only describes the situation when -funsafe-math-optimizations is on.
2806 (vect_float_strict): Document.
2808 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2810 PR tree-optimization/83965
2811 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2812 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2813 instead of checking only for a reduction.
2814 (vect_recog_widen_sum_pattern): Likewise.
2816 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2818 * predict.c (probably_never_executed): Only use precise profile info.
2819 (compute_function_frequency): Skip after inlining hack since we now
2820 have quality checking.
2822 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2824 * profile-count.h (profile_probability::very_unlikely,
2825 profile_probability::unlikely, profile_probability::even): Set
2826 precision to guessed.
2828 2018-01-23 Richard Biener <rguenther@suse.de>
2830 PR tree-optimization/83963
2831 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2832 Properly terminate dominator walk when crossing the exit edge not
2833 when visiting its source block.
2835 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2838 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2839 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2841 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2843 PR tree-optimization/83957
2844 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2845 semicolon after for body surrounded by braces.
2847 PR tree-optimization/83081
2848 * profile-count.h (profile_probability::split): New method.
2849 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2850 Use profile_probability::split.
2851 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2852 when splitting a single conditional jump into 2.
2854 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2856 PR tree-optimization/69452
2857 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2860 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2862 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2863 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2864 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2866 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2868 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2869 * config/rl78/rl78.md (movdi): New define_expand.
2870 * config/rl78/rl78.c (rl78_split_movdi): New function.
2872 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2875 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2877 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2878 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2879 128-bit to produce an UNSPEC move to get the double word with the
2880 signbit and then a shift directly to do signbit.
2881 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2882 implementation with a new version that just does either a direct
2883 move or a regular move. Move memory interface to separate insns.
2884 Move insns so they are next to the expander.
2885 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2886 with signbit move. Split big and little endian case.
2887 (signbit<mode>2_dm_mem_le): Likewise.
2888 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2889 (signbit<mode>2_dm2): Likewise.
2891 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2893 * config/rl78/rl78.md (anddi3): New define_expand.
2895 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2897 * config/rl78/rl78.md (umindi3): New define_expand.
2899 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2901 * config/rl78/rl78.md (smindi3): New define_expand.
2903 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2905 * config/rl78/rl78.md (smaxdi3): New define_expand.
2907 2018-01-22 Carl Love <cel@us.ibm.com>
2909 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2910 LVX_V1TI): Add macro expansion.
2911 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2912 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2913 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2914 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2915 Change check to determine if the instruction is a byte reversing
2916 entry. Fix typo in comment.
2917 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2918 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2919 Add def_builtin calls for new builtins.
2920 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2921 Add define_insn expansion.
2923 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2925 * config/rl78/rl78.md (umaxdi3): New define_expand.
2927 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2929 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2930 for non-QImode registers.
2932 2018-01-22 Richard Biener <rguenther@suse.de>
2934 PR tree-optimization/83963
2935 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2936 including the loop exit block.
2937 (scop_detection::merge_sese): Likewise.
2938 (scop_detection::add_scop): Do it here instead.
2940 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2942 * doc/sourcebuild.texi (arm_softfloat): Document.
2944 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2947 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2948 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2949 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2951 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2952 David Edelsohn <dje.gcc@gmail.com>
2955 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2956 Change "crset eq" to "crset 2".
2957 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2958 (*call_indirect_aix<mode>_nospec): Likewise.
2959 (*call_value_indirect_aix<mode>_nospec): Likewise.
2960 (*call_indirect_elfv2<mode>_nospec): Likewise.
2961 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2962 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2963 change assembly output from . to $.
2964 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2965 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2966 (*tablejump<mode>_internal1_nospec): Likewise.
2968 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2971 * config/sh/sh_optimize_sett_clrt.cc:
2972 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2974 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2976 PR tree-optimization/83940
2977 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2978 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2979 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2980 use it to pass back the definition type.
2981 (vect_check_store_rhs): Likewise rhs_dt_out.
2982 (vect_build_gather_load_calls): Add a mask_dt argument and use
2983 it instead of a call to vect_is_simple_use.
2984 (vectorizable_store): Update calls to vect_check_load_store_mask
2985 and vect_check_store_rhs. Use the dt returned by the latter instead
2986 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2987 instead of calls to vect_is_simple_use. Pass the scalar rather
2988 than the vector operand to vect_is_simple_use when handling
2989 second and subsequent copies of an rhs value.
2990 (vectorizable_load): Update calls to vect_check_load_store_mask
2991 and vect_build_gather_load_calls. Use the cached mask_dt and
2992 gs_info.offset_dt instead of calls to vect_is_simple_use.
2994 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2997 * tree-emutls.c: Include gimplify.h.
2998 (lower_emutls_2): New function.
2999 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
3000 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
3001 it before further processing.
3004 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
3005 UINTVAL (trueop1) instead of INTVAL (op1).
3007 2018-01-19 Jakub Jelinek <jakub@redhat.com>
3011 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
3012 INCOMING_FRAME_SP_OFFSET if not defined.
3013 (scan_trace): Add ENTRY argument. If true and
3014 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
3015 emit a note to adjust the CFA offset.
3016 (create_cfi_notes): Adjust scan_trace callers.
3017 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
3018 INCOMING_FRAME_SP_OFFSET in the CIE.
3019 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
3020 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
3022 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
3023 * doc/tm.texi: Regenerated.
3025 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
3027 PR rtl-optimization/83147
3028 * lra-constraints.c (remove_inheritance_pseudos): Use
3029 lra_substitute_pseudo_within_insn.
3031 2018-01-19 Tom de Vries <tom@codesourcery.com>
3032 Cesar Philippidis <cesar@codesourcery.com>
3035 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
3037 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
3040 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
3041 spaces for function labels.
3043 2018-01-19 Martin Liska <mliska@suse.cz>
3045 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
3046 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
3047 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
3048 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
3049 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
3050 (PRED_CONST_RETURN): Change from 69 to 65.
3051 (PRED_NULL_RETURN): Change from 91 to 71.
3052 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
3053 (PRED_LOOP_GUARD): Change from 66 to 73.
3055 2018-01-19 Martin Liska <mliska@suse.cz>
3057 * predict.c (predict_insn_def): Add new assert.
3058 (struct branch_predictor): Change type to signed integer.
3059 (test_prediction_value_range): Amend test to cover
3061 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
3062 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
3063 (PRED_LOOP_ITERATIONS_MAX): Likewise.
3064 (PRED_LOOP_IV_COMPARE): Likewise.
3065 * predict.h (PROB_UNINITIALIZED): Define new constant.
3067 2018-01-19 Martin Liska <mliska@suse.cz>
3069 * predict.c (dump_prediction): Add new format for
3070 analyze_brprob.py script which is enabled with -details
3072 * profile-count.h (precise_p): New function.
3074 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
3076 PR tree-optimization/83922
3077 * tree-vect-loop.c (vect_verify_full_masking): Return false if
3078 there are no statements that need masking.
3079 (vect_active_double_reduction_p): New function.
3080 (vect_analyze_loop_operations): Use it when handling phis that
3081 are not in the loop header.
3083 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
3085 PR tree-optimization/83914
3086 * tree-vect-loop.c (vectorizable_induction): Don't convert
3087 init_expr or apply the peeling adjustment for inductions
3088 that are nested within the vectorized loop.
3090 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3092 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
3095 2018-01-18 Jakub Jelinek <jakub@redhat.com>
3099 * function.h (gimplify_parameters): Add gimple_seq * argument.
3100 * function.c: Include gimple.h and options.h.
3101 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
3102 for the added local temporaries if needed.
3103 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
3104 if there are any parameter cleanups, wrap whole body into a
3105 try/finally with the cleanups.
3107 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
3110 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
3111 Use GET_MODE_CLASS for scalar floating point.
3113 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
3117 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
3118 Fix call of call_cgraph_insertion_hooks.
3120 2018-01-18 Martin Sebor <msebor@redhat.com>
3122 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
3124 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
3127 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
3130 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
3133 * common.opt: (-ffile-prefix-map): New option.
3134 * opts.c (common_handle_option): Defer it.
3135 * opts-global.c (handle_common_deferred_options): Handle it.
3136 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
3137 * file-prefix-map.h: New file.
3138 (remap_debug_filename, add_debug_prefix_map): ...here.
3139 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
3140 * final.c (debug_prefix_map, add_debug_prefix_map
3141 remap_debug_filename): Move to...
3142 * file-prefix-map.c: New file.
3143 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
3144 generalize, get rid of alloca(), use strrchr() instead of strchr().
3145 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
3146 Implement in terms of add_prefix_map().
3147 (remap_macro_filename, remap_debug_filename): Implement in term of
3149 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
3150 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
3151 * dbxout.c: Include file-prefix-map.h.
3152 * varasm.c: Likewise.
3153 * vmsdbgout.c: Likewise.
3154 * xcoffout.c: Likewise.
3155 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
3156 * doc/cppopts.texi (-fmacro-prefix-map): Document.
3157 * doc/invoke.texi (-ffile-prefix-map): Document.
3158 (-fdebug-prefix-map): Update description.
3160 2018-01-18 Martin Liska <mliska@suse.cz>
3162 * config/i386/i386.c (indirect_thunk_name): Document that also
3164 (output_indirect_thunk): Document why both instructions
3165 (pause and lfence) are generated.
3167 2018-01-18 Richard Biener <rguenther@suse.de>
3169 PR tree-optimization/83887
3170 * graphite-scop-detection.c
3171 (scop_detection::get_nearest_dom_with_single_entry): Remove.
3172 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
3173 (scop_detection::merge_sese): Re-implement with a flood-fill
3174 algorithm that properly finds a SESE region if it exists.
3176 2018-01-18 Jakub Jelinek <jakub@redhat.com>
3179 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
3180 pointer_diff optimizations use view_convert instead of convert.
3182 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3184 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
3185 Generate different code for -mno-speculate-indirect-jumps.
3186 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
3187 (*call_indirect_aix<mode>): Disable for
3188 -mno-speculate-indirect-jumps.
3189 (*call_indirect_aix<mode>_nospec): New define_insn.
3190 (*call_value_indirect_aix<mode>): Disable for
3191 -mno-speculate-indirect-jumps.
3192 (*call_value_indirect_aix<mode>_nospec): New define_insn.
3193 (*sibcall_nonlocal_sysv<mode>): Generate different code for
3194 -mno-speculate-indirect-jumps.
3195 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
3197 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
3199 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
3200 long double type, set the flags for noting the default long double
3201 type, even if we don't pass or return a long double type.
3203 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
3206 * ipa-inline.c (flatten_function): Do not overwrite final inlining
3209 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
3211 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
3212 support for merge[hl].
3213 (fold_mergehl_helper): New helper function.
3214 (tree-vector-builder.h): New #include for tree_vector_builder usage.
3215 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
3216 (altivec_vmrglw_direct): Add xxmrglw insn.
3218 2018-01-17 Andrew Waterman <andrew@sifive.com>
3220 * config/riscv/riscv.c (riscv_conditional_register_usage): If
3221 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
3223 2018-01-17 David Malcolm <dmalcolm@redhat.com>
3226 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
3227 call the lto_location_cache before reading the
3228 DECL_SOURCE_LOCATION of the types.
3230 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
3231 Richard Sandiford <richard.sandiford@linaro.org>
3233 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
3234 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
3235 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
3236 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
3237 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
3239 * config/aarch64/constraints.md (aarch64_movti_operand):
3241 * config/aarch64/predicates.md (Uti): Add new constraint.
3243 2018-01-17 Carl Love <cel@us.ibm.com>
3245 * config/rs6000/vsx.md (define_expand xl_len_r,
3246 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
3247 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
3249 (define_expand, define_insn): Move the shift left from the
3250 define_insn to the define_expand for lxvl and stxvl instructions.
3251 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
3252 and XL_LEN_R definitions to PURE.
3254 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
3256 * config/i386/i386.c (indirect_thunk_name): Declare regno
3257 as unsigned int. Compare regno with INVALID_REGNUM.
3258 (output_indirect_thunk): Ditto.
3259 (output_indirect_thunk_function): Ditto.
3260 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
3261 in the call to output_indirect_thunk_function.
3263 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
3266 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
3267 rather than the size of inner_type to determine the stack slot size
3268 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
3270 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
3273 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
3276 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
3278 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
3279 endian Linux systems to optionally enable multilibs for selecting
3280 the long double type if the user configured an explicit type.
3281 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
3282 have no long double multilibs if not defined.
3283 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
3284 warn if the user used -mabi={ieee,ibm}longdouble and we built
3285 multilibs for long double.
3286 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
3287 appropriate multilib option.
3288 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
3290 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
3291 for building long double multilibs.
3292 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
3294 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
3296 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
3299 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
3301 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
3304 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
3307 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
3310 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3312 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
3313 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
3315 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
3317 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
3318 different rtl trees depending on TARGET_64BIT.
3319 (rs6000_gen_lvx): Likewise.
3321 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3323 * config/visium/visium.md (nop): Tweak comment.
3324 (hazard_nop): Likewise.
3326 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3328 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
3329 -mspeculate-indirect-jumps.
3330 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
3331 for -mno-speculate-indirect-jumps.
3332 (*call_indirect_elfv2<mode>_nospec): New define_insn.
3333 (*call_value_indirect_elfv2<mode>): Disable for
3334 -mno-speculate-indirect-jumps.
3335 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
3336 (indirect_jump): Emit different RTL for
3337 -mno-speculate-indirect-jumps.
3338 (*indirect_jump<mode>): Disable for
3339 -mno-speculate-indirect-jumps.
3340 (*indirect_jump<mode>_nospec): New define_insn.
3341 (tablejump): Emit different RTL for
3342 -mno-speculate-indirect-jumps.
3343 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
3344 (tablejumpsi_nospec): New define_expand.
3345 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
3346 (tablejumpdi_nospec): New define_expand.
3347 (*tablejump<mode>_internal1): Disable for
3348 -mno-speculate-indirect-jumps.
3349 (*tablejump<mode>_internal1_nospec): New define_insn.
3350 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
3353 2018-01-16 Artyom Skrobov tyomitch@gmail.com
3355 * caller-save.c (insert_save): Drop unnecessary parameter. All
3358 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3359 Richard Biener <rguenth@suse.de>
3362 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
3363 return early, inline manually is_gimple_sizepos. Make sure if we
3364 call gimplify_expr we don't end up with a gimple constant.
3365 * tree.c (variably_modified_type_p): Don't return true for
3366 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
3367 * gimplify.h (is_gimple_sizepos): Remove.
3369 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3371 PR tree-optimization/83857
3372 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
3373 vectorizable_live_operation for pure SLP statements.
3374 (vectorizable_live_operation): Handle PHIs.
3376 2018-01-16 Richard Biener <rguenther@suse.de>
3378 PR tree-optimization/83867
3379 * tree-vect-stmts.c (vect_transform_stmt): Precompute
3380 nested_in_vect_loop_p since the scalar stmt may get invalidated.
3382 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3385 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
3386 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
3387 If off is not INTEGER_CST, issue a may not be aligned warning
3388 rather than isn't aligned. Use isn%'t rather than isn't.
3389 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
3391 <case MULT_EXPR>: Improve the case when bottom and one of the
3392 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
3393 operand, in that case check if the other operand is multiple of
3394 bottom divided by the INTEGER_CST operand.
3396 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3399 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
3400 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
3401 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
3402 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
3403 * config/pa/pa.c (pa_function_arg_advance): Likewise.
3404 (pa_function_arg, pa_arg_partial_bytes): Likewise.
3405 (pa_function_arg_size): New function.
3407 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3409 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
3410 in a separate statement.
3412 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3414 PR tree-optimization/83847
3415 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
3416 group gathers and scatters.
3418 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3420 PR rtl-optimization/86620
3421 * params.def (max-sched-ready-insns): Bump minimum value to 1.
3423 PR rtl-optimization/83213
3424 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
3425 to last if both are JUMP_INSNs.
3427 PR tree-optimization/83843
3428 * gimple-ssa-store-merging.c
3429 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
3430 store_immediate_info for bswap/nop orig_stores.
3432 2018-01-15 Andrew Waterman <andrew@sifive.com>
3434 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
3436 <UDIV>: Increase cost if !TARGET_DIV.
3438 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
3440 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
3441 (define_attr "cr_logical_3op"): New.
3442 (cceq_ior_compare): Adjust.
3443 (cceq_ior_compare_complement): Adjust.
3444 (*cceq_rev_compare): Adjust.
3445 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
3446 (is_cracked_insn): Adjust.
3447 (insn_must_be_first_in_group): Adjust.
3448 * config/rs6000/40x.md: Adjust.
3449 * config/rs6000/440.md: Adjust.
3450 * config/rs6000/476.md: Adjust.
3451 * config/rs6000/601.md: Adjust.
3452 * config/rs6000/603.md: Adjust.
3453 * config/rs6000/6xx.md: Adjust.
3454 * config/rs6000/7450.md: Adjust.
3455 * config/rs6000/7xx.md: Adjust.
3456 * config/rs6000/8540.md: Adjust.
3457 * config/rs6000/cell.md: Adjust.
3458 * config/rs6000/e300c2c3.md: Adjust.
3459 * config/rs6000/e500mc.md: Adjust.
3460 * config/rs6000/e500mc64.md: Adjust.
3461 * config/rs6000/e5500.md: Adjust.
3462 * config/rs6000/e6500.md: Adjust.
3463 * config/rs6000/mpc.md: Adjust.
3464 * config/rs6000/power4.md: Adjust.
3465 * config/rs6000/power5.md: Adjust.
3466 * config/rs6000/power6.md: Adjust.
3467 * config/rs6000/power7.md: Adjust.
3468 * config/rs6000/power8.md: Adjust.
3469 * config/rs6000/power9.md: Adjust.
3470 * config/rs6000/rs64.md: Adjust.
3471 * config/rs6000/titan.md: Adjust.
3473 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3475 * config/i386/predicates.md (indirect_branch_operand): Rewrite
3476 ix86_indirect_branch_register logic.
3478 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3480 * config/i386/constraints.md (Bs): Update
3481 ix86_indirect_branch_register check. Don't check
3482 ix86_indirect_branch_register with GOT_memory_operand.
3484 * config/i386/predicates.md (GOT_memory_operand): Don't check
3485 ix86_indirect_branch_register here.
3486 (GOT32_symbol_operand): Likewise.
3488 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3490 * config/i386/predicates.md (constant_call_address_operand):
3491 Rewrite ix86_indirect_branch_register logic.
3492 (sibcall_insn_operand): Likewise.
3494 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3496 * config/i386/constraints.md (Bs): Replace
3497 ix86_indirect_branch_thunk_register with
3498 ix86_indirect_branch_register.
3500 * config/i386/i386.md (indirect_jump): Likewise.
3501 (tablejump): Likewise.
3502 (*sibcall_memory): Likewise.
3503 (*sibcall_value_memory): Likewise.
3504 Peepholes of indirect call and jump via memory: Likewise.
3505 * config/i386/i386.opt: Likewise.
3506 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3507 (GOT_memory_operand): Likewise.
3508 (call_insn_operand): Likewise.
3509 (sibcall_insn_operand): Likewise.
3510 (GOT32_symbol_operand): Likewise.
3512 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3515 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
3516 type rather than type addr's type points to.
3517 (expand_omp_atomic_mutex): Likewise.
3518 (expand_omp_atomic): Likewise.
3520 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3523 * config/i386/i386.c (output_indirect_thunk_function): Use
3524 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
3525 for __x86_return_thunk.
3527 2018-01-15 Richard Biener <rguenther@suse.de>
3530 * expmed.c (extract_bit_field_1): Fix typo.
3532 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3535 * config/arm/iterators.md (VF): New mode iterator.
3536 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
3537 Remove integer-related logic from pattern.
3538 (neon_vabd<mode>_3): Likewise.
3540 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3543 * common.opt (fstrict-overflow): No longer an alias.
3544 (fwrapv-pointer): New option.
3545 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
3546 also for pointer types based on flag_wrapv_pointer.
3547 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
3548 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
3549 opts->x_flag_wrapv got set.
3550 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
3551 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
3552 POINTER_TYPE_OVERFLOW_UNDEFINED.
3553 * match.pd: Likewise in address comparison pattern.
3554 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
3556 2018-01-15 Richard Biener <rguenther@suse.de>
3559 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
3560 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
3561 Reset type names to their identifier if their TYPE_DECL doesn't
3562 have linkage (and thus is used for ODR and devirt).
3563 (save_debug_info_for_decl): Remove.
3564 (save_debug_info_for_type): Likewise.
3565 (add_tree_to_fld_list): Adjust.
3566 * tree-pretty-print.c (dump_generic_node): Make dumping of
3567 type names more robust.
3569 2018-01-15 Richard Biener <rguenther@suse.de>
3571 * BASE-VER: Bump to 8.0.1.
3573 2018-01-14 Martin Sebor <msebor@redhat.com>
3576 * builtins.c (check_access): Avoid warning when the no-warning bit
3579 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
3581 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
3582 * ira-color (allocno_hard_regs_compare): Likewise.
3584 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
3587 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
3588 Use .pushsection/.popsection.
3590 2018-01-14 Martin Sebor <msebor@redhat.com>
3593 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
3595 2018-01-14 Jakub Jelinek <jakub@redhat.com>
3597 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
3598 entry from extra_headers.
3599 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
3600 extra_headers, make the list bitwise identical to the i?86-*-* one.
3602 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3604 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
3605 -mcmodel=large with -mindirect-branch=thunk,
3606 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
3607 -mfunction-return=thunk-extern.
3608 * doc/invoke.texi: Document -mcmodel=large is incompatible with
3609 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
3610 -mfunction-return=thunk and -mfunction-return=thunk-extern.
3612 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3614 * config/i386/i386.c (print_reg): Print the name of the full
3615 integer register without '%'.
3616 (ix86_print_operand): Handle 'V'.
3617 * doc/extend.texi: Document 'V' modifier.
3619 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3621 * config/i386/constraints.md (Bs): Disallow memory operand for
3622 -mindirect-branch-register.
3624 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3625 (GOT_memory_operand): Likewise.
3626 (call_insn_operand): Likewise.
3627 (sibcall_insn_operand): Likewise.
3628 (GOT32_symbol_operand): Likewise.
3629 * config/i386/i386.md (indirect_jump): Call convert_memory_address
3630 for -mindirect-branch-register.
3631 (tablejump): Likewise.
3632 (*sibcall_memory): Likewise.
3633 (*sibcall_value_memory): Likewise.
3634 Disallow peepholes of indirect call and jump via memory for
3635 -mindirect-branch-register.
3636 (*call_pop): Replace m with Bw.
3637 (*call_value_pop): Likewise.
3638 (*sibcall_pop_memory): Replace m with Bs.
3639 * config/i386/i386.opt (mindirect-branch-register): New option.
3640 * doc/invoke.texi: Document -mindirect-branch-register option.
3642 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3644 * config/i386/i386-protos.h (ix86_output_function_return): New.
3645 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
3646 set function_return_type.
3647 (indirect_thunk_name): Add ret_p to indicate thunk for function
3649 (output_indirect_thunk_function): Pass false to
3650 indirect_thunk_name.
3651 (ix86_output_indirect_branch_via_reg): Likewise.
3652 (ix86_output_indirect_branch_via_push): Likewise.
3653 (output_indirect_thunk_function): Create alias for function
3654 return thunk if regno < 0.
3655 (ix86_output_function_return): New function.
3656 (ix86_handle_fndecl_attribute): Handle function_return.
3657 (ix86_attribute_table): Add function_return.
3658 * config/i386/i386.h (machine_function): Add
3659 function_return_type.
3660 * config/i386/i386.md (simple_return_internal): Use
3661 ix86_output_function_return.
3662 (simple_return_internal_long): Likewise.
3663 * config/i386/i386.opt (mfunction-return=): New option.
3664 (indirect_branch): Mention -mfunction-return=.
3665 * doc/extend.texi: Document function_return function attribute.
3666 * doc/invoke.texi: Document -mfunction-return= option.
3668 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3670 * config/i386/i386-opts.h (indirect_branch): New.
3671 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
3672 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
3673 with local indirect jump when converting indirect call and jump.
3674 (ix86_set_indirect_branch_type): New.
3675 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
3676 (indirectlabelno): New.
3677 (indirect_thunk_needed): Likewise.
3678 (indirect_thunk_bnd_needed): Likewise.
3679 (indirect_thunks_used): Likewise.
3680 (indirect_thunks_bnd_used): Likewise.
3681 (INDIRECT_LABEL): Likewise.
3682 (indirect_thunk_name): Likewise.
3683 (output_indirect_thunk): Likewise.
3684 (output_indirect_thunk_function): Likewise.
3685 (ix86_output_indirect_branch_via_reg): Likewise.
3686 (ix86_output_indirect_branch_via_push): Likewise.
3687 (ix86_output_indirect_branch): Likewise.
3688 (ix86_output_indirect_jmp): Likewise.
3689 (ix86_code_end): Call output_indirect_thunk_function if needed.
3690 (ix86_output_call_insn): Call ix86_output_indirect_branch if
3692 (ix86_handle_fndecl_attribute): Handle indirect_branch.
3693 (ix86_attribute_table): Add indirect_branch.
3694 * config/i386/i386.h (machine_function): Add indirect_branch_type
3695 and has_local_indirect_jump.
3696 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
3698 (tablejump): Likewise.
3699 (*indirect_jump): Use ix86_output_indirect_jmp.
3700 (*tablejump_1): Likewise.
3701 (simple_return_indirect_internal): Likewise.
3702 * config/i386/i386.opt (mindirect-branch=): New option.
3703 (indirect_branch): New.
3706 (thunk-inline): Likewise.
3707 (thunk-extern): Likewise.
3708 * doc/extend.texi: Document indirect_branch function attribute.
3709 * doc/invoke.texi: Document -mindirect-branch= option.
3711 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
3714 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
3716 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
3718 * ipa-inline.c (want_inline_small_function_p): Return false if
3719 inlining has already failed with CIF_FINAL_ERROR.
3720 (update_caller_keys): Call want_inline_small_function_p before
3722 (update_callee_keys): Likewise.
3724 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
3726 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
3728 (rs6000_quadword_masked_address_p): Likewise.
3729 (quad_aligned_load_p): Likewise.
3730 (quad_aligned_store_p): Likewise.
3731 (const_load_sequence_p): Add comment to describe the outer-most loop.
3732 (mimic_memory_attributes_and_flags): New function.
3733 (rs6000_gen_stvx): Likewise.
3734 (replace_swapped_aligned_store): Likewise.
3735 (rs6000_gen_lvx): Likewise.
3736 (replace_swapped_aligned_load): Likewise.
3737 (replace_swapped_load_constant): Capitalize argument name in
3738 comment describing this function.
3739 (rs6000_analyze_swaps): Add a third pass to search for vector loads
3740 and stores that access quad-word aligned addresses and replace
3741 with stvx or lvx instructions when appropriate.
3742 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
3743 New function prototype.
3744 (rs6000_quadword_masked_address_p): Likewise.
3745 (rs6000_gen_lvx): Likewise.
3746 (rs6000_gen_stvx): Likewise.
3747 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3748 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3749 when memory address is aligned.
3750 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3751 this split to select lvx instruction when memory address is aligned.
3752 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3753 instruction when memory address is aligned.
3754 (*vsx_le_perm_load_v16qi): Likewise.
3755 (four unnamed splitters): Modify to select the stvx instruction
3756 when memory is aligned.
3758 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3760 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3761 which appears in the queue multiple times.
3763 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3764 Alan Hayward <alan.hayward@arm.com>
3765 David Sherwood <david.sherwood@arm.com>
3767 * tree-vectorizer.h (vec_lower_bound): New structure.
3768 (_loop_vec_info): Add check_nonzero and lower_bounds.
3769 (LOOP_VINFO_CHECK_NONZERO): New macro.
3770 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3771 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3772 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3773 fields. Make seg_len the distance travelled, not including the
3775 (dr_direction_indicator): Declare.
3776 (dr_zero_step_indicator): Likewise.
3777 (dr_known_forward_stride_p): Likewise.
3778 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3780 (runtime_alias_check_p): Allow runtime alias checks with
3782 (operator ==): Compare access_size and align.
3783 (prune_runtime_alias_test_list): Rework for new distinction between
3784 the access_size and seg_len.
3785 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3787 (get_segment_min_max): New function.
3788 (create_intersect_range_checks): Use it.
3789 (dr_step_indicator): New function.
3790 (dr_direction_indicator): Likewise.
3791 (dr_zero_step_indicator): Likewise.
3792 (dr_known_forward_stride_p): Likewise.
3793 * tree-loop-distribution.c (data_ref_segment_size): Return
3794 DR_STEP * (niters - 1).
3795 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3797 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3798 (vect_preserves_scalar_order_p): New function, split out from...
3799 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3800 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3801 (vect_vfa_access_size): New function.
3802 (vect_vfa_align): Likewise.
3803 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3804 (dump_lower_bound): New function.
3805 (vect_check_lower_bound): Likewise.
3806 (vect_small_gap_p): Likewise.
3807 (vectorizable_with_step_bound_p): Likewise.
3808 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3809 depencies if the vectorization factor is 1. Convert the checks
3810 for nonzero steps into checks on the bounds of DR_STEP. Try using
3811 a bunds check for variable steps if the minimum required step is
3812 relatively small. Update calls to the dr_with_seg_len
3813 constructor and to vect_compile_time_alias.
3814 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3816 (vect_loop_versioning): Call it.
3817 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3819 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3821 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3822 Alan Hayward <alan.hayward@arm.com>
3823 David Sherwood <david.sherwood@arm.com>
3825 * doc/sourcebuild.texi (vect_scatter_store): Document.
3826 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3828 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3830 * genopinit.c (main): Add supports_vec_scatter_store and
3831 supports_vec_scatter_store_cached to target_optabs.
3832 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3833 IFN_MASK_SCATTER_STORE.
3834 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3836 * internal-fn.h (internal_store_fn_p): Declare.
3837 (internal_fn_stored_value_index): Likewise.
3838 * internal-fn.c (scatter_store_direct): New macro.
3839 (expand_scatter_store_optab_fn): New function.
3840 (direct_scatter_store_optab_supported_p): New macro.
3841 (internal_store_fn_p): New function.
3842 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3843 IFN_MASK_SCATTER_STORE.
3844 (internal_fn_mask_index): Likewise.
3845 (internal_fn_stored_value_index): New function.
3846 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3848 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3849 * optabs-query.c (supports_vec_scatter_store_p): New function.
3850 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3851 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3852 true for scatter stores.
3853 (vect_gather_scatter_fn_p): Handle scatter stores too.
3854 (vect_check_gather_scatter): Consider using scatter stores if
3855 supports_vec_scatter_store_p.
3856 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3858 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3859 internal_fn_stored_value_index.
3860 (check_load_store_masking): Handle scatter stores too.
3861 (vect_get_store_rhs): Make public.
3862 (vectorizable_call): Use internal_store_fn_p.
3863 (vectorizable_store): Handle scatter store internal functions.
3864 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3865 when deciding whether the end of the group has been reached.
3866 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3867 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3868 (mask_scatter_store<mode>): New insns.
3870 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3871 Alan Hayward <alan.hayward@arm.com>
3872 David Sherwood <david.sherwood@arm.com>
3874 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3875 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3876 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3878 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3879 Use vect_truncate_gather_scatter_offset if we can't treat the
3880 operation as a normal gather load or scatter store.
3881 (get_group_load_store_type): Take the gather_scatter_info
3882 as argument. Try using a gather load or scatter store for
3883 single-element groups.
3884 (get_load_store_type): Update calls to get_group_load_store_type
3885 and vect_use_strided_gather_scatters_p.
3887 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3888 Alan Hayward <alan.hayward@arm.com>
3889 David Sherwood <david.sherwood@arm.com>
3891 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3892 optional tree argument.
3893 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3895 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3896 but continue to use the current value as a fallback.
3897 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3898 to compare the updates.
3899 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3900 (get_load_store_type): Use it when handling a strided access.
3901 (vect_get_strided_load_store_ops): New function.
3902 (vect_get_data_ptr_increment): Likewise.
3903 (vectorizable_load): Handle strided gather loads. Always pass
3904 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3906 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3907 Alan Hayward <alan.hayward@arm.com>
3908 David Sherwood <david.sherwood@arm.com>
3910 * doc/md.texi (gather_load@var{m}): Document.
3911 (mask_gather_load@var{m}): Likewise.
3912 * genopinit.c (main): Add supports_vec_gather_load and
3913 supports_vec_gather_load_cached to target_optabs.
3914 * optabs-tree.c (init_tree_optimization_optabs): Use
3915 ggc_cleared_alloc to allocate target_optabs.
3916 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3917 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3919 * internal-fn.h (internal_load_fn_p): Declare.
3920 (internal_gather_scatter_fn_p): Likewise.
3921 (internal_fn_mask_index): Likewise.
3922 (internal_gather_scatter_fn_supported_p): Likewise.
3923 * internal-fn.c (gather_load_direct): New macro.
3924 (expand_gather_load_optab_fn): New function.
3925 (direct_gather_load_optab_supported_p): New macro.
3926 (direct_internal_fn_optab): New function.
3927 (internal_load_fn_p): Likewise.
3928 (internal_gather_scatter_fn_p): Likewise.
3929 (internal_fn_mask_index): Likewise.
3930 (internal_gather_scatter_fn_supported_p): Likewise.
3931 * optabs-query.c (supports_at_least_one_mode_p): New function.
3932 (supports_vec_gather_load_p): Likewise.
3933 * optabs-query.h (supports_vec_gather_load_p): Declare.
3934 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3935 and memory_type field.
3936 (NUM_PATTERNS): Bump to 15.
3937 * tree-vect-data-refs.c: Include internal-fn.h.
3938 (vect_gather_scatter_fn_p): New function.
3939 (vect_describe_gather_scatter_call): Likewise.
3940 (vect_check_gather_scatter): Try using internal functions for
3941 gather loads. Recognize existing calls to a gather load function.
3942 (vect_analyze_data_refs): Consider using gather loads if
3943 supports_vec_gather_load_p.
3944 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3945 (vect_get_gather_scatter_offset_type): Likewise.
3946 (vect_convert_mask_for_vectype): Likewise.
3947 (vect_add_conversion_to_patterm): Likewise.
3948 (vect_try_gather_scatter_pattern): Likewise.
3949 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3950 (vect_vect_recog_func_ptrs): Add it.
3951 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3952 internal_fn_mask_index and internal_gather_scatter_fn_p.
3953 (check_load_store_masking): Take the gather_scatter_info as an
3954 argument and handle gather loads.
3955 (vect_get_gather_scatter_ops): New function.
3956 (vectorizable_call): Check internal_load_fn_p.
3957 (vectorizable_load): Likewise. Handle gather load internal
3959 (vectorizable_store): Update call to check_load_store_masking.
3960 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3961 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3962 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3963 (aarch64_gather_scale_operand_d): New predicates.
3964 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3965 (mask_gather_load<mode>): New insns.
3967 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3968 Alan Hayward <alan.hayward@arm.com>
3969 David Sherwood <david.sherwood@arm.com>
3971 * optabs.def (fold_left_plus_optab): New optab.
3972 * doc/md.texi (fold_left_plus_@var{m}): Document.
3973 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3974 * internal-fn.c (fold_left_direct): Define.
3975 (expand_fold_left_optab_fn): Likewise.
3976 (direct_fold_left_optab_supported_p): Likewise.
3977 * fold-const-call.c (fold_const_fold_left): New function.
3978 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3979 * tree-parloops.c (valid_reduction_p): New function.
3980 (gather_scalar_reductions): Use it.
3981 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3982 (vect_finish_replace_stmt): Declare.
3983 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3984 (needs_fold_left_reduction_p): New function, split out from...
3985 (vect_is_simple_reduction): ...here. Accept reductions that
3986 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3987 (vect_force_simple_reduction): Also store the reduction type in
3988 the assignment's STMT_VINFO_REDUC_TYPE.
3989 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3990 (merge_with_identity): New function.
3991 (vect_expand_fold_left): Likewise.
3992 (vectorize_fold_left_reduction): Likewise.
3993 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3994 scalar phi in place for it. Check for target support and reject
3995 cases that would reassociate the operation. Defer the transform
3996 phase to vectorize_fold_left_reduction.
3997 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3998 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3999 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
4001 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4003 * tree-if-conv.c (predicate_mem_writes): Remove redundant
4004 call to ifc_temp_var.
4006 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4007 Alan Hayward <alan.hayward@arm.com>
4008 David Sherwood <david.sherwood@arm.com>
4010 * target.def (legitimize_address_displacement): Take the original
4011 offset as a poly_int.
4012 * targhooks.h (default_legitimize_address_displacement): Update
4014 * targhooks.c (default_legitimize_address_displacement): Likewise.
4015 * doc/tm.texi: Regenerate.
4016 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
4017 as an argument, moving assert of ad->disp == ad->disp_term to...
4018 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
4019 Try calling targetm.legitimize_address_displacement before expanding
4020 the address rather than afterwards, and adjust for the new interface.
4021 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
4022 Match the new hook interface. Handle SVE addresses.
4023 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
4026 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4028 * Makefile.in (OBJS): Add early-remat.o.
4029 * target.def (select_early_remat_modes): New hook.
4030 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
4031 * doc/tm.texi: Regenerate.
4032 * targhooks.h (default_select_early_remat_modes): Declare.
4033 * targhooks.c (default_select_early_remat_modes): New function.
4034 * timevar.def (TV_EARLY_REMAT): New timevar.
4035 * passes.def (pass_early_remat): New pass.
4036 * tree-pass.h (make_pass_early_remat): Declare.
4037 * early-remat.c: New file.
4038 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
4040 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
4042 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4043 Alan Hayward <alan.hayward@arm.com>
4044 David Sherwood <david.sherwood@arm.com>
4046 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
4047 vfm1 with a bound_epilog parameter.
4048 (vect_do_peeling): Update calls accordingly, and move the prologue
4049 call earlier in the function. Treat the base bound_epilog as 0 for
4050 fully-masked loops and retain vf - 1 for other loops. Add 1 to
4051 this base when peeling for gaps.
4052 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
4053 with fully-masked loops.
4054 (vect_estimate_min_profitable_iters): Handle the single peeled
4055 iteration in that case.
4057 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4058 Alan Hayward <alan.hayward@arm.com>
4059 David Sherwood <david.sherwood@arm.com>
4061 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
4062 single-element interleaving even if the size is not a power of 2.
4063 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
4064 accesses for single-element interleaving if the group size is
4067 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4068 Alan Hayward <alan.hayward@arm.com>
4069 David Sherwood <david.sherwood@arm.com>
4071 * doc/md.texi (fold_extract_last_@var{m}): Document.
4072 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
4073 * optabs.def (fold_extract_last_optab): New optab.
4074 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
4075 * internal-fn.c (fold_extract_direct): New macro.
4076 (expand_fold_extract_optab_fn): Likewise.
4077 (direct_fold_extract_optab_supported_p): Likewise.
4078 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
4079 * tree-vect-loop.c (vect_model_reduction_cost): Handle
4080 EXTRACT_LAST_REDUCTION.
4081 (get_initial_def_for_reduction): Do not create an initial vector
4082 for EXTRACT_LAST_REDUCTION reductions.
4083 (vectorizable_reduction): Leave the scalar phi in place for
4084 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
4085 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
4086 epilogue code for EXTRACT_LAST_REDUCTION and defer the
4087 transform phase to vectorizable_condition.
4088 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
4090 (vect_finish_stmt_generation): ...here.
4091 (vect_finish_replace_stmt): New function.
4092 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
4093 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
4095 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
4097 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4098 Alan Hayward <alan.hayward@arm.com>
4099 David Sherwood <david.sherwood@arm.com>
4101 * doc/md.texi (extract_last_@var{m}): Document.
4102 * optabs.def (extract_last_optab): New optab.
4103 * internal-fn.def (EXTRACT_LAST): New internal function.
4104 * internal-fn.c (cond_unary_direct): New macro.
4105 (expand_cond_unary_optab_fn): Likewise.
4106 (direct_cond_unary_optab_supported_p): Likewise.
4107 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
4108 loops using EXTRACT_LAST.
4109 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
4110 (extract_last_<mode>): ...this optab.
4111 (vec_extract<mode><Vel>): Update accordingly.
4113 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4114 Alan Hayward <alan.hayward@arm.com>
4115 David Sherwood <david.sherwood@arm.com>
4117 * target.def (empty_mask_is_expensive): New hook.
4118 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
4119 * doc/tm.texi: Regenerate.
4120 * targhooks.h (default_empty_mask_is_expensive): Declare.
4121 * targhooks.c (default_empty_mask_is_expensive): New function.
4122 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
4123 if the target says that empty masks are expensive.
4124 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
4126 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
4128 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4129 Alan Hayward <alan.hayward@arm.com>
4130 David Sherwood <david.sherwood@arm.com>
4132 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
4133 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
4134 (vect_use_loop_mask_for_alignment_p): New function.
4135 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
4136 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
4137 niters_skip argument. Make sure that the first niters_skip elements
4138 of the first iteration are inactive.
4139 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
4140 Update call to vect_set_loop_masks_directly.
4141 (get_misalign_in_elems): New function, split out from...
4142 (vect_gen_prolog_loop_niters): ...here.
4143 (vect_update_init_of_dr): Take a code argument that specifies whether
4144 the adjustment should be added or subtracted.
4145 (vect_update_init_of_drs): Likewise.
4146 (vect_prepare_for_masked_peels): New function.
4147 (vect_do_peeling): Skip prologue peeling if we're using a mask
4148 instead. Update call to vect_update_inits_of_drs.
4149 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4151 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
4152 alignment. Do not include the number of peeled iterations in
4153 the minimum threshold in that case.
4154 (vectorizable_induction): Adjust the start value down by
4155 LOOP_VINFO_MASK_SKIP_NITERS iterations.
4156 (vect_transform_loop): Call vect_prepare_for_masked_peels.
4157 Take the number of skipped iterations into account when calculating
4159 * tree-vect-stmts.c (vect_gen_while_not): New function.
4161 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4162 Alan Hayward <alan.hayward@arm.com>
4163 David Sherwood <david.sherwood@arm.com>
4165 * doc/sourcebuild.texi (vect_fully_masked): Document.
4166 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
4168 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
4170 (vect_analyze_loop_2): ...here. Don't check the vectorization
4171 factor against the number of loop iterations if the loop is
4174 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4175 Alan Hayward <alan.hayward@arm.com>
4176 David Sherwood <david.sherwood@arm.com>
4178 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
4179 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
4180 (dump_groups): Update accordingly.
4181 (iv_use::mem_type): New member variable.
4182 (address_p): New function.
4183 (record_use): Add a mem_type argument and initialize the new
4185 (record_group_use): Add a mem_type argument. Use address_p.
4186 Remove obsolete null checks of base_object. Update call to record_use.
4187 (find_interesting_uses_op): Update call to record_group_use.
4188 (find_interesting_uses_cond): Likewise.
4189 (find_interesting_uses_address): Likewise.
4190 (get_mem_type_for_internal_fn): New function.
4191 (find_address_like_use): Likewise.
4192 (find_interesting_uses_stmt): Try find_address_like_use before
4193 calling find_interesting_uses_op.
4194 (addr_offset_valid_p): Use the iv mem_type field as the type
4195 of the addressed memory.
4196 (add_autoinc_candidates): Likewise.
4197 (get_address_cost): Likewise.
4198 (split_small_address_groups_p): Use address_p.
4199 (split_address_groups): Likewise.
4200 (add_iv_candidate_for_use): Likewise.
4201 (autoinc_possible_for_pair): Likewise.
4202 (rewrite_groups): Likewise.
4203 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
4204 (determine_group_iv_cost): Update after split of USE_ADDRESS.
4205 (get_alias_ptr_type_for_ptr_address): New function.
4206 (rewrite_use_address): Rewrite address uses in calls that were
4207 identified by find_address_like_use.
4209 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4210 Alan Hayward <alan.hayward@arm.com>
4211 David Sherwood <david.sherwood@arm.com>
4213 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
4215 * gimple-expr.h (is_gimple_addressable: Likewise.
4216 * gimple-expr.c (is_gimple_address): Likewise.
4217 * internal-fn.c (expand_call_mem_ref): New function.
4218 (expand_mask_load_optab_fn): Use it.
4219 (expand_mask_store_optab_fn): Likewise.
4221 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4222 Alan Hayward <alan.hayward@arm.com>
4223 David Sherwood <david.sherwood@arm.com>
4225 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
4226 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
4227 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
4228 (cond_umax@var{mode}): Document.
4229 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
4230 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
4231 (cond_umin_optab, cond_umax_optab): New optabs.
4232 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
4233 (COND_IOR, COND_XOR): New internal functions.
4234 * internal-fn.h (get_conditional_internal_fn): Declare.
4235 * internal-fn.c (cond_binary_direct): New macro.
4236 (expand_cond_binary_optab_fn): Likewise.
4237 (direct_cond_binary_optab_supported_p): Likewise.
4238 (get_conditional_internal_fn): New function.
4239 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
4240 Cope with reduction statements that are vectorized as calls rather
4242 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
4243 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
4244 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
4245 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
4246 (UNSPEC_COND_EOR): New unspecs.
4247 (optab): Add mappings for them.
4248 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
4249 (sve_int_op, sve_fp_op): New int attributes.
4251 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4252 Alan Hayward <alan.hayward@arm.com>
4253 David Sherwood <david.sherwood@arm.com>
4255 * optabs.def (while_ult_optab): New optab.
4256 * doc/md.texi (while_ult@var{m}@var{n}): Document.
4257 * internal-fn.def (WHILE_ULT): New internal function.
4258 * internal-fn.h (direct_internal_fn_supported_p): New override
4259 that takes two types as argument.
4260 * internal-fn.c (while_direct): New macro.
4261 (expand_while_optab_fn): New function.
4262 (convert_optab_supported_p): Likewise.
4263 (direct_while_optab_supported_p): New macro.
4264 * wide-int.h (wi::udiv_ceil): New function.
4265 * tree-vectorizer.h (rgroup_masks): New structure.
4266 (vec_loop_masks): New typedef.
4267 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
4269 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
4270 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
4271 (vect_max_vf): New function.
4272 (slpeel_make_loop_iterate_ntimes): Delete.
4273 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
4274 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
4275 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
4276 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
4277 internal-fn.h, stor-layout.h and optabs-query.h.
4278 (vect_set_loop_mask): New function.
4279 (add_preheader_seq): Likewise.
4280 (add_header_seq): Likewise.
4281 (interleave_supported_p): Likewise.
4282 (vect_maybe_permute_loop_masks): Likewise.
4283 (vect_set_loop_masks_directly): Likewise.
4284 (vect_set_loop_condition_masked): Likewise.
4285 (vect_set_loop_condition_unmasked): New function, split out from
4286 slpeel_make_loop_iterate_ntimes.
4287 (slpeel_make_loop_iterate_ntimes): Rename to..
4288 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
4289 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
4290 (vect_do_peeling): Update call accordingly.
4291 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
4293 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4294 mask_compare_type, can_fully_mask_p and fully_masked_p.
4295 (release_vec_loop_masks): New function.
4296 (_loop_vec_info): Use it to free the loop masks.
4297 (can_produce_all_loop_masks_p): New function.
4298 (vect_get_max_nscalars_per_iter): Likewise.
4299 (vect_verify_full_masking): Likewise.
4300 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
4301 retries, and free the mask rgroups before retrying. Check loop-wide
4302 reasons for disallowing fully-masked loops. Make the final decision
4303 about whether use a fully-masked loop or not.
4304 (vect_estimate_min_profitable_iters): Do not assume that peeling
4305 for the number of iterations will be needed for fully-masked loops.
4306 (vectorizable_reduction): Disable fully-masked loops.
4307 (vectorizable_live_operation): Likewise.
4308 (vect_halve_mask_nunits): New function.
4309 (vect_double_mask_nunits): Likewise.
4310 (vect_record_loop_mask): Likewise.
4311 (vect_get_loop_mask): Likewise.
4312 (vect_transform_loop): Handle the case in which the final loop
4313 iteration might handle a partial vector. Call vect_set_loop_condition
4314 instead of slpeel_make_loop_iterate_ntimes.
4315 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
4316 (check_load_store_masking): New function.
4317 (prepare_load_store_mask): Likewise.
4318 (vectorizable_store): Handle fully-masked loops.
4319 (vectorizable_load): Likewise.
4320 (supportable_widening_operation): Use vect_halve_mask_nunits for
4322 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
4323 (vect_gen_while): New function.
4324 * config/aarch64/aarch64.md (umax<mode>3): New expander.
4325 (aarch64_uqdec<mode>): New insn.
4327 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4328 Alan Hayward <alan.hayward@arm.com>
4329 David Sherwood <david.sherwood@arm.com>
4331 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
4332 (reduc_xor_scal_optab): New optabs.
4333 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
4334 (reduc_xor_scal_@var{m}): Document.
4335 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
4336 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
4338 * fold-const-call.c (fold_const_call): Handle them.
4339 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
4340 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
4341 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
4342 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
4343 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
4344 (UNSPEC_XORV): New unspecs.
4345 (optab): Add entries for them.
4346 (BITWISEV): New int iterator.
4347 (bit_reduc_op): New int attributes.
4349 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4350 Alan Hayward <alan.hayward@arm.com>
4351 David Sherwood <david.sherwood@arm.com>
4353 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
4354 * internal-fn.def (VEC_SHL_INSERT): New internal function.
4355 * optabs.def (vec_shl_insert_optab): New optab.
4356 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
4357 (duplicate_and_interleave): Likewise.
4358 * tree-vect-loop.c: Include internal-fn.h.
4359 (neutral_op_for_slp_reduction): New function, split out from
4360 get_initial_defs_for_reduction.
4361 (get_initial_def_for_reduction): Handle option 2 for variable-length
4362 vectors by loading the neutral value into a vector and then shifting
4363 the initial value into element 0.
4364 (get_initial_defs_for_reduction): Replace the code argument with
4365 the neutral value calculated by neutral_op_for_slp_reduction.
4366 Use gimple_build_vector for constant-length vectors.
4367 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
4368 but the first group_size elements have a neutral value.
4369 Use duplicate_and_interleave otherwise.
4370 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
4371 Update call to get_initial_defs_for_reduction. Handle SLP
4372 reductions for variable-length vectors by creating one vector
4373 result for each scalar result, with the elements associated
4374 with other scalar results stubbed out with the neutral value.
4375 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
4376 Require IFN_VEC_SHL_INSERT for double reductions on
4377 variable-length vectors, or SLP reductions that have
4378 a neutral value. Require can_duplicate_and_interleave_p
4379 support for variable-length unchained SLP reductions if there
4380 is no neutral value, such as for MIN/MAX reductions. Also require
4381 the number of vector elements to be a multiple of the number of
4382 SLP statements when doing variable-length unchained SLP reductions.
4383 Update call to vect_create_epilog_for_reduction.
4384 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
4385 and remove initial values.
4386 (duplicate_and_interleave): Make public.
4387 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
4388 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
4390 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4391 Alan Hayward <alan.hayward@arm.com>
4392 David Sherwood <david.sherwood@arm.com>
4394 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
4395 (can_duplicate_and_interleave_p): New function.
4396 (vect_get_and_check_slp_defs): Take the vector of statements
4397 rather than just the current one. Remove excess parentheses.
4398 Restriction rejectinon of vect_constant_def and vect_external_def
4399 for variable-length vectors to boolean types, or types for which
4400 can_duplicate_and_interleave_p is false.
4401 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
4402 (duplicate_and_interleave): New function.
4403 (vect_get_constant_vectors): Use gimple_build_vector for
4404 constant-length vectors and suitable variable-length constant
4405 vectors. Use duplicate_and_interleave for other variable-length
4406 vectors. Don't defer the update when inserting new statements.
4408 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4409 Alan Hayward <alan.hayward@arm.com>
4410 David Sherwood <david.sherwood@arm.com>
4412 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
4413 min_profitable_iters doesn't go negative.
4415 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4416 Alan Hayward <alan.hayward@arm.com>
4417 David Sherwood <david.sherwood@arm.com>
4419 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
4420 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
4421 * optabs.def (vec_mask_load_lanes_optab): New optab.
4422 (vec_mask_store_lanes_optab): Likewise.
4423 * internal-fn.def (MASK_LOAD_LANES): New internal function.
4424 (MASK_STORE_LANES): Likewise.
4425 * internal-fn.c (mask_load_lanes_direct): New macro.
4426 (mask_store_lanes_direct): Likewise.
4427 (expand_mask_load_optab_fn): Handle masked operations.
4428 (expand_mask_load_lanes_optab_fn): New macro.
4429 (expand_mask_store_optab_fn): Handle masked operations.
4430 (expand_mask_store_lanes_optab_fn): New macro.
4431 (direct_mask_load_lanes_optab_supported_p): Likewise.
4432 (direct_mask_store_lanes_optab_supported_p): Likewise.
4433 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
4435 (vect_load_lanes_supported): Likewise.
4436 * tree-vect-data-refs.c (strip_conversion): New function.
4437 (can_group_stmts_p): Likewise.
4438 (vect_analyze_data_ref_accesses): Use it instead of checking
4439 for a pair of assignments.
4440 (vect_store_lanes_supported): Take a masked_p parameter.
4441 (vect_load_lanes_supported): Likewise.
4442 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
4443 vect_store_lanes_supported and vect_load_lanes_supported.
4444 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
4445 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
4446 parameter. Don't allow gaps for masked accesses.
4447 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
4448 and vect_load_lanes_supported.
4449 (get_load_store_type): Take a masked_p parameter and update
4450 call to get_group_load_store_type.
4451 (vectorizable_store): Update call to get_load_store_type.
4452 Handle IFN_MASK_STORE_LANES.
4453 (vectorizable_load): Update call to get_load_store_type.
4454 Handle IFN_MASK_LOAD_LANES.
4456 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4457 Alan Hayward <alan.hayward@arm.com>
4458 David Sherwood <david.sherwood@arm.com>
4460 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
4462 * config/aarch64/aarch64-protos.h
4463 (aarch64_sve_struct_memory_operand_p): Declare.
4464 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
4465 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
4466 (VPRED, vpred): Handle SVE structure modes.
4467 * config/aarch64/constraints.md (Utx): New constraint.
4468 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
4469 (aarch64_sve_struct_nonimmediate_operand): New predicates.
4470 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
4471 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
4472 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
4473 structure modes. Split into pieces after RA.
4474 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
4475 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
4477 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
4478 SVE structure modes.
4479 (aarch64_classify_address): Likewise.
4480 (sizetochar): Move earlier in file.
4481 (aarch64_print_operand): Handle SVE register lists.
4482 (aarch64_array_mode): New function.
4483 (aarch64_sve_struct_memory_operand_p): Likewise.
4484 (TARGET_ARRAY_MODE): Redefine.
4486 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4487 Alan Hayward <alan.hayward@arm.com>
4488 David Sherwood <david.sherwood@arm.com>
4490 * target.def (array_mode): New target hook.
4491 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
4492 * doc/tm.texi: Regenerate.
4493 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
4494 * hooks.c (hook_optmode_mode_uhwi_none): New function.
4495 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
4497 * stor-layout.c (mode_for_array): Likewise. Support polynomial
4500 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4501 Alan Hayward <alan.hayward@arm.com>
4502 David Sherwood <david.sherwood@arm.com>
4504 * fold-const.c (fold_binary_loc): Check the argument types
4505 rather than the result type when testing for a vector operation.
4507 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4509 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
4510 * doc/tm.texi: Regenerate.
4512 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4513 Alan Hayward <alan.hayward@arm.com>
4514 David Sherwood <david.sherwood@arm.com>
4516 * doc/invoke.texi (-msve-vector-bits=): Document new option.
4517 (sve): Document new AArch64 extension.
4518 * doc/md.texi (w): Extend the description of the AArch64
4519 constraint to include SVE vectors.
4520 (Upl, Upa): Document new AArch64 predicate constraints.
4521 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
4523 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
4524 (msve-vector-bits=): New option.
4525 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
4526 SVE when these are disabled.
4527 (sve): New extension.
4528 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
4529 modes. Adjust their number of units based on aarch64_sve_vg.
4530 (MAX_BITSIZE_MODE_ANY_MODE): Define.
4531 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
4532 aarch64_addr_query_type.
4533 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
4534 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
4535 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
4536 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
4537 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
4538 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
4539 (aarch64_simd_imm_zero_p): Delete.
4540 (aarch64_check_zero_based_sve_index_immediate): Declare.
4541 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4542 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4543 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4544 (aarch64_sve_float_mul_immediate_p): Likewise.
4545 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4547 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
4548 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
4549 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
4550 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
4551 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
4552 (aarch64_regmode_natural_size): Likewise.
4553 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
4554 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
4556 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
4557 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
4558 for VG and the SVE predicate registers.
4559 (V_ALIASES): Add a "z"-prefixed alias.
4560 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
4561 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
4562 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
4563 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
4564 (REG_CLASS_NAMES): Add entries for them.
4565 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
4566 and the predicate registers.
4567 (aarch64_sve_vg): Declare.
4568 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
4569 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
4570 (REGMODE_NATURAL_SIZE): Define.
4571 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
4573 * config/aarch64/aarch64.c: Include cfgrtl.h.
4574 (simd_immediate_info): Add a constructor for series vectors,
4575 and an associated step field.
4576 (aarch64_sve_vg): New variable.
4577 (aarch64_dbx_register_number): Handle VG and the predicate registers.
4578 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
4579 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
4580 (VEC_ANY_DATA, VEC_STRUCT): New constants.
4581 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
4582 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
4583 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
4584 (aarch64_get_mask_mode): New functions.
4585 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
4586 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
4587 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
4588 predicate modes and predicate registers. Explicitly restrict
4589 GPRs to modes of 16 bytes or smaller. Only allow FP registers
4590 to store a vector mode if it is recognized by
4591 aarch64_classify_vector_mode.
4592 (aarch64_regmode_natural_size): New function.
4593 (aarch64_hard_regno_caller_save_mode): Return the original mode
4595 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
4596 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
4597 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
4598 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
4600 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
4601 does not overlap dest if the function is frame-related. Handle
4603 (aarch64_split_add_offset): New function.
4604 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
4605 them aarch64_add_offset.
4606 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
4607 and update call to aarch64_sub_sp.
4608 (aarch64_add_cfa_expression): New function.
4609 (aarch64_expand_prologue): Pass extra temporary registers to the
4610 functions above. Handle the case in which we need to emit new
4611 DW_CFA_expressions for registers that were originally saved
4612 relative to the stack pointer, but now have to be expressed
4613 relative to the frame pointer.
4614 (aarch64_output_mi_thunk): Pass extra temporary registers to the
4616 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
4617 IP0 and IP1 values for SVE frames.
4618 (aarch64_expand_vec_series): New function.
4619 (aarch64_expand_sve_widened_duplicate): Likewise.
4620 (aarch64_expand_sve_const_vector): Likewise.
4621 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
4622 Handle SVE constants. Use emit_move_insn to move a force_const_mem
4623 into the register, rather than emitting a SET directly.
4624 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
4625 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
4626 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
4627 (offset_9bit_signed_scaled_p): New functions.
4628 (aarch64_replicate_bitmask_imm): New function.
4629 (aarch64_bitmask_imm): Use it.
4630 (aarch64_cannot_force_const_mem): Reject expressions involving
4631 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
4632 (aarch64_classify_index): Handle SVE indices, by requiring
4633 a plain register index with a scale that matches the element size.
4634 (aarch64_classify_address): Handle SVE addresses. Assert that
4635 the mode of the address is VOIDmode or an integer mode.
4636 Update call to aarch64_classify_symbol.
4637 (aarch64_classify_symbolic_expression): Update call to
4638 aarch64_classify_symbol.
4639 (aarch64_const_vec_all_in_range_p): New function.
4640 (aarch64_print_vector_float_operand): Likewise.
4641 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
4642 "vN" for FP registers with SVE modes. Handle (const ...) vectors
4643 and the FP immediates 1.0 and 0.5.
4644 (aarch64_print_address_internal): Handle SVE addresses.
4645 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
4646 (aarch64_regno_regclass): Handle predicate registers.
4647 (aarch64_secondary_reload): Handle big-endian reloads of SVE
4649 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
4650 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
4651 (aarch64_convert_sve_vector_bits): New function.
4652 (aarch64_override_options): Use it to handle -msve-vector-bits=.
4653 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4655 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
4656 Handle SVE vector and predicate modes. Accept VL-based constants
4657 that need only one temporary register, and VL offsets that require
4658 no temporary registers.
4659 (aarch64_conditional_register_usage): Mark the predicate registers
4660 as fixed if SVE isn't available.
4661 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
4662 Return true for SVE vector and predicate modes.
4663 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
4664 rather than an unsigned int. Handle SVE modes.
4665 (aarch64_preferred_simd_mode): Update call accordingly. Handle
4667 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
4669 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4670 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4671 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4672 (aarch64_sve_float_mul_immediate_p): New functions.
4673 (aarch64_sve_valid_immediate): New function.
4674 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
4675 Explicitly reject structure modes. Check for INDEX constants.
4676 Handle PTRUE and PFALSE constants.
4677 (aarch64_check_zero_based_sve_index_immediate): New function.
4678 (aarch64_simd_imm_zero_p): Delete.
4679 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
4680 vector modes. Accept constants in the range of CNT[BHWD].
4681 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
4682 ask for an Advanced SIMD mode.
4683 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
4684 (aarch64_simd_vector_alignment): Handle SVE predicates.
4685 (aarch64_vectorize_preferred_vector_alignment): New function.
4686 (aarch64_simd_vector_alignment_reachable): Use it instead of
4688 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
4689 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
4691 (MAX_VECT_LEN): Delete.
4692 (expand_vec_perm_d): Add a vec_flags field.
4693 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
4694 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
4695 (aarch64_evpc_ext): Don't apply a big-endian lane correction
4697 (aarch64_evpc_rev): Rename to...
4698 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
4699 (aarch64_evpc_rev_global): New function.
4700 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
4701 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
4703 (aarch64_evpc_sve_tbl): New function.
4704 (aarch64_expand_vec_perm_const_1): Update after rename of
4705 aarch64_evpc_rev. Handle SVE permutes too, trying
4706 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
4707 than aarch64_evpc_tbl.
4708 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
4709 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
4710 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
4711 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
4712 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
4713 (aarch64_expand_sve_vcond): New functions.
4714 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
4715 of aarch64_vector_mode_p.
4716 (aarch64_dwarf_poly_indeterminate_value): New function.
4717 (aarch64_compute_pressure_classes): Likewise.
4718 (aarch64_can_change_mode_class): Likewise.
4719 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
4720 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
4721 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
4722 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
4723 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
4724 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
4725 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
4726 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
4728 (Dn, Dl, Dr): Accept const as well as const_vector.
4729 (Dz): Likewise. Compare against CONST0_RTX.
4730 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
4731 of "vector" where appropriate.
4732 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
4733 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
4734 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
4735 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
4736 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
4737 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
4738 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
4739 (v_int_equiv): Extend to SVE modes.
4740 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
4742 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
4743 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4744 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4745 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4746 (SVE_COND_FP_CMP): New int iterators.
4747 (perm_hilo): Handle the new unpack unspecs.
4748 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4750 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4751 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4752 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4753 (aarch64_equality_operator, aarch64_constant_vector_operand)
4754 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4755 (aarch64_sve_nonimmediate_operand): Likewise.
4756 (aarch64_sve_general_operand): Likewise.
4757 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4758 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4759 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4760 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4761 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4762 (aarch64_sve_float_arith_immediate): Likewise.
4763 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4764 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4765 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4766 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4767 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4768 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4769 (aarch64_sve_float_arith_operand): Likewise.
4770 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4771 (aarch64_sve_float_mul_operand): Likewise.
4772 (aarch64_sve_vec_perm_operand): Likewise.
4773 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4774 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4775 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4776 as well as const_vector.
4777 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4778 in file. Use CONST0_RTX and CONSTM1_RTX.
4779 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4780 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4781 Use aarch64_simd_imm_zero.
4782 * config/aarch64/aarch64-sve.md: New file.
4783 * config/aarch64/aarch64.md: Include it.
4784 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4785 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4786 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4787 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4788 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4789 (sve): New attribute.
4790 (enabled): Disable instructions with the sve attribute unless
4792 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4793 aarch64_expand_mov_immediate.
4794 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4795 CNT[BHSD] immediates.
4796 (movti): Split CONST_POLY_INT moves into two halves.
4797 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4798 Split additions that need a temporary here if the destination
4799 is the stack pointer.
4800 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4801 (*add<mode>3_poly_1): New instruction.
4802 (set_clobber_cc): New expander.
4804 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4806 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4807 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4808 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4809 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4810 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4811 Change innermode from fixed_mode_size to machine_mode.
4812 (simplify_subreg): Update call accordingly. Handle a constant-sized
4813 subreg of a variable-length CONST_VECTOR.
4815 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4816 Alan Hayward <alan.hayward@arm.com>
4817 David Sherwood <david.sherwood@arm.com>
4819 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4820 (add_offset_to_base): New function, split out from...
4821 (create_mem_ref): ...here. When handling a scale other than 1,
4822 check first whether the address is valid without the offset.
4823 Add it into the base if so, leaving the index and scale as-is.
4825 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4828 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4829 fold_for_warn before checking if arg2 is INTEGER_CST.
4831 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4833 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4834 (store_multiple_operation): Delete.
4835 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4836 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4837 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4838 guarded by TARGET_STRING.
4839 (rs6000_output_load_multiple): Delete.
4840 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4841 OPTION_MASK_STRING / TARGET_STRING handling.
4842 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4843 (const rs6000_opt_masks) <"string">: Change mask to 0.
4844 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4845 (MASK_STRING): Delete.
4846 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4848 (load_multiple): Delete.
4855 (store_multiple): Delete.
4862 (movmemsi_8reg): Delete.
4863 (corresponding unnamed define_insn): Delete.
4864 (movmemsi_6reg): Delete.
4865 (corresponding unnamed define_insn): Delete.
4866 (movmemsi_4reg): Delete.
4867 (corresponding unnamed define_insn): Delete.
4868 (movmemsi_2reg): Delete.
4869 (corresponding unnamed define_insn): Delete.
4870 (movmemsi_1reg): Delete.
4871 (corresponding unnamed define_insn): Delete.
4872 * config/rs6000/rs6000.opt (mno-string): New.
4873 (mstring): Replace by deprecation warning stub.
4874 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4876 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4878 * regrename.c (regrename_do_replace): If replacing the same
4879 reg multiple times, try to reuse last created gen_raw_REG.
4882 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4883 main to workaround a bug in GDB.
4885 2018-01-12 Tom de Vries <tom@codesourcery.com>
4888 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4890 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4892 PR rtl-optimization/80481
4893 * ira-color.c (get_cap_member): New function.
4894 (allocnos_conflict_by_live_ranges_p): Use it.
4895 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4896 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4898 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4901 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4902 (*saddl_se_1): Ditto.
4904 (*ssubl_se_1): Ditto.
4906 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4908 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4909 rather than wi::to_widest for DR_INITs.
4910 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4911 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4912 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4914 (vect_analyze_group_access_1): Note that here.
4916 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4918 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4919 polynomial type sizes.
4921 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4923 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4924 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4925 (gimple_add_tmp_var): Likewise.
4927 2018-01-12 Martin Liska <mliska@suse.cz>
4929 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4930 (gimple_alloc_sizes): Likewise.
4931 (dump_gimple_statistics): Use PRIu64 in printf format.
4932 * gimple.h: Change uint64_t to int.
4934 2018-01-12 Martin Liska <mliska@suse.cz>
4936 * tree-core.h: Use uint64_t instead of int.
4937 * tree.c (tree_node_counts): Likewise.
4938 (tree_node_sizes): Likewise.
4939 (dump_tree_statistics): Use PRIu64 in printf format.
4941 2018-01-12 Martin Liska <mliska@suse.cz>
4943 * Makefile.in: As qsort_chk is implemented in vec.c, add
4944 vec.o to linkage of gencfn-macros.
4945 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4946 passing the info to record_node_allocation_statistics.
4947 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4949 * ggc-common.c (struct ggc_usage): Add operator== and use
4950 it in operator< and compare function.
4951 * mem-stats.h (struct mem_usage): Likewise.
4952 * vec.c (struct vec_usage): Remove operator< and compare
4953 function. Can be simply inherited.
4955 2018-01-12 Martin Jambor <mjambor@suse.cz>
4958 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4959 * tree-ssa-math-opts.c: Include domwalk.h.
4960 (convert_mult_to_fma_1): New function.
4961 (fma_transformation_info): New type.
4962 (fma_deferring_state): Likewise.
4963 (cancel_fma_deferring): New function.
4964 (result_of_phi): Likewise.
4965 (last_fma_candidate_feeds_initial_phi): Likewise.
4966 (convert_mult_to_fma): Added deferring logic, split actual
4967 transformation to convert_mult_to_fma_1.
4968 (math_opts_dom_walker): New type.
4969 (math_opts_dom_walker::after_dom_children): New method, body moved
4970 here from pass_optimize_widening_mul::execute, added deferring logic
4972 (pass_optimize_widening_mul::execute): Moved most of code to
4973 math_opts_dom_walker::after_dom_children.
4974 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4975 * config/i386/i386.c (ix86_option_override_internal): Added
4976 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4978 2018-01-12 Richard Biener <rguenther@suse.de>
4981 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4982 inline instance vars.
4984 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4987 * config/rx/rx.c (rx_is_restricted_memory_address):
4990 2018-01-12 Richard Biener <rguenther@suse.de>
4992 PR tree-optimization/80846
4993 * target.def (split_reduction): New target hook.
4994 * targhooks.c (default_split_reduction): New function.
4995 * targhooks.h (default_split_reduction): Declare.
4996 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4997 target requests first reduce vectors by combining low and high
4999 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
5000 (get_vectype_for_scalar_type_and_size): Export.
5001 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
5002 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
5003 * doc/tm.texi: Regenerate.
5004 * config/i386/i386.c (ix86_split_reduction): Implement
5005 TARGET_VECTORIZE_SPLIT_REDUCTION.
5007 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
5010 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
5011 in PIC mode except for TARGET_VXWORKS_RTP.
5012 * config/sparc/sparc.c: Include cfgrtl.h.
5013 (TARGET_INIT_PIC_REG): Define.
5014 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
5015 (sparc_pic_register_p): New predicate.
5016 (sparc_legitimate_address_p): Use it.
5017 (sparc_legitimize_pic_address): Likewise.
5018 (sparc_delegitimize_address): Likewise.
5019 (sparc_mode_dependent_address_p): Likewise.
5020 (gen_load_pcrel_sym): Remove 4th parameter.
5021 (load_got_register): Adjust call to above. Remove obsolete stuff.
5022 (sparc_expand_prologue): Do not call load_got_register here.
5023 (sparc_flat_expand_prologue): Likewise.
5024 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
5025 (sparc_use_pseudo_pic_reg): New function.
5026 (sparc_init_pic_reg): Likewise.
5027 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
5028 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
5030 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
5032 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
5033 Add item for branch_cost.
5035 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
5037 PR rtl-optimization/83565
5038 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
5039 not extend the result to a larger mode for rotate operations.
5040 (num_sign_bit_copies1): Likewise.
5042 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5045 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
5047 Use values-Xc.o for -pedantic.
5048 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
5050 2018-01-12 Martin Liska <mliska@suse.cz>
5053 * ipa-devirt.c (final_warning_record::grow_type_warnings):
5055 (possible_polymorphic_call_targets): Use it.
5056 (ipa_devirt): Likewise.
5058 2018-01-12 Martin Liska <mliska@suse.cz>
5060 * profile-count.h (enum profile_quality): Use 0 as invalid
5061 enum value of profile_quality.
5063 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
5065 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
5066 -mext-string options.
5068 2018-01-12 Richard Biener <rguenther@suse.de>
5070 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
5071 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
5072 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
5074 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
5076 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
5078 * configure.ac (--with-long-double-format): Add support for the
5079 configuration option to change the default long double format on
5081 * config.gcc (powerpc*-linux*-*): Likewise.
5082 * configure: Regenerate.
5083 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
5084 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
5085 used without modification.
5087 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5089 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
5090 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
5091 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
5092 MISC_BUILTIN_SPEC_BARRIER.
5093 (rs6000_init_builtins): Likewise.
5094 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
5096 (speculation_barrier): New define_insn.
5097 * doc/extend.texi: Document __builtin_speculation_barrier.
5099 2018-01-11 Jakub Jelinek <jakub@redhat.com>
5102 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
5103 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
5104 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
5106 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
5107 integral modes instead of "ss" and "sd".
5108 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
5109 vectors with 32-bit and 64-bit elements.
5110 (vecdupssescalarmodesuffix): New mode attribute.
5111 (vec_dup<mode>): Use it.
5113 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
5116 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
5117 frame if argument is passed on stack.
5119 2018-01-11 Jakub Jelinek <jakub@redhat.com>
5122 * ree.c (combine_reaching_defs): Optimize also
5123 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
5124 reg2=any_extend(exp); reg1=reg2;, formatting fix.
5126 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
5129 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
5131 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
5134 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
5135 after they are computed.
5137 2018-01-11 Bin Cheng <bin.cheng@arm.com>
5139 PR tree-optimization/83695
5140 * gimple-loop-linterchange.cc
5141 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
5142 reset cached scev information after interchange.
5143 (pass_linterchange::execute): Remove call to scev_reset_htab.
5145 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5147 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
5148 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
5149 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
5150 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
5151 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
5152 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
5153 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
5154 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
5155 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
5156 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
5157 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
5158 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
5159 (V_lane_reg): Likewise.
5160 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
5162 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
5163 (vfmal_lane_low<mode>_intrinsic,
5164 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
5165 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
5166 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
5167 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
5168 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
5169 vfmsl_lane_high<mode>_intrinsic): New define_insns.
5171 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5173 * config/arm/arm-cpus.in (fp16fml): New feature.
5174 (ALL_SIMD): Add fp16fml.
5175 (armv8.2-a): Add fp16fml as an option.
5176 (armv8.3-a): Likewise.
5177 (armv8.4-a): Add fp16fml as part of fp16.
5178 * config/arm/arm.h (TARGET_FP16FML): Define.
5179 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
5181 * config/arm/arm-modes.def (V2HF): Define.
5182 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
5183 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
5184 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
5185 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
5186 vfmsl_low, vfmsl_high): New set of builtins.
5187 * config/arm/iterators.md (PLUSMINUS): New code iterator.
5188 (vfml_op): New code attribute.
5189 (VFMLHALVES): New int iterator.
5190 (VFML, VFMLSEL): New mode attributes.
5191 (V_reg): Define mapping for V2HF.
5192 (V_hi, V_lo): New mode attributes.
5193 (VF_constraint): Likewise.
5194 (vfml_half, vfml_half_selector): New int attributes.
5195 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
5197 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
5198 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
5200 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
5201 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
5202 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
5203 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
5205 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
5206 Document new effective target and option set.
5208 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5210 * config/arm/arm-cpus.in (armv8_4): New feature.
5211 (ARMv8_4a): New fgroup.
5212 (armv8.4-a): New arch.
5213 * config/arm/arm-tables.opt: Regenerate.
5214 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
5215 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
5216 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
5217 Add matching rules for -march=armv8.4-a and extensions.
5218 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
5220 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
5223 * config/rx/rx.md (BW): New mode attribute.
5224 (sync_lock_test_and_setsi): Add mode suffix to insn output.
5226 2018-01-11 Richard Biener <rguenther@suse.de>
5228 PR tree-optimization/83435
5229 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
5230 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
5231 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
5233 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5234 Alan Hayward <alan.hayward@arm.com>
5235 David Sherwood <david.sherwood@arm.com>
5237 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
5239 (aarch64_classify_address): Initialize it. Track polynomial offsets.
5240 (aarch64_print_address_internal): Use it to check for a zero offset.
5242 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5243 Alan Hayward <alan.hayward@arm.com>
5244 David Sherwood <david.sherwood@arm.com>
5246 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
5247 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
5248 Return a poly_int64 rather than a HOST_WIDE_INT.
5249 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
5250 rather than a HOST_WIDE_INT.
5251 * config/aarch64/aarch64.h (aarch64_frame): Protect with
5252 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
5253 hard_fp_offset, frame_size, initial_adjust, callee_offset and
5254 final_offset from HOST_WIDE_INT to poly_int64.
5255 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
5256 to_constant when getting the number of units in an Advanced SIMD
5258 (aarch64_builtin_vectorized_function): Check for a constant number
5260 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
5262 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
5263 attribute instead of GET_MODE_NUNITS.
5264 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
5265 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
5266 GET_MODE_SIZE for fixed-size registers.
5267 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
5268 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
5269 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
5270 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
5271 (aarch64_print_operand, aarch64_print_address_internal)
5272 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
5273 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
5274 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
5275 Handle polynomial GET_MODE_SIZE.
5276 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
5277 wider than SImode without modification.
5278 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
5279 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
5280 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
5281 passing and returning SVE modes.
5282 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
5283 rather than GEN_INT.
5284 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
5285 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
5286 (aarch64_allocate_and_probe_stack_space): Likewise.
5287 (aarch64_layout_frame): Cope with polynomial offsets.
5288 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
5289 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
5291 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
5292 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
5293 poly_int64 rather than a HOST_WIDE_INT.
5294 (aarch64_get_separate_components, aarch64_process_components)
5295 (aarch64_expand_prologue, aarch64_expand_epilogue)
5296 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
5297 (aarch64_anchor_offset): New function, split out from...
5298 (aarch64_legitimize_address): ...here.
5299 (aarch64_builtin_vectorization_cost): Handle polynomial
5300 TYPE_VECTOR_SUBPARTS.
5301 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
5303 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
5304 number of elements from the PARALLEL rather than the mode.
5305 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
5306 rather than GET_MODE_BITSIZE.
5307 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
5308 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
5309 (aarch64_expand_vec_perm_const_1): Handle polynomial
5310 d->perm.length () and d->perm elements.
5311 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
5312 Apply to_constant to d->perm elements.
5313 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
5314 polynomial CONST_VECTOR_NUNITS.
5315 (aarch64_move_pointer): Take amount as a poly_int64 rather
5317 (aarch64_progress_pointer): Avoid temporary variable.
5318 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
5319 the mode attribute instead of GET_MODE.
5321 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5322 Alan Hayward <alan.hayward@arm.com>
5323 David Sherwood <david.sherwood@arm.com>
5325 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
5326 x exists before using it.
5327 (aarch64_add_constant_internal): Rename to...
5328 (aarch64_add_offset_1): ...this. Replace regnum with separate
5329 src and dest rtxes. Handle the case in which they're different,
5330 including when the offset is zero. Replace scratchreg with an rtx.
5331 Use 2 additions if there is no spare register into which we can
5332 move a 16-bit constant.
5333 (aarch64_add_constant): Delete.
5334 (aarch64_add_offset): Replace reg with separate src and dest
5335 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
5336 Use aarch64_add_offset_1.
5337 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
5338 an rtx rather than an int. Take the delta as a poly_int64
5339 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
5340 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
5341 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
5342 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
5343 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
5345 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
5346 aarch64_add_constant.
5348 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5350 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
5351 Use scalar_float_mode.
5353 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5355 * config/aarch64/aarch64-simd.md
5356 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
5357 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
5358 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
5359 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
5360 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
5361 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
5362 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
5363 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
5364 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
5365 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
5367 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5370 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
5371 targ_options->x_arm_arch_string is non NULL.
5373 2018-01-11 Tamar Christina <tamar.christina@arm.com>
5375 * config/aarch64/aarch64.h
5376 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
5378 2018-01-11 Sudakshina Das <sudi.das@arm.com>
5381 * expmed.c (emit_store_flag_force): Swap if const op0
5382 and change VOIDmode to mode of op0.
5384 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5386 PR rtl-optimization/83761
5387 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
5388 than bytes to mode_for_size.
5390 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5393 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
5394 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
5397 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5400 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
5401 when in layout mode.
5402 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
5403 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
5406 2018-01-10 Michael Collison <michael.collison@arm.com>
5408 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
5409 * config/aarch64/aarch64-option-extension.def: Add
5410 AARCH64_OPT_EXTENSION of 'fp16fml'.
5411 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5412 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
5413 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
5414 * config/aarch64/constraints.md (Ui7): New constraint.
5415 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
5416 (VFMLA_SEL_W): Ditto.
5419 (VFMLA16_LOW): New int iterator.
5420 (VFMLA16_HIGH): Ditto.
5421 (UNSPEC_FMLAL): New unspec.
5422 (UNSPEC_FMLSL): Ditto.
5423 (UNSPEC_FMLAL2): Ditto.
5424 (UNSPEC_FMLSL2): Ditto.
5425 (f16mac): New code attribute.
5426 * config/aarch64/aarch64-simd-builtins.def
5427 (aarch64_fmlal_lowv2sf): Ditto.
5428 (aarch64_fmlsl_lowv2sf): Ditto.
5429 (aarch64_fmlalq_lowv4sf): Ditto.
5430 (aarch64_fmlslq_lowv4sf): Ditto.
5431 (aarch64_fmlal_highv2sf): Ditto.
5432 (aarch64_fmlsl_highv2sf): Ditto.
5433 (aarch64_fmlalq_highv4sf): Ditto.
5434 (aarch64_fmlslq_highv4sf): Ditto.
5435 (aarch64_fmlal_lane_lowv2sf): Ditto.
5436 (aarch64_fmlsl_lane_lowv2sf): Ditto.
5437 (aarch64_fmlal_laneq_lowv2sf): Ditto.
5438 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
5439 (aarch64_fmlalq_lane_lowv4sf): Ditto.
5440 (aarch64_fmlsl_lane_lowv4sf): Ditto.
5441 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
5442 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
5443 (aarch64_fmlal_lane_highv2sf): Ditto.
5444 (aarch64_fmlsl_lane_highv2sf): Ditto.
5445 (aarch64_fmlal_laneq_highv2sf): Ditto.
5446 (aarch64_fmlsl_laneq_highv2sf): Ditto.
5447 (aarch64_fmlalq_lane_highv4sf): Ditto.
5448 (aarch64_fmlsl_lane_highv4sf): Ditto.
5449 (aarch64_fmlalq_laneq_highv4sf): Ditto.
5450 (aarch64_fmlsl_laneq_highv4sf): Ditto.
5451 * config/aarch64/aarch64-simd.md:
5452 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
5453 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5454 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
5455 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5456 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
5457 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
5458 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
5459 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
5460 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
5461 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
5462 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
5463 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
5464 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
5465 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
5466 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
5467 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
5468 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
5469 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
5470 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
5471 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
5472 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
5473 (vfmlsl_low_u32): Ditto.
5474 (vfmlalq_low_u32): Ditto.
5475 (vfmlslq_low_u32): Ditto.
5476 (vfmlal_high_u32): Ditto.
5477 (vfmlsl_high_u32): Ditto.
5478 (vfmlalq_high_u32): Ditto.
5479 (vfmlslq_high_u32): Ditto.
5480 (vfmlal_lane_low_u32): Ditto.
5481 (vfmlsl_lane_low_u32): Ditto.
5482 (vfmlal_laneq_low_u32): Ditto.
5483 (vfmlsl_laneq_low_u32): Ditto.
5484 (vfmlalq_lane_low_u32): Ditto.
5485 (vfmlslq_lane_low_u32): Ditto.
5486 (vfmlalq_laneq_low_u32): Ditto.
5487 (vfmlslq_laneq_low_u32): Ditto.
5488 (vfmlal_lane_high_u32): Ditto.
5489 (vfmlsl_lane_high_u32): Ditto.
5490 (vfmlal_laneq_high_u32): Ditto.
5491 (vfmlsl_laneq_high_u32): Ditto.
5492 (vfmlalq_lane_high_u32): Ditto.
5493 (vfmlslq_lane_high_u32): Ditto.
5494 (vfmlalq_laneq_high_u32): Ditto.
5495 (vfmlslq_laneq_high_u32): Ditto.
5496 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
5497 (AARCH64_FL_FOR_ARCH8_4): New.
5498 (AARCH64_ISA_F16FML): New ISA flag.
5499 (TARGET_F16FML): New feature flag for fp16fml.
5500 (doc/invoke.texi): Document new fp16fml option.
5502 2018-01-10 Michael Collison <michael.collison@arm.com>
5504 * config/aarch64/aarch64-builtins.c:
5505 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
5506 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5507 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
5508 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
5509 (AARCH64_ISA_SHA3): New ISA flag.
5510 (TARGET_SHA3): New feature flag for sha3.
5511 * config/aarch64/iterators.md (sha512_op): New int attribute.
5512 (CRYPTO_SHA512): New int iterator.
5513 (UNSPEC_SHA512H): New unspec.
5514 (UNSPEC_SHA512H2): Ditto.
5515 (UNSPEC_SHA512SU0): Ditto.
5516 (UNSPEC_SHA512SU1): Ditto.
5517 * config/aarch64/aarch64-simd-builtins.def
5518 (aarch64_crypto_sha512hqv2di): New builtin.
5519 (aarch64_crypto_sha512h2qv2di): Ditto.
5520 (aarch64_crypto_sha512su0qv2di): Ditto.
5521 (aarch64_crypto_sha512su1qv2di): Ditto.
5522 (aarch64_eor3qv8hi): Ditto.
5523 (aarch64_rax1qv2di): Ditto.
5524 (aarch64_xarqv2di): Ditto.
5525 (aarch64_bcaxqv8hi): Ditto.
5526 * config/aarch64/aarch64-simd.md:
5527 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
5528 (aarch64_crypto_sha512su0qv2di): Ditto.
5529 (aarch64_crypto_sha512su1qv2di): Ditto.
5530 (aarch64_eor3qv8hi): Ditto.
5531 (aarch64_rax1qv2di): Ditto.
5532 (aarch64_xarqv2di): Ditto.
5533 (aarch64_bcaxqv8hi): Ditto.
5534 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
5535 (vsha512h2q_u64): Ditto.
5536 (vsha512su0q_u64): Ditto.
5537 (vsha512su1q_u64): Ditto.
5538 (veor3q_u16): Ditto.
5539 (vrax1q_u64): Ditto.
5541 (vbcaxq_u16): Ditto.
5542 * config/arm/types.md (crypto_sha512): New type attribute.
5543 (crypto_sha3): Ditto.
5544 (doc/invoke.texi): Document new sha3 option.
5546 2018-01-10 Michael Collison <michael.collison@arm.com>
5548 * config/aarch64/aarch64-builtins.c:
5549 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
5550 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5551 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
5552 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
5553 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
5554 (AARCH64_ISA_SM4): New ISA flag.
5555 (TARGET_SM4): New feature flag for sm4.
5556 * config/aarch64/aarch64-simd-builtins.def
5557 (aarch64_sm3ss1qv4si): Ditto.
5558 (aarch64_sm3tt1aq4si): Ditto.
5559 (aarch64_sm3tt1bq4si): Ditto.
5560 (aarch64_sm3tt2aq4si): Ditto.
5561 (aarch64_sm3tt2bq4si): Ditto.
5562 (aarch64_sm3partw1qv4si): Ditto.
5563 (aarch64_sm3partw2qv4si): Ditto.
5564 (aarch64_sm4eqv4si): Ditto.
5565 (aarch64_sm4ekeyqv4si): Ditto.
5566 * config/aarch64/aarch64-simd.md:
5567 (aarch64_sm3ss1qv4si): Ditto.
5568 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
5569 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
5570 (aarch64_sm4eqv4si): Ditto.
5571 (aarch64_sm4ekeyqv4si): Ditto.
5572 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
5573 (sm3part_op): Ditto.
5574 (CRYPTO_SM3TT): Ditto.
5575 (CRYPTO_SM3PART): Ditto.
5576 (UNSPEC_SM3SS1): New unspec.
5577 (UNSPEC_SM3TT1A): Ditto.
5578 (UNSPEC_SM3TT1B): Ditto.
5579 (UNSPEC_SM3TT2A): Ditto.
5580 (UNSPEC_SM3TT2B): Ditto.
5581 (UNSPEC_SM3PARTW1): Ditto.
5582 (UNSPEC_SM3PARTW2): Ditto.
5583 (UNSPEC_SM4E): Ditto.
5584 (UNSPEC_SM4EKEY): Ditto.
5585 * config/aarch64/constraints.md (Ui2): New constraint.
5586 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
5587 * config/arm/types.md (crypto_sm3): New type attribute.
5588 (crypto_sm4): Ditto.
5589 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
5590 (vsm3tt1aq_u32): Ditto.
5591 (vsm3tt1bq_u32): Ditto.
5592 (vsm3tt2aq_u32): Ditto.
5593 (vsm3tt2bq_u32): Ditto.
5594 (vsm3partw1q_u32): Ditto.
5595 (vsm3partw2q_u32): Ditto.
5596 (vsm4eq_u32): Ditto.
5597 (vsm4ekeyq_u32): Ditto.
5598 (doc/invoke.texi): Document new sm4 option.
5600 2018-01-10 Michael Collison <michael.collison@arm.com>
5602 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
5603 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
5604 (AARCH64_FL_FOR_ARCH8_4): New.
5605 (AARCH64_FL_V8_4): New flag.
5606 (doc/invoke.texi): Document new armv8.4-a option.
5608 2018-01-10 Michael Collison <michael.collison@arm.com>
5610 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5611 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
5612 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
5613 * config/aarch64/aarch64-option-extension.def: Add
5614 AARCH64_OPT_EXTENSION of 'sha2'.
5615 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
5616 (crypto): Disable sha2 and aes if crypto disabled.
5617 (crypto): Enable aes and sha2 if enabled.
5618 (simd): Disable sha2 and aes if simd disabled.
5619 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
5621 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
5622 (TARGET_SHA2): New feature flag for sha2.
5623 (TARGET_AES): New feature flag for aes.
5624 * config/aarch64/aarch64-simd.md:
5625 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
5626 conditional on TARGET_AES.
5627 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
5628 (aarch64_crypto_sha1hsi): Make pattern conditional
5630 (aarch64_crypto_sha1hv4si): Ditto.
5631 (aarch64_be_crypto_sha1hv4si): Ditto.
5632 (aarch64_crypto_sha1su1v4si): Ditto.
5633 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
5634 (aarch64_crypto_sha1su0v4si): Ditto.
5635 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
5636 (aarch64_crypto_sha256su0v4si): Ditto.
5637 (aarch64_crypto_sha256su1v4si): Ditto.
5638 (doc/invoke.texi): Document new aes and sha2 options.
5640 2018-01-10 Martin Sebor <msebor@redhat.com>
5642 PR tree-optimization/83781
5643 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
5646 2018-01-11 Martin Sebor <msebor@gmail.com>
5647 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5649 PR tree-optimization/83501
5650 PR tree-optimization/81703
5652 * tree-ssa-strlen.c (get_string_cst): Rename...
5653 (get_string_len): ...to this. Handle global constants.
5654 (handle_char_store): Adjust.
5656 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
5657 Jim Wilson <jimw@sifive.com>
5659 * config/riscv/riscv-protos.h (riscv_output_return): New.
5660 * config/riscv/riscv.c (struct machine_function): New naked_p field.
5661 (riscv_attribute_table, riscv_output_return),
5662 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
5663 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
5664 (riscv_compute_frame_info): Only compute frame->mask if not a naked
5666 (riscv_expand_prologue): Add early return for naked function.
5667 (riscv_expand_epilogue): Likewise.
5668 (riscv_function_ok_for_sibcall): Return false for naked function.
5669 (riscv_set_current_function): New.
5670 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
5671 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
5672 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
5673 * doc/extend.texi (RISC-V Function Attributes): New.
5675 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
5677 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
5678 check for 128-bit long double before checking TCmode.
5679 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
5680 128-bit long doubles before checking TFmode or TCmode.
5681 (FLOAT128_IBM_P): Likewise.
5683 2018-01-10 Martin Sebor <msebor@redhat.com>
5685 PR tree-optimization/83671
5686 * builtins.c (c_strlen): Unconditionally return zero for the empty
5688 Use -Warray-bounds for warnings.
5689 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
5690 for non-constant array indices with COMPONENT_REF, arrays of
5691 arrays, and pointers to arrays.
5692 (gimple_fold_builtin_strlen): Determine and set length range for
5693 non-constant character arrays.
5695 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
5698 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
5701 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
5703 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
5705 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5708 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
5709 VECTOR_MEM_ALTIVEC_OR_VSX_P.
5710 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
5711 indexed_or_indirect_operand predicate.
5712 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
5713 (*vsx_le_perm_load_v8hi): Likewise.
5714 (*vsx_le_perm_load_v16qi): Likewise.
5715 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
5716 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
5717 (*vsx_le_perm_store_v8hi): Likewise.
5718 (*vsx_le_perm_store_v16qi): Likewise.
5719 (eight unnamed splitters): Likewise.
5721 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5723 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
5724 * config/rs6000/emmintrin.h: Likewise.
5725 * config/rs6000/mmintrin.h: Likewise.
5726 * config/rs6000/xmmintrin.h: Likewise.
5728 2018-01-10 David Malcolm <dmalcolm@redhat.com>
5731 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
5733 * tree.c (tree_nop_conversion): Return true for location wrapper
5735 (maybe_wrap_with_location): New function.
5736 (selftest::check_strip_nops): New function.
5737 (selftest::test_location_wrappers): New function.
5738 (selftest::tree_c_tests): Call it.
5739 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
5740 (maybe_wrap_with_location): New decl.
5741 (EXPR_LOCATION_WRAPPER_P): New macro.
5742 (location_wrapper_p): New inline function.
5743 (tree_strip_any_location_wrapper): New inline function.
5745 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5748 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5749 stack_realign_offset for the largest alignment of stack slot
5751 (ix86_find_max_used_stack_alignment): New function.
5752 (ix86_finalize_stack_frame_flags): Use it. Set
5753 max_used_stack_alignment if we don't realign stack.
5754 * config/i386/i386.h (machine_function): Add
5755 max_used_stack_alignment.
5757 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5759 * config/arm/arm.opt (-mbranch-cost): New option.
5760 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5763 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5766 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5767 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5769 2018-01-10 Richard Biener <rguenther@suse.de>
5772 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5773 early out so it also covers the case where we have a non-NULL
5776 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5778 PR tree-optimization/83753
5779 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5780 for non-strided grouped accesses if the number of elements is 1.
5782 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5785 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5786 * i386.h (TARGET_USE_GATHER): Define.
5787 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5789 2018-01-10 Martin Liska <mliska@suse.cz>
5792 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5793 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5795 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5796 CLEANUP_NO_PARTITIONING is not set.
5798 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5800 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5801 for vectors, as a partial revert of r254296.
5802 * rtl.h (const_vec_p): Delete.
5803 (const_vec_duplicate_p): Don't test for vector CONSTs.
5804 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5805 * expmed.c (make_tree): Likewise.
5808 * common.md (E, F): Use CONSTANT_P instead of checking for
5810 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5811 checking for CONST_VECTOR.
5813 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5816 * predict.c (force_edge_cold): Handle in more sane way edges
5819 2018-01-09 Carl Love <cel@us.ibm.com>
5821 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5823 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5824 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5825 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5826 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5827 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5828 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5829 * config/rs6000/rs6000-protos.h: Add extern defition for
5830 rs6000_generate_float2_double_code.
5831 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5833 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5834 (float2_v2df): Add define_expand.
5836 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5839 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5840 op_mode in the force_to_mode call.
5842 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5844 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5845 instead of checking each element individually.
5846 (aarch64_evpc_uzp): Likewise.
5847 (aarch64_evpc_zip): Likewise.
5848 (aarch64_evpc_ext): Likewise.
5849 (aarch64_evpc_rev): Likewise.
5850 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5851 instead of checking each element individually. Return true without
5853 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5854 whether all selected elements come from the same input, instead of
5855 checking each element individually. Remove calls to gen_rtx_REG,
5856 start_sequence and end_sequence and instead assert that no rtl is
5859 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5861 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5862 order of HIGH and CONST checks.
5864 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5866 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5867 if the destination isn't an SSA_NAME.
5869 2018-01-09 Richard Biener <rguenther@suse.de>
5871 PR tree-optimization/83668
5872 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5874 (canonicalize_loop_form): ... here, renamed from ...
5875 (canonicalize_loop_closed_ssa_form): ... this and amended to
5876 swap successor edges for loop exit blocks to make us use
5877 the RPO order we need for initial schedule generation.
5879 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5881 PR tree-optimization/64811
5882 * match.pd: When optimizing comparisons with Inf, avoid
5883 introducing or losing exceptions from comparisons with NaN.
5885 2018-01-09 Martin Liska <mliska@suse.cz>
5888 * asan.c (shadow_mem_size): Add gcc_assert.
5890 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5892 Don't save registers in main().
5895 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5896 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5897 * config/avr/avr.c (avr_set_current_function): Don't error if
5898 naked, OS_task or OS_main are specified at the same time.
5899 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5901 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5903 * common/config/avr/avr-common.c (avr_option_optimization_table):
5904 Switch on -mmain-is-OS_task for optimizing compilations.
5906 2018-01-09 Richard Biener <rguenther@suse.de>
5908 PR tree-optimization/83572
5909 * graphite.c: Include cfganal.h.
5910 (graphite_transform_loops): Connect infinite loops to exit
5911 and remove fake edges at the end.
5913 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5915 * ipa-inline.c (edge_badness): Revert accidental checkin.
5917 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5920 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5921 symbols; not inline clones.
5923 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5926 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5927 hard registers. Formatting fixes.
5929 PR preprocessor/83722
5930 * gcc.c (try_generate_repro): Pass
5931 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5932 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5935 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5936 Kito Cheng <kito.cheng@gmail.com>
5938 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5939 (riscv_leaf_function_p): Delete.
5940 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5942 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5944 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5946 (do_ifelse): New function.
5947 (do_isel): New function.
5948 (do_sub3): New function.
5949 (do_add3): New function.
5950 (do_load_mask_compare): New function.
5951 (do_overlap_load_compare): New function.
5952 (expand_compare_loop): New function.
5953 (expand_block_compare): Call expand_compare_loop() when appropriate.
5954 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5956 (-mblock-compare-inline-loop-limit): New option.
5958 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5961 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5962 Reverse order of second and third operands in first alternative.
5963 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5964 of first and second elements in UNSPEC_VPERMR vector.
5965 (altivec_expand_vec_perm_le): Likewise.
5967 2018-01-08 Jeff Law <law@redhat.com>
5969 PR rtl-optimizatin/81308
5970 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5971 (process_switch): If group_case_labels makes a change, then set
5973 (pass_convert_switch::execute): If a switch is converted, then
5974 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5976 PR rtl-optimization/81308
5977 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5980 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5982 PR target/83663 - Revert r255946
5983 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5984 generation for cases where splatting a value is not useful.
5985 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5986 across a vec_duplicate and a paradoxical subreg forming a vector
5987 mode to a vec_concat.
5989 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5991 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5992 -march=armv8.3-a variants.
5993 * config/arm/t-multilib: Likewise.
5994 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5996 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5998 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
6000 (cceq_ior_compare_complement): Give it a name so I can use it, and
6001 change boolean_or_operator predicate to boolean_operator so it can
6002 be used to generate a crand.
6003 (eqne): New code iterator.
6004 (bd/bd_neg): New code_attrs.
6005 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
6006 a single define_insn.
6007 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
6008 decrement (bdnzt/bdnzf/bdzt/bdzf).
6009 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
6010 with the new names of the branch decrement patterns, and added the
6011 names of the branch decrement conditional patterns.
6013 2018-01-08 Richard Biener <rguenther@suse.de>
6015 PR tree-optimization/83563
6016 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
6019 2018-01-08 Richard Biener <rguenther@suse.de>
6022 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
6024 2018-01-08 Richard Biener <rguenther@suse.de>
6026 PR tree-optimization/83685
6027 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
6028 references to abnormals.
6030 2018-01-08 Richard Biener <rguenther@suse.de>
6033 * dwarf2out.c (output_indirect_strings): Handle empty
6034 skeleton_debug_str_hash.
6035 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
6037 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
6039 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
6040 (emit_store_direct): Likewise.
6041 (arc_trampoline_adjust_address): Likewise.
6042 (arc_asm_trampoline_template): New function.
6043 (arc_initialize_trampoline): Use asm_trampoline_template.
6044 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
6045 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
6046 * config/arc/arc.md (flush_icache): Delete pattern.
6048 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
6050 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
6051 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
6054 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
6057 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
6058 by not USED_FOR_TARGET.
6059 (make_pass_resolve_sw_modes): Likewise.
6061 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
6063 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
6066 2018-01-08 Richard Biener <rguenther@suse.de>
6069 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
6071 2018-01-08 Richard Biener <rguenther@suse.de>
6074 * match.pd ((t * 2) / 2) -> t): Add missing :c.
6076 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
6079 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
6080 basic blocks with a small number of successors.
6081 (convert_control_dep_chain_into_preds): Improve handling of
6083 (dump_predicates): Split apart into...
6084 (dump_pred_chain): ...here...
6085 (dump_pred_info): ...and here.
6086 (can_one_predicate_be_invalidated_p): Add debugging printfs.
6087 (can_chain_union_be_invalidated_p): Improve check for invalidation
6089 (uninit_uses_cannot_happen): Avoid unnecessary if
6090 convert_control_dep_chain_into_preds yielded nothing.
6092 2018-01-06 Martin Sebor <msebor@redhat.com>
6094 PR tree-optimization/83640
6095 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
6096 subtracting negative offset from size.
6097 (builtin_access::overlap): Adjust offset bounds of the access to fall
6098 within the size of the object if possible.
6100 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
6102 PR rtl-optimization/83699
6103 * expmed.c (extract_bit_field_1): Restrict the vector usage of
6104 extract_bit_field_as_subreg to cases in which the extracted
6105 value is also a vector.
6107 * lra-constraints.c (process_alt_operands): Test for the equivalence
6108 substitutions when detecting a possible reload cycle.
6110 2018-01-06 Jakub Jelinek <jakub@redhat.com>
6113 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
6114 by default if flag_selective_schedling{,2}. Formatting fixes.
6116 PR rtl-optimization/83682
6117 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
6118 if it has non-VECTOR_MODE element mode.
6119 (vec_duplicate_p): Likewise.
6122 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
6123 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
6125 2018-01-05 Jakub Jelinek <jakub@redhat.com>
6128 * config/i386/i386-builtin.def
6129 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
6130 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
6131 Require also OPTION_MASK_ISA_AVX512F in addition to
6132 OPTION_MASK_ISA_GFNI.
6133 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
6134 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
6135 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
6136 to OPTION_MASK_ISA_GFNI.
6137 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
6138 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
6139 OPTION_MASK_ISA_AVX512BW.
6140 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
6141 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
6142 addition to OPTION_MASK_ISA_GFNI.
6143 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
6144 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
6145 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
6146 to OPTION_MASK_ISA_GFNI.
6147 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
6148 a requirement for all ISAs rather than any of them with a few
6150 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
6152 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
6153 bitmasks to be enabled with 3 exceptions, instead of requiring any
6154 enabled ISA with lots of exceptions.
6155 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
6156 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
6157 Change avx512bw in isa attribute to avx512f.
6158 * config/i386/sgxintrin.h: Add license boilerplate.
6159 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
6160 to __AVX512F__ and __AVX512VL to __AVX512VL__.
6161 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
6162 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
6164 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
6165 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
6166 temporarily sse2 rather than sse if not enabled already.
6169 * config/i386/sse.md (VI248_VLBW): Rename to ...
6170 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
6171 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
6172 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
6173 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
6174 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
6175 mode iterator instead of VI248_VLBW.
6177 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
6179 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
6180 (record_modified): Skip clobbers; add debug output.
6181 (param_change_prob): Use sreal frequencies.
6183 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
6185 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
6186 punt for user-aligned variables.
6188 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
6190 * tree-chrec.c (chrec_contains_symbols): Return true for
6193 2018-01-05 Sudakshina Das <sudi.das@arm.com>
6196 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
6197 of (x|y) == x for BICS pattern.
6199 2018-01-05 Jakub Jelinek <jakub@redhat.com>
6201 PR tree-optimization/83605
6202 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
6203 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
6206 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
6208 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
6209 * config/epiphany/rtems.h: New file.
6211 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6212 Uros Bizjak <ubizjak@gmail.com>
6215 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
6216 QIreg_operand instead of register_operand predicate.
6217 * config/i386/i386.c (ix86_rop_should_change_byte_p,
6218 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
6219 comments instead of -fmitigate[-_]rop.
6221 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
6224 * cgraphunit.c (symbol_table::compile): Switch to text_section
6225 before calling assembly_start debug hook.
6226 * run-rtl-passes.c (run_rtl_passes): Likewise.
6229 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6231 * tree-vrp.c (extract_range_from_binary_expr_1): Check
6232 range_int_cst_p rather than !symbolic_range_p before calling
6233 extract_range_from_multiplicative_op_1.
6235 2018-01-04 Jeff Law <law@redhat.com>
6237 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
6238 redundant test in assertion.
6240 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6242 * doc/rtl.texi: Document machine_mode wrapper classes.
6244 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6246 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
6249 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6251 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
6252 the VEC_PERM_EXPR fold to fail.
6254 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6257 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
6258 to switched_sections.
6260 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6263 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
6266 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
6269 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
6270 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
6272 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6275 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
6276 is BLKmode and bitpos not zero or mode change is needed.
6278 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6281 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
6284 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
6287 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
6288 instead of MULT rtx. Update all corresponding splitters.
6290 (*ssub<modesuffix>): Ditto.
6292 (*cmp_sadd_di): Update split patterns.
6293 (*cmp_sadd_si): Ditto.
6294 (*cmp_sadd_sidi): Ditto.
6295 (*cmp_ssub_di): Ditto.
6296 (*cmp_ssub_si): Ditto.
6297 (*cmp_ssub_sidi): Ditto.
6298 * config/alpha/predicates.md (const23_operand): New predicate.
6299 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
6300 Look for ASHIFT, not MULT inner operand.
6301 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
6303 2018-01-04 Martin Liska <mliska@suse.cz>
6305 PR gcov-profile/83669
6306 * gcov.c (output_intermediate_file): Add version to intermediate
6308 * doc/gcov.texi: Document new field 'version' in intermediate
6309 file format. Fix location of '-k' option of gcov command.
6311 2018-01-04 Martin Liska <mliska@suse.cz>
6314 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
6316 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6318 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
6320 2018-01-03 Martin Sebor <msebor@redhat.com>
6322 PR tree-optimization/83655
6323 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
6324 checking calls with invalid arguments.
6326 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6328 * tree-vect-stmts.c (vect_get_store_rhs): New function.
6329 (vectorizable_mask_load_store): Delete.
6330 (vectorizable_call): Return false for masked loads and stores.
6331 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
6332 instead of gimple_assign_rhs1.
6333 (vectorizable_load): Handle IFN_MASK_LOAD.
6334 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
6336 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6338 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
6340 (vectorizable_mask_load_store): ...here.
6341 (vectorizable_load): ...and here.
6343 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6345 * tree-vect-stmts.c (vect_build_all_ones_mask)
6346 (vect_build_zero_merge_argument): New functions, split out from...
6347 (vectorizable_load): ...here.
6349 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6351 * tree-vect-stmts.c (vect_check_store_rhs): New function,
6353 (vectorizable_mask_load_store): ...here.
6354 (vectorizable_store): ...and here.
6356 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6358 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
6360 (vectorizable_mask_load_store): ...here.
6362 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6364 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
6365 (vect_model_store_cost): Take a vec_load_store_type instead of a
6367 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
6368 (vect_model_store_cost): Take a vec_load_store_type instead of a
6370 (vectorizable_mask_load_store): Update accordingly.
6371 (vectorizable_store): Likewise.
6372 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
6374 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6376 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
6377 IFN_MASK_LOAD calls here rather than...
6378 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
6380 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6381 Alan Hayward <alan.hayward@arm.com>
6382 David Sherwood <david.sherwood@arm.com>
6384 * expmed.c (extract_bit_field_1): For vector extracts,
6385 fall back to extract_bit_field_as_subreg if vec_extract
6388 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6389 Alan Hayward <alan.hayward@arm.com>
6390 David Sherwood <david.sherwood@arm.com>
6392 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
6393 they are variable or constant sized.
6394 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
6395 slots for constant-sized data.
6397 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6398 Alan Hayward <alan.hayward@arm.com>
6399 David Sherwood <david.sherwood@arm.com>
6401 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
6402 handling COND_EXPRs with boolean comparisons, try to find a better
6403 basis for the mask type than the boolean itself.
6405 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6407 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
6408 is calculated and how it can be overridden.
6409 * genmodes.c (max_bitsize_mode_any_mode): New variable.
6410 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
6412 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
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 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
6420 Remove the mode argument.
6421 (aarch64_simd_valid_immediate): Remove the mode and inverse
6423 * config/aarch64/iterators.md (bitsize): New iterator.
6424 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
6425 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
6426 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
6427 aarch64_simd_valid_immediate.
6428 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
6429 (aarch64_reg_or_bic_imm): Likewise.
6430 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
6431 with an insn_type enum and msl with a modifier_type enum.
6432 Replace element_width with a scalar_mode. Change the shift
6433 to unsigned int. Add constructors for scalar_float_mode and
6434 scalar_int_mode elements.
6435 (aarch64_vect_float_const_representable_p): Delete.
6436 (aarch64_can_const_movi_rtx_p)
6437 (aarch64_simd_scalar_immediate_valid_for_move)
6438 (aarch64_simd_make_constant): Update call to
6439 aarch64_simd_valid_immediate.
6440 (aarch64_advsimd_valid_immediate_hs): New function.
6441 (aarch64_advsimd_valid_immediate): Likewise.
6442 (aarch64_simd_valid_immediate): Remove mode and inverse
6443 arguments. Rewrite to use the above. Use const_vec_duplicate_p
6444 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
6445 and aarch64_float_const_representable_p on the result.
6446 (aarch64_output_simd_mov_immediate): Remove mode argument.
6447 Update call to aarch64_simd_valid_immediate and use of
6448 simd_immediate_info.
6449 (aarch64_output_scalar_simd_mov_immediate): Update call
6452 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6453 Alan Hayward <alan.hayward@arm.com>
6454 David Sherwood <david.sherwood@arm.com>
6456 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
6457 (mode_nunits): Likewise CONST_MODE_NUNITS.
6458 * machmode.def (ADJUST_NUNITS): Document.
6459 * genmodes.c (mode_data::need_nunits_adj): New field.
6460 (blank_mode): Update accordingly.
6461 (adj_nunits): New variable.
6462 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
6464 (emit_mode_size_inline): Set need_bytesize_adj for all modes
6465 listed in adj_nunits.
6466 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
6467 listed in adj_nunits. Don't emit case statements for such modes.
6468 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
6469 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
6470 nothing if adj_nunits is nonnull.
6471 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
6472 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
6473 (emit_mode_fbit): Update use of print_maybe_const_decl.
6474 (emit_move_size): Likewise. Treat the array as non-const
6476 (emit_mode_adjustments): Handle adj_nunits.
6478 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6480 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
6481 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
6482 (VECTOR_MODES): Use it.
6483 (make_vector_modes): Take the prefix as an argument.
6485 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6486 Alan Hayward <alan.hayward@arm.com>
6487 David Sherwood <david.sherwood@arm.com>
6489 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
6490 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
6491 for MODE_VECTOR_BOOL.
6492 * machmode.def (VECTOR_BOOL_MODE): Document.
6493 * genmodes.c (VECTOR_BOOL_MODE): New macro.
6494 (make_vector_bool_mode): New function.
6495 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
6497 * lto-streamer-in.c (lto_input_mode_table): Likewise.
6498 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
6500 * stor-layout.c (int_mode_for_mode): Likewise.
6501 * tree.c (build_vector_type_for_mode): Likewise.
6502 * varasm.c (output_constant_pool_2): Likewise.
6503 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
6504 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
6505 for MODE_VECTOR_BOOL.
6506 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
6507 of mode class checks.
6508 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
6509 instead of a list of mode class checks.
6510 (expand_vector_scalar_condition): Likewise.
6511 (type_for_widest_vector_mode): Handle BImode as an inner mode.
6513 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6514 Alan Hayward <alan.hayward@arm.com>
6515 David Sherwood <david.sherwood@arm.com>
6517 * machmode.h (mode_size): Change from unsigned short to
6519 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
6520 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6521 or if measurement_type is not polynomial.
6522 (fixed_size_mode::includes_p): Check for constant-sized modes.
6523 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
6524 return a poly_uint16 rather than an unsigned short.
6525 (emit_mode_size): Change the type of mode_size from unsigned short
6526 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
6527 (emit_mode_adjustments): Cope with polynomial vector sizes.
6528 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6530 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6532 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
6533 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
6534 * caller-save.c (setup_save_areas): Likewise.
6535 (replace_reg_with_saved_mem): Likewise.
6536 * calls.c (emit_library_call_value_1): Likewise.
6537 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
6538 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
6539 (gen_lowpart_for_combine): Likewise.
6540 * convert.c (convert_to_integer_1): Likewise.
6541 * cse.c (equiv_constant, cse_insn): Likewise.
6542 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
6543 (cselib_subst_to_values): Likewise.
6544 * dce.c (word_dce_process_block): Likewise.
6545 * df-problems.c (df_word_lr_mark_ref): Likewise.
6546 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
6547 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
6548 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
6549 (rtl_for_decl_location): Likewise.
6550 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
6551 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
6552 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
6553 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
6554 (expand_expr_real_1): Likewise.
6555 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
6556 (pad_below): Likewise.
6557 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6558 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
6559 * ira.c (get_subreg_tracking_sizes): Likewise.
6560 * ira-build.c (ira_create_allocno_objects): Likewise.
6561 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
6562 (ira_sort_regnos_for_alter_reg): Likewise.
6563 * ira-costs.c (record_operand_costs): Likewise.
6564 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
6565 (resolve_simple_move): Likewise.
6566 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
6567 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
6568 (lra_constraints): Likewise.
6569 (CONST_POOL_OK_P): Reject variable-sized modes.
6570 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
6571 (add_pseudo_to_slot, lra_spill): Likewise.
6572 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6573 * optabs-query.c (get_best_extraction_insn): Likewise.
6574 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6575 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
6576 (expand_mult_highpart, valid_multiword_target_p): Likewise.
6577 * recog.c (offsettable_address_addr_space_p): Likewise.
6578 * regcprop.c (maybe_mode_change): Likewise.
6579 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
6580 * regrename.c (build_def_use): Likewise.
6581 * regstat.c (dump_reg_info): Likewise.
6582 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
6583 (find_reloads, find_reloads_subreg_address): Likewise.
6584 * reload1.c (eliminate_regs_1): Likewise.
6585 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
6586 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
6587 (simplify_binary_operation_1, simplify_subreg): Likewise.
6588 * targhooks.c (default_function_arg_padding): Likewise.
6589 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
6590 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
6591 (verify_gimple_assign_ternary): Likewise.
6592 * tree-inline.c (estimate_move_cost): Likewise.
6593 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6594 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
6595 (get_address_cost_ainc): Likewise.
6596 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
6597 (vect_supportable_dr_alignment): Likewise.
6598 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6599 (vectorizable_reduction): Likewise.
6600 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
6601 (vectorizable_operation, vectorizable_load): Likewise.
6602 * tree.c (build_same_sized_truth_vector_type): Likewise.
6603 * valtrack.c (cleanup_auto_inc_dec): Likewise.
6604 * var-tracking.c (emit_note_insn_var_location): Likewise.
6605 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
6606 (ADDR_VEC_ALIGN): Likewise.
6608 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6609 Alan Hayward <alan.hayward@arm.com>
6610 David Sherwood <david.sherwood@arm.com>
6612 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
6614 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6615 or if measurement_type is polynomial.
6616 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
6617 * combine.c (make_extraction): Likewise.
6618 * dse.c (find_shift_sequence): Likewise.
6619 * dwarf2out.c (mem_loc_descriptor): Likewise.
6620 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
6621 (extract_bit_field, extract_low_bits): Likewise.
6622 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
6623 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
6624 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
6625 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
6626 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6627 * reload.c (find_reloads): Likewise.
6628 * reload1.c (alter_reg): Likewise.
6629 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
6630 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
6631 * tree-if-conv.c (predicate_mem_writes): Likewise.
6632 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
6633 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
6634 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
6635 * valtrack.c (dead_debug_insert_temp): Likewise.
6636 * varasm.c (mergeable_constant_section): Likewise.
6637 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
6639 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6640 Alan Hayward <alan.hayward@arm.com>
6641 David Sherwood <david.sherwood@arm.com>
6643 * expr.c (expand_assignment): Cope with polynomial mode sizes
6644 when assigning to a CONCAT.
6646 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6647 Alan Hayward <alan.hayward@arm.com>
6648 David Sherwood <david.sherwood@arm.com>
6650 * machmode.h (mode_precision): Change from unsigned short to
6652 (mode_to_precision): Return a poly_uint16 rather than an unsigned
6654 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
6655 or if measurement_type is not polynomial.
6656 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
6657 in which the mode is already known to be a scalar_int_mode.
6658 * genmodes.c (emit_mode_precision): Change the type of mode_precision
6659 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
6661 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6662 for GET_MODE_PRECISION.
6663 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6664 for GET_MODE_PRECISION.
6665 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
6667 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
6668 (expand_field_assignment, make_extraction): Likewise.
6669 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
6670 (get_last_value): Likewise.
6671 * convert.c (convert_to_integer_1): Likewise.
6672 * cse.c (cse_insn): Likewise.
6673 * expr.c (expand_expr_real_1): Likewise.
6674 * lra-constraints.c (simplify_operand_subreg): Likewise.
6675 * optabs-query.c (can_atomic_load_p): Likewise.
6676 * optabs.c (expand_atomic_load): Likewise.
6677 (expand_atomic_store): Likewise.
6678 * ree.c (combine_reaching_defs): Likewise.
6679 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
6680 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
6681 * tree.h (type_has_mode_precision_p): Likewise.
6682 * ubsan.c (instrument_si_overflow): Likewise.
6684 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6685 Alan Hayward <alan.hayward@arm.com>
6686 David Sherwood <david.sherwood@arm.com>
6688 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
6689 polynomial numbers of units.
6690 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
6691 (valid_vector_subparts_p): New function.
6692 (build_vector_type): Remove temporary shim and take the number
6693 of units as a poly_uint64 rather than an int.
6694 (build_opaque_vector_type): Take the number of units as a
6695 poly_uint64 rather than an int.
6696 * tree.c (build_vector_from_ctor): Handle polynomial
6697 TYPE_VECTOR_SUBPARTS.
6698 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
6699 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
6700 (build_vector_from_val): If the number of units is variable,
6701 use build_vec_duplicate_cst for constant operands and
6702 VEC_DUPLICATE_EXPR otherwise.
6703 (make_vector_type): Remove temporary is_constant ().
6704 (build_vector_type, build_opaque_vector_type): Take the number of
6705 units as a poly_uint64 rather than an int.
6706 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
6708 * cfgexpand.c (expand_debug_expr): Likewise.
6709 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
6710 (store_constructor, expand_expr_real_1): Likewise.
6711 (const_scalar_mask_from_tree): Likewise.
6712 * fold-const-call.c (fold_const_reduction): Likewise.
6713 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
6714 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
6715 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
6716 (fold_relational_const): Likewise.
6717 (native_interpret_vector): Likewise. Change the size from an
6718 int to an unsigned int.
6719 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
6720 TYPE_VECTOR_SUBPARTS.
6721 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
6722 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
6723 duplicating a non-constant operand into a variable-length vector.
6724 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
6725 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
6726 * ipa-icf.c (sem_variable::equals): Likewise.
6727 * match.pd: Likewise.
6728 * omp-simd-clone.c (simd_clone_subparts): Likewise.
6729 * print-tree.c (print_node): Likewise.
6730 * stor-layout.c (layout_type): Likewise.
6731 * targhooks.c (default_builtin_vectorization_cost): Likewise.
6732 * tree-cfg.c (verify_gimple_comparison): Likewise.
6733 (verify_gimple_assign_binary): Likewise.
6734 (verify_gimple_assign_ternary): Likewise.
6735 (verify_gimple_assign_single): Likewise.
6736 * tree-pretty-print.c (dump_generic_node): Likewise.
6737 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6738 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
6739 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
6740 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
6741 (vect_shift_permute_load_chain): Likewise.
6742 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
6743 (expand_vector_condition, optimize_vector_constructor): Likewise.
6744 (lower_vec_perm, get_compute_type): Likewise.
6745 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6746 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6747 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6748 (vect_recog_mask_conversion_pattern): Likewise.
6749 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6750 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6751 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6752 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6753 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6754 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6755 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6756 (supportable_widening_operation): Likewise.
6757 (supportable_narrowing_operation): Likewise.
6758 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6760 * varasm.c (output_constant): Likewise.
6762 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6763 Alan Hayward <alan.hayward@arm.com>
6764 David Sherwood <david.sherwood@arm.com>
6766 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6767 so that both the length == 3 and length != 3 cases set up their
6768 own permute vectors. Add comments explaining why we know the
6769 number of elements is constant.
6770 (vect_permute_load_chain): Likewise.
6772 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6773 Alan Hayward <alan.hayward@arm.com>
6774 David Sherwood <david.sherwood@arm.com>
6776 * machmode.h (mode_nunits): Change from unsigned char to
6778 (ONLY_FIXED_SIZE_MODES): New macro.
6779 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6780 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6781 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6783 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6784 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6785 or if measurement_type is not polynomial.
6786 * genmodes.c (ZERO_COEFFS): New macro.
6787 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6789 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6790 Use ZERO_COEFFS when emitting initializers.
6791 * data-streamer.h (bp_pack_poly_value): New function.
6792 (bp_unpack_poly_value): Likewise.
6793 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6794 for GET_MODE_NUNITS.
6795 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6796 for GET_MODE_NUNITS.
6797 * tree.c (make_vector_type): Remove temporary shim and make
6798 the real function take the number of units as a poly_uint64
6800 (build_vector_type_for_mode): Handle polynomial nunits.
6801 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6802 * emit-rtl.c (const_vec_series_p_1): Likewise.
6803 (gen_rtx_CONST_VECTOR): Likewise.
6804 * fold-const.c (test_vec_duplicate_folding): Likewise.
6805 * genrecog.c (validate_pattern): Likewise.
6806 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6807 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6808 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6809 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6810 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6811 * rtlanal.c (subreg_get_info): Likewise.
6812 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6813 (vect_grouped_load_supported): Likewise.
6814 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6815 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6816 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6817 (simplify_const_unary_operation, simplify_binary_operation_1)
6818 (simplify_const_binary_operation, simplify_ternary_operation)
6819 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6820 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6821 instead of CONST_VECTOR_NUNITS.
6822 * varasm.c (output_constant_pool_2): Likewise.
6823 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6824 explicit-encoded elements in the XVEC for variable-length vectors.
6826 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6828 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6830 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6831 Alan Hayward <alan.hayward@arm.com>
6832 David Sherwood <david.sherwood@arm.com>
6834 * coretypes.h (fixed_size_mode): Declare.
6835 (fixed_size_mode_pod): New typedef.
6836 * builtins.h (target_builtins::x_apply_args_mode)
6837 (target_builtins::x_apply_result_mode): Change type to
6838 fixed_size_mode_pod.
6839 * builtins.c (apply_args_size, apply_result_size, result_vector)
6840 (expand_builtin_apply_args_1, expand_builtin_apply)
6841 (expand_builtin_return): Update accordingly.
6843 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6845 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6846 * cselib.c (cselib_hash_rtx): Likewise.
6847 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6848 CONST_VECTOR encoding.
6850 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6851 Jeff Law <law@redhat.com>
6854 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6855 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6856 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6857 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6860 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6861 explicitly probe *sp in a noreturn function if there were any callee
6862 register saves or frame pointer is needed.
6864 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6867 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6868 BLKmode for ternary, binary or unary expressions.
6871 * var-tracking.c (delete_vta_debug_insn): New inline function.
6872 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6873 insns from get_insns () to NULL instead of each bb separately.
6874 Use delete_vta_debug_insn. No longer static.
6875 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6876 delete_vta_debug_insns callers.
6877 * rtl.h (delete_vta_debug_insns): Declare.
6878 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6879 instead of variable_tracking_main.
6881 2018-01-03 Martin Sebor <msebor@redhat.com>
6883 PR tree-optimization/83603
6884 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6885 arguments past the endof the argument list in functions declared
6886 without a prototype.
6887 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6888 Avoid checking when arguments are null.
6890 2018-01-03 Martin Sebor <msebor@redhat.com>
6893 * doc/extend.texi (attribute const): Fix a typo.
6894 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6895 issuing -Wsuggest-attribute for void functions.
6897 2018-01-03 Martin Sebor <msebor@redhat.com>
6899 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6900 offset_int::from instead of wide_int::to_shwi.
6901 (maybe_diag_overlap): Remove assertion.
6902 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6903 * gimple-ssa-sprintf.c (format_directive): Same.
6904 (parse_directive): Same.
6905 (sprintf_dom_walker::compute_format_length): Same.
6906 (try_substitute_return_value): Same.
6908 2018-01-03 Jeff Law <law@redhat.com>
6911 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6912 non-constant residual for zero at runtime and avoid probing in
6913 that case. Reorganize code for trailing problem to mirror handling
6916 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6918 PR tree-optimization/83501
6919 * tree-ssa-strlen.c (get_string_cst): New.
6920 (handle_char_store): Call get_string_cst.
6922 2018-01-03 Martin Liska <mliska@suse.cz>
6924 PR tree-optimization/83593
6925 * tree-ssa-strlen.c: Include tree-cfg.h.
6926 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6927 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6928 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6930 (strlen_dom_walker::before_dom_children): Call
6931 gimple_purge_dead_eh_edges. Dump tranformation with details
6933 (strlen_dom_walker::before_dom_children): Update call by adding
6934 new argument cleanup_eh.
6935 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6937 2018-01-03 Martin Liska <mliska@suse.cz>
6940 * cif-code.def (VARIADIC_THUNK): New enum value.
6941 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6944 2018-01-03 Jan Beulich <jbeulich@suse.com>
6946 * sse.md (mov<mode>_internal): Tighten condition for when to use
6947 vmovdqu<ssescalarsize> for TI and OI modes.
6949 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6951 Update copyright years.
6953 2018-01-03 Martin Liska <mliska@suse.cz>
6956 * ipa-visibility.c (function_and_variable_visibility): Skip
6957 functions with noipa attribure.
6959 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6961 * gcc.c (process_command): Update copyright notice dates.
6962 * gcov-dump.c (print_version): Ditto.
6963 * gcov.c (print_version): Ditto.
6964 * gcov-tool.c (print_version): Ditto.
6965 * gengtype.c (create_file): Ditto.
6966 * doc/cpp.texi: Bump @copying's copyright year.
6967 * doc/cppinternals.texi: Ditto.
6968 * doc/gcc.texi: Ditto.
6969 * doc/gccint.texi: Ditto.
6970 * doc/gcov.texi: Ditto.
6971 * doc/install.texi: Ditto.
6972 * doc/invoke.texi: Ditto.
6974 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6976 * vector-builder.h (vector_builder::m_full_nelts): Change from
6977 unsigned int to poly_uint64.
6978 (vector_builder::full_nelts): Update prototype accordingly.
6979 (vector_builder::new_vector): Likewise.
6980 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6981 (vector_builder::operator ==): Likewise.
6982 (vector_builder::finalize): Likewise.
6983 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6984 Take the number of elements as a poly_uint64 rather than an
6986 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6987 from unsigned int to poly_uint64.
6988 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6989 (vec_perm_indices::new_vector): Likewise.
6990 (vec_perm_indices::length): Likewise.
6991 (vec_perm_indices::nelts_per_input): Likewise.
6992 (vec_perm_indices::input_nelts): Likewise.
6993 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6994 number of elements per input as a poly_uint64 rather than an
6995 unsigned int. Use the original encoding for variable-length
6996 vectors, rather than clamping each individual element.
6997 For the second and subsequent elements in each pattern,
6998 clamp the step and base before clamping their sum.
6999 (vec_perm_indices::series_p): Handle polynomial element counts.
7000 (vec_perm_indices::all_in_range_p): Likewise.
7001 (vec_perm_indices_to_tree): Likewise.
7002 (vec_perm_indices_to_rtx): Likewise.
7003 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
7004 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
7005 (tree_vector_builder::new_binary_operation): Handle polynomial
7006 element counts. Return false if we need to know the number
7007 of elements at compile time.
7008 * fold-const.c (fold_vec_perm): Punt if the number of elements
7009 isn't known at compile time.
7011 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7013 * vec-perm-indices.h (vec_perm_builder): Change element type
7014 from HOST_WIDE_INT to poly_int64.
7015 (vec_perm_indices::element_type): Update accordingly.
7016 (vec_perm_indices::clamp): Handle polynomial element_types.
7017 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7018 (vec_perm_indices::all_in_range_p): Likewise.
7019 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
7021 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
7022 polynomial vec_perm_indices element types.
7023 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
7024 * fold-const.c (fold_vec_perm): Likewise.
7025 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
7026 * tree-vect-generic.c (lower_vec_perm): Likewise.
7027 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7028 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
7029 element type to HOST_WIDE_INT.
7031 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7032 Alan Hayward <alan.hayward@arm.com>
7033 David Sherwood <david.sherwood@arm.com>
7035 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
7036 rather than an int. Use plus_constant.
7037 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
7038 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
7040 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7041 Alan Hayward <alan.hayward@arm.com>
7042 David Sherwood <david.sherwood@arm.com>
7044 * calls.c (emit_call_1, expand_call): Change struct_value_size from
7045 a HOST_WIDE_INT to a poly_int64.
7047 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7048 Alan Hayward <alan.hayward@arm.com>
7049 David Sherwood <david.sherwood@arm.com>
7051 * calls.c (load_register_parameters): Cope with polynomial
7052 mode sizes. Require a constant size for BLKmode parameters
7053 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
7054 forces a parameter to be padded at the lsb end in order to
7055 fill a complete number of words, require the parameter size
7056 to be ordered wrt UNITS_PER_WORD.
7058 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7059 Alan Hayward <alan.hayward@arm.com>
7060 David Sherwood <david.sherwood@arm.com>
7062 * reload1.c (spill_stack_slot_width): Change element type
7063 from unsigned int to poly_uint64_pod.
7064 (alter_reg): Treat mode sizes as polynomial.
7066 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7067 Alan Hayward <alan.hayward@arm.com>
7068 David Sherwood <david.sherwood@arm.com>
7070 * reload.c (complex_word_subreg_p): New function.
7071 (reload_inner_reg_of_subreg, push_reload): Use it.
7073 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7074 Alan Hayward <alan.hayward@arm.com>
7075 David Sherwood <david.sherwood@arm.com>
7077 * lra-constraints.c (process_alt_operands): Reject matched
7078 operands whose sizes aren't ordered.
7079 (match_reload): Refer to this check here.
7081 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7082 Alan Hayward <alan.hayward@arm.com>
7083 David Sherwood <david.sherwood@arm.com>
7085 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
7086 that the mode size is in the set {1, 2, 4, 8, 16}.
7088 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7089 Alan Hayward <alan.hayward@arm.com>
7090 David Sherwood <david.sherwood@arm.com>
7092 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
7093 Use plus_constant instead of gen_rtx_PLUS.
7095 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7096 Alan Hayward <alan.hayward@arm.com>
7097 David Sherwood <david.sherwood@arm.com>
7099 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
7100 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
7101 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
7102 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
7103 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
7104 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
7105 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
7106 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
7107 * config/i386/i386.c (ix86_push_rounding): ...this new function.
7108 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
7110 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
7111 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
7112 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
7113 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
7114 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
7115 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
7116 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
7117 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
7118 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
7119 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
7121 * expr.c (emit_move_resolve_push): Treat the input and result
7122 of PUSH_ROUNDING as a poly_int64.
7123 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
7124 (emit_push_insn): Likewise.
7125 * lra-eliminations.c (mark_not_eliminable): Likewise.
7126 * recog.c (push_operand): Likewise.
7127 * reload1.c (elimination_effects): Likewise.
7128 * rtlanal.c (nonzero_bits1): Likewise.
7129 * calls.c (store_one_arg): Likewise. Require the padding to be
7130 known at compile time.
7132 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7133 Alan Hayward <alan.hayward@arm.com>
7134 David Sherwood <david.sherwood@arm.com>
7136 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
7137 Use plus_constant instead of gen_rtx_PLUS.
7139 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7140 Alan Hayward <alan.hayward@arm.com>
7141 David Sherwood <david.sherwood@arm.com>
7143 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
7146 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7147 Alan Hayward <alan.hayward@arm.com>
7148 David Sherwood <david.sherwood@arm.com>
7150 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
7151 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
7152 via stack temporaries. Treat the mode size as polynomial too.
7154 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7155 Alan Hayward <alan.hayward@arm.com>
7156 David Sherwood <david.sherwood@arm.com>
7158 * expr.c (expand_expr_real_2): When handling conversions involving
7159 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
7160 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
7161 as a poly_uint64 too.
7163 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7164 Alan Hayward <alan.hayward@arm.com>
7165 David Sherwood <david.sherwood@arm.com>
7167 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
7169 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7170 Alan Hayward <alan.hayward@arm.com>
7171 David Sherwood <david.sherwood@arm.com>
7173 * combine.c (can_change_dest_mode): Handle polynomial
7174 REGMODE_NATURAL_SIZE.
7175 * expmed.c (store_bit_field_1): Likewise.
7176 * expr.c (store_constructor): Likewise.
7177 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
7178 and polynomial REGMODE_NATURAL_SIZE.
7179 (gen_lowpart_common): Likewise.
7180 * reginfo.c (record_subregs_of_mode): Likewise.
7181 * rtlanal.c (read_modify_subreg_p): Likewise.
7183 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7184 Alan Hayward <alan.hayward@arm.com>
7185 David Sherwood <david.sherwood@arm.com>
7187 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
7188 numbers of elements.
7190 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7191 Alan Hayward <alan.hayward@arm.com>
7192 David Sherwood <david.sherwood@arm.com>
7194 * match.pd: Cope with polynomial numbers of vector elements.
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 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
7201 in a POINTER_PLUS_EXPR.
7203 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7204 Alan Hayward <alan.hayward@arm.com>
7205 David Sherwood <david.sherwood@arm.com>
7207 * omp-simd-clone.c (simd_clone_subparts): New function.
7208 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
7209 (ipa_simd_modify_function_body): Likewise.
7211 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7212 Alan Hayward <alan.hayward@arm.com>
7213 David Sherwood <david.sherwood@arm.com>
7215 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
7216 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
7217 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
7218 (expand_vector_condition, vector_element): Likewise.
7219 (subparts_gt): New function.
7220 (get_compute_type): Use subparts_gt.
7221 (count_type_subparts): Delete.
7222 (expand_vector_operations_1): Use subparts_gt instead of
7223 count_type_subparts.
7225 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7226 Alan Hayward <alan.hayward@arm.com>
7227 David Sherwood <david.sherwood@arm.com>
7229 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
7230 (vect_compile_time_alias): ...this new function. Do the calculation
7231 on poly_ints rather than trees.
7232 (vect_prune_runtime_alias_test_list): Update call accordingly.
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-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
7240 (vect_schedule_slp_instance): Likewise.
7242 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7243 Alan Hayward <alan.hayward@arm.com>
7244 David Sherwood <david.sherwood@arm.com>
7246 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
7247 constant and extern definitions for variable-length vectors.
7248 (vect_get_constant_vectors): Note that the number of units
7249 is known to be constant.
7251 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7252 Alan Hayward <alan.hayward@arm.com>
7253 David Sherwood <david.sherwood@arm.com>
7255 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
7256 of units as polynomial. Choose between WIDE and NARROW based
7259 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7260 Alan Hayward <alan.hayward@arm.com>
7261 David Sherwood <david.sherwood@arm.com>
7263 * tree-vect-stmts.c (simd_clone_subparts): New function.
7264 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
7266 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7267 Alan Hayward <alan.hayward@arm.com>
7268 David Sherwood <david.sherwood@arm.com>
7270 * tree-vect-stmts.c (vectorizable_call): Treat the number of
7271 vectors as polynomial. Use build_index_vector for
7274 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7275 Alan Hayward <alan.hayward@arm.com>
7276 David Sherwood <david.sherwood@arm.com>
7278 * tree-vect-stmts.c (get_load_store_type): Treat the number of
7279 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
7280 for variable-length vectors.
7281 (vectorizable_mask_load_store): Treat the number of units as
7282 polynomial, asserting that it is constant if the condition has
7283 already been enforced.
7284 (vectorizable_store, vectorizable_load): Likewise.
7286 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7287 Alan Hayward <alan.hayward@arm.com>
7288 David Sherwood <david.sherwood@arm.com>
7290 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
7291 of units as polynomial. Punt if we can't tell at compile time
7292 which vector contains the final result.
7294 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7295 Alan Hayward <alan.hayward@arm.com>
7296 David Sherwood <david.sherwood@arm.com>
7298 * tree-vect-loop.c (vectorizable_induction): Treat the number
7299 of units as polynomial. Punt on SLP inductions. Use an integer
7300 VEC_SERIES_EXPR for variable-length integer reductions. Use a
7301 cast of such a series for variable-length floating-point
7304 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7305 Alan Hayward <alan.hayward@arm.com>
7306 David Sherwood <david.sherwood@arm.com>
7308 * tree.h (build_index_vector): Declare.
7309 * tree.c (build_index_vector): New function.
7310 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
7311 of units as polynomial, forcibly converting it to a constant if
7312 vectorizable_reduction has already enforced the condition.
7313 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
7314 to create a {1,2,3,...} vector.
7315 (vectorizable_reduction): Treat the number of units as polynomial.
7316 Choose vectype_in based on the largest scalar element size rather
7317 than the smallest number of units. Enforce the restrictions
7320 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7321 Alan Hayward <alan.hayward@arm.com>
7322 David Sherwood <david.sherwood@arm.com>
7324 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
7325 number of units as polynomial.
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 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
7332 * target.def (autovectorize_vector_sizes): Return the vector sizes
7333 by pointer, using vector_sizes rather than a bitmask.
7334 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
7335 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
7336 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
7338 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
7339 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
7340 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
7341 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
7342 * omp-general.c (omp_max_vf): Likewise.
7343 * omp-low.c (omp_clause_aligned_alignment): Likewise.
7344 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
7345 * tree-vect-loop.c (vect_analyze_loop): Likewise.
7346 * tree-vect-slp.c (vect_slp_bb): Likewise.
7347 * doc/tm.texi: Regenerate.
7348 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
7350 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
7351 the vector size as a poly_uint64 rather than an unsigned int.
7352 (current_vector_size): Change from an unsigned int to a poly_uint64.
7353 (get_vectype_for_scalar_type): Update accordingly.
7354 * tree.h (build_truth_vector_type): Take the size and number of
7355 units as a poly_uint64 rather than an unsigned int.
7356 (build_vector_type): Add a temporary overload that takes
7357 the number of units as a poly_uint64 rather than an unsigned int.
7358 * tree.c (make_vector_type): Likewise.
7359 (build_truth_vector_type): Take the number of units as a poly_uint64
7360 rather than an unsigned int.
7362 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7363 Alan Hayward <alan.hayward@arm.com>
7364 David Sherwood <david.sherwood@arm.com>
7366 * target.def (get_mask_mode): Take the number of units and length
7367 as poly_uint64s rather than unsigned ints.
7368 * targhooks.h (default_get_mask_mode): Update accordingly.
7369 * targhooks.c (default_get_mask_mode): Likewise.
7370 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
7371 * doc/tm.texi: Regenerate.
7373 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7374 Alan Hayward <alan.hayward@arm.com>
7375 David Sherwood <david.sherwood@arm.com>
7377 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
7378 * omp-general.c (omp_max_vf): Likewise.
7379 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
7380 (expand_omp_simd): Handle polynomial safelen.
7381 * omp-low.c (omplow_simd_context): Add a default constructor.
7382 (omplow_simd_context::max_vf): Change from int to poly_uint64.
7383 (lower_rec_simd_input_clauses): Update accordingly.
7384 (lower_rec_input_clauses): Likewise.
7386 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7387 Alan Hayward <alan.hayward@arm.com>
7388 David Sherwood <david.sherwood@arm.com>
7390 * tree-vectorizer.h (vect_nunits_for_cost): New function.
7391 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
7392 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
7393 (vect_analyze_slp_cost): Likewise.
7394 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
7395 (vect_model_load_cost): Likewise.
7397 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7398 Alan Hayward <alan.hayward@arm.com>
7399 David Sherwood <david.sherwood@arm.com>
7401 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
7402 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
7403 from an unsigned int * to a poly_uint64_pod *.
7404 (calculate_unrolling_factor): New function.
7405 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
7407 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7408 Alan Hayward <alan.hayward@arm.com>
7409 David Sherwood <david.sherwood@arm.com>
7411 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
7412 from an unsigned int to a poly_uint64.
7413 (_loop_vec_info::slp_unrolling_factor): Likewise.
7414 (_loop_vec_info::vectorization_factor): Change from an int
7416 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
7417 (vect_get_num_vectors): New function.
7418 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
7419 (vect_get_num_copies): Use vect_get_num_vectors.
7420 (vect_analyze_data_ref_dependences): Change max_vf from an int *
7421 to an unsigned int *.
7422 (vect_analyze_data_refs): Change min_vf from an int * to a
7424 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7425 than an unsigned HOST_WIDE_INT.
7426 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
7427 (vect_analyze_data_ref_dependence): Change max_vf from an int *
7428 to an unsigned int *.
7429 (vect_analyze_data_ref_dependences): Likewise.
7430 (vect_compute_data_ref_alignment): Handle polynomial vf.
7431 (vect_enhance_data_refs_alignment): Likewise.
7432 (vect_prune_runtime_alias_test_list): Likewise.
7433 (vect_shift_permute_load_chain): Likewise.
7434 (vect_supportable_dr_alignment): Likewise.
7435 (dependence_distance_ge_vf): Take the vectorization factor as a
7436 poly_uint64 rather than an unsigned HOST_WIDE_INT.
7437 (vect_analyze_data_refs): Change min_vf from an int * to a
7439 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
7440 vfm1 as a poly_uint64 rather than an int. Make the same change
7441 for the returned bound_scalar.
7442 (vect_gen_vector_loop_niters): Handle polynomial vf.
7443 (vect_do_peeling): Likewise. Update call to
7444 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
7445 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
7447 * tree-vect-loop.c (vect_determine_vectorization_factor)
7448 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
7449 (vect_get_known_peeling_cost): Likewise.
7450 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
7451 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
7452 (vect_transform_loop): Likewise. Use the lowest possible VF when
7453 updating the upper bounds of the loop.
7454 (vect_min_worthwhile_factor): Make static. Return an unsigned int
7456 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
7457 polynomial unroll factors.
7458 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
7459 (vect_make_slp_decision): Likewise.
7460 (vect_supported_load_permutation_p): Likewise, and polynomial
7462 (vect_analyze_slp_cost): Handle polynomial vf.
7463 (vect_slp_analyze_node_operations): Likewise.
7464 (vect_slp_analyze_bb_1): Likewise.
7465 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7466 than an unsigned HOST_WIDE_INT.
7467 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
7468 (vectorizable_load): Handle polynomial vf.
7469 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
7471 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
7473 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7474 Alan Hayward <alan.hayward@arm.com>
7475 David Sherwood <david.sherwood@arm.com>
7477 * match.pd: Handle bit operations involving three constants
7478 and try to fold one pair.
7480 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7482 * tree-vect-loop-manip.c: Include gimple-fold.h.
7483 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
7484 niters_maybe_zero parameters. Handle other cases besides a step of 1.
7485 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
7486 Add a path that uses a step of VF instead of 1, but disable it
7488 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
7489 and niters_no_overflow parameters. Update calls to
7490 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
7491 Create a new SSA name if the latter choses to use a ste other
7492 than zero, and return it via niters_vector_mult_vf_var.
7493 * tree-vect-loop.c (vect_transform_loop): Update calls to
7494 vect_do_peeling, vect_gen_vector_loop_niters and
7495 slpeel_make_loop_iterate_ntimes.
7496 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
7497 (vect_gen_vector_loop_niters): Update declarations after above changes.
7499 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
7501 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
7502 128-bit round to integer instructions.
7503 (ceil<mode>2): Likewise.
7504 (btrunc<mode>2): Likewise.
7505 (round<mode>2): Likewise.
7507 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7509 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
7510 unaligned VSX load/store on P8/P9.
7511 (expand_block_clear): Allow the use of unaligned VSX
7512 load/store on P8/P9.
7514 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
7516 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
7518 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
7519 swap associated with both a load and a store.
7521 2018-01-02 Andrew Waterman <andrew@sifive.com>
7523 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
7524 * config/riscv/riscv.md (clear_cache): Use it.
7526 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
7528 * web.c: Remove out-of-date comment.
7530 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7532 * expr.c (fixup_args_size_notes): Check that any existing
7533 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
7534 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
7535 (emit_single_push_insn): ...here.
7537 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7539 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
7540 (const_vector_encoded_nelts): New function.
7541 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
7542 (const_vector_int_elt, const_vector_elt): Declare.
7543 * emit-rtl.c (const_vector_int_elt_1): New function.
7544 (const_vector_elt): Likewise.
7545 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
7546 of CONST_VECTOR_ELT.
7548 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7550 * expr.c: Include rtx-vector-builder.h.
7551 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
7552 directly on the tree encoding.
7553 (const_vector_from_tree): Likewise.
7554 * optabs.c: Include rtx-vector-builder.h.
7555 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
7556 sequence of "u" values.
7557 * vec-perm-indices.c: Include rtx-vector-builder.h.
7558 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
7559 directly on the vec_perm_indices encoding.
7561 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7563 * doc/rtl.texi (const_vector): Describe new encoding scheme.
7564 * Makefile.in (OBJS): Add rtx-vector-builder.o.
7565 * rtx-vector-builder.h: New file.
7566 * rtx-vector-builder.c: Likewise.
7567 * rtl.h (rtx_def::u2): Add a const_vector field.
7568 (CONST_VECTOR_NPATTERNS): New macro.
7569 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
7570 (CONST_VECTOR_DUPLICATE_P): Likewise.
7571 (CONST_VECTOR_STEPPED_P): Likewise.
7572 (CONST_VECTOR_ENCODED_ELT): Likewise.
7573 (const_vec_duplicate_p): Check for a duplicated vector encoding.
7574 (unwrap_const_vec_duplicate): Likewise.
7575 (const_vec_series_p): Check for a non-duplicated vector encoding.
7576 Say that the function only returns true for integer vectors.
7577 * emit-rtl.c: Include rtx-vector-builder.h.
7578 (gen_const_vec_duplicate_1): Delete.
7579 (gen_const_vector): Call gen_const_vec_duplicate instead of
7580 gen_const_vec_duplicate_1.
7581 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
7582 (gen_const_vec_duplicate): Use rtx_vector_builder.
7583 (gen_const_vec_series): Likewise.
7584 (gen_rtx_CONST_VECTOR): Likewise.
7585 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
7586 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7587 Build a new vector rather than modifying a CONST_VECTOR in-place.
7588 (handle_special_swappables): Update call accordingly.
7589 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
7590 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7591 Build a new vector rather than modifying a CONST_VECTOR in-place.
7592 (handle_special_swappables): Update call accordingly.
7594 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7596 * simplify-rtx.c (simplify_const_binary_operation): Use
7597 CONST_VECTOR_ELT instead of XVECEXP.
7599 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7601 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
7602 the selector elements to be different from the data elements
7603 if the selector is a VECTOR_CST.
7604 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
7605 ssizetype for the selector.
7607 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7609 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
7610 before testing each element individually.
7611 * tree-vect-generic.c (lower_vec_perm): Likewise.
7613 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7615 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
7616 * selftest-run-tests.c (selftest::run_tests): Call it.
7617 * vector-builder.h (vector_builder::operator ==): New function.
7618 (vector_builder::operator !=): Likewise.
7619 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
7620 (vec_perm_indices::all_from_input_p): New function.
7621 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7622 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
7623 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
7624 instead of reading the VECTOR_CST directly. Detect whether both
7625 vector inputs are the same before constructing the vec_perm_indices,
7626 and update the number of inputs argument accordingly. Use the
7627 utility functions added above. Only construct sel2 if we need to.
7629 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7631 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
7632 the broadcast of the low byte.
7633 (expand_mult_highpart): Use an explicit encoding for the permutes.
7634 * optabs-query.c (can_mult_highpart_p): Likewise.
7635 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
7636 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7637 (vectorizable_bswap): Likewise.
7638 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
7639 explicit encoding for the power-of-2 permutes.
7640 (vect_permute_store_chain): Likewise.
7641 (vect_grouped_load_supported): Likewise.
7642 (vect_permute_load_chain): Likewise.
7644 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7646 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
7647 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
7648 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
7649 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7650 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
7651 (vect_gen_perm_mask_any): Likewise.
7653 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7655 * int-vector-builder.h: New file.
7656 * vec-perm-indices.h: Include int-vector-builder.h.
7657 (vec_perm_indices): Redefine as an int_vector_builder.
7658 (auto_vec_perm_indices): Delete.
7659 (vec_perm_builder): Redefine as a stand-alone class.
7660 (vec_perm_indices::vec_perm_indices): New function.
7661 (vec_perm_indices::clamp): Likewise.
7662 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
7663 (vec_perm_indices::new_vector): New function.
7664 (vec_perm_indices::new_expanded_vector): Update for new
7665 vec_perm_indices class.
7666 (vec_perm_indices::rotate_inputs): New function.
7667 (vec_perm_indices::all_in_range_p): Operate directly on the
7668 encoded form, without computing elided elements.
7669 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
7670 encoding. Update for new vec_perm_indices class.
7671 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
7672 the given vec_perm_builder.
7673 (expand_vec_perm_var): Update vec_perm_builder constructor.
7674 (expand_mult_highpart): Use vec_perm_builder instead of
7675 auto_vec_perm_indices.
7676 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
7677 vec_perm_indices instead of auto_vec_perm_indices. Use a single
7678 or double series encoding as appropriate.
7679 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
7680 vec_perm_indices instead of auto_vec_perm_indices.
7681 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7682 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7683 (vect_permute_store_chain): Likewise.
7684 (vect_grouped_load_supported): Likewise.
7685 (vect_permute_load_chain): Likewise.
7686 (vect_shift_permute_load_chain): Likewise.
7687 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7688 (vect_transform_slp_perm_load): Likewise.
7689 (vect_schedule_slp_instance): Likewise.
7690 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7691 (vectorizable_mask_load_store): Likewise.
7692 (vectorizable_bswap): Likewise.
7693 (vectorizable_store): Likewise.
7694 (vectorizable_load): Likewise.
7695 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
7696 vec_perm_indices instead of auto_vec_perm_indices. Use
7697 tree_to_vec_perm_builder to read the vector from a tree.
7698 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
7699 vec_perm_builder instead of a vec_perm_indices.
7700 (have_whole_vector_shift): Use vec_perm_builder and
7701 vec_perm_indices instead of auto_vec_perm_indices. Leave the
7702 truncation to calc_vec_perm_mask_for_shift.
7703 (vect_create_epilog_for_reduction): Likewise.
7704 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
7705 from auto_vec_perm_indices to vec_perm_indices.
7706 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7707 instead of changing individual elements.
7708 (aarch64_vectorize_vec_perm_const): Use new_vector to install
7709 the vector in d.perm.
7710 * config/arm/arm.c (expand_vec_perm_d::perm): Change
7711 from auto_vec_perm_indices to vec_perm_indices.
7712 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7713 instead of changing individual elements.
7714 (arm_vectorize_vec_perm_const): Use new_vector to install
7715 the vector in d.perm.
7716 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
7717 Update vec_perm_builder constructor.
7718 (rs6000_expand_interleave): Likewise.
7719 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
7720 (rs6000_expand_interleave): Likewise.
7722 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7724 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
7725 to qimode could truncate the indices.
7726 * optabs.c (expand_vec_perm_var): Likewise.
7728 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7730 * Makefile.in (OBJS): Add vec-perm-indices.o.
7731 * vec-perm-indices.h: New file.
7732 * vec-perm-indices.c: Likewise.
7733 * target.h (vec_perm_indices): Replace with a forward class
7735 (auto_vec_perm_indices): Move to vec-perm-indices.h.
7736 * optabs.h: Include vec-perm-indices.h.
7737 (expand_vec_perm): Delete.
7738 (selector_fits_mode_p, expand_vec_perm_var): Declare.
7739 (expand_vec_perm_const): Declare.
7740 * target.def (vec_perm_const_ok): Replace with...
7741 (vec_perm_const): ...this new hook.
7742 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
7743 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7744 * doc/tm.texi: Regenerate.
7745 * optabs.def (vec_perm_const): Delete.
7746 * doc/md.texi (vec_perm_const): Likewise.
7747 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7748 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7749 expand_vec_perm for constant permutation vectors. Assert that
7750 the mode of variable permutation vectors is the integer equivalent
7751 of the mode that is being permuted.
7752 * optabs-query.h (selector_fits_mode_p): Declare.
7753 * optabs-query.c: Include vec-perm-indices.h.
7754 (selector_fits_mode_p): New function.
7755 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7756 is defined, instead of checking whether the vec_perm_const_optab
7757 exists. Use targetm.vectorize.vec_perm_const instead of
7758 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7759 fit in the vector mode before using a variable permute.
7760 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7761 vec_perm_indices instead of an rtx.
7762 (expand_vec_perm): Replace with...
7763 (expand_vec_perm_const): ...this new function. Take the selector
7764 as a vec_perm_indices rather than an rtx. Also take the mode of
7765 the selector. Update call to shift_amt_for_vec_perm_mask.
7766 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7767 Use vec_perm_indices::new_expanded_vector to expand the original
7768 selector into bytes. Check whether the indices fit in the vector
7769 mode before using a variable permute.
7770 (expand_vec_perm_var): Make global.
7771 (expand_mult_highpart): Use expand_vec_perm_const.
7772 * fold-const.c: Includes vec-perm-indices.h.
7773 * tree-ssa-forwprop.c: Likewise.
7774 * tree-vect-data-refs.c: Likewise.
7775 * tree-vect-generic.c: Likewise.
7776 * tree-vect-loop.c: Likewise.
7777 * tree-vect-slp.c: Likewise.
7778 * tree-vect-stmts.c: Likewise.
7779 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7781 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7782 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7783 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7784 (aarch64_vectorize_vec_perm_const): ...this new function.
7785 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7786 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7787 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7788 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7789 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7790 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7791 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7793 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7794 check for NEON modes.
7795 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7796 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7797 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7798 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7800 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7801 the old VEC_PERM_CONST conditions.
7802 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7803 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7804 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7805 (ia64_vectorize_vec_perm_const_ok): Merge into...
7806 (ia64_vectorize_vec_perm_const): ...this new function.
7807 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7808 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7809 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7810 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7811 * config/mips/mips.c (mips_expand_vec_perm_const)
7812 (mips_vectorize_vec_perm_const_ok): Merge into...
7813 (mips_vectorize_vec_perm_const): ...this new function.
7814 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7815 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7816 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7817 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7818 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7819 (rs6000_expand_vec_perm_const): Delete.
7820 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7822 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7823 (altivec_expand_vec_perm_const_le): Take each operand individually.
7824 Operate on constant selectors rather than rtxes.
7825 (altivec_expand_vec_perm_const): Likewise. Update call to
7826 altivec_expand_vec_perm_const_le.
7827 (rs6000_expand_vec_perm_const): Delete.
7828 (rs6000_vectorize_vec_perm_const_ok): Delete.
7829 (rs6000_vectorize_vec_perm_const): New function.
7830 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7831 an element count and rtx array.
7832 (rs6000_expand_extract_even): Update call accordingly.
7833 (rs6000_expand_interleave): Likewise.
7834 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7835 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7836 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7837 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7838 (rs6000_expand_vec_perm_const): Delete.
7839 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7840 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7841 (altivec_expand_vec_perm_const_le): Take each operand individually.
7842 Operate on constant selectors rather than rtxes.
7843 (altivec_expand_vec_perm_const): Likewise. Update call to
7844 altivec_expand_vec_perm_const_le.
7845 (rs6000_expand_vec_perm_const): Delete.
7846 (rs6000_vectorize_vec_perm_const_ok): Delete.
7847 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7848 reference to the SPE evmerge intructions.
7849 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7850 an element count and rtx array.
7851 (rs6000_expand_extract_even): Update call accordingly.
7852 (rs6000_expand_interleave): Likewise.
7853 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7854 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7856 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7858 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7860 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7861 vector mode and that that mode matches the mode of the data
7863 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7864 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7865 directly using expand_vec_perm_1 when forcing selectors into
7867 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7869 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7871 * optabs-query.h (can_vec_perm_p): Delete.
7872 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7873 * optabs-query.c (can_vec_perm_p): Split into...
7874 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7875 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7876 particular selector is valid.
7877 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7878 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7879 (vect_grouped_load_supported): Likewise.
7880 (vect_shift_permute_load_chain): Likewise.
7881 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7882 (vect_transform_slp_perm_load): Likewise.
7883 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7884 (vectorizable_bswap): Likewise.
7885 (vect_gen_perm_mask_checked): Likewise.
7886 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7887 implementations of variable permutation vectors into account
7888 when deciding which selector to use.
7889 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7890 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7891 with a false third argument.
7892 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7893 to test whether the constant selector is valid and can_vec_perm_var_p
7894 to test whether a variable selector is valid.
7896 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7898 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7899 * optabs-query.c (can_vec_perm_p): Likewise.
7900 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7901 instead of vec_perm_indices.
7902 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7903 (vect_gen_perm_mask_checked): Likewise,
7904 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7905 (vect_gen_perm_mask_checked): Likewise,
7907 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7909 * optabs-query.h (qimode_for_vec_perm): Declare.
7910 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7911 (qimode_for_vec_perm): ...this new function.
7912 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7914 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7916 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7917 does not have a conditional at the top.
7919 2018-01-02 Richard Biener <rguenther@suse.de>
7921 * ipa-inline.c (big_speedup_p): Fix expression.
7923 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7926 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7929 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7933 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7934 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7935 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7936 cond_taken_branch_cost 3->4.
7938 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7940 PR tree-optimization/83581
7941 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7942 TODO_cleanup_cfg if any changes have been made.
7945 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7946 convert_modes if target mode has the right side, but different mode
7950 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7951 last argument when extracting from CONCAT. If either from_real or
7952 from_imag is NULL, use expansion through memory. If result is not
7953 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7954 the parts directly to inner mode, if even that fails, use expansion
7958 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7959 check for bswap in mode rather than HImode and use that in expand_unop
7962 Copyright (C) 2018 Free Software Foundation, Inc.
7964 Copying and distribution of this file, with or without modification,
7965 are permitted in any medium without royalty provided the copyright
7966 notice and this notice are preserved.