1 2018-03-01 Martin Liska <mliska@suse.cz>
4 * sanopt.c (sanitize_rewrite_addressable_params): Do not handle
7 2018-03-01 Richard Biener <rguenther@suse.de>
10 * dwarf2out.c (gen_variable_die): Properly handle late VLA
11 type annotation with LTO when debug was disabled at compile-time.
13 2018-03-01 Matthew Fortune <mfortune@gmail.com>
15 * config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
17 (mips_final_postscan_insn): Likewise.
19 2018-03-01 Richard Sandiford <richard.sandiford@linaro.org>
21 PR rtl-optimization/84528
22 * alias.c (init_alias_target): Add commentary.
23 (init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
24 a unique base value if the frame pointer is not eliminated
27 2018-03-01 Tom de Vries <tom@codesourcery.com>
29 PR rtl-optimization/83327
30 * lra-int.h (hard_regs_spilled_into): Declare.
31 * lra.c (hard_regs_spilled_into): Define.
32 (init_reg_info): Init hard_regs_spilled_into.
33 * lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
34 * lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
35 (process_bb_lives): Handle hard_regs_spilled_into.
36 (lra_create_live_ranges_1): Before doing liveness propagation, clear
37 regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.
39 2018-02-28 David Edelsohn <dje.gcc@gmail.com>
41 * config.gcc (powerpc-ibm-aix7.1.*): New stanza.
42 (powerpc-ibm-aix[789]*): Default to AIX 7.2.
43 * config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
44 * config/rs6000/aix72.h: New file.
46 2018-02-28 Jakub Jelinek <jakub@redhat.com>
48 * gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
49 instead of warning_at with conditional singular and plural messages
53 * stor-layout.c (update_alignment_for_field): For
54 targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
55 && !DECL_PACKED (field), do the alignment update, just use
56 only desired_align instead of MAX (type_align, desired_align)
58 (place_field): Don't do known_align < desired_align handling
59 early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
60 is non-NULL, instead do it after rli->prev_field handling and
61 only if not within a bitfield word. For DECL_PACKED (field)
62 use type_align of BITS_PER_UNIT.
64 2018-02-28 Eric Botcazou <ebotcazou@adacore.com>
66 * config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
67 superfluous parentheses and trailing spaces.
69 2018-02-28 Richard Biener <rguenther@suse.de>
71 PR tree-optimization/84584
72 * graphite-scop-detection.c (scop_detection::add_scop): Discard
73 SCoPs with fake exit edge.
75 2018-02-28 Martin Liska <mliska@suse.cz>
78 * timevar.c (timer::print): Fix format to properly print 100%
81 2018-02-28 Richard Biener <rguenther@suse.de>
84 * genmatch.c (capture_info::walk_match): Do not mark
85 captured expressions without operands as expr_p given
86 they act more like predicates and should be subject to
87 "lost tail" side-effect preserving.
89 2018-02-28 Alexandre Oliva <aoliva@redhat.com>
91 PR rtl-optimization/81611
92 * auto-inc-dec.c (attempt_change): Move dead note from
93 mem_insn if it's the next use of regno
94 (find_address): Take address use of reg holding
95 non-incremented value. Add parm to limit search to the named
97 (merge_in_block): Attempt to use a mem insn that is the next
98 use of the original regno.
100 2018-02-27 Martin Sebor <msebor@redhat.com>
103 * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
104 * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
106 2018-02-27 Martin Sebor <msebor@redhat.com>
109 * diagnostic-core.h (warning_n, error_n, inform_n): Change
110 n argument to unsigned HOST_WIDE_INT.
111 * diagnostic.c (warning_n, error_n, inform_n): Ditto.
112 (diagnostic_n_impl): Ditto. Handle arguments in excess of LONG_MAX.
113 * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
114 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
116 2018-02-27 Richard Biener <rguenther@suse.de>
118 PR tree-optimization/84512
119 * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
120 Do not use the estimate returned from record_stmt_cost for
121 the scalar iteration cost but sum properly using add_stmt_cost.
123 2018-02-27 Richard Biener <rguenther@suse.de>
125 PR tree-optimization/84466
126 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
127 Adjust last change to less strictly validate use operands.
129 2018-02-27 Martin Liska <mliska@suse.cz>
131 PR gcov-profile/84548
132 * gcov.c (process_file): Allow partial overlap and consider it
133 also as group functions.
134 (output_lines): Properly calculate range of lines for a group.
136 2018-02-27 Martin Liska <mliska@suse.cz>
138 * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
139 'ggc' suffixes. Change first column width.
140 (timer::print): Fix formatting of the column.
142 2018-02-27 Alexandre Oliva <aoliva@redhat.com>
144 * tree-ssa-live.c (remove_unused_scope_block_p): Do not
145 preserve inline entry blocks for the sake of debug inline
146 entry point markers alone.
147 (remove_unused_locals): Suggest in comments a better place to
148 force the preservation of inline entry blocks that are
149 otherwise unused, but do not preserve them.
151 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
153 * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
155 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
158 * config/i386/constraints.md (Bs): Replace
159 ix86_indirect_branch_register with
160 TARGET_INDIRECT_BRANCH_REGISTER.
162 * config/i386/i386.md (indirect_jump): Likewise.
163 (tablejump): Likewise.
164 (*sibcall_memory): Likewise.
165 (*sibcall_value_memory): Likewise.
166 Peepholes of indirect call and jump via memory: Likewise.
167 (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
168 (*sibcall_value_GOT_32): Likewise.
169 * config/i386/predicates.md (indirect_branch_operand): Likewise.
170 (GOT_memory_operand): Likewise.
171 (call_insn_operand): Likewise.
172 (sibcall_insn_operand): Likewise.
173 (GOT32_symbol_operand): Likewise.
174 * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
176 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
178 PR rtl-optimization/83496
179 * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
180 booleans to RTXes. Call fix_reg_dead_note on every non-null element.
181 (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
182 redundant insn, if any.
183 (relax_delay_slots): Likewise.
184 (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
186 2018-02-26 Richard Sandiford <richard.sandiford@linaro.org>
188 PR tree-optimization/83965
189 * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
190 that grouped statements are part of a reduction chain. Return
191 true if the statement is not marked as a reduction itself but
193 (vect_recog_dot_prod_pattern): Don't check whether the statement
194 is part of a group here.
195 (vect_recog_sad_pattern): Likewise.
196 (vect_recog_widen_sum_pattern): Likewise.
198 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
201 * final.c (rest_of_clean_state): Also look for calls inside sequences.
203 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
206 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
208 (ix86_output_indirect_function_return): New prototype.
209 (ix86_split_simple_return_pop_internal): Likewise.
210 * config/i386/i386.c (indirect_return_via_cx): New.
211 (indirect_return_via_cx_bnd): Likewise.
212 (indirect_thunk_name): Handle return va CX_REG.
213 (output_indirect_thunk_function): Create alias for
214 __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
215 (ix86_output_indirect_jmp): Remove the bool argument.
216 (ix86_output_indirect_function_return): New function.
217 (ix86_split_simple_return_pop_internal): Likewise.
218 * config/i386/i386.md (*indirect_jump): Don't pass false
219 to ix86_output_indirect_jmp.
220 (*tablejump_1): Likewise.
221 (simple_return_pop_internal): Change it to define_insn_and_split.
222 Call ix86_split_simple_return_pop_internal to split it for
224 (simple_return_indirect_internal): Call
225 ix86_output_indirect_function_return instead of
226 ix86_output_indirect_jmp.
228 2018-02-26 Jakub Jelinek <jakub@redhat.com>
231 * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
232 memset and value initialization afterwards.
234 2018-02-26 Christophe Lyon <christophe.lyon@linaro.org>
236 * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
238 2018-02-26 Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
241 * common/config/aarch64/aarch64-common.c
242 (aarch_option_optimization_table[]): Switch
243 off fomit-frame-pointer
245 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
246 Chung-Ju Wu <jasonwucj@gmail.com>
248 * config/nds32/nds32-multiple.md (load_multiple): Disallow
250 (store_multiple): Ditto.
252 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
254 * config.gcc: Add --with-cpu support for nds32 target.
255 * config/nds32/nds32-opts.h (nds32_cpu_type): New.
256 * config/nds32/nds32.opt: Add -mcpu= option.
258 2018-02-25 Segher Boessenkool <segher@kernel.crashing.org>
260 * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
261 isel=yes): Warn for these deprecated options.
263 2018-02-23 David Edelsohn <dje.gcc@gmail.com>
265 * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
266 ISA_2_5_MASKS_EMBEDDED.
268 2018-02-23 Jakub Jelinek <jakub@redhat.com>
270 * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
271 p->max as pointers rather than using iterative_hash_expr.
273 2018-02-23 Carl Love <cel@us.ibm.com>
275 * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
276 macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
278 * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
279 P8V_BUILTIN_VEC_VSIGNED2. Change VSX_BUILTIN_VEC_VUNSIGNED2 to
280 P8V_BUILTIN_VEC_VUNSIGNED2.
282 2018-02-22 Vladimir Makarov <vmakarov@redhat.com>
285 * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
286 * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
288 * lra-constraints.c (curr_insn_transform): Set up
289 LRA_NON_CLOBBERED_ALT for moves processed on the fast path. Use
291 (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
292 * lra-eliminations.c (spill_pseudos): Ditto.
293 (process_insn_for_elimination): Ditto.
294 * lra-lives.c (reg_early_clobber_p): Use the new macros.
295 * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
296 LRA_NON_CLOBBERED_ALT.
298 2018-02-22 Martin Sebor <msebor@redhat.com>
300 PR tree-optimization/84480
301 * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
302 to maybe_diag_stxncpy_trunc. Call it.
303 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
304 from gimple_fold_builtin_strcpy. Print inlining stack.
305 (handle_builtin_stxncpy): Print inlining stack.
306 * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
308 2018-02-22 H.J. Lu <hongjiu.lu@intel.com>
311 * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
312 error when -mindirect-branch=thunk-extern, -fcf-protection=branch
313 and -fcheck-pointer-bounds are used together.
314 (indirect_thunk_prefix): New enum.
315 (indirect_thunk_need_prefix): New function.
316 (indirect_thunk_name): Replace need_bnd_p with need_prefix. Use
317 "_nt" instead of "_bnd" for NOTRACK prefix.
318 (output_indirect_thunk): Replace need_bnd_p with need_prefix.
319 (output_indirect_thunk_function): Likewise.
321 (ix86_code_end): Update output_indirect_thunk_function calls.
322 (ix86_output_indirect_branch_via_reg): Replace
323 ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
324 (ix86_output_indirect_branch_via_push): Likewise.
325 (ix86_output_function_return): Likewise.
326 * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
327 incompatible with -fcf-protection=branch and
328 -fcheck-pointer-bounds.
330 2018-02-22 Steve Ellcey <sellcey@cavium.com>
333 * config/aarch64/aarch64.c (aarch64_print_address_internal):
334 Change gcc_assert call to output_operand_lossage.
336 2018-02-22 Steve Ellcey <sellcey@cavium.com>
338 * doc/extend.texi (__builtin_extend_pointer): Document builtin.
340 2018-02-22 DJ Delorie <dj@redhat.com>
341 Sebastian Perta <sebastian.perta@renesas.com>
342 Oleg Endo <olegendo@gcc.gnu.org>
344 * config/rx/rx.c (rx_rtx_costs): New function.
345 (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
347 2018-02-22 Thomas Preud'homme <thomas.preudhomme@arm.com>
349 * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
351 2018-02-22 Martin Liska <mliska@suse.cz>
354 * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
355 Add "native" as a possible value.
357 2018-02-22 Martin Liska <mliska@suse.cz>
360 * config/i386/i386.c (ix86_option_override_internal):
361 Add "native" as a possible value for -march and -mtune.
363 2018-02-22 Jakub Jelinek <jakub@redhat.com>
366 * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
367 to all type variants.
369 PR tree-optimization/84503
370 * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
371 width as info->bitpos + info->bitsize - start.
372 (merged_store_group::merge_overlapping): Simplify width computation.
373 (check_no_overlap): New function.
374 (imm_store_chain_info::try_coalesce_bswap): Compute expected
375 start + width and last_order of the group, fail if check_no_overlap
377 (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
378 to group if check_no_overlap fails.
380 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
382 * config/rs6000/altivec.md: Delete contraint arguments to
383 define_expand, define_split, and define_peephole2, and in
384 define_insn_and_split if always unused.
385 * config/rs6000/darwin.md: Ditto.
386 * config/rs6000/dfp.md: Ditto.
387 * config/rs6000/rs6000.md: Ditto.
388 * config/rs6000/sync.md: Ditto.
389 * config/rs6000/vector.md: Ditto.
390 * config/rs6000/vsx.md: Ditto.
392 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
394 * config/rs6000/altivec.md: Write output control strings as braced
395 blocks instead of double-quoted strings.
396 * config/rs6000/darwin.md: Ditto.
397 * config/rs6000/rs6000.md: Ditto.
398 * config/rs6000/vector.md: Ditto.
399 * config/rs6000/vsx.md: Ditto.
401 2018-02-21 Jason Merrill <jason@redhat.com>
403 PR c++/84314 - ICE with templates and fastcall attribute.
404 * attribs.c (build_type_attribute_qual_variant): Remove assert.
406 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
408 * ipa-cp.c (determine_versionability): Fix comment typos.
410 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
413 * ipa-cp.c (determine_versionability): Do not version functions caling
416 2018-02-21 Martin Liska <mliska@suse.cz>
419 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
420 Add "native" as a possible value.
421 * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT): Define
422 the macro when native cpu detection is available.
424 2018-02-21 Martin Liska <mliska@suse.cz>
427 * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
428 Add "native" as a possible value.
429 * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
430 when native cpu detection is available.
432 2018-02-21 Jakub Jelinek <jakub@redhat.com>
433 Martin Sebor <msebor@redhat.com>
435 PR tree-optimization/84478
436 * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
438 * gimple-fold.c (get_range_strlen): Make minlen const and assume it
439 can't be NULL. Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
440 support which is conservatively correct, for 2 only stay conservative
441 for maxlen. Formatting and comment capitalization fixes. Add STRICT
442 argument to the 2 argument get_range_strlen, adjust 6 arg
443 get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
445 (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
446 (gimple_fold_builtin_strlen): Pass true as last argument to
449 2018-02-20 Martin Sebor <msebor@redhat.com>
452 * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
453 (builtin_memref::set_base_and_offset): Same. Handle inner references.
454 (builtin_memref::builtin_memref): Factor out parts into
455 set_base_and_offset and call it.
457 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
460 * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
461 is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
462 greater precision. If to_mode is a MODE_PARTIAL_INT, stop the
463 search at the associated MODE_INT.
465 2018-02-20 Jeff Law <law@redhat.com>
468 PR tree-optimization/81592
470 * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
471 for range data rather than using global data.
472 * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
473 range data rather than using global data.
474 * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
475 pass it to children as needed.
476 (struct directive::fmtresult): Similarly.
477 (struct directive::set_width): Similarly.
478 (struct directive::set_precision): Similarly.
479 (format_integer, format_directive, parse_directive): Similarly.
480 (format_none): Accept unnamed vr_values parameter.
481 (format_percent, format_floating, format_character): Similarly.
482 (format_string, format_plain): Similarly.
483 * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
484 the EVRP range analyzer for range data rather than using global data.
485 * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
486 gimple-ssa-evrp-analyze.h
487 (class sprintf_dom_walker): Add after_dom_children member function.
488 Add evrp_range_analyzer member.
489 (sprintf_dom_walker::before_dom_children): Call into the EVRP
490 range analyzer as needed.
491 (sprintf_dom_walker::after_dom_children): New member function.
492 * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
494 (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
495 (evrp_range_analyzer::pop_to_marker): Likewise.
497 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
499 PR tree-optimization/84419
500 * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
501 with the required type if its current type is compatible but
504 2018-02-20 Jakub Jelinek <jakub@redhat.com>
507 * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
510 2018-02-20 Martin Liska <mliska@suse.cz>
513 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
514 possible values if we don't have a hint.
516 2018-02-20 Martin Liska <mliska@suse.cz>
520 * final.c (shorten_branches): Build align_tab array with one
522 * opts.c (finish_options): Add alignment option limit check.
523 (MAX_CODE_ALIGN): Likewise.
524 (MAX_CODE_ALIGN_VALUE): Likewise.
525 * doc/invoke.texi: Document maximum allowed option value for
526 all -falign-* options.
528 2018-02-19 Jakub Jelinek <jakub@redhat.com>
531 * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
532 * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
533 * var-tracking.c (emit_note_insn_var_location): Remove all references
534 to NOTE_INSN_CALL_ARG_LOCATION.
535 (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
536 the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
537 Use copy_rtx_if_shared.
538 * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
539 NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
540 (dwarf2out_var_location): Remove handling of
541 NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
543 * final.c (final_scan_insn): Remove all references to
544 NOTE_INSN_CALL_ARG_LOCATION.
545 (rest_of_clean_state): Likewise. Remove REG_CALL_ARG_LOCATION notes
546 before dumping final insns.
547 * except.c (emit_note_eh_region_end): Remove all references to
548 NOTE_INSN_CALL_ARG_LOCATION.
549 * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
550 * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
551 * config/arc/arc.c (hwloop_optimize): Likewise.
552 * config/arm/arm.c (create_fix_barrier): Likewise.
553 * config/s390/s390.c (s390_chunkify_start): Likewise.
554 * config/sh/sh.c (find_barrier): Likewise.
555 * config/i386/i386.c (rest_of_insert_endbranch,
556 ix86_seh_fixup_eh_fallthru): Likewise.
557 * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
558 * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
559 * config/frv/frv.c (frv_function_prologue): Likewise.
560 * emit-rtl.c (try_split): Likewise. Copy over REG_CALL_ARG_LOCATION
562 (note_outside_basic_block_p): Remove all references to
563 NOTE_INSN_CALL_ARG_LOCATION.
564 * gengtype.c (adjust_field_rtx_def): Likewise.
565 * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
567 * jump.c (cleanup_barriers, delete_related_insns): Likewise.
568 * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
571 * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
574 PR tree-optimization/84452
575 * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
576 expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
579 2018-02-19 Martin Liska <mliska@suse.cz>
582 * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
584 2018-02-19 Martin Liska <mliska@suse.cz>
585 Richard Sandiford <richard.sandiford@linaro.org>
587 PR tree-optimization/82491
588 * gimple-fold.c (get_base_constructor): Make earlier bail out
591 2018-02-19 Carl Love <cel@us.ibm.com>
593 * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
594 BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
596 * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
599 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
601 * config/rl78/rl78.md (movdf): New define expand.
603 2018-02-19 Martin Liska <mliska@suse.cz>
606 * doc/invoke.texi: Fix typo.
607 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
609 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
611 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
612 handle rs6000_single_float and rs6000_double_float specially for
615 2018-02-16 Jeff Law <law@redhat.com>
617 * config/rx/rx.c (add_pop_cfi_notes): New function.;
620 2018-02-16 Jakub Jelinek <jakub@redhat.com>
623 * ipa-inline.c (inline_small_functions): Fix a typo.
625 2018-02-16 Nathan Sidwell <nathan@acm.org>
627 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
629 2018-02-16 Carl Love <cel@us.ibm.com>
631 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
632 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
633 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
634 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
635 expansion to P8V_BUILTIN_VEC_FLOAT2.
637 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
639 PR rtl-optimization/70023
640 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
641 src_regno into account.
643 2018-02-16 Carl Love <cel@us.ibm.com>
645 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
646 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
647 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
648 * config/rs6000/rs6000.c: Remove case statements for
649 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
650 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
651 and P9V_BUILTIN_VEC_VINSERT4B.
652 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
653 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
654 * config/rs6000/vsx.md:
655 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
658 2018-02-16 Carl Love <cel@us.ibm.com>
660 * config/rs6000/altivec.h: Add builtin names vec_extract4b
662 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
664 * config/rs6000/rs6000-c.c: Add the definitions for
665 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
666 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
667 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
668 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
669 definition for insert4b and define insn *insert3b_internal.
670 * doc/extend.texi: Add documentation for vec_extract4b.
672 2018-02-16 Nathan Sidwell <nathan@acm.org>
674 * doc/extend.texi (Backwards Compatibility): Mention friend
675 injection. Note for-scope is deprecated.
676 * doc/invoke.texi (-ffriend-injection): Deprecate.
678 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
680 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
681 that moved to I2, also allow destinations that are a paradoxical
682 subreg (instead of a normal reg).
684 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
687 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
690 2018-02-16 Richard Biener <rguenther@suse.de>
692 PR tree-optimization/84037
693 PR tree-optimization/84016
695 * config/i386/i386.c (ix86_builtin_vectorization_cost):
696 Adjust vec_construct for the fact we need additional higher latency
697 128bit inserts for AVX256 and AVX512 vector builds.
698 (ix86_add_stmt_cost): Scale vector construction cost for
701 2018-02-16 Richard Biener <rguenther@suse.de>
703 PR tree-optimization/84417
704 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
705 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
706 (non_rewritable_lvalue_p): Likewise, use poly-ints.
708 2018-02-16 Martin Liska <mliska@suse.cz>
711 * internal-fn.def (ASAN_CHECK): Set proper flags.
712 (ASAN_MARK): Likewise.
714 2018-02-16 Julia Koval <julia.koval@intel.com>
716 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
719 2018-02-16 Jakub Jelinek <jakub@redhat.com>
722 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
723 Use ++iter rather than iter++ for std::list iterators.
724 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
725 defer deleting them until all nodes in the forest are processed. Do
726 free even leaf nodes. Change to_process into auto_vec.
729 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
730 * vec.h (vec_default_construct): Use memset instead of placement new
731 if BROKEN_VALUE_INITIALIZATION is defined.
732 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
733 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
736 PR rtl-optimization/83723
737 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
738 * lra.c (lra_substitute_pseudo): Likewise. If true, use
739 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
741 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
743 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
745 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
747 PR rtl-optimization/81443
748 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
749 from inner REGs to paradoxical SUBREGs.
751 2018-02-16 Richard Biener <rguenther@suse.de>
753 PR tree-optimization/84399
754 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
755 For operands we can analyze at their definition make sure we can
756 analyze them at each use as well.
758 2018-02-16 Richard Biener <rguenther@suse.de>
760 PR tree-optimization/84190
761 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
762 volatile accesses if the decl isn't volatile.
764 2018-02-15 Jason Merrill <jason@redhat.com>
766 PR c++/84314 - ICE with templates and fastcall attribute.
767 * attribs.c (build_type_attribute_qual_variant): Don't clobber
768 TYPE_CANONICAL on an existing type.
770 2018-02-15 Jakub Jelinek <jakub@redhat.com>
772 PR tree-optimization/84383
773 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
774 dstoff nor call operand_equal_p if dstbase is NULL.
776 PR tree-optimization/84334
777 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
778 also a CONSTANT_CLASS_P, punt.
780 2018-02-14 Jim Wilson <jimw@sifive.com>
782 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
783 first SMALL_OPERAND check. New local min_second_step. Move assert
784 to where locals are set. Add TARGET_RVC support.
785 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
787 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
789 * doc/invoke.texi: Correct -Wformat-overflow code sample.
791 2018-02-14 Martin Sebor <msebor@redhat.com>
793 PR tree-optimization/83698
794 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
795 arrays constrain the offset range to their bounds.
796 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
797 (builtin_access::overlap): Avoid setting the size of overlap if it's
799 (maybe_diag_overlap): Also consider arrays when deciding what values
800 of offsets to include in diagnostics.
802 2018-02-14 Martin Sebor <msebor@redhat.com>
805 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
806 that correspond to the kind of a declaration.
808 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
811 * config/pa/pa.md: Load address of PIC label using the linkage table
812 if the label is nonlocal.
814 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
816 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
817 warning message if user requests -maltivec=be.
818 * doc/invoke.texi: Document deprecation of -maltivec=be.
820 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
823 * config/rs6000/rs6000-c.c: Update definitions for
824 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
825 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
827 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
830 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
831 add _get_ssp intrinsics. Remove argument from
832 __builtin_ia32_rdssp[d|q].
833 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
834 * config/i386/i386-builtin.def: Remove argument from
835 __builtin_ia32_rdssp[d|q].
836 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
837 ix86_expand_special_args_builtin for _rdssp[d|q].
838 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
839 Clear register before usage.
840 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
841 Add documentation for new _get_ssp and _inc_ssp intrinsics.
843 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
845 PR tree-optimization/84357
846 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
847 operand 1 of an ARRAY_REF too.
849 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
852 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
853 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
855 (set_of_reg): New struct.
856 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
857 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
858 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
860 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
861 Split into bitclr, bitset, bitinvert patterns if appropriate.
862 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
863 use rx_fuse_in_memory_bitop.
864 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
865 to named insn, correct maximum insn length.
867 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
870 * machmode.def: Define a complex mode for PARTIAL_INT.
871 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
873 * doc/rtl.texi: Document CSPImode.
874 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
876 (msp430_hard_regno_nregs_with_padding): Likewise.
878 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
881 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
883 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
885 PR rtl-optimization/84169
886 * combine.c (try_combine): New variable split_i2i3. Set it to true if
887 we generated a parallel as new i3 and we split that to new i2 and i3
888 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
889 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
890 those to i2, not i1. Partially rewrite this scan code.
892 2018-02-13 Jakub Jelinek <jakub@redhat.com>
895 * stor-layout.c (place_field): For variable length fields, adjust
896 offset_align afterwards not just based on the field's alignment,
897 but also on the size.
900 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
901 of exps and logs in the use_exp2 case.
903 2018-02-13 Jeff Law <law@redhat.com>
905 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
908 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
911 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
915 * common.opt (gas-loc-support, gas-locview-support): New.
916 (ginline-points, ginternal-reset-location-views): New.
917 * doc/invoke.texi: Document them. Use @itemx where intended.
918 (gvariable-location-views): Adjust.
919 * target.def (reset_location_view): New.
920 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
921 (TARGET_RESET_LOCATION_VIEW): New.
922 * doc/tm.texi: Rebuilt.
923 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
924 (dwarf2out_default_as_locview_support): New.
925 (output_asm_line_debug_info): Use option variables.
926 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
927 (output_loc_list): Likewise.
928 (add_high_low_attributes): Check option variables.
929 Don't output entry view attribute in strict mode.
930 (gen_inlined_subroutine_die): Check option variables.
931 (dwarf2out_inline_entry): Likewise.
932 (init_sections_and_labels): Likewise.
933 (dwarf2out_early_finish): Likewise.
934 (maybe_reset_location_view): New, from...
935 (dwarf2out_var_location): ... here. Call it.
936 * debug.h (dwarf2out_default_as_loc_support): Declare.
937 (dwarf2out_default_as_locview_support): Declare.
938 * hooks.c (hook_int_rtx_insn_0): New.
939 * hooks.h (hook_int_rtx_insn_0): Declare.
940 * toplev.c (process_options): Take -gas-loc-support and
941 -gas-locview-support from dwarf2out. Enable
942 -gvariable-location-views by default only with locview
943 assembler support. Enable -ginternal-reset-location-views by
944 default only if the target defines the corresponding hook.
945 Enable -ginline-points by default if location views are
946 enabled; force it disabled if statement frontiers are
948 * tree-inline.c (expand_call_inline): Check option variables.
949 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
951 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
953 PR tree-optimization/84321
954 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
955 handling. Also check whether the anti-range contains any values
956 that satisfy the mask; switch to a VR_RANGE if not.
958 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
961 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
963 2018-02-13 Martin Jambor <mjambor@suse.cz>
966 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
967 of call statements, also set location of a load to a temporary.
969 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
971 * config/rl78/rl78.c (add_vector_labels): New function.
972 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
973 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
974 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
975 which checks that no arguments are passed.
976 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
977 * doc/extend.texi: Documentation for the new attribute.
979 2018-02-13 Andreas Schwab <schwab@suse.de>
981 * config/riscv/linux.h (CPP_SPEC): Define.
983 2018-02-13 Jakub Jelinek <jakub@redhat.com>
986 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
987 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
988 OPTION_MASK_ISA_AES as first argument to def_builtin_const
989 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
990 instead of OPTION_MASK_ISA_PCLMUL as first argument to
991 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
992 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
993 temporarily for AES and PCLMUL builtins.
995 PR tree-optimization/84339
996 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
997 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
1001 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
1002 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
1003 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
1005 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
1007 * omp-simd-clone.h: New file.
1008 * omp-simd-clone.c: Include omp-simd-clone.h.
1009 (expand_simd_clones): No longer static.
1010 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
1011 cgraph.h and omp-simd-clone.h.
1012 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
1013 (vect_recog_widen_shift_pattern): Formatting fix.
1014 (vect_pattern_recog_1): Don't check optab for calls.
1017 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
1018 operands[2] into a REG before using gen_lowpart on it.
1020 2018-02-12 Jeff Law <law@redhat.com>
1023 * config/sh/sh.c (find_barrier): Consider a sibling call
1026 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
1027 successfully back substituting a reg.
1029 2018-02-12 Richard Biener <rguenther@suse.de>
1031 PR tree-optimization/84037
1032 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
1033 parameter, move visited init to caller.
1034 (vect_slp_analyze_operations): Separate cost from validity
1035 check, initialize visited once for all instances.
1036 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
1038 * tree-vect-stmts.c (vect_model_simple_cost): Make early
1040 (vect_model_promotion_demotion_cost): Likewise.
1041 (vectorizable_bswap): Guard cost modeling with !slp_node
1042 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
1044 (vectorizable_call): Likewise.
1045 (vectorizable_conversion): Likewise.
1046 (vectorizable_assignment): Likewise.
1047 (vectorizable_shift): Likewise.
1048 (vectorizable_operation): Likewise.
1049 (vectorizable_store): Likewise.
1050 (vectorizable_load): Likewise.
1051 (vectorizable_condition): Likewise.
1052 (vectorizable_comparison): Likewise.
1054 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
1057 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
1058 (ASAN_MARK): Fix fnspec to account for return value, change pointer
1059 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
1061 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1064 * params.def (inline-min-speedup): Increase from 8 to 15.
1065 (max-inline-insns-auto): Decrease from 40 to 30.
1066 * ipa-split.c (consider_split): Add some buffer for function to
1067 be considered inlining candidate.
1068 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
1071 2018-02-12 Richard Biener <rguenther@suse.de>
1073 PR tree-optimization/84037
1074 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
1075 matched stmts if we cannot swap the non-matched ones.
1077 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
1079 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
1080 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
1081 _mm_maskz_scalef_round_ss): New intrinsics.
1082 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
1083 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
1084 __builtin_ia32_scalefss_round): Remove.
1085 (__builtin_ia32_scalefsd_mask_round,
1086 __builtin_ia32_scalefss_mask_round): New intrinsics.
1087 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
1088 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
1089 ((match_operand:VF_128 2 "<round_nimm_predicate>"
1090 "<round_constraint>")): Changed to ...
1091 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
1092 "<round_scalar_constraint>")): ... this.
1093 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
1094 %0, %1, %2<round_op3>}"): Changed to ...
1095 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
1096 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
1097 %2<round_scalar_mask_op3>}"): ... this.
1098 * config/i386/subst.md (round_scalar_nimm_predicate): New.
1100 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
1102 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
1103 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
1104 (_mm_maskz_sqrt_round_ss): New intrinsics.
1105 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
1106 (__builtin_ia32_sqrtsd_mask_round)
1107 (__builtin_ia32_sqrtss_mask_round): New builtins.
1108 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
1109 (__builtin_ia32_sqrtss_round): Remove.
1110 (__builtin_ia32_sqrtsd_mask_round)
1111 (__builtin_ia32_sqrtss_mask_round): New builtins.
1112 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
1113 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
1114 ((match_operand:VF_128 1 "vector_operand"
1115 "xBm,<round_constraint>")): Changed to ...
1116 ((match_operand:VF_128 1 "vector_operand"
1117 "xBm,<round_scalar_constraint>")): ... this.
1118 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
1119 %0, %2, %<iptr>1<round_op3>}): Changed to ...
1120 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
1121 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
1122 %<iptr>1<round_scalar_mask_op3>}): ... this.
1123 ((set_attr "prefix" "<round_prefix>")): Changed to ...
1124 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
1126 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
1129 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
1130 Cast vec_cmpeq result to correct type.
1131 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
1132 Cast vec_cmpgt result to correct type.
1134 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
1136 * final.c (final_scan_insn_1): Renamed from...
1137 (final_scan_insn): ... this. New wrapper, to recover
1138 seen from the outermost call in recursive ones.
1139 * config/sparc/sparc.c (output_return): Drop seen from call.
1140 (output_sibcall): Likewise.
1141 * config/visium/visium.c (output_branch): Likewise.
1143 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
1145 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
1148 2018-02-10 Alan Modra <amodra@gmail.com>
1151 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
1152 Specify LR as an input.
1154 2018-02-10 Jakub Jelinek <jakub@redhat.com>
1157 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
1158 remove_member_access_dummy_vars): New functions.
1159 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
1160 lower_omp_1, execute_lower_omp): Use them.
1162 PR rtl-optimization/84308
1163 * shrink-wrap.c (spread_components): Release todo vector.
1165 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
1167 PR rtl-optimization/57193
1168 * ira-color.c (struct allocno_color_data): Add member
1169 conflict_allocno_hard_prefs.
1170 (update_conflict_allocno_hard_prefs): New.
1171 (bucket_allocno_compare_func): Add a preference based on
1172 conflict_allocno_hard_prefs.
1173 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
1174 (color_allocnos): Remove a dead code. Initiate
1175 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
1177 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1180 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
1181 constraint from =wa to wa. Avoid a subreg on the output operand,
1182 instead use a pseudo and subreg it in a move.
1183 (p9_xxbrd_<mode>): Changed to ...
1184 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
1185 (p9_xxbrd_v2df): New expander.
1186 (p9_xxbrw_<mode>): Changed to ...
1187 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
1188 (p9_xxbrw_v4sf): New expander.
1190 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1192 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
1194 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
1197 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
1198 multiply in 32-bit mode.
1199 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
1200 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
1203 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1205 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
1206 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
1207 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
1208 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
1210 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
1213 * dwarf2out.c (is_trivial_indirect_ref): New function.
1214 (dwarf2out_late_global_decl): Do not generate a location
1215 attribute for variables that have a non-trivial DECL_VALUE_EXPR
1216 and that are not defined in the current unit.
1218 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1220 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
1221 instead of a libcall for UNORDERED.
1223 2018-02-09 Tamar Christina <tamar.christina@arm.com>
1226 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
1227 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
1229 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1232 * config/s390/s390.c (s390_set_current_function): Invoke
1233 s390_indirect_branch_settings also if fndecl didn't change.
1235 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1237 * config/rs6000/rs6000.md (blockage): Set length to zero.
1239 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1241 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
1243 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1246 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
1247 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
1251 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
1252 PARALLEL incoming that failed vt_get_decl_and_offset check.
1255 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
1256 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
1258 (get_variable_section): For decls in named .bss* sections pass true as
1259 second argument to bss_initializer_p.
1261 2018-02-09 Marek Polacek <polacek@redhat.com>
1262 Jakub Jelinek <jakub@redhat.com>
1265 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
1266 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
1267 Sync some changes from cxx_fold_indirect_ref.
1269 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1271 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
1273 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
1274 (BLOCK_INLINE_ENTRY_LABEL): New.
1275 (dwarf2out_var_location): Disregard inline entry markers.
1276 (inline_entry_data): New struct.
1277 (inline_entry_data_hasher): New hashtable type.
1278 (inline_entry_data_hasher::hash): New.
1279 (inline_entry_data_hasher::equal): New.
1280 (inline_entry_data_table): New variable.
1281 (add_high_low_attributes): Add DW_AT_entry_pc and
1282 DW_AT_GNU_entry_view attributes if a pending entry is found
1283 in inline_entry_data_table. Add old entry_pc attribute only
1284 if debug nonbinding markers are disabled.
1285 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
1286 markers are enabled.
1287 (block_within_block_p, dwarf2out_inline_entry): New.
1288 (dwarf2out_finish): Check that no entries remained in
1289 inline_entry_data_table.
1290 * final.c (reemit_insn_block_notes): Handle inline entry notes.
1291 (final_scan_insn, notice_source_line): Likewise.
1292 (rest_of_clean_state): Skip inline entry markers.
1293 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
1295 * gimple.c (gimple_build_debug_inline_entry): New.
1296 * gimple.h (enum gimple_debug_subcode): Add
1297 GIMPLE_DEBUG_INLINE_ENTRY.
1298 (gimple_build_debug_inline_entry): Declare.
1299 (gimple_debug_inline_entry_p): New.
1300 (gimple_debug_nonbind_marker_p): Adjust.
1301 * insn-notes.def (INLINE_ENTRY): New.
1302 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
1303 inline entry marker notes.
1304 (print_insn): Likewise.
1305 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
1306 (INSN_DEBUG_MARKER_KIND): Likewise.
1307 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
1308 * tree-inline.c (expand_call_inline): Build and insert
1309 debug_inline_entry stmt.
1310 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
1311 inline entry blocks early, if nonbind markers are enabled.
1312 (dump_scope_block): Dump fragment info.
1313 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
1314 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
1315 (gimple_build_debug_inline_entry): New.
1316 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
1317 Enable/disable inline entry points too.
1318 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
1319 (DEBUG_INSN): Describe inline entry markers.
1321 * common.opt (gvariable-location-views): New.
1322 (gvariable-location-views=incompat5): New.
1323 * config.in: Rebuilt.
1324 * configure: Rebuilt.
1325 * configure.ac: Test assembler for view support.
1326 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
1327 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
1328 * dwarf2out.c (var_loc_view): New typedef.
1329 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
1330 (dwarf2out_locviews_in_attribute): New.
1331 (dwarf2out_locviews_in_loclist): New.
1332 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
1333 (enum dw_line_info_opcode): Add LI_adv_address.
1334 (struct dw_line_info_table): Add view.
1335 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
1336 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
1337 (zero_view_p): New variable.
1338 (ZERO_VIEW_P): New macro.
1339 (output_asm_line_debug_info): New.
1340 (struct var_loc_node): Add view.
1341 (add_AT_view_list, AT_loc_list): New.
1342 (add_var_loc_to_decl): Add view param. Test it against last.
1343 (new_loc_list): Add view params. Record them.
1344 (AT_loc_list_ptr): Handle loc and view lists.
1345 (view_list_to_loc_list_val_node): New.
1346 (print_dw_val): Handle dw_val_class_view_list.
1347 (size_of_die): Likewise.
1348 (value_format): Likewise.
1349 (loc_list_has_views): New.
1350 (gen_llsym): Set vl_symbol too.
1351 (maybe_gen_llsym, skip_loc_list_entry): New.
1352 (dwarf2out_maybe_output_loclist_view_pair): New.
1353 (output_loc_list): Output view list or entries too.
1354 (output_view_list_offset): New.
1355 (output_die): Handle dw_val_class_view_list.
1356 (output_dwarf_version): New.
1357 (output_compilation_unit_header): Use it.
1358 (output_skeleton_debug_sections): Likewise.
1359 (output_rnglists, output_line_info): Likewise.
1360 (output_pubnames, output_aranges): Update version comments.
1361 (output_one_line_info_table): Output view numbers in asm comments.
1362 (dw_loc_list): Determine current endview, pass it to new_loc_list.
1363 Call maybe_gen_llsym.
1364 (loc_list_from_tree_1): Adjust.
1365 (add_AT_location_description): Create view list attribute if
1366 needed, check it's absent otherwise.
1367 (convert_cfa_to_fb_loc_list): Adjust.
1368 (maybe_emit_file): Call output_asm_line_debug_info for test.
1369 (dwarf2out_var_location): Reset views as needed. Precompute
1370 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
1371 attribute. Set view.
1372 (new_line_info_table): Reset next view.
1373 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
1374 (dwarf2out_source_line): Likewise. Output view resets and labels to
1375 the assembler, or select appropriate line info opcodes.
1376 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
1377 (optimize_string_length): Catch it. Adjust.
1378 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
1379 dw_val_class_view_list, and remove it if no longer needed.
1380 (hash_loc_list): Hash view numbers.
1381 (loc_list_hasher::equal): Compare them.
1382 (optimize_location_lists): Check whether a view list symbol is
1383 needed, and whether the locview attribute is present, and
1384 whether they match. Remove the locview attribute if no longer
1386 (index_location_lists): Call skip_loc_list_entry for test.
1387 (dwarf2out_finish): Call output_asm_line_debug_info for test.
1388 Use output_dwarf_version.
1389 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
1390 (struct dw_val_node): Add val_view_list.
1391 * final.c (SEEN_NEXT_VIEW): New.
1392 (set_next_view_needed): New.
1393 (clear_next_view_needed): New.
1394 (maybe_output_next_view): New.
1395 (final_start_function): Rename to...
1396 (final_start_function_1): ... this. Take pointer to FIRST,
1397 add SEEN parameter. Emit param bindings in the initial view.
1398 (final_start_function): Reintroduce SEEN-less interface.
1399 (final): Rename to...
1400 (final_1): ... this. Take SEEN parameter. Output final pending
1401 next view at the end.
1402 (final): Reintroduce seen-less interface.
1403 (final_scan_insn): Output pending next view before switching
1404 sections or ending a block. Mark the next view as needed when
1405 outputting variable locations. Notify debug backend of section
1406 changes, and of location view changes.
1407 (rest_of_handle_final): Adjust.
1408 * toplev.c (process_options): Autodetect value for debug variable
1409 location views option. Warn on incompat5 without -gdwarf-5.
1410 * doc/invoke.texi (gvariable-location-views): New.
1411 (gvariable-location-views=incompat5): New.
1412 (gno-variable-location-views): New.
1414 2018-02-08 David Malcolm <dmalcolm@redhat.com>
1416 PR tree-optimization/84136
1417 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
1418 that the result of find_edge is non-NULL.
1420 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1423 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
1424 storing integer register in SImode. Fix cost of 256 and 512
1425 byte aligned SSE register store.
1427 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1429 * config/i386/i386.c (ix86_multiplication_cost): Fix
1430 multiplication cost for TARGET_AVX512DQ.
1432 2018-02-08 Marek Polacek <polacek@redhat.com>
1434 PR tree-optimization/84238
1435 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
1438 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1440 PR tree-optimization/84265
1441 * tree-vect-stmts.c (vectorizable_store): Don't treat
1442 VMAT_CONTIGUOUS accesses as grouped.
1443 (vectorizable_load): Likewise.
1445 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1447 PR tree-optimization/81635
1448 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
1449 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
1450 (test_round_for_mask): New functions.
1451 (wide_int_cc_tests): Call test_round_for_mask.
1452 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
1453 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
1454 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
1455 range returned by get_range_info.
1457 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1460 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
1461 * symtab.c: Include builtins.h
1462 (symtab_node::output_to_lto_symbol_table_p): Move here
1463 from lto-streamer-out.c:output_symbol_p.
1464 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
1465 (output_symbol_p): Move all logic to symtab.c
1466 (produce_symtab): Update.
1468 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1470 * config/s390/s390-opts.h (enum indirect_branch): Define.
1471 * config/s390/s390-protos.h (s390_return_addr_from_memory)
1472 (s390_indirect_branch_via_thunk)
1473 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
1474 (enum s390_indirect_branch_type): Define.
1475 * config/s390/s390.c (struct s390_frame_layout, struct
1476 machine_function): Remove.
1477 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
1478 (indirect_branch_table_label_no, indirect_branch_table_name):
1480 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
1481 (enum s390_indirect_branch_option): Define.
1482 (s390_return_addr_from_memory): New function.
1483 (s390_handle_string_attribute): New function.
1484 (s390_attribute_table): Add new attribute handler.
1485 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
1486 (s390_indirect_branch_via_thunk): New function.
1487 (s390_indirect_branch_via_inline_thunk): New function.
1488 (s390_function_ok_for_sibcall): When jumping via thunk disallow
1489 sibling call optimization for non z10 compiles.
1490 (s390_emit_call): Force indirect branch target to be a single
1491 register. Add r1 clobber for non-z10 compiles.
1492 (s390_emit_epilogue): Emit return jump via return_use expander.
1493 (s390_reorg): Handle JUMP_INSNs as execute targets.
1494 (s390_option_override_internal): Perform validity checks for the
1495 new command line options.
1496 (s390_indirect_branch_attrvalue): New function.
1497 (s390_indirect_branch_settings): New function.
1498 (s390_set_current_function): Invoke s390_indirect_branch_settings.
1499 (s390_output_indirect_thunk_function): New function.
1500 (s390_code_end): Implement target hook.
1501 (s390_case_values_threshold): Implement target hook.
1502 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
1504 * config/s390/s390.h (struct s390_frame_layout)
1505 (struct machine_function): Move here from s390.c.
1506 (TARGET_INDIRECT_BRANCH_NOBP_RET)
1507 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
1508 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
1509 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
1510 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
1511 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
1512 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
1513 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
1514 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
1515 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
1516 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
1517 (mnemonic attribute): Add values which aren't recognized
1519 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
1520 pattern for branch conversion. Fix mnemonic attribute.
1521 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
1522 indirect branch via thunk if requested.
1523 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
1524 ("*indirect_jump"): Disable for branch conversion using out of
1526 ("indirect_jump_via_thunk<mode>_z10")
1527 ("indirect_jump_via_thunk<mode>")
1528 ("indirect_jump_via_inlinethunk<mode>_z10")
1529 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
1530 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
1531 ("casesi_jump_via_inlinethunk<mode>_z10")
1532 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
1533 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
1534 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
1535 ("*indirect2_jump"): Disable for branch conversion.
1536 ("casesi_jump"): Turn into expander and expand patterns for branch
1538 ("return_use"): New expander.
1539 ("*return"): Emit return via thunk and rename it to ...
1540 ("*return<mode>"): ... this one.
1541 * config/s390/s390.opt: Add new options and and enum for the
1544 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1546 * lra-constraints.c (match_reload): Unconditionally use
1547 gen_lowpart_SUBREG, rather than selecting between that
1548 and equivalent gen_rtx_SUBREG code.
1550 2018-02-08 Richard Biener <rguenther@suse.de>
1552 PR tree-optimization/84233
1553 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
1554 changed flag instead of boguously re-using phi_inserted.
1556 2018-02-08 Martin Jambor <mjambor@suse.cz>
1558 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
1559 static local variables.
1561 2018-02-08 Richard Biener <rguenther@suse.de>
1563 PR tree-optimization/84278
1564 * tree-vect-stmts.c (vectorizable_store): When looking for
1565 smaller vector types to perform grouped strided loads/stores
1566 make sure the mode is supported by the target.
1567 (vectorizable_load): Likewise.
1569 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1571 * config/aarch64/aarch64.c (aarch64_components_for_bb):
1572 Increase LDP/STP opportunities by adding adjacent callee-saves.
1574 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1576 PR rtl-optimization/84068
1577 PR rtl-optimization/83459
1578 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
1580 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
1582 PR tree-optimization/84224
1583 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
1584 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
1587 2018-02-07 Iain Sandoe <iain@codesourcery.com>
1590 * config/rs6000/altivec.md (*restore_world): Remove LR use.
1591 * config/rs6000/predicates.md (restore_world_operation): Adjust op
1592 count, remove one USE.
1594 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
1596 * doc/install.texi (Configuration): Document the
1597 --with-long-double-format={ibm,ieee} PowerPC configuration
1601 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
1602 Convert from define_expand to be define_insn_and_split. Rework
1603 float/double/_Float128 conversions to QI/HI/SImode to work with
1604 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
1605 conversions to QI/HImode types did a store and then a load to
1606 truncate the value. For conversions to VSX registers, don't split
1607 the insn, instead emit the code directly. Use the code iterator
1608 any_fix to combine signed and unsigned conversions.
1609 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
1610 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
1611 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1612 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
1613 (fix_<mode>di2_hw): Likewise.
1614 (fixuns_<mode>di2_hw): Likewise.
1615 (fix_<mode>si2_hw): Likewise.
1616 (fixuns_<mode>si2_hw): Likewise.
1617 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
1618 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1619 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1620 fix<uns>_trunc<SFDF:mode>si2_p8.
1621 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1623 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1624 (fix<uns>_<mode>_mem): Likewise.
1625 (fctiw<u>z_<mode>_mem): Likewise.
1626 (fix<uns>_<mode>_mem): Likewise.
1627 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1628 the register allocator from doing a direct move to the GPRs to do
1629 a store, and instead use the ISA 3.0 store byte/half-word from
1630 vector register instruction. For IEEE 128-bit floating point,
1631 also optimize stores of 32-bit ints.
1632 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1634 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1636 * genextract.c (push_pathstr_operand): New function to support
1638 (walk_rtx): Call push_pathstr_operand.
1639 (print_path): Support [a-zA-Z].
1641 2018-02-07 Richard Biener <rguenther@suse.de>
1643 PR tree-optimization/84037
1644 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1645 (cse_and_gimplify_to_preheader): Declare.
1646 (vect_get_place_in_interleaving_chain): Likewise.
1647 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1649 (_loop_vec_info::~_loop_vec_info): Delete it.
1650 (cse_and_gimplify_to_preheader): New function.
1651 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1652 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1653 (vectorizable_load): Likewise. For grouped stores always base
1654 the IV on the first element.
1655 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1656 condition before gimplifying.
1658 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1660 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1661 *DIV_EXPR and *MOD_EXPR.
1663 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1666 * config/i386/i386.c (ix86_option_override_internal): Mask out
1667 the CF_SET bit when checking -fcf-protection.
1669 2018-02-07 Tom de Vries <tom@codesourcery.com>
1672 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1675 2018-02-07 Richard Biener <rguenther@suse.de>
1677 PR tree-optimization/84204
1678 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1681 PR tree-optimization/84205
1682 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1683 special-case isl_ast_op_zdiv_r.
1685 PR tree-optimization/84223
1686 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1687 Only add conditions from within the region.
1688 (gather_bbs::after_dom_children): Adjust.
1690 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1693 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1694 * config/avr/avr.md: Only post-reload split REG-REG moves if
1695 either register is REGERAL_REG_P.
1697 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1699 PR tree-optimization/84235
1700 * tree-ssa-scopedtables.c
1701 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1702 if the subtraction is performed in floating point type where NaNs are
1703 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1704 build 1. Formatting fix.
1706 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1709 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1710 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1711 and skip it regardless of bb boundaries. Use CALL_P macro,
1712 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1713 unnecessarily, formatting fix.
1715 2018-02-06 Michael Collison <michael.collison@arm.com>
1717 * config/arm/thumb2.md:
1718 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1719 (*thumb_mov_notscc): Ditto.
1721 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1724 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1725 unsigned_fix, not "s".
1727 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1729 * configure.ac (gcc_fn_eh_frame_ro): New function.
1730 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1731 correct .eh_frame permissions.
1732 * configure: Regenerate.
1734 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1736 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1739 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1741 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1742 Display warning message for -mno-speculate-indirect-jumps.
1744 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1746 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1748 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1750 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1752 PR tree-optimization/84225
1753 * tree-eh.c (find_trapping_overflow): Only call
1754 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1756 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1759 * config/i386/i386.c: Reimplement the check of possible options
1760 -mibt/-mshstk conbination. Change error messages.
1761 * doc/invoke.texi: Fix a typo: remove extra '='.
1763 2018-02-06 Marek Polacek <polacek@redhat.com>
1765 PR tree-optimization/84228
1766 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1768 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1771 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1772 emitted arch directives.
1773 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1774 __ARM_FEATURE_COPROC before changing architectures.
1776 2018-02-06 Richard Biener <rguenther@suse.de>
1778 * config/i386/i386.c (print_reg): Fix typo.
1779 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1781 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1783 * configure: Regenerate.
1785 2018-02-05 Martin Sebor <msebor@redhat.com>
1787 PR tree-optimization/83369
1788 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1791 2018-02-05 Martin Liska <mliska@suse.cz>
1793 * doc/invoke.texi: Cherry-pick upstream r323995.
1795 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1797 * ira.c (ira_init_register_move_cost): Adjust comment.
1799 2018-02-05 Martin Liska <mliska@suse.cz>
1801 PR gcov-profile/84137
1802 * doc/gcov.texi: Fix typo in documentation.
1804 2018-02-05 Martin Liska <mliska@suse.cz>
1806 PR gcov-profile/83879
1807 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1808 using dlopen functionality.
1810 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1812 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1813 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1814 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1815 _mm_maskz_range_round_ss): New intrinsics.
1816 (__builtin_ia32_rangesd128_round)
1817 (__builtin_ia32_rangess128_round): Remove.
1818 (__builtin_ia32_rangesd128_mask_round,
1819 __builtin_ia32_rangess128_mask_round): New builtins.
1820 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1821 __builtin_ia32_rangess128_round): Remove.
1822 (__builtin_ia32_rangesd128_mask_round,
1823 __builtin_ia32_rangess128_mask_round): New builtins.
1824 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1825 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1826 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1827 "<round_saeonly_constraint>")): Changed to ...
1828 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1829 "<round_saeonly_scalar_constraint>")): ... this.
1830 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1831 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1832 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1833 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1834 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1836 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1838 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1840 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1841 Remove all values except native, 8540 and 8548.
1843 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1845 * config/i386/i386.c (ix86_output_function_return): Pass
1846 INVALID_REGNUM, instead of -1, as invalid register number to
1847 indirect_thunk_name and output_indirect_thunk.
1849 2018-02-02 Julia Koval <julia.koval@intel.com>
1851 * config.gcc: Add -march=icelake.
1852 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1853 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1854 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1855 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1856 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1857 (processor_target_table): Add icelake.
1858 (ix86_option_override_internal): Handle new PTAs.
1859 (get_builtin_code_for_version): Handle icelake.
1860 (M_INTEL_COREI7_ICELAKE): New.
1861 (fold_builtin_cpu): Handle icelake.
1862 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1863 * doc/invoke.texi: Add -march=icelake.
1865 2018-02-02 Julia Koval <julia.koval@intel.com>
1867 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1868 to wide_int_bitmask.
1869 * wide-int-bitmask.h: New.
1871 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1874 * config/i386/i386.md: Replace Pmode with word_mode in
1875 builtin_setjmp_setup and builtin_longjmp to support x32.
1877 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1881 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1883 (rs6000_supported_cpu_names): New static variable.
1884 (linux_cpu_translation_table): Likewise.
1885 (elf_platform) <cpu>: Define new static variable and use it.
1886 Translate kernel AT_PLATFORM name to canonical name if needed.
1887 Error if platform name is unknown.
1889 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1892 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1894 2018-02-01 Jeff Law <law@redhat.com>
1897 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1898 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1899 the scratch if RELEASE_VIA_POP is false.
1900 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1901 If we have to save a temporary register, decrement SIZE appropriately.
1902 Pass new arguments to release_scratch_register_on_entry.
1903 (ix86_adjust_stack_and_probe): Likewise.
1904 (ix86_emit_probe_stack_range): Pass new arguments to
1905 release_scratch_register_on_entry.
1907 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1909 PR rtl-optimization/84157
1910 * combine.c (change_zero_ext): Use REG_P predicate in
1911 front of HARD_REGISTER_P predicate.
1913 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1915 * config/avr/avr.c (avr_option_override): Move disabling of
1916 -fdelete-null-pointer-checks to...
1917 * common/config/avr/avr-common.c (avr_option_optimization_table):
1920 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1922 PR tree-optimization/81635
1923 * tree-data-ref.c (split_constant_offset_1): For types that
1924 wrap on overflow, try to use range info to prove that wrapping
1927 2018-02-01 Renlin Li <renlin.li@arm.com>
1930 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1932 (aarch64_register_move_cost): Likewise.
1933 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1935 (REG_CLASS_NAMES): Likewise.
1936 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1937 TAILCALL_ADDR_REGS. Remove IP registers.
1938 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1940 2018-02-01 Richard Biener <rguenther@suse.de>
1942 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1943 for specifying RPO order and allow NULL for that.
1944 * domwalk.c (dom_walker::dom_walker): Likewise.
1945 (dom_walker::walk): Handle NULL RPO order.
1946 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1948 (rewrite_update_dom_walker): Likewise.
1949 (mark_def_dom_walker): Likewise.
1951 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1953 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1954 (aarch64_maybe_expand_sve_subreg_move): Declare.
1955 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1956 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1958 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1959 that are semantically a reverse operation.
1960 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1961 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1962 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1964 (aarch64_can_change_mode_class): For big-endian, forbid changes
1965 between two SVE modes if they have different element sizes.
1967 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1969 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1970 the TImode handling for big-endian targets.
1972 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1974 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1975 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1977 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1978 Remove BSWAP handing for big-endian targets and use the form of
1979 LD1RQ appropariate for the mode.
1981 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1983 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1984 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1987 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1990 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1991 check for operands that need to go through aarch64_sve_reload_be.
1993 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1995 PR tree-optimization/81661
1996 PR tree-optimization/84117
1997 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1998 * tree-eh.c: Include gimplify.h.
1999 (find_trapping_overflow, replace_trapping_overflow,
2000 rewrite_to_non_trapping_overflow): New functions.
2001 * tree-vect-loop.c: Include tree-eh.h.
2002 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
2003 * tree-data-ref.c: Include tree-eh.h.
2004 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
2006 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
2008 PR rtl-optimization/84123
2009 * combine.c (change_zero_ext): Check if hard register satisfies
2010 can_change_dest_mode before calling gen_lowpart_SUBREG.
2012 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
2015 * ira.c (ira_init_register_move_cost): Remove assert.
2017 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2019 PR rtl-optimization/84071
2020 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
2021 * doc/tm.texi: Regenerate.
2023 2018-01-31 Richard Biener <rguenther@suse.de>
2025 PR tree-optimization/84132
2026 * tree-data-ref.c (analyze_miv_subscript): Properly
2027 check whether evolution_function_is_affine_multivariate_p
2028 before calling gcd_of_steps_may_divide_p.
2030 2018-01-31 Julia Koval <julia.koval@intel.com>
2033 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
2034 * config/i386/i386.md (rdpid_rex64) New.
2035 (rdpid): Make 32bit only.
2037 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
2040 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
2041 an IDENTIFIER_NODE for FUNCTION_TYPE's.
2043 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2046 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2048 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2050 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2052 PR rtl-optimization/84071
2053 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
2054 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
2056 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
2058 * config/arc/arc.c (arc_handle_aux_attribute): New function.
2059 (arc_attribute_table): Add 'aux' attribute.
2060 (arc_in_small_data_p): Consider aux like variables.
2061 (arc_is_aux_reg_p): New function.
2062 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
2063 (arc_get_aux_arg): New function.
2064 (prepare_move_operands): Handle aux-register access.
2065 (arc_handle_aux_attribute): New function.
2066 * doc/extend.texi (ARC Variable attributes): Add subsection.
2068 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
2070 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
2071 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
2072 (arc_attribute_table): Add 'uncached' attribute.
2073 (arc_print_operand): Print '.di' flag for uncached memory
2075 (arc_in_small_data_p): Do not consider for small data the uncached
2077 (arc_is_uncached_mem_p): New function.
2078 * config/arc/predicates.md (compact_store_memory_operand): Check
2079 for uncached memory accesses.
2080 (nonvol_nonimm_operand): Likewise.
2081 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
2083 2018-01-31 Jakub Jelinek <jakub@redhat.com>
2086 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
2087 falign-loops=): Add Optimization flag.
2089 2018-01-30 Jeff Law <law@redhat.com>
2092 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
2093 INT_REGISTERS_SAVED. Check it prior to calling
2094 get_scratch_register_on_entry.
2095 (ix86_adjust_stack_and_probe): Similarly.
2096 (ix86_emit_probe_stack_range): Similarly.
2097 (ix86_expand_prologue): Corresponding changes.
2099 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2102 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
2103 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
2105 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
2108 * lra-constraints.c (curr_insn_transform): Process AND in the
2111 2018-01-30 Jakub Jelinek <jakub@redhat.com>
2113 PR rtl-optimization/83986
2114 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
2115 dependence against last_pending_memory_flush in addition to
2118 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
2120 PR tree-optimization/81611
2121 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
2124 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2127 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
2130 2018-01-30 Richard Biener <rguenther@suse.de>
2131 Jakub Jelinek <jakub@redhat.com>
2133 PR tree-optimization/84111
2134 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
2135 inner loops added during recursion, as they don't have up-to-date
2138 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2141 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
2142 (can_inline_edge_by_limits_p): ... here.
2143 (can_early_inline_edge_p, check_callers,
2144 update_caller_keys, update_callee_keys, recursive_inlining,
2145 add_new_edges_to_heap, speculation_useful_p,
2146 inline_small_functions,
2147 inline_small_functions, flatten_function,
2148 inline_to_all_callers_1): Update.
2150 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2152 * profile-count.c (profile_count::combine_with_ipa_count): Handle
2155 2018-01-30 Richard Biener <rguenther@suse.de>
2157 PR tree-optimization/83008
2158 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
2159 invariant and constant vector uses in stmts when they need
2162 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2165 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
2166 * configure: Regenerate.
2168 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2170 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
2172 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
2173 Use gen_rtx_REG rather than gen_lowpart.
2175 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2177 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
2178 rather than 0 when creating partial subregs.
2180 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2182 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
2185 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
2188 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
2189 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
2190 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
2191 flags. This restores the settings used before the 2017-07-24.
2192 Turning off pre increment/decrement/modify allows IVOPTS to
2193 optimize DF/SF loops where the index is an int.
2195 2018-01-29 Richard Biener <rguenther@suse.de>
2196 Kelvin Nilsen <kelvin@gcc.gnu.org>
2199 * tree-vect-stmts.c (vectorizable_call): Don't call
2200 targetm.vectorize_builtin_md_vectorized_function if callee is
2203 2018-01-22 Carl Love <cel@us.ibm.com>
2205 * doc/extend.tex: Fix typo in second arg in
2206 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
2208 2018-01-29 Richard Biener <rguenther@suse.de>
2210 PR tree-optimization/84086
2211 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
2212 (flush_ssaname_freelist): When SSA names were released reset
2213 the SCEV hash table.
2215 2018-01-29 Richard Biener <rguenther@suse.de>
2217 PR tree-optimization/84057
2218 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
2219 removed paths when removing edges.
2221 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
2223 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
2224 -mfunction-return=@var{choice}.
2226 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
2229 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
2231 (layout::print_source_line): Likewise.
2232 (test_get_line_width_without_trailing_whitespace): Add test cases.
2234 2018-01-27 Jakub Jelinek <jakub@redhat.com>
2237 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
2240 2018-01-26 Jim Wilson <jimw@sifive.com>
2242 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
2244 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
2247 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2249 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
2250 and CMP + SUB-immediate -> SUBS.
2252 2018-01-26 Martin Sebor <msebor@redhat.com>
2254 PR tree-optimization/83896
2255 * tree-ssa-strlen.c (get_string_len): Rename...
2256 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
2257 Avoid assuming length is constant.
2258 (handle_char_store): Use HOST_WIDE_INT for string length.
2260 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
2263 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
2264 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
2266 2018-01-26 Richard Biener <rguenther@suse.de>
2268 PR rtl-optimization/84003
2269 * dse.c (record_store): Only record redundant stores when
2270 the earlier store aliases at least all accesses the later one does.
2272 2018-01-26 Jakub Jelinek <jakub@redhat.com>
2274 PR rtl-optimization/83985
2275 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
2276 REG_CFA_RESTORE insns.
2277 (delete_unmarked_insns): Don't ignore separate shrink wrapping
2278 REG_CFA_RESTORE insns here.
2281 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
2282 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
2284 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2286 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
2287 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
2288 (arc_init): Likewise.
2289 (arc_override_options): Likewise.
2290 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
2292 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
2294 * config/arc/arc.h (TARGET_DBNZ): Define.
2295 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
2296 properly set the tune attribute.
2297 (dbnz): Use TARGET_DBNZ guard.
2298 * config/arc/arc.opt (mtune): Add core3 option.
2300 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2302 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
2303 recognize new pic like addresses.
2304 (arc_delegitimize_address): Clean up.
2306 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2308 * config/arc/arc-arches.def: Option mrf16 valid for all
2310 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
2311 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
2312 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
2313 * config/arc/arc-tables.opt: Regenerate.
2314 * config/arc/arc.c (arc_conditional_register_usage): Handle
2315 reduced register file case.
2316 (arc_file_start): Set must have build attributes.
2317 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
2319 * config/arc/arc.opt (mrf16): Add new option.
2320 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
2321 * config/arc/genmultilib.awk: Handle new mrf16 option.
2322 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
2323 * config/arc/t-multilib: Regenerate.
2324 * doc/invoke.texi (ARC Options): Document mrf16 option.
2326 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2328 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
2329 * config/arc/arc.c (arc_handle_secure_attribute): New function.
2330 (arc_attribute_table): Add 'secure_call' attribute.
2331 (arc_print_operand): Print secure call operand.
2332 (arc_function_ok_for_sibcall): Don't optimize tail calls when
2334 (arc_is_secure_call_p): New function. * config/arc/arc.md
2335 (call_i): Add support for sjli instruction.
2336 (call_value_i): Likewise.
2337 * config/arc/constraints.md (Csc): New constraint.
2339 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2340 John Eric Martin <John.Martin@emmicro-us.com>
2342 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
2343 * config/arc/arc.c (_arc_jli_section): New struct.
2344 (arc_jli_section): New type.
2345 (rc_jli_sections): New static variable.
2346 (arc_handle_jli_attribute): New function.
2347 (arc_attribute_table): Add jli_always and jli_fixed attribute.
2348 (arc_file_end): New function.
2349 (TARGET_ASM_FILE_END): Define.
2350 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
2351 (arc_add_jli_section): New function.
2352 (jli_call_scan): Likewise.
2353 (arc_reorg): Call jli_call_scan.
2354 (arc_output_addsi): Remove 'S' from printing asm operand.
2355 (arc_is_jli_call_p): New function.
2356 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
2358 (movhi_insn): Likewise.
2359 (movsi_insn): Likewise.
2360 (movsi_set_cc_insn): Likewise.
2361 (loadqi_update): Likewise.
2362 (load_zeroextendqisi_update): Likewise.
2363 (load_signextendqisi_update): Likewise.
2364 (loadhi_update): Likewise.
2365 (load_zeroextendhisi_update): Likewise.
2366 (load_signextendhisi_update): Likewise.
2367 (loadsi_update): Likewise.
2368 (loadsf_update): Likewise.
2369 (movsicc_insn): Likewise.
2370 (bset_insn): Likewise.
2371 (bxor_insn): Likewise.
2372 (bclr_insn): Likewise.
2373 (bmsk_insn): Likewise.
2374 (bicsi3_insn): Likewise.
2375 (cmpsi_cc_c_insn): Likewise.
2376 (movsi_ne): Likewise.
2377 (movsi_cond_exec): Likewise.
2378 (clrsbsi2): Likewise.
2389 (call_i): Remove 'S' asm letter, add jli instruction.
2390 (call_value_i): Likewise.
2391 * config/arc/arc.op (mjli-always): New option.
2392 * config/arc/constraints.md (Cji): New constraint.
2393 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
2395 (subsf3_fpx): Likewise.
2396 (mulsf3_fpx): Likewise.
2397 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
2399 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
2401 * doc/invoke.texi (ARC): Document mjli-always option.
2403 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
2405 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
2406 avoid addition with 0 and use incw and decw where possible.
2408 2018-01-26 Richard Biener <rguenther@suse.de>
2410 PR tree-optimization/81082
2411 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
2412 association if it requires casting to unsigned.
2413 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
2414 from fold_plusminus_mult_expr to catch important cases late when
2415 range info is available.
2417 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2419 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
2420 * configure.ac (hidden_linkonce): New test.
2421 * configure: Regenerate.
2422 * config.in: Regenerate.
2424 2018-01-26 Julia Koval <julia.koval@intel.com>
2426 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
2427 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
2428 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
2429 _mm_mask_bitshuffle_epi64_mask): Fix type.
2430 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
2431 USI_FTYPE_V4DI_V4DI_USI): Remove.
2432 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
2433 __builtin_ia32_vpshufbitqmb256_mask,
2434 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
2435 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
2436 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
2438 2018-01-26 Alan Modra <amodra@gmail.com>
2441 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
2442 UNSPEC_VBPERMQ. Sort other unspecs.
2444 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
2446 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
2448 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
2451 * predict.c (drop_profile): Do not push/pop cfun; update also
2453 (handle_missing_profiles): Fix logic looking for zero profiles.
2455 2018-01-25 Jakub Jelinek <jakub@redhat.com>
2458 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
2459 on functions with #pragma omp declare simd or functions with simd
2461 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
2462 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
2463 Remove trailing \n from warning_at calls.
2465 2018-01-25 Tom de Vries <tom@codesourcery.com>
2468 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2469 for neutered workers.
2471 2018-01-24 Joseph Myers <joseph@codesourcery.com>
2474 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
2475 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
2477 2018-01-24 Jeff Law <law@redhat.com>
2480 * i386.c (get_probe_interval): Move to earlier point.
2481 (ix86_compute_frame_layout): If -fstack-clash-protection and
2482 the frame is larger than the probe interval, then use pushes
2483 to save registers rather than reg->mem moves.
2484 (ix86_expand_prologue): Remove conditional for int_registers_saved
2487 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
2490 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
2491 min/max for never referenced object.
2493 2018-01-24 Jakub Jelinek <jakub@redhat.com>
2496 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
2498 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
2499 attributes from DECL_ATTRIBUTES (decl) without affecting
2500 DECL_ATTRIBUTES (current_function_decl).
2501 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
2502 functions with non-NULL DECL_ABSTRACT_ORIGIN.
2504 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
2506 PR tree-optimization/83979
2507 * fold-const.c (fold_comparison): Use constant_boolean_node
2508 instead of boolean_{true,false}_node.
2510 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
2512 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
2515 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2517 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2518 Simplify the clause that sets the length attribute.
2519 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2520 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
2521 clause that sets the length attribute.
2522 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2524 2018-01-24 Tom de Vries <tom@codesourcery.com>
2527 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
2528 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
2529 Add strict parameter.
2530 (prevent_branch_around_nothing): Insert dummy insn between branch to
2531 label and label with no ptx insn inbetween.
2532 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
2534 2018-01-24 Tom de Vries <tom@codesourcery.com>
2537 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2538 for neutered threads in warp.
2539 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
2541 2018-01-24 Richard Biener <rguenther@suse.de>
2543 PR tree-optimization/83176
2544 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
2547 2018-01-24 Richard Biener <rguenther@suse.de>
2549 PR tree-optimization/82819
2550 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
2551 code generating pluses that are no-ops in the target precision.
2553 2018-01-24 Richard Biener <rguenther@suse.de>
2556 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
2558 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2560 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
2561 to merge probabilities.
2562 * predict.c (probably_never_executed): Also mark as cold functions
2563 with global 0 profile and guessed local profile.
2564 * profile-count.c (profile_probability::combine_with_count): New
2566 * profile-count.h (profile_probability::operator*,
2567 profile_probability::operator*=, profile_probability::operator/,
2568 profile_probability::operator/=): Reduce precision to adjusted
2569 and set value to guessed on contradictory divisions.
2570 (profile_probability::combine_with_freq): Remove.
2571 (profile_probability::combine_wiht_count): Declare.
2572 (profile_count::force_nonzero):: Set to adjusted.
2573 (profile_count::probability_in):: Set quality to adjusted.
2574 * tree-ssa-tail-merge.c (replace_block_by): Use
2577 2018-01-23 Andrew Waterman <andrew@sifive.com>
2578 Jim Wilson <jimw@sifive.com>
2580 * config/riscv/riscv.c (riscv_stack_boundary): New.
2581 (riscv_option_override): Set riscv_stack_boundary. Handle
2582 riscv_preferred_stack_boundary_arg.
2583 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
2584 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
2585 (STACK_BOUNDARY): Set to riscv_stack_boundary.
2586 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
2587 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
2588 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
2590 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
2593 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
2594 of struct ix86_frame.
2595 (ix86_expand_epilogue): Likewise. Add a local variable for
2596 the reg_save_offset field in struct ix86_frame.
2598 2018-01-23 Bin Cheng <bin.cheng@arm.com>
2600 PR tree-optimization/82604
2601 * tree-loop-distribution.c (enum partition_kind): New enum item
2602 PKIND_PARTIAL_MEMSET.
2603 (partition_builtin_p): Support above new enum item.
2604 (generate_code_for_partition): Ditto.
2605 (compute_access_range): Differentiate cases that equality can be
2606 proven at all loops, the innermost loops or no loops.
2607 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
2608 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
2609 (finalize_partitions, distribute_loop): Don't fuse partition of
2610 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
2611 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
2614 2018-01-23 Martin Liska <mliska@suse.cz>
2616 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
2617 order to ignore the predictor.
2618 (PRED_POLYMORPHIC_CALL): Likewise.
2619 (PRED_RECURSIVE_CALL): Likewise.
2621 2018-01-23 Martin Liska <mliska@suse.cz>
2623 * tree-profile.c (tree_profiling): Print function header to
2624 aware reader which function we are working on.
2625 * value-prof.c (gimple_find_values_to_profile): Do not print
2626 not interesting value histograms.
2628 2018-01-23 Martin Liska <mliska@suse.cz>
2630 * profile-count.h (enum profile_quality): Add
2631 profile_uninitialized as the first value. Do not number values
2632 as they are zero based.
2633 (profile_count::verify): Update sanity check.
2634 (profile_probability::verify): Likewise.
2636 2018-01-23 Nathan Sidwell <nathan@acm.org>
2638 * doc/invoke.texi (ffor-scope): Deprecate.
2640 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2642 PR tree-optimization/83510
2643 * domwalk.c (set_all_edges_as_executable): New function.
2644 (dom_walker::dom_walker): Convert bool param
2645 "skip_unreachable_blocks" to enum reachability. Move setup of
2646 edge flags to set_all_edges_as_executable and only do it when
2647 reachability is REACHABLE_BLOCKS.
2648 * domwalk.h (enum dom_walker::reachability): New enum.
2649 (dom_walker::dom_walker): Convert bool param
2650 "skip_unreachable_blocks" to enum reachability.
2651 (set_all_edges_as_executable): New decl.
2652 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2653 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2655 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2656 but converting true to REACHABLE_BLOCKS.
2657 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2659 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2660 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2661 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2663 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2664 if check_all_array_refs will be called.
2666 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2668 * tree.c (selftest::test_location_wrappers): Add more test
2671 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2673 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2674 (selftest::test_bit_in_range): Likewise.
2676 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2679 * doc/sourcebuild.texi (vect_float): Say that the selector
2680 only describes the situation when -funsafe-math-optimizations is on.
2681 (vect_float_strict): Document.
2683 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2685 PR tree-optimization/83965
2686 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2687 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2688 instead of checking only for a reduction.
2689 (vect_recog_widen_sum_pattern): Likewise.
2691 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2693 * predict.c (probably_never_executed): Only use precise profile info.
2694 (compute_function_frequency): Skip after inlining hack since we now
2695 have quality checking.
2697 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2699 * profile-count.h (profile_probability::very_unlikely,
2700 profile_probability::unlikely, profile_probability::even): Set
2701 precision to guessed.
2703 2018-01-23 Richard Biener <rguenther@suse.de>
2705 PR tree-optimization/83963
2706 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2707 Properly terminate dominator walk when crossing the exit edge not
2708 when visiting its source block.
2710 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2713 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2714 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2716 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2718 PR tree-optimization/83957
2719 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2720 semicolon after for body surrounded by braces.
2722 PR tree-optimization/83081
2723 * profile-count.h (profile_probability::split): New method.
2724 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2725 Use profile_probability::split.
2726 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2727 when splitting a single conditional jump into 2.
2729 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2731 PR tree-optimization/69452
2732 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2735 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2737 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2738 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2739 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2741 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2743 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2744 * config/rl78/rl78.md (movdi): New define_expand.
2745 * config/rl78/rl78.c (rl78_split_movdi): New function.
2747 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2750 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2752 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2753 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2754 128-bit to produce an UNSPEC move to get the double word with the
2755 signbit and then a shift directly to do signbit.
2756 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2757 implementation with a new version that just does either a direct
2758 move or a regular move. Move memory interface to separate insns.
2759 Move insns so they are next to the expander.
2760 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2761 with signbit move. Split big and little endian case.
2762 (signbit<mode>2_dm_mem_le): Likewise.
2763 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2764 (signbit<mode>2_dm2): Likewise.
2766 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2768 * config/rl78/rl78.md (anddi3): New define_expand.
2770 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2772 * config/rl78/rl78.md (umindi3): New define_expand.
2774 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2776 * config/rl78/rl78.md (smindi3): New define_expand.
2778 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2780 * config/rl78/rl78.md (smaxdi3): New define_expand.
2782 2018-01-22 Carl Love <cel@us.ibm.com>
2784 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2785 LVX_V1TI): Add macro expansion.
2786 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2787 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2788 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2789 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2790 Change check to determine if the instruction is a byte reversing
2791 entry. Fix typo in comment.
2792 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2793 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2794 Add def_builtin calls for new builtins.
2795 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2796 Add define_insn expansion.
2798 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2800 * config/rl78/rl78.md (umaxdi3): New define_expand.
2802 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2804 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2805 for non-QImode registers.
2807 2018-01-22 Richard Biener <rguenther@suse.de>
2809 PR tree-optimization/83963
2810 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2811 including the loop exit block.
2812 (scop_detection::merge_sese): Likewise.
2813 (scop_detection::add_scop): Do it here instead.
2815 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2817 * doc/sourcebuild.texi (arm_softfloat): Document.
2819 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2822 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2823 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2824 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2826 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2827 David Edelsohn <dje.gcc@gmail.com>
2830 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2831 Change "crset eq" to "crset 2".
2832 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2833 (*call_indirect_aix<mode>_nospec): Likewise.
2834 (*call_value_indirect_aix<mode>_nospec): Likewise.
2835 (*call_indirect_elfv2<mode>_nospec): Likewise.
2836 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2837 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2838 change assembly output from . to $.
2839 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2840 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2841 (*tablejump<mode>_internal1_nospec): Likewise.
2843 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2846 * config/sh/sh_optimize_sett_clrt.cc:
2847 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2849 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2851 PR tree-optimization/83940
2852 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2853 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2854 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2855 use it to pass back the definition type.
2856 (vect_check_store_rhs): Likewise rhs_dt_out.
2857 (vect_build_gather_load_calls): Add a mask_dt argument and use
2858 it instead of a call to vect_is_simple_use.
2859 (vectorizable_store): Update calls to vect_check_load_store_mask
2860 and vect_check_store_rhs. Use the dt returned by the latter instead
2861 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2862 instead of calls to vect_is_simple_use. Pass the scalar rather
2863 than the vector operand to vect_is_simple_use when handling
2864 second and subsequent copies of an rhs value.
2865 (vectorizable_load): Update calls to vect_check_load_store_mask
2866 and vect_build_gather_load_calls. Use the cached mask_dt and
2867 gs_info.offset_dt instead of calls to vect_is_simple_use.
2869 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2872 * tree-emutls.c: Include gimplify.h.
2873 (lower_emutls_2): New function.
2874 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2875 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2876 it before further processing.
2879 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2880 UINTVAL (trueop1) instead of INTVAL (op1).
2882 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2886 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2887 INCOMING_FRAME_SP_OFFSET if not defined.
2888 (scan_trace): Add ENTRY argument. If true and
2889 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2890 emit a note to adjust the CFA offset.
2891 (create_cfi_notes): Adjust scan_trace callers.
2892 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2893 INCOMING_FRAME_SP_OFFSET in the CIE.
2894 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2895 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2897 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2898 * doc/tm.texi: Regenerated.
2900 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2902 PR rtl-optimization/83147
2903 * lra-constraints.c (remove_inheritance_pseudos): Use
2904 lra_substitute_pseudo_within_insn.
2906 2018-01-19 Tom de Vries <tom@codesourcery.com>
2907 Cesar Philippidis <cesar@codesourcery.com>
2910 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2912 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2915 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2916 spaces for function labels.
2918 2018-01-19 Martin Liska <mliska@suse.cz>
2920 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2921 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2922 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2923 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2924 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2925 (PRED_CONST_RETURN): Change from 69 to 65.
2926 (PRED_NULL_RETURN): Change from 91 to 71.
2927 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2928 (PRED_LOOP_GUARD): Change from 66 to 73.
2930 2018-01-19 Martin Liska <mliska@suse.cz>
2932 * predict.c (predict_insn_def): Add new assert.
2933 (struct branch_predictor): Change type to signed integer.
2934 (test_prediction_value_range): Amend test to cover
2936 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2937 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2938 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2939 (PRED_LOOP_IV_COMPARE): Likewise.
2940 * predict.h (PROB_UNINITIALIZED): Define new constant.
2942 2018-01-19 Martin Liska <mliska@suse.cz>
2944 * predict.c (dump_prediction): Add new format for
2945 analyze_brprob.py script which is enabled with -details
2947 * profile-count.h (precise_p): New function.
2949 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2951 PR tree-optimization/83922
2952 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2953 there are no statements that need masking.
2954 (vect_active_double_reduction_p): New function.
2955 (vect_analyze_loop_operations): Use it when handling phis that
2956 are not in the loop header.
2958 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2960 PR tree-optimization/83914
2961 * tree-vect-loop.c (vectorizable_induction): Don't convert
2962 init_expr or apply the peeling adjustment for inductions
2963 that are nested within the vectorized loop.
2965 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2967 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2970 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2974 * function.h (gimplify_parameters): Add gimple_seq * argument.
2975 * function.c: Include gimple.h and options.h.
2976 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2977 for the added local temporaries if needed.
2978 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2979 if there are any parameter cleanups, wrap whole body into a
2980 try/finally with the cleanups.
2982 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2985 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2986 Use GET_MODE_CLASS for scalar floating point.
2988 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2992 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2993 Fix call of call_cgraph_insertion_hooks.
2995 2018-01-18 Martin Sebor <msebor@redhat.com>
2997 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2999 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
3002 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
3005 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
3008 * common.opt: (-ffile-prefix-map): New option.
3009 * opts.c (common_handle_option): Defer it.
3010 * opts-global.c (handle_common_deferred_options): Handle it.
3011 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
3012 * file-prefix-map.h: New file.
3013 (remap_debug_filename, add_debug_prefix_map): ...here.
3014 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
3015 * final.c (debug_prefix_map, add_debug_prefix_map
3016 remap_debug_filename): Move to...
3017 * file-prefix-map.c: New file.
3018 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
3019 generalize, get rid of alloca(), use strrchr() instead of strchr().
3020 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
3021 Implement in terms of add_prefix_map().
3022 (remap_macro_filename, remap_debug_filename): Implement in term of
3024 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
3025 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
3026 * dbxout.c: Include file-prefix-map.h.
3027 * varasm.c: Likewise.
3028 * vmsdbgout.c: Likewise.
3029 * xcoffout.c: Likewise.
3030 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
3031 * doc/cppopts.texi (-fmacro-prefix-map): Document.
3032 * doc/invoke.texi (-ffile-prefix-map): Document.
3033 (-fdebug-prefix-map): Update description.
3035 2018-01-18 Martin Liska <mliska@suse.cz>
3037 * config/i386/i386.c (indirect_thunk_name): Document that also
3039 (output_indirect_thunk): Document why both instructions
3040 (pause and lfence) are generated.
3042 2018-01-18 Richard Biener <rguenther@suse.de>
3044 PR tree-optimization/83887
3045 * graphite-scop-detection.c
3046 (scop_detection::get_nearest_dom_with_single_entry): Remove.
3047 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
3048 (scop_detection::merge_sese): Re-implement with a flood-fill
3049 algorithm that properly finds a SESE region if it exists.
3051 2018-01-18 Jakub Jelinek <jakub@redhat.com>
3054 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
3055 pointer_diff optimizations use view_convert instead of convert.
3057 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3059 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
3060 Generate different code for -mno-speculate-indirect-jumps.
3061 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
3062 (*call_indirect_aix<mode>): Disable for
3063 -mno-speculate-indirect-jumps.
3064 (*call_indirect_aix<mode>_nospec): New define_insn.
3065 (*call_value_indirect_aix<mode>): Disable for
3066 -mno-speculate-indirect-jumps.
3067 (*call_value_indirect_aix<mode>_nospec): New define_insn.
3068 (*sibcall_nonlocal_sysv<mode>): Generate different code for
3069 -mno-speculate-indirect-jumps.
3070 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
3072 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
3074 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
3075 long double type, set the flags for noting the default long double
3076 type, even if we don't pass or return a long double type.
3078 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
3081 * ipa-inline.c (flatten_function): Do not overwrite final inlining
3084 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
3086 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
3087 support for merge[hl].
3088 (fold_mergehl_helper): New helper function.
3089 (tree-vector-builder.h): New #include for tree_vector_builder usage.
3090 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
3091 (altivec_vmrglw_direct): Add xxmrglw insn.
3093 2018-01-17 Andrew Waterman <andrew@sifive.com>
3095 * config/riscv/riscv.c (riscv_conditional_register_usage): If
3096 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
3098 2018-01-17 David Malcolm <dmalcolm@redhat.com>
3101 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
3102 call the lto_location_cache before reading the
3103 DECL_SOURCE_LOCATION of the types.
3105 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
3106 Richard Sandiford <richard.sandiford@linaro.org>
3108 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
3109 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
3110 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
3111 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
3112 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
3114 * config/aarch64/constraints.md (aarch64_movti_operand):
3116 * config/aarch64/predicates.md (Uti): Add new constraint.
3118 2018-01-17 Carl Love <cel@us.ibm.com>
3120 * config/rs6000/vsx.md (define_expand xl_len_r,
3121 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
3122 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
3124 (define_expand, define_insn): Move the shift left from the
3125 define_insn to the define_expand for lxvl and stxvl instructions.
3126 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
3127 and XL_LEN_R definitions to PURE.
3129 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
3131 * config/i386/i386.c (indirect_thunk_name): Declare regno
3132 as unsigned int. Compare regno with INVALID_REGNUM.
3133 (output_indirect_thunk): Ditto.
3134 (output_indirect_thunk_function): Ditto.
3135 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
3136 in the call to output_indirect_thunk_function.
3138 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
3141 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
3142 rather than the size of inner_type to determine the stack slot size
3143 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
3145 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
3148 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
3151 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
3153 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
3154 endian Linux systems to optionally enable multilibs for selecting
3155 the long double type if the user configured an explicit type.
3156 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
3157 have no long double multilibs if not defined.
3158 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
3159 warn if the user used -mabi={ieee,ibm}longdouble and we built
3160 multilibs for long double.
3161 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
3162 appropriate multilib option.
3163 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
3165 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
3166 for building long double multilibs.
3167 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
3169 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
3171 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
3174 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
3176 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
3179 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
3182 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
3185 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3187 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
3188 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
3190 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
3192 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
3193 different rtl trees depending on TARGET_64BIT.
3194 (rs6000_gen_lvx): Likewise.
3196 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3198 * config/visium/visium.md (nop): Tweak comment.
3199 (hazard_nop): Likewise.
3201 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3203 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
3204 -mspeculate-indirect-jumps.
3205 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
3206 for -mno-speculate-indirect-jumps.
3207 (*call_indirect_elfv2<mode>_nospec): New define_insn.
3208 (*call_value_indirect_elfv2<mode>): Disable for
3209 -mno-speculate-indirect-jumps.
3210 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
3211 (indirect_jump): Emit different RTL for
3212 -mno-speculate-indirect-jumps.
3213 (*indirect_jump<mode>): Disable for
3214 -mno-speculate-indirect-jumps.
3215 (*indirect_jump<mode>_nospec): New define_insn.
3216 (tablejump): Emit different RTL for
3217 -mno-speculate-indirect-jumps.
3218 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
3219 (tablejumpsi_nospec): New define_expand.
3220 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
3221 (tablejumpdi_nospec): New define_expand.
3222 (*tablejump<mode>_internal1): Disable for
3223 -mno-speculate-indirect-jumps.
3224 (*tablejump<mode>_internal1_nospec): New define_insn.
3225 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
3228 2018-01-16 Artyom Skrobov tyomitch@gmail.com
3230 * caller-save.c (insert_save): Drop unnecessary parameter. All
3233 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3234 Richard Biener <rguenth@suse.de>
3237 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
3238 return early, inline manually is_gimple_sizepos. Make sure if we
3239 call gimplify_expr we don't end up with a gimple constant.
3240 * tree.c (variably_modified_type_p): Don't return true for
3241 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
3242 * gimplify.h (is_gimple_sizepos): Remove.
3244 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3246 PR tree-optimization/83857
3247 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
3248 vectorizable_live_operation for pure SLP statements.
3249 (vectorizable_live_operation): Handle PHIs.
3251 2018-01-16 Richard Biener <rguenther@suse.de>
3253 PR tree-optimization/83867
3254 * tree-vect-stmts.c (vect_transform_stmt): Precompute
3255 nested_in_vect_loop_p since the scalar stmt may get invalidated.
3257 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3260 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
3261 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
3262 If off is not INTEGER_CST, issue a may not be aligned warning
3263 rather than isn't aligned. Use isn%'t rather than isn't.
3264 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
3266 <case MULT_EXPR>: Improve the case when bottom and one of the
3267 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
3268 operand, in that case check if the other operand is multiple of
3269 bottom divided by the INTEGER_CST operand.
3271 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3274 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
3275 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
3276 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
3277 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
3278 * config/pa/pa.c (pa_function_arg_advance): Likewise.
3279 (pa_function_arg, pa_arg_partial_bytes): Likewise.
3280 (pa_function_arg_size): New function.
3282 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3284 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
3285 in a separate statement.
3287 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3289 PR tree-optimization/83847
3290 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
3291 group gathers and scatters.
3293 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3295 PR rtl-optimization/86620
3296 * params.def (max-sched-ready-insns): Bump minimum value to 1.
3298 PR rtl-optimization/83213
3299 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
3300 to last if both are JUMP_INSNs.
3302 PR tree-optimization/83843
3303 * gimple-ssa-store-merging.c
3304 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
3305 store_immediate_info for bswap/nop orig_stores.
3307 2018-01-15 Andrew Waterman <andrew@sifive.com>
3309 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
3311 <UDIV>: Increase cost if !TARGET_DIV.
3313 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
3315 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
3316 (define_attr "cr_logical_3op"): New.
3317 (cceq_ior_compare): Adjust.
3318 (cceq_ior_compare_complement): Adjust.
3319 (*cceq_rev_compare): Adjust.
3320 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
3321 (is_cracked_insn): Adjust.
3322 (insn_must_be_first_in_group): Adjust.
3323 * config/rs6000/40x.md: Adjust.
3324 * config/rs6000/440.md: Adjust.
3325 * config/rs6000/476.md: Adjust.
3326 * config/rs6000/601.md: Adjust.
3327 * config/rs6000/603.md: Adjust.
3328 * config/rs6000/6xx.md: Adjust.
3329 * config/rs6000/7450.md: Adjust.
3330 * config/rs6000/7xx.md: Adjust.
3331 * config/rs6000/8540.md: Adjust.
3332 * config/rs6000/cell.md: Adjust.
3333 * config/rs6000/e300c2c3.md: Adjust.
3334 * config/rs6000/e500mc.md: Adjust.
3335 * config/rs6000/e500mc64.md: Adjust.
3336 * config/rs6000/e5500.md: Adjust.
3337 * config/rs6000/e6500.md: Adjust.
3338 * config/rs6000/mpc.md: Adjust.
3339 * config/rs6000/power4.md: Adjust.
3340 * config/rs6000/power5.md: Adjust.
3341 * config/rs6000/power6.md: Adjust.
3342 * config/rs6000/power7.md: Adjust.
3343 * config/rs6000/power8.md: Adjust.
3344 * config/rs6000/power9.md: Adjust.
3345 * config/rs6000/rs64.md: Adjust.
3346 * config/rs6000/titan.md: Adjust.
3348 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3350 * config/i386/predicates.md (indirect_branch_operand): Rewrite
3351 ix86_indirect_branch_register logic.
3353 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3355 * config/i386/constraints.md (Bs): Update
3356 ix86_indirect_branch_register check. Don't check
3357 ix86_indirect_branch_register with GOT_memory_operand.
3359 * config/i386/predicates.md (GOT_memory_operand): Don't check
3360 ix86_indirect_branch_register here.
3361 (GOT32_symbol_operand): Likewise.
3363 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3365 * config/i386/predicates.md (constant_call_address_operand):
3366 Rewrite ix86_indirect_branch_register logic.
3367 (sibcall_insn_operand): Likewise.
3369 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3371 * config/i386/constraints.md (Bs): Replace
3372 ix86_indirect_branch_thunk_register with
3373 ix86_indirect_branch_register.
3375 * config/i386/i386.md (indirect_jump): Likewise.
3376 (tablejump): Likewise.
3377 (*sibcall_memory): Likewise.
3378 (*sibcall_value_memory): Likewise.
3379 Peepholes of indirect call and jump via memory: Likewise.
3380 * config/i386/i386.opt: Likewise.
3381 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3382 (GOT_memory_operand): Likewise.
3383 (call_insn_operand): Likewise.
3384 (sibcall_insn_operand): Likewise.
3385 (GOT32_symbol_operand): Likewise.
3387 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3390 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
3391 type rather than type addr's type points to.
3392 (expand_omp_atomic_mutex): Likewise.
3393 (expand_omp_atomic): Likewise.
3395 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3398 * config/i386/i386.c (output_indirect_thunk_function): Use
3399 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
3400 for __x86_return_thunk.
3402 2018-01-15 Richard Biener <rguenther@suse.de>
3405 * expmed.c (extract_bit_field_1): Fix typo.
3407 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3410 * config/arm/iterators.md (VF): New mode iterator.
3411 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
3412 Remove integer-related logic from pattern.
3413 (neon_vabd<mode>_3): Likewise.
3415 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3418 * common.opt (fstrict-overflow): No longer an alias.
3419 (fwrapv-pointer): New option.
3420 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
3421 also for pointer types based on flag_wrapv_pointer.
3422 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
3423 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
3424 opts->x_flag_wrapv got set.
3425 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
3426 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
3427 POINTER_TYPE_OVERFLOW_UNDEFINED.
3428 * match.pd: Likewise in address comparison pattern.
3429 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
3431 2018-01-15 Richard Biener <rguenther@suse.de>
3434 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
3435 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
3436 Reset type names to their identifier if their TYPE_DECL doesn't
3437 have linkage (and thus is used for ODR and devirt).
3438 (save_debug_info_for_decl): Remove.
3439 (save_debug_info_for_type): Likewise.
3440 (add_tree_to_fld_list): Adjust.
3441 * tree-pretty-print.c (dump_generic_node): Make dumping of
3442 type names more robust.
3444 2018-01-15 Richard Biener <rguenther@suse.de>
3446 * BASE-VER: Bump to 8.0.1.
3448 2018-01-14 Martin Sebor <msebor@redhat.com>
3451 * builtins.c (check_access): Avoid warning when the no-warning bit
3454 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
3456 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
3457 * ira-color (allocno_hard_regs_compare): Likewise.
3459 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
3462 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
3463 Use .pushsection/.popsection.
3465 2018-01-14 Martin Sebor <msebor@redhat.com>
3468 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
3470 2018-01-14 Jakub Jelinek <jakub@redhat.com>
3472 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
3473 entry from extra_headers.
3474 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
3475 extra_headers, make the list bitwise identical to the i?86-*-* one.
3477 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3479 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
3480 -mcmodel=large with -mindirect-branch=thunk,
3481 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
3482 -mfunction-return=thunk-extern.
3483 * doc/invoke.texi: Document -mcmodel=large is incompatible with
3484 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
3485 -mfunction-return=thunk and -mfunction-return=thunk-extern.
3487 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3489 * config/i386/i386.c (print_reg): Print the name of the full
3490 integer register without '%'.
3491 (ix86_print_operand): Handle 'V'.
3492 * doc/extend.texi: Document 'V' modifier.
3494 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3496 * config/i386/constraints.md (Bs): Disallow memory operand for
3497 -mindirect-branch-register.
3499 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3500 (GOT_memory_operand): Likewise.
3501 (call_insn_operand): Likewise.
3502 (sibcall_insn_operand): Likewise.
3503 (GOT32_symbol_operand): Likewise.
3504 * config/i386/i386.md (indirect_jump): Call convert_memory_address
3505 for -mindirect-branch-register.
3506 (tablejump): Likewise.
3507 (*sibcall_memory): Likewise.
3508 (*sibcall_value_memory): Likewise.
3509 Disallow peepholes of indirect call and jump via memory for
3510 -mindirect-branch-register.
3511 (*call_pop): Replace m with Bw.
3512 (*call_value_pop): Likewise.
3513 (*sibcall_pop_memory): Replace m with Bs.
3514 * config/i386/i386.opt (mindirect-branch-register): New option.
3515 * doc/invoke.texi: Document -mindirect-branch-register option.
3517 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3519 * config/i386/i386-protos.h (ix86_output_function_return): New.
3520 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
3521 set function_return_type.
3522 (indirect_thunk_name): Add ret_p to indicate thunk for function
3524 (output_indirect_thunk_function): Pass false to
3525 indirect_thunk_name.
3526 (ix86_output_indirect_branch_via_reg): Likewise.
3527 (ix86_output_indirect_branch_via_push): Likewise.
3528 (output_indirect_thunk_function): Create alias for function
3529 return thunk if regno < 0.
3530 (ix86_output_function_return): New function.
3531 (ix86_handle_fndecl_attribute): Handle function_return.
3532 (ix86_attribute_table): Add function_return.
3533 * config/i386/i386.h (machine_function): Add
3534 function_return_type.
3535 * config/i386/i386.md (simple_return_internal): Use
3536 ix86_output_function_return.
3537 (simple_return_internal_long): Likewise.
3538 * config/i386/i386.opt (mfunction-return=): New option.
3539 (indirect_branch): Mention -mfunction-return=.
3540 * doc/extend.texi: Document function_return function attribute.
3541 * doc/invoke.texi: Document -mfunction-return= option.
3543 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3545 * config/i386/i386-opts.h (indirect_branch): New.
3546 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
3547 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
3548 with local indirect jump when converting indirect call and jump.
3549 (ix86_set_indirect_branch_type): New.
3550 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
3551 (indirectlabelno): New.
3552 (indirect_thunk_needed): Likewise.
3553 (indirect_thunk_bnd_needed): Likewise.
3554 (indirect_thunks_used): Likewise.
3555 (indirect_thunks_bnd_used): Likewise.
3556 (INDIRECT_LABEL): Likewise.
3557 (indirect_thunk_name): Likewise.
3558 (output_indirect_thunk): Likewise.
3559 (output_indirect_thunk_function): Likewise.
3560 (ix86_output_indirect_branch_via_reg): Likewise.
3561 (ix86_output_indirect_branch_via_push): Likewise.
3562 (ix86_output_indirect_branch): Likewise.
3563 (ix86_output_indirect_jmp): Likewise.
3564 (ix86_code_end): Call output_indirect_thunk_function if needed.
3565 (ix86_output_call_insn): Call ix86_output_indirect_branch if
3567 (ix86_handle_fndecl_attribute): Handle indirect_branch.
3568 (ix86_attribute_table): Add indirect_branch.
3569 * config/i386/i386.h (machine_function): Add indirect_branch_type
3570 and has_local_indirect_jump.
3571 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
3573 (tablejump): Likewise.
3574 (*indirect_jump): Use ix86_output_indirect_jmp.
3575 (*tablejump_1): Likewise.
3576 (simple_return_indirect_internal): Likewise.
3577 * config/i386/i386.opt (mindirect-branch=): New option.
3578 (indirect_branch): New.
3581 (thunk-inline): Likewise.
3582 (thunk-extern): Likewise.
3583 * doc/extend.texi: Document indirect_branch function attribute.
3584 * doc/invoke.texi: Document -mindirect-branch= option.
3586 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
3589 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
3591 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
3593 * ipa-inline.c (want_inline_small_function_p): Return false if
3594 inlining has already failed with CIF_FINAL_ERROR.
3595 (update_caller_keys): Call want_inline_small_function_p before
3597 (update_callee_keys): Likewise.
3599 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
3601 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
3603 (rs6000_quadword_masked_address_p): Likewise.
3604 (quad_aligned_load_p): Likewise.
3605 (quad_aligned_store_p): Likewise.
3606 (const_load_sequence_p): Add comment to describe the outer-most loop.
3607 (mimic_memory_attributes_and_flags): New function.
3608 (rs6000_gen_stvx): Likewise.
3609 (replace_swapped_aligned_store): Likewise.
3610 (rs6000_gen_lvx): Likewise.
3611 (replace_swapped_aligned_load): Likewise.
3612 (replace_swapped_load_constant): Capitalize argument name in
3613 comment describing this function.
3614 (rs6000_analyze_swaps): Add a third pass to search for vector loads
3615 and stores that access quad-word aligned addresses and replace
3616 with stvx or lvx instructions when appropriate.
3617 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
3618 New function prototype.
3619 (rs6000_quadword_masked_address_p): Likewise.
3620 (rs6000_gen_lvx): Likewise.
3621 (rs6000_gen_stvx): Likewise.
3622 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3623 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3624 when memory address is aligned.
3625 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3626 this split to select lvx instruction when memory address is aligned.
3627 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3628 instruction when memory address is aligned.
3629 (*vsx_le_perm_load_v16qi): Likewise.
3630 (four unnamed splitters): Modify to select the stvx instruction
3631 when memory is aligned.
3633 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3635 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3636 which appears in the queue multiple times.
3638 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3639 Alan Hayward <alan.hayward@arm.com>
3640 David Sherwood <david.sherwood@arm.com>
3642 * tree-vectorizer.h (vec_lower_bound): New structure.
3643 (_loop_vec_info): Add check_nonzero and lower_bounds.
3644 (LOOP_VINFO_CHECK_NONZERO): New macro.
3645 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3646 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3647 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3648 fields. Make seg_len the distance travelled, not including the
3650 (dr_direction_indicator): Declare.
3651 (dr_zero_step_indicator): Likewise.
3652 (dr_known_forward_stride_p): Likewise.
3653 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3655 (runtime_alias_check_p): Allow runtime alias checks with
3657 (operator ==): Compare access_size and align.
3658 (prune_runtime_alias_test_list): Rework for new distinction between
3659 the access_size and seg_len.
3660 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3662 (get_segment_min_max): New function.
3663 (create_intersect_range_checks): Use it.
3664 (dr_step_indicator): New function.
3665 (dr_direction_indicator): Likewise.
3666 (dr_zero_step_indicator): Likewise.
3667 (dr_known_forward_stride_p): Likewise.
3668 * tree-loop-distribution.c (data_ref_segment_size): Return
3669 DR_STEP * (niters - 1).
3670 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3672 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3673 (vect_preserves_scalar_order_p): New function, split out from...
3674 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3675 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3676 (vect_vfa_access_size): New function.
3677 (vect_vfa_align): Likewise.
3678 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3679 (dump_lower_bound): New function.
3680 (vect_check_lower_bound): Likewise.
3681 (vect_small_gap_p): Likewise.
3682 (vectorizable_with_step_bound_p): Likewise.
3683 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3684 depencies if the vectorization factor is 1. Convert the checks
3685 for nonzero steps into checks on the bounds of DR_STEP. Try using
3686 a bunds check for variable steps if the minimum required step is
3687 relatively small. Update calls to the dr_with_seg_len
3688 constructor and to vect_compile_time_alias.
3689 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3691 (vect_loop_versioning): Call it.
3692 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3694 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3696 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3697 Alan Hayward <alan.hayward@arm.com>
3698 David Sherwood <david.sherwood@arm.com>
3700 * doc/sourcebuild.texi (vect_scatter_store): Document.
3701 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3703 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3705 * genopinit.c (main): Add supports_vec_scatter_store and
3706 supports_vec_scatter_store_cached to target_optabs.
3707 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3708 IFN_MASK_SCATTER_STORE.
3709 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3711 * internal-fn.h (internal_store_fn_p): Declare.
3712 (internal_fn_stored_value_index): Likewise.
3713 * internal-fn.c (scatter_store_direct): New macro.
3714 (expand_scatter_store_optab_fn): New function.
3715 (direct_scatter_store_optab_supported_p): New macro.
3716 (internal_store_fn_p): New function.
3717 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3718 IFN_MASK_SCATTER_STORE.
3719 (internal_fn_mask_index): Likewise.
3720 (internal_fn_stored_value_index): New function.
3721 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3723 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3724 * optabs-query.c (supports_vec_scatter_store_p): New function.
3725 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3726 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3727 true for scatter stores.
3728 (vect_gather_scatter_fn_p): Handle scatter stores too.
3729 (vect_check_gather_scatter): Consider using scatter stores if
3730 supports_vec_scatter_store_p.
3731 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3733 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3734 internal_fn_stored_value_index.
3735 (check_load_store_masking): Handle scatter stores too.
3736 (vect_get_store_rhs): Make public.
3737 (vectorizable_call): Use internal_store_fn_p.
3738 (vectorizable_store): Handle scatter store internal functions.
3739 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3740 when deciding whether the end of the group has been reached.
3741 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3742 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3743 (mask_scatter_store<mode>): New insns.
3745 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3746 Alan Hayward <alan.hayward@arm.com>
3747 David Sherwood <david.sherwood@arm.com>
3749 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3750 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3751 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3753 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3754 Use vect_truncate_gather_scatter_offset if we can't treat the
3755 operation as a normal gather load or scatter store.
3756 (get_group_load_store_type): Take the gather_scatter_info
3757 as argument. Try using a gather load or scatter store for
3758 single-element groups.
3759 (get_load_store_type): Update calls to get_group_load_store_type
3760 and vect_use_strided_gather_scatters_p.
3762 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3763 Alan Hayward <alan.hayward@arm.com>
3764 David Sherwood <david.sherwood@arm.com>
3766 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3767 optional tree argument.
3768 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3770 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3771 but continue to use the current value as a fallback.
3772 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3773 to compare the updates.
3774 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3775 (get_load_store_type): Use it when handling a strided access.
3776 (vect_get_strided_load_store_ops): New function.
3777 (vect_get_data_ptr_increment): Likewise.
3778 (vectorizable_load): Handle strided gather loads. Always pass
3779 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3781 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3782 Alan Hayward <alan.hayward@arm.com>
3783 David Sherwood <david.sherwood@arm.com>
3785 * doc/md.texi (gather_load@var{m}): Document.
3786 (mask_gather_load@var{m}): Likewise.
3787 * genopinit.c (main): Add supports_vec_gather_load and
3788 supports_vec_gather_load_cached to target_optabs.
3789 * optabs-tree.c (init_tree_optimization_optabs): Use
3790 ggc_cleared_alloc to allocate target_optabs.
3791 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3792 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3794 * internal-fn.h (internal_load_fn_p): Declare.
3795 (internal_gather_scatter_fn_p): Likewise.
3796 (internal_fn_mask_index): Likewise.
3797 (internal_gather_scatter_fn_supported_p): Likewise.
3798 * internal-fn.c (gather_load_direct): New macro.
3799 (expand_gather_load_optab_fn): New function.
3800 (direct_gather_load_optab_supported_p): New macro.
3801 (direct_internal_fn_optab): New function.
3802 (internal_load_fn_p): Likewise.
3803 (internal_gather_scatter_fn_p): Likewise.
3804 (internal_fn_mask_index): Likewise.
3805 (internal_gather_scatter_fn_supported_p): Likewise.
3806 * optabs-query.c (supports_at_least_one_mode_p): New function.
3807 (supports_vec_gather_load_p): Likewise.
3808 * optabs-query.h (supports_vec_gather_load_p): Declare.
3809 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3810 and memory_type field.
3811 (NUM_PATTERNS): Bump to 15.
3812 * tree-vect-data-refs.c: Include internal-fn.h.
3813 (vect_gather_scatter_fn_p): New function.
3814 (vect_describe_gather_scatter_call): Likewise.
3815 (vect_check_gather_scatter): Try using internal functions for
3816 gather loads. Recognize existing calls to a gather load function.
3817 (vect_analyze_data_refs): Consider using gather loads if
3818 supports_vec_gather_load_p.
3819 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3820 (vect_get_gather_scatter_offset_type): Likewise.
3821 (vect_convert_mask_for_vectype): Likewise.
3822 (vect_add_conversion_to_patterm): Likewise.
3823 (vect_try_gather_scatter_pattern): Likewise.
3824 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3825 (vect_vect_recog_func_ptrs): Add it.
3826 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3827 internal_fn_mask_index and internal_gather_scatter_fn_p.
3828 (check_load_store_masking): Take the gather_scatter_info as an
3829 argument and handle gather loads.
3830 (vect_get_gather_scatter_ops): New function.
3831 (vectorizable_call): Check internal_load_fn_p.
3832 (vectorizable_load): Likewise. Handle gather load internal
3834 (vectorizable_store): Update call to check_load_store_masking.
3835 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3836 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3837 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3838 (aarch64_gather_scale_operand_d): New predicates.
3839 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3840 (mask_gather_load<mode>): New insns.
3842 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3843 Alan Hayward <alan.hayward@arm.com>
3844 David Sherwood <david.sherwood@arm.com>
3846 * optabs.def (fold_left_plus_optab): New optab.
3847 * doc/md.texi (fold_left_plus_@var{m}): Document.
3848 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3849 * internal-fn.c (fold_left_direct): Define.
3850 (expand_fold_left_optab_fn): Likewise.
3851 (direct_fold_left_optab_supported_p): Likewise.
3852 * fold-const-call.c (fold_const_fold_left): New function.
3853 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3854 * tree-parloops.c (valid_reduction_p): New function.
3855 (gather_scalar_reductions): Use it.
3856 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3857 (vect_finish_replace_stmt): Declare.
3858 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3859 (needs_fold_left_reduction_p): New function, split out from...
3860 (vect_is_simple_reduction): ...here. Accept reductions that
3861 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3862 (vect_force_simple_reduction): Also store the reduction type in
3863 the assignment's STMT_VINFO_REDUC_TYPE.
3864 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3865 (merge_with_identity): New function.
3866 (vect_expand_fold_left): Likewise.
3867 (vectorize_fold_left_reduction): Likewise.
3868 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3869 scalar phi in place for it. Check for target support and reject
3870 cases that would reassociate the operation. Defer the transform
3871 phase to vectorize_fold_left_reduction.
3872 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3873 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3874 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3876 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3878 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3879 call to ifc_temp_var.
3881 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3882 Alan Hayward <alan.hayward@arm.com>
3883 David Sherwood <david.sherwood@arm.com>
3885 * target.def (legitimize_address_displacement): Take the original
3886 offset as a poly_int.
3887 * targhooks.h (default_legitimize_address_displacement): Update
3889 * targhooks.c (default_legitimize_address_displacement): Likewise.
3890 * doc/tm.texi: Regenerate.
3891 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3892 as an argument, moving assert of ad->disp == ad->disp_term to...
3893 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3894 Try calling targetm.legitimize_address_displacement before expanding
3895 the address rather than afterwards, and adjust for the new interface.
3896 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3897 Match the new hook interface. Handle SVE addresses.
3898 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3901 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3903 * Makefile.in (OBJS): Add early-remat.o.
3904 * target.def (select_early_remat_modes): New hook.
3905 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3906 * doc/tm.texi: Regenerate.
3907 * targhooks.h (default_select_early_remat_modes): Declare.
3908 * targhooks.c (default_select_early_remat_modes): New function.
3909 * timevar.def (TV_EARLY_REMAT): New timevar.
3910 * passes.def (pass_early_remat): New pass.
3911 * tree-pass.h (make_pass_early_remat): Declare.
3912 * early-remat.c: New file.
3913 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3915 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3917 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3918 Alan Hayward <alan.hayward@arm.com>
3919 David Sherwood <david.sherwood@arm.com>
3921 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3922 vfm1 with a bound_epilog parameter.
3923 (vect_do_peeling): Update calls accordingly, and move the prologue
3924 call earlier in the function. Treat the base bound_epilog as 0 for
3925 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3926 this base when peeling for gaps.
3927 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3928 with fully-masked loops.
3929 (vect_estimate_min_profitable_iters): Handle the single peeled
3930 iteration in that case.
3932 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3933 Alan Hayward <alan.hayward@arm.com>
3934 David Sherwood <david.sherwood@arm.com>
3936 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3937 single-element interleaving even if the size is not a power of 2.
3938 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3939 accesses for single-element interleaving if the group size is
3942 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3943 Alan Hayward <alan.hayward@arm.com>
3944 David Sherwood <david.sherwood@arm.com>
3946 * doc/md.texi (fold_extract_last_@var{m}): Document.
3947 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3948 * optabs.def (fold_extract_last_optab): New optab.
3949 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3950 * internal-fn.c (fold_extract_direct): New macro.
3951 (expand_fold_extract_optab_fn): Likewise.
3952 (direct_fold_extract_optab_supported_p): Likewise.
3953 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3954 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3955 EXTRACT_LAST_REDUCTION.
3956 (get_initial_def_for_reduction): Do not create an initial vector
3957 for EXTRACT_LAST_REDUCTION reductions.
3958 (vectorizable_reduction): Leave the scalar phi in place for
3959 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3960 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3961 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3962 transform phase to vectorizable_condition.
3963 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3965 (vect_finish_stmt_generation): ...here.
3966 (vect_finish_replace_stmt): New function.
3967 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3968 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3970 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3972 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3973 Alan Hayward <alan.hayward@arm.com>
3974 David Sherwood <david.sherwood@arm.com>
3976 * doc/md.texi (extract_last_@var{m}): Document.
3977 * optabs.def (extract_last_optab): New optab.
3978 * internal-fn.def (EXTRACT_LAST): New internal function.
3979 * internal-fn.c (cond_unary_direct): New macro.
3980 (expand_cond_unary_optab_fn): Likewise.
3981 (direct_cond_unary_optab_supported_p): Likewise.
3982 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3983 loops using EXTRACT_LAST.
3984 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3985 (extract_last_<mode>): ...this optab.
3986 (vec_extract<mode><Vel>): Update accordingly.
3988 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3989 Alan Hayward <alan.hayward@arm.com>
3990 David Sherwood <david.sherwood@arm.com>
3992 * target.def (empty_mask_is_expensive): New hook.
3993 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3994 * doc/tm.texi: Regenerate.
3995 * targhooks.h (default_empty_mask_is_expensive): Declare.
3996 * targhooks.c (default_empty_mask_is_expensive): New function.
3997 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3998 if the target says that empty masks are expensive.
3999 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
4001 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
4003 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4004 Alan Hayward <alan.hayward@arm.com>
4005 David Sherwood <david.sherwood@arm.com>
4007 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
4008 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
4009 (vect_use_loop_mask_for_alignment_p): New function.
4010 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
4011 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
4012 niters_skip argument. Make sure that the first niters_skip elements
4013 of the first iteration are inactive.
4014 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
4015 Update call to vect_set_loop_masks_directly.
4016 (get_misalign_in_elems): New function, split out from...
4017 (vect_gen_prolog_loop_niters): ...here.
4018 (vect_update_init_of_dr): Take a code argument that specifies whether
4019 the adjustment should be added or subtracted.
4020 (vect_update_init_of_drs): Likewise.
4021 (vect_prepare_for_masked_peels): New function.
4022 (vect_do_peeling): Skip prologue peeling if we're using a mask
4023 instead. Update call to vect_update_inits_of_drs.
4024 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4026 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
4027 alignment. Do not include the number of peeled iterations in
4028 the minimum threshold in that case.
4029 (vectorizable_induction): Adjust the start value down by
4030 LOOP_VINFO_MASK_SKIP_NITERS iterations.
4031 (vect_transform_loop): Call vect_prepare_for_masked_peels.
4032 Take the number of skipped iterations into account when calculating
4034 * tree-vect-stmts.c (vect_gen_while_not): New function.
4036 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4037 Alan Hayward <alan.hayward@arm.com>
4038 David Sherwood <david.sherwood@arm.com>
4040 * doc/sourcebuild.texi (vect_fully_masked): Document.
4041 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
4043 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
4045 (vect_analyze_loop_2): ...here. Don't check the vectorization
4046 factor against the number of loop iterations if the loop is
4049 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4050 Alan Hayward <alan.hayward@arm.com>
4051 David Sherwood <david.sherwood@arm.com>
4053 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
4054 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
4055 (dump_groups): Update accordingly.
4056 (iv_use::mem_type): New member variable.
4057 (address_p): New function.
4058 (record_use): Add a mem_type argument and initialize the new
4060 (record_group_use): Add a mem_type argument. Use address_p.
4061 Remove obsolete null checks of base_object. Update call to record_use.
4062 (find_interesting_uses_op): Update call to record_group_use.
4063 (find_interesting_uses_cond): Likewise.
4064 (find_interesting_uses_address): Likewise.
4065 (get_mem_type_for_internal_fn): New function.
4066 (find_address_like_use): Likewise.
4067 (find_interesting_uses_stmt): Try find_address_like_use before
4068 calling find_interesting_uses_op.
4069 (addr_offset_valid_p): Use the iv mem_type field as the type
4070 of the addressed memory.
4071 (add_autoinc_candidates): Likewise.
4072 (get_address_cost): Likewise.
4073 (split_small_address_groups_p): Use address_p.
4074 (split_address_groups): Likewise.
4075 (add_iv_candidate_for_use): Likewise.
4076 (autoinc_possible_for_pair): Likewise.
4077 (rewrite_groups): Likewise.
4078 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
4079 (determine_group_iv_cost): Update after split of USE_ADDRESS.
4080 (get_alias_ptr_type_for_ptr_address): New function.
4081 (rewrite_use_address): Rewrite address uses in calls that were
4082 identified by find_address_like_use.
4084 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4085 Alan Hayward <alan.hayward@arm.com>
4086 David Sherwood <david.sherwood@arm.com>
4088 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
4090 * gimple-expr.h (is_gimple_addressable: Likewise.
4091 * gimple-expr.c (is_gimple_address): Likewise.
4092 * internal-fn.c (expand_call_mem_ref): New function.
4093 (expand_mask_load_optab_fn): Use it.
4094 (expand_mask_store_optab_fn): Likewise.
4096 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4097 Alan Hayward <alan.hayward@arm.com>
4098 David Sherwood <david.sherwood@arm.com>
4100 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
4101 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
4102 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
4103 (cond_umax@var{mode}): Document.
4104 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
4105 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
4106 (cond_umin_optab, cond_umax_optab): New optabs.
4107 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
4108 (COND_IOR, COND_XOR): New internal functions.
4109 * internal-fn.h (get_conditional_internal_fn): Declare.
4110 * internal-fn.c (cond_binary_direct): New macro.
4111 (expand_cond_binary_optab_fn): Likewise.
4112 (direct_cond_binary_optab_supported_p): Likewise.
4113 (get_conditional_internal_fn): New function.
4114 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
4115 Cope with reduction statements that are vectorized as calls rather
4117 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
4118 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
4119 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
4120 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
4121 (UNSPEC_COND_EOR): New unspecs.
4122 (optab): Add mappings for them.
4123 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
4124 (sve_int_op, sve_fp_op): New int attributes.
4126 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4127 Alan Hayward <alan.hayward@arm.com>
4128 David Sherwood <david.sherwood@arm.com>
4130 * optabs.def (while_ult_optab): New optab.
4131 * doc/md.texi (while_ult@var{m}@var{n}): Document.
4132 * internal-fn.def (WHILE_ULT): New internal function.
4133 * internal-fn.h (direct_internal_fn_supported_p): New override
4134 that takes two types as argument.
4135 * internal-fn.c (while_direct): New macro.
4136 (expand_while_optab_fn): New function.
4137 (convert_optab_supported_p): Likewise.
4138 (direct_while_optab_supported_p): New macro.
4139 * wide-int.h (wi::udiv_ceil): New function.
4140 * tree-vectorizer.h (rgroup_masks): New structure.
4141 (vec_loop_masks): New typedef.
4142 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
4144 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
4145 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
4146 (vect_max_vf): New function.
4147 (slpeel_make_loop_iterate_ntimes): Delete.
4148 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
4149 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
4150 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
4151 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
4152 internal-fn.h, stor-layout.h and optabs-query.h.
4153 (vect_set_loop_mask): New function.
4154 (add_preheader_seq): Likewise.
4155 (add_header_seq): Likewise.
4156 (interleave_supported_p): Likewise.
4157 (vect_maybe_permute_loop_masks): Likewise.
4158 (vect_set_loop_masks_directly): Likewise.
4159 (vect_set_loop_condition_masked): Likewise.
4160 (vect_set_loop_condition_unmasked): New function, split out from
4161 slpeel_make_loop_iterate_ntimes.
4162 (slpeel_make_loop_iterate_ntimes): Rename to..
4163 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
4164 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
4165 (vect_do_peeling): Update call accordingly.
4166 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
4168 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4169 mask_compare_type, can_fully_mask_p and fully_masked_p.
4170 (release_vec_loop_masks): New function.
4171 (_loop_vec_info): Use it to free the loop masks.
4172 (can_produce_all_loop_masks_p): New function.
4173 (vect_get_max_nscalars_per_iter): Likewise.
4174 (vect_verify_full_masking): Likewise.
4175 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
4176 retries, and free the mask rgroups before retrying. Check loop-wide
4177 reasons for disallowing fully-masked loops. Make the final decision
4178 about whether use a fully-masked loop or not.
4179 (vect_estimate_min_profitable_iters): Do not assume that peeling
4180 for the number of iterations will be needed for fully-masked loops.
4181 (vectorizable_reduction): Disable fully-masked loops.
4182 (vectorizable_live_operation): Likewise.
4183 (vect_halve_mask_nunits): New function.
4184 (vect_double_mask_nunits): Likewise.
4185 (vect_record_loop_mask): Likewise.
4186 (vect_get_loop_mask): Likewise.
4187 (vect_transform_loop): Handle the case in which the final loop
4188 iteration might handle a partial vector. Call vect_set_loop_condition
4189 instead of slpeel_make_loop_iterate_ntimes.
4190 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
4191 (check_load_store_masking): New function.
4192 (prepare_load_store_mask): Likewise.
4193 (vectorizable_store): Handle fully-masked loops.
4194 (vectorizable_load): Likewise.
4195 (supportable_widening_operation): Use vect_halve_mask_nunits for
4197 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
4198 (vect_gen_while): New function.
4199 * config/aarch64/aarch64.md (umax<mode>3): New expander.
4200 (aarch64_uqdec<mode>): New insn.
4202 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4203 Alan Hayward <alan.hayward@arm.com>
4204 David Sherwood <david.sherwood@arm.com>
4206 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
4207 (reduc_xor_scal_optab): New optabs.
4208 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
4209 (reduc_xor_scal_@var{m}): Document.
4210 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
4211 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
4213 * fold-const-call.c (fold_const_call): Handle them.
4214 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
4215 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
4216 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
4217 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
4218 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
4219 (UNSPEC_XORV): New unspecs.
4220 (optab): Add entries for them.
4221 (BITWISEV): New int iterator.
4222 (bit_reduc_op): New int attributes.
4224 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4225 Alan Hayward <alan.hayward@arm.com>
4226 David Sherwood <david.sherwood@arm.com>
4228 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
4229 * internal-fn.def (VEC_SHL_INSERT): New internal function.
4230 * optabs.def (vec_shl_insert_optab): New optab.
4231 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
4232 (duplicate_and_interleave): Likewise.
4233 * tree-vect-loop.c: Include internal-fn.h.
4234 (neutral_op_for_slp_reduction): New function, split out from
4235 get_initial_defs_for_reduction.
4236 (get_initial_def_for_reduction): Handle option 2 for variable-length
4237 vectors by loading the neutral value into a vector and then shifting
4238 the initial value into element 0.
4239 (get_initial_defs_for_reduction): Replace the code argument with
4240 the neutral value calculated by neutral_op_for_slp_reduction.
4241 Use gimple_build_vector for constant-length vectors.
4242 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
4243 but the first group_size elements have a neutral value.
4244 Use duplicate_and_interleave otherwise.
4245 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
4246 Update call to get_initial_defs_for_reduction. Handle SLP
4247 reductions for variable-length vectors by creating one vector
4248 result for each scalar result, with the elements associated
4249 with other scalar results stubbed out with the neutral value.
4250 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
4251 Require IFN_VEC_SHL_INSERT for double reductions on
4252 variable-length vectors, or SLP reductions that have
4253 a neutral value. Require can_duplicate_and_interleave_p
4254 support for variable-length unchained SLP reductions if there
4255 is no neutral value, such as for MIN/MAX reductions. Also require
4256 the number of vector elements to be a multiple of the number of
4257 SLP statements when doing variable-length unchained SLP reductions.
4258 Update call to vect_create_epilog_for_reduction.
4259 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
4260 and remove initial values.
4261 (duplicate_and_interleave): Make public.
4262 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
4263 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
4265 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4266 Alan Hayward <alan.hayward@arm.com>
4267 David Sherwood <david.sherwood@arm.com>
4269 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
4270 (can_duplicate_and_interleave_p): New function.
4271 (vect_get_and_check_slp_defs): Take the vector of statements
4272 rather than just the current one. Remove excess parentheses.
4273 Restriction rejectinon of vect_constant_def and vect_external_def
4274 for variable-length vectors to boolean types, or types for which
4275 can_duplicate_and_interleave_p is false.
4276 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
4277 (duplicate_and_interleave): New function.
4278 (vect_get_constant_vectors): Use gimple_build_vector for
4279 constant-length vectors and suitable variable-length constant
4280 vectors. Use duplicate_and_interleave for other variable-length
4281 vectors. Don't defer the update when inserting new statements.
4283 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4284 Alan Hayward <alan.hayward@arm.com>
4285 David Sherwood <david.sherwood@arm.com>
4287 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
4288 min_profitable_iters doesn't go negative.
4290 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4291 Alan Hayward <alan.hayward@arm.com>
4292 David Sherwood <david.sherwood@arm.com>
4294 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
4295 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
4296 * optabs.def (vec_mask_load_lanes_optab): New optab.
4297 (vec_mask_store_lanes_optab): Likewise.
4298 * internal-fn.def (MASK_LOAD_LANES): New internal function.
4299 (MASK_STORE_LANES): Likewise.
4300 * internal-fn.c (mask_load_lanes_direct): New macro.
4301 (mask_store_lanes_direct): Likewise.
4302 (expand_mask_load_optab_fn): Handle masked operations.
4303 (expand_mask_load_lanes_optab_fn): New macro.
4304 (expand_mask_store_optab_fn): Handle masked operations.
4305 (expand_mask_store_lanes_optab_fn): New macro.
4306 (direct_mask_load_lanes_optab_supported_p): Likewise.
4307 (direct_mask_store_lanes_optab_supported_p): Likewise.
4308 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
4310 (vect_load_lanes_supported): Likewise.
4311 * tree-vect-data-refs.c (strip_conversion): New function.
4312 (can_group_stmts_p): Likewise.
4313 (vect_analyze_data_ref_accesses): Use it instead of checking
4314 for a pair of assignments.
4315 (vect_store_lanes_supported): Take a masked_p parameter.
4316 (vect_load_lanes_supported): Likewise.
4317 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
4318 vect_store_lanes_supported and vect_load_lanes_supported.
4319 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
4320 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
4321 parameter. Don't allow gaps for masked accesses.
4322 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
4323 and vect_load_lanes_supported.
4324 (get_load_store_type): Take a masked_p parameter and update
4325 call to get_group_load_store_type.
4326 (vectorizable_store): Update call to get_load_store_type.
4327 Handle IFN_MASK_STORE_LANES.
4328 (vectorizable_load): Update call to get_load_store_type.
4329 Handle IFN_MASK_LOAD_LANES.
4331 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4332 Alan Hayward <alan.hayward@arm.com>
4333 David Sherwood <david.sherwood@arm.com>
4335 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
4337 * config/aarch64/aarch64-protos.h
4338 (aarch64_sve_struct_memory_operand_p): Declare.
4339 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
4340 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
4341 (VPRED, vpred): Handle SVE structure modes.
4342 * config/aarch64/constraints.md (Utx): New constraint.
4343 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
4344 (aarch64_sve_struct_nonimmediate_operand): New predicates.
4345 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
4346 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
4347 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
4348 structure modes. Split into pieces after RA.
4349 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
4350 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
4352 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
4353 SVE structure modes.
4354 (aarch64_classify_address): Likewise.
4355 (sizetochar): Move earlier in file.
4356 (aarch64_print_operand): Handle SVE register lists.
4357 (aarch64_array_mode): New function.
4358 (aarch64_sve_struct_memory_operand_p): Likewise.
4359 (TARGET_ARRAY_MODE): Redefine.
4361 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4362 Alan Hayward <alan.hayward@arm.com>
4363 David Sherwood <david.sherwood@arm.com>
4365 * target.def (array_mode): New target hook.
4366 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
4367 * doc/tm.texi: Regenerate.
4368 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
4369 * hooks.c (hook_optmode_mode_uhwi_none): New function.
4370 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
4372 * stor-layout.c (mode_for_array): Likewise. Support polynomial
4375 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4376 Alan Hayward <alan.hayward@arm.com>
4377 David Sherwood <david.sherwood@arm.com>
4379 * fold-const.c (fold_binary_loc): Check the argument types
4380 rather than the result type when testing for a vector operation.
4382 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4384 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
4385 * doc/tm.texi: Regenerate.
4387 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4388 Alan Hayward <alan.hayward@arm.com>
4389 David Sherwood <david.sherwood@arm.com>
4391 * doc/invoke.texi (-msve-vector-bits=): Document new option.
4392 (sve): Document new AArch64 extension.
4393 * doc/md.texi (w): Extend the description of the AArch64
4394 constraint to include SVE vectors.
4395 (Upl, Upa): Document new AArch64 predicate constraints.
4396 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
4398 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
4399 (msve-vector-bits=): New option.
4400 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
4401 SVE when these are disabled.
4402 (sve): New extension.
4403 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
4404 modes. Adjust their number of units based on aarch64_sve_vg.
4405 (MAX_BITSIZE_MODE_ANY_MODE): Define.
4406 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
4407 aarch64_addr_query_type.
4408 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
4409 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
4410 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
4411 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
4412 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
4413 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
4414 (aarch64_simd_imm_zero_p): Delete.
4415 (aarch64_check_zero_based_sve_index_immediate): Declare.
4416 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4417 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4418 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4419 (aarch64_sve_float_mul_immediate_p): Likewise.
4420 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4422 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
4423 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
4424 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
4425 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
4426 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
4427 (aarch64_regmode_natural_size): Likewise.
4428 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
4429 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
4431 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
4432 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
4433 for VG and the SVE predicate registers.
4434 (V_ALIASES): Add a "z"-prefixed alias.
4435 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
4436 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
4437 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
4438 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
4439 (REG_CLASS_NAMES): Add entries for them.
4440 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
4441 and the predicate registers.
4442 (aarch64_sve_vg): Declare.
4443 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
4444 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
4445 (REGMODE_NATURAL_SIZE): Define.
4446 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
4448 * config/aarch64/aarch64.c: Include cfgrtl.h.
4449 (simd_immediate_info): Add a constructor for series vectors,
4450 and an associated step field.
4451 (aarch64_sve_vg): New variable.
4452 (aarch64_dbx_register_number): Handle VG and the predicate registers.
4453 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
4454 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
4455 (VEC_ANY_DATA, VEC_STRUCT): New constants.
4456 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
4457 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
4458 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
4459 (aarch64_get_mask_mode): New functions.
4460 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
4461 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
4462 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
4463 predicate modes and predicate registers. Explicitly restrict
4464 GPRs to modes of 16 bytes or smaller. Only allow FP registers
4465 to store a vector mode if it is recognized by
4466 aarch64_classify_vector_mode.
4467 (aarch64_regmode_natural_size): New function.
4468 (aarch64_hard_regno_caller_save_mode): Return the original mode
4470 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
4471 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
4472 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
4473 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
4475 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
4476 does not overlap dest if the function is frame-related. Handle
4478 (aarch64_split_add_offset): New function.
4479 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
4480 them aarch64_add_offset.
4481 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
4482 and update call to aarch64_sub_sp.
4483 (aarch64_add_cfa_expression): New function.
4484 (aarch64_expand_prologue): Pass extra temporary registers to the
4485 functions above. Handle the case in which we need to emit new
4486 DW_CFA_expressions for registers that were originally saved
4487 relative to the stack pointer, but now have to be expressed
4488 relative to the frame pointer.
4489 (aarch64_output_mi_thunk): Pass extra temporary registers to the
4491 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
4492 IP0 and IP1 values for SVE frames.
4493 (aarch64_expand_vec_series): New function.
4494 (aarch64_expand_sve_widened_duplicate): Likewise.
4495 (aarch64_expand_sve_const_vector): Likewise.
4496 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
4497 Handle SVE constants. Use emit_move_insn to move a force_const_mem
4498 into the register, rather than emitting a SET directly.
4499 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
4500 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
4501 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
4502 (offset_9bit_signed_scaled_p): New functions.
4503 (aarch64_replicate_bitmask_imm): New function.
4504 (aarch64_bitmask_imm): Use it.
4505 (aarch64_cannot_force_const_mem): Reject expressions involving
4506 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
4507 (aarch64_classify_index): Handle SVE indices, by requiring
4508 a plain register index with a scale that matches the element size.
4509 (aarch64_classify_address): Handle SVE addresses. Assert that
4510 the mode of the address is VOIDmode or an integer mode.
4511 Update call to aarch64_classify_symbol.
4512 (aarch64_classify_symbolic_expression): Update call to
4513 aarch64_classify_symbol.
4514 (aarch64_const_vec_all_in_range_p): New function.
4515 (aarch64_print_vector_float_operand): Likewise.
4516 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
4517 "vN" for FP registers with SVE modes. Handle (const ...) vectors
4518 and the FP immediates 1.0 and 0.5.
4519 (aarch64_print_address_internal): Handle SVE addresses.
4520 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
4521 (aarch64_regno_regclass): Handle predicate registers.
4522 (aarch64_secondary_reload): Handle big-endian reloads of SVE
4524 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
4525 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
4526 (aarch64_convert_sve_vector_bits): New function.
4527 (aarch64_override_options): Use it to handle -msve-vector-bits=.
4528 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4530 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
4531 Handle SVE vector and predicate modes. Accept VL-based constants
4532 that need only one temporary register, and VL offsets that require
4533 no temporary registers.
4534 (aarch64_conditional_register_usage): Mark the predicate registers
4535 as fixed if SVE isn't available.
4536 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
4537 Return true for SVE vector and predicate modes.
4538 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
4539 rather than an unsigned int. Handle SVE modes.
4540 (aarch64_preferred_simd_mode): Update call accordingly. Handle
4542 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
4544 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4545 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4546 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4547 (aarch64_sve_float_mul_immediate_p): New functions.
4548 (aarch64_sve_valid_immediate): New function.
4549 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
4550 Explicitly reject structure modes. Check for INDEX constants.
4551 Handle PTRUE and PFALSE constants.
4552 (aarch64_check_zero_based_sve_index_immediate): New function.
4553 (aarch64_simd_imm_zero_p): Delete.
4554 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
4555 vector modes. Accept constants in the range of CNT[BHWD].
4556 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
4557 ask for an Advanced SIMD mode.
4558 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
4559 (aarch64_simd_vector_alignment): Handle SVE predicates.
4560 (aarch64_vectorize_preferred_vector_alignment): New function.
4561 (aarch64_simd_vector_alignment_reachable): Use it instead of
4563 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
4564 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
4566 (MAX_VECT_LEN): Delete.
4567 (expand_vec_perm_d): Add a vec_flags field.
4568 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
4569 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
4570 (aarch64_evpc_ext): Don't apply a big-endian lane correction
4572 (aarch64_evpc_rev): Rename to...
4573 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
4574 (aarch64_evpc_rev_global): New function.
4575 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
4576 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
4578 (aarch64_evpc_sve_tbl): New function.
4579 (aarch64_expand_vec_perm_const_1): Update after rename of
4580 aarch64_evpc_rev. Handle SVE permutes too, trying
4581 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
4582 than aarch64_evpc_tbl.
4583 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
4584 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
4585 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
4586 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
4587 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
4588 (aarch64_expand_sve_vcond): New functions.
4589 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
4590 of aarch64_vector_mode_p.
4591 (aarch64_dwarf_poly_indeterminate_value): New function.
4592 (aarch64_compute_pressure_classes): Likewise.
4593 (aarch64_can_change_mode_class): Likewise.
4594 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
4595 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
4596 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
4597 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
4598 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
4599 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
4600 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
4601 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
4603 (Dn, Dl, Dr): Accept const as well as const_vector.
4604 (Dz): Likewise. Compare against CONST0_RTX.
4605 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
4606 of "vector" where appropriate.
4607 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
4608 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
4609 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
4610 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
4611 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
4612 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
4613 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
4614 (v_int_equiv): Extend to SVE modes.
4615 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
4617 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
4618 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4619 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4620 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4621 (SVE_COND_FP_CMP): New int iterators.
4622 (perm_hilo): Handle the new unpack unspecs.
4623 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4625 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4626 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4627 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4628 (aarch64_equality_operator, aarch64_constant_vector_operand)
4629 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4630 (aarch64_sve_nonimmediate_operand): Likewise.
4631 (aarch64_sve_general_operand): Likewise.
4632 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4633 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4634 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4635 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4636 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4637 (aarch64_sve_float_arith_immediate): Likewise.
4638 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4639 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4640 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4641 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4642 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4643 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4644 (aarch64_sve_float_arith_operand): Likewise.
4645 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4646 (aarch64_sve_float_mul_operand): Likewise.
4647 (aarch64_sve_vec_perm_operand): Likewise.
4648 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4649 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4650 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4651 as well as const_vector.
4652 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4653 in file. Use CONST0_RTX and CONSTM1_RTX.
4654 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4655 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4656 Use aarch64_simd_imm_zero.
4657 * config/aarch64/aarch64-sve.md: New file.
4658 * config/aarch64/aarch64.md: Include it.
4659 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4660 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4661 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4662 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4663 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4664 (sve): New attribute.
4665 (enabled): Disable instructions with the sve attribute unless
4667 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4668 aarch64_expand_mov_immediate.
4669 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4670 CNT[BHSD] immediates.
4671 (movti): Split CONST_POLY_INT moves into two halves.
4672 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4673 Split additions that need a temporary here if the destination
4674 is the stack pointer.
4675 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4676 (*add<mode>3_poly_1): New instruction.
4677 (set_clobber_cc): New expander.
4679 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4681 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4682 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4683 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4684 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4685 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4686 Change innermode from fixed_mode_size to machine_mode.
4687 (simplify_subreg): Update call accordingly. Handle a constant-sized
4688 subreg of a variable-length CONST_VECTOR.
4690 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4691 Alan Hayward <alan.hayward@arm.com>
4692 David Sherwood <david.sherwood@arm.com>
4694 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4695 (add_offset_to_base): New function, split out from...
4696 (create_mem_ref): ...here. When handling a scale other than 1,
4697 check first whether the address is valid without the offset.
4698 Add it into the base if so, leaving the index and scale as-is.
4700 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4703 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4704 fold_for_warn before checking if arg2 is INTEGER_CST.
4706 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4708 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4709 (store_multiple_operation): Delete.
4710 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4711 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4712 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4713 guarded by TARGET_STRING.
4714 (rs6000_output_load_multiple): Delete.
4715 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4716 OPTION_MASK_STRING / TARGET_STRING handling.
4717 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4718 (const rs6000_opt_masks) <"string">: Change mask to 0.
4719 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4720 (MASK_STRING): Delete.
4721 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4723 (load_multiple): Delete.
4730 (store_multiple): Delete.
4737 (movmemsi_8reg): Delete.
4738 (corresponding unnamed define_insn): Delete.
4739 (movmemsi_6reg): Delete.
4740 (corresponding unnamed define_insn): Delete.
4741 (movmemsi_4reg): Delete.
4742 (corresponding unnamed define_insn): Delete.
4743 (movmemsi_2reg): Delete.
4744 (corresponding unnamed define_insn): Delete.
4745 (movmemsi_1reg): Delete.
4746 (corresponding unnamed define_insn): Delete.
4747 * config/rs6000/rs6000.opt (mno-string): New.
4748 (mstring): Replace by deprecation warning stub.
4749 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4751 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4753 * regrename.c (regrename_do_replace): If replacing the same
4754 reg multiple times, try to reuse last created gen_raw_REG.
4757 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4758 main to workaround a bug in GDB.
4760 2018-01-12 Tom de Vries <tom@codesourcery.com>
4763 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4765 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4767 PR rtl-optimization/80481
4768 * ira-color.c (get_cap_member): New function.
4769 (allocnos_conflict_by_live_ranges_p): Use it.
4770 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4771 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4773 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4776 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4777 (*saddl_se_1): Ditto.
4779 (*ssubl_se_1): Ditto.
4781 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4783 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4784 rather than wi::to_widest for DR_INITs.
4785 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4786 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4787 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4789 (vect_analyze_group_access_1): Note that here.
4791 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4793 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4794 polynomial type sizes.
4796 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4798 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4799 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4800 (gimple_add_tmp_var): Likewise.
4802 2018-01-12 Martin Liska <mliska@suse.cz>
4804 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4805 (gimple_alloc_sizes): Likewise.
4806 (dump_gimple_statistics): Use PRIu64 in printf format.
4807 * gimple.h: Change uint64_t to int.
4809 2018-01-12 Martin Liska <mliska@suse.cz>
4811 * tree-core.h: Use uint64_t instead of int.
4812 * tree.c (tree_node_counts): Likewise.
4813 (tree_node_sizes): Likewise.
4814 (dump_tree_statistics): Use PRIu64 in printf format.
4816 2018-01-12 Martin Liska <mliska@suse.cz>
4818 * Makefile.in: As qsort_chk is implemented in vec.c, add
4819 vec.o to linkage of gencfn-macros.
4820 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4821 passing the info to record_node_allocation_statistics.
4822 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4824 * ggc-common.c (struct ggc_usage): Add operator== and use
4825 it in operator< and compare function.
4826 * mem-stats.h (struct mem_usage): Likewise.
4827 * vec.c (struct vec_usage): Remove operator< and compare
4828 function. Can be simply inherited.
4830 2018-01-12 Martin Jambor <mjambor@suse.cz>
4833 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4834 * tree-ssa-math-opts.c: Include domwalk.h.
4835 (convert_mult_to_fma_1): New function.
4836 (fma_transformation_info): New type.
4837 (fma_deferring_state): Likewise.
4838 (cancel_fma_deferring): New function.
4839 (result_of_phi): Likewise.
4840 (last_fma_candidate_feeds_initial_phi): Likewise.
4841 (convert_mult_to_fma): Added deferring logic, split actual
4842 transformation to convert_mult_to_fma_1.
4843 (math_opts_dom_walker): New type.
4844 (math_opts_dom_walker::after_dom_children): New method, body moved
4845 here from pass_optimize_widening_mul::execute, added deferring logic
4847 (pass_optimize_widening_mul::execute): Moved most of code to
4848 math_opts_dom_walker::after_dom_children.
4849 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4850 * config/i386/i386.c (ix86_option_override_internal): Added
4851 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4853 2018-01-12 Richard Biener <rguenther@suse.de>
4856 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4857 inline instance vars.
4859 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4862 * config/rx/rx.c (rx_is_restricted_memory_address):
4865 2018-01-12 Richard Biener <rguenther@suse.de>
4867 PR tree-optimization/80846
4868 * target.def (split_reduction): New target hook.
4869 * targhooks.c (default_split_reduction): New function.
4870 * targhooks.h (default_split_reduction): Declare.
4871 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4872 target requests first reduce vectors by combining low and high
4874 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4875 (get_vectype_for_scalar_type_and_size): Export.
4876 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4877 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4878 * doc/tm.texi: Regenerate.
4879 * config/i386/i386.c (ix86_split_reduction): Implement
4880 TARGET_VECTORIZE_SPLIT_REDUCTION.
4882 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4885 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4886 in PIC mode except for TARGET_VXWORKS_RTP.
4887 * config/sparc/sparc.c: Include cfgrtl.h.
4888 (TARGET_INIT_PIC_REG): Define.
4889 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4890 (sparc_pic_register_p): New predicate.
4891 (sparc_legitimate_address_p): Use it.
4892 (sparc_legitimize_pic_address): Likewise.
4893 (sparc_delegitimize_address): Likewise.
4894 (sparc_mode_dependent_address_p): Likewise.
4895 (gen_load_pcrel_sym): Remove 4th parameter.
4896 (load_got_register): Adjust call to above. Remove obsolete stuff.
4897 (sparc_expand_prologue): Do not call load_got_register here.
4898 (sparc_flat_expand_prologue): Likewise.
4899 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4900 (sparc_use_pseudo_pic_reg): New function.
4901 (sparc_init_pic_reg): Likewise.
4902 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4903 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4905 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4907 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4908 Add item for branch_cost.
4910 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4912 PR rtl-optimization/83565
4913 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4914 not extend the result to a larger mode for rotate operations.
4915 (num_sign_bit_copies1): Likewise.
4917 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4920 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4922 Use values-Xc.o for -pedantic.
4923 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4925 2018-01-12 Martin Liska <mliska@suse.cz>
4928 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4930 (possible_polymorphic_call_targets): Use it.
4931 (ipa_devirt): Likewise.
4933 2018-01-12 Martin Liska <mliska@suse.cz>
4935 * profile-count.h (enum profile_quality): Use 0 as invalid
4936 enum value of profile_quality.
4938 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4940 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4941 -mext-string options.
4943 2018-01-12 Richard Biener <rguenther@suse.de>
4945 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4946 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4947 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4949 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4951 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4953 * configure.ac (--with-long-double-format): Add support for the
4954 configuration option to change the default long double format on
4956 * config.gcc (powerpc*-linux*-*): Likewise.
4957 * configure: Regenerate.
4958 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4959 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4960 used without modification.
4962 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4964 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4965 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4966 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4967 MISC_BUILTIN_SPEC_BARRIER.
4968 (rs6000_init_builtins): Likewise.
4969 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4971 (speculation_barrier): New define_insn.
4972 * doc/extend.texi: Document __builtin_speculation_barrier.
4974 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4977 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4978 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4979 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4981 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4982 integral modes instead of "ss" and "sd".
4983 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4984 vectors with 32-bit and 64-bit elements.
4985 (vecdupssescalarmodesuffix): New mode attribute.
4986 (vec_dup<mode>): Use it.
4988 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4991 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4992 frame if argument is passed on stack.
4994 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4997 * ree.c (combine_reaching_defs): Optimize also
4998 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4999 reg2=any_extend(exp); reg1=reg2;, formatting fix.
5001 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
5004 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
5006 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
5009 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
5010 after they are computed.
5012 2018-01-11 Bin Cheng <bin.cheng@arm.com>
5014 PR tree-optimization/83695
5015 * gimple-loop-linterchange.cc
5016 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
5017 reset cached scev information after interchange.
5018 (pass_linterchange::execute): Remove call to scev_reset_htab.
5020 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5022 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
5023 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
5024 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
5025 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
5026 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
5027 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
5028 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
5029 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
5030 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
5031 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
5032 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
5033 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
5034 (V_lane_reg): Likewise.
5035 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
5037 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
5038 (vfmal_lane_low<mode>_intrinsic,
5039 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
5040 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
5041 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
5042 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
5043 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
5044 vfmsl_lane_high<mode>_intrinsic): New define_insns.
5046 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5048 * config/arm/arm-cpus.in (fp16fml): New feature.
5049 (ALL_SIMD): Add fp16fml.
5050 (armv8.2-a): Add fp16fml as an option.
5051 (armv8.3-a): Likewise.
5052 (armv8.4-a): Add fp16fml as part of fp16.
5053 * config/arm/arm.h (TARGET_FP16FML): Define.
5054 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
5056 * config/arm/arm-modes.def (V2HF): Define.
5057 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
5058 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
5059 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
5060 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
5061 vfmsl_low, vfmsl_high): New set of builtins.
5062 * config/arm/iterators.md (PLUSMINUS): New code iterator.
5063 (vfml_op): New code attribute.
5064 (VFMLHALVES): New int iterator.
5065 (VFML, VFMLSEL): New mode attributes.
5066 (V_reg): Define mapping for V2HF.
5067 (V_hi, V_lo): New mode attributes.
5068 (VF_constraint): Likewise.
5069 (vfml_half, vfml_half_selector): New int attributes.
5070 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
5072 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
5073 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
5075 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
5076 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
5077 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
5078 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
5080 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
5081 Document new effective target and option set.
5083 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5085 * config/arm/arm-cpus.in (armv8_4): New feature.
5086 (ARMv8_4a): New fgroup.
5087 (armv8.4-a): New arch.
5088 * config/arm/arm-tables.opt: Regenerate.
5089 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
5090 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
5091 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
5092 Add matching rules for -march=armv8.4-a and extensions.
5093 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
5095 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
5098 * config/rx/rx.md (BW): New mode attribute.
5099 (sync_lock_test_and_setsi): Add mode suffix to insn output.
5101 2018-01-11 Richard Biener <rguenther@suse.de>
5103 PR tree-optimization/83435
5104 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
5105 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
5106 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
5108 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5109 Alan Hayward <alan.hayward@arm.com>
5110 David Sherwood <david.sherwood@arm.com>
5112 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
5114 (aarch64_classify_address): Initialize it. Track polynomial offsets.
5115 (aarch64_print_address_internal): Use it to check for a zero offset.
5117 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5118 Alan Hayward <alan.hayward@arm.com>
5119 David Sherwood <david.sherwood@arm.com>
5121 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
5122 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
5123 Return a poly_int64 rather than a HOST_WIDE_INT.
5124 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
5125 rather than a HOST_WIDE_INT.
5126 * config/aarch64/aarch64.h (aarch64_frame): Protect with
5127 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
5128 hard_fp_offset, frame_size, initial_adjust, callee_offset and
5129 final_offset from HOST_WIDE_INT to poly_int64.
5130 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
5131 to_constant when getting the number of units in an Advanced SIMD
5133 (aarch64_builtin_vectorized_function): Check for a constant number
5135 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
5137 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
5138 attribute instead of GET_MODE_NUNITS.
5139 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
5140 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
5141 GET_MODE_SIZE for fixed-size registers.
5142 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
5143 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
5144 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
5145 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
5146 (aarch64_print_operand, aarch64_print_address_internal)
5147 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
5148 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
5149 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
5150 Handle polynomial GET_MODE_SIZE.
5151 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
5152 wider than SImode without modification.
5153 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
5154 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
5155 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
5156 passing and returning SVE modes.
5157 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
5158 rather than GEN_INT.
5159 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
5160 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
5161 (aarch64_allocate_and_probe_stack_space): Likewise.
5162 (aarch64_layout_frame): Cope with polynomial offsets.
5163 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
5164 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
5166 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
5167 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
5168 poly_int64 rather than a HOST_WIDE_INT.
5169 (aarch64_get_separate_components, aarch64_process_components)
5170 (aarch64_expand_prologue, aarch64_expand_epilogue)
5171 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
5172 (aarch64_anchor_offset): New function, split out from...
5173 (aarch64_legitimize_address): ...here.
5174 (aarch64_builtin_vectorization_cost): Handle polynomial
5175 TYPE_VECTOR_SUBPARTS.
5176 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
5178 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
5179 number of elements from the PARALLEL rather than the mode.
5180 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
5181 rather than GET_MODE_BITSIZE.
5182 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
5183 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
5184 (aarch64_expand_vec_perm_const_1): Handle polynomial
5185 d->perm.length () and d->perm elements.
5186 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
5187 Apply to_constant to d->perm elements.
5188 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
5189 polynomial CONST_VECTOR_NUNITS.
5190 (aarch64_move_pointer): Take amount as a poly_int64 rather
5192 (aarch64_progress_pointer): Avoid temporary variable.
5193 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
5194 the mode attribute instead of GET_MODE.
5196 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5197 Alan Hayward <alan.hayward@arm.com>
5198 David Sherwood <david.sherwood@arm.com>
5200 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
5201 x exists before using it.
5202 (aarch64_add_constant_internal): Rename to...
5203 (aarch64_add_offset_1): ...this. Replace regnum with separate
5204 src and dest rtxes. Handle the case in which they're different,
5205 including when the offset is zero. Replace scratchreg with an rtx.
5206 Use 2 additions if there is no spare register into which we can
5207 move a 16-bit constant.
5208 (aarch64_add_constant): Delete.
5209 (aarch64_add_offset): Replace reg with separate src and dest
5210 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
5211 Use aarch64_add_offset_1.
5212 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
5213 an rtx rather than an int. Take the delta as a poly_int64
5214 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
5215 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
5216 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
5217 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
5218 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
5220 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
5221 aarch64_add_constant.
5223 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5225 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
5226 Use scalar_float_mode.
5228 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5230 * config/aarch64/aarch64-simd.md
5231 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
5232 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
5233 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
5234 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
5235 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
5236 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
5237 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
5238 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
5239 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
5240 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
5242 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5245 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
5246 targ_options->x_arm_arch_string is non NULL.
5248 2018-01-11 Tamar Christina <tamar.christina@arm.com>
5250 * config/aarch64/aarch64.h
5251 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
5253 2018-01-11 Sudakshina Das <sudi.das@arm.com>
5256 * expmed.c (emit_store_flag_force): Swap if const op0
5257 and change VOIDmode to mode of op0.
5259 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5261 PR rtl-optimization/83761
5262 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
5263 than bytes to mode_for_size.
5265 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5268 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
5269 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
5272 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5275 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
5276 when in layout mode.
5277 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
5278 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
5281 2018-01-10 Michael Collison <michael.collison@arm.com>
5283 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
5284 * config/aarch64/aarch64-option-extension.def: Add
5285 AARCH64_OPT_EXTENSION of 'fp16fml'.
5286 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5287 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
5288 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
5289 * config/aarch64/constraints.md (Ui7): New constraint.
5290 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
5291 (VFMLA_SEL_W): Ditto.
5294 (VFMLA16_LOW): New int iterator.
5295 (VFMLA16_HIGH): Ditto.
5296 (UNSPEC_FMLAL): New unspec.
5297 (UNSPEC_FMLSL): Ditto.
5298 (UNSPEC_FMLAL2): Ditto.
5299 (UNSPEC_FMLSL2): Ditto.
5300 (f16mac): New code attribute.
5301 * config/aarch64/aarch64-simd-builtins.def
5302 (aarch64_fmlal_lowv2sf): Ditto.
5303 (aarch64_fmlsl_lowv2sf): Ditto.
5304 (aarch64_fmlalq_lowv4sf): Ditto.
5305 (aarch64_fmlslq_lowv4sf): Ditto.
5306 (aarch64_fmlal_highv2sf): Ditto.
5307 (aarch64_fmlsl_highv2sf): Ditto.
5308 (aarch64_fmlalq_highv4sf): Ditto.
5309 (aarch64_fmlslq_highv4sf): Ditto.
5310 (aarch64_fmlal_lane_lowv2sf): Ditto.
5311 (aarch64_fmlsl_lane_lowv2sf): Ditto.
5312 (aarch64_fmlal_laneq_lowv2sf): Ditto.
5313 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
5314 (aarch64_fmlalq_lane_lowv4sf): Ditto.
5315 (aarch64_fmlsl_lane_lowv4sf): Ditto.
5316 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
5317 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
5318 (aarch64_fmlal_lane_highv2sf): Ditto.
5319 (aarch64_fmlsl_lane_highv2sf): Ditto.
5320 (aarch64_fmlal_laneq_highv2sf): Ditto.
5321 (aarch64_fmlsl_laneq_highv2sf): Ditto.
5322 (aarch64_fmlalq_lane_highv4sf): Ditto.
5323 (aarch64_fmlsl_lane_highv4sf): Ditto.
5324 (aarch64_fmlalq_laneq_highv4sf): Ditto.
5325 (aarch64_fmlsl_laneq_highv4sf): Ditto.
5326 * config/aarch64/aarch64-simd.md:
5327 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
5328 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5329 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
5330 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5331 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
5332 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
5333 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
5334 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
5335 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
5336 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
5337 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
5338 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
5339 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
5340 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
5341 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
5342 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
5343 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
5344 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
5345 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
5346 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
5347 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
5348 (vfmlsl_low_u32): Ditto.
5349 (vfmlalq_low_u32): Ditto.
5350 (vfmlslq_low_u32): Ditto.
5351 (vfmlal_high_u32): Ditto.
5352 (vfmlsl_high_u32): Ditto.
5353 (vfmlalq_high_u32): Ditto.
5354 (vfmlslq_high_u32): Ditto.
5355 (vfmlal_lane_low_u32): Ditto.
5356 (vfmlsl_lane_low_u32): Ditto.
5357 (vfmlal_laneq_low_u32): Ditto.
5358 (vfmlsl_laneq_low_u32): Ditto.
5359 (vfmlalq_lane_low_u32): Ditto.
5360 (vfmlslq_lane_low_u32): Ditto.
5361 (vfmlalq_laneq_low_u32): Ditto.
5362 (vfmlslq_laneq_low_u32): Ditto.
5363 (vfmlal_lane_high_u32): Ditto.
5364 (vfmlsl_lane_high_u32): Ditto.
5365 (vfmlal_laneq_high_u32): Ditto.
5366 (vfmlsl_laneq_high_u32): Ditto.
5367 (vfmlalq_lane_high_u32): Ditto.
5368 (vfmlslq_lane_high_u32): Ditto.
5369 (vfmlalq_laneq_high_u32): Ditto.
5370 (vfmlslq_laneq_high_u32): Ditto.
5371 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
5372 (AARCH64_FL_FOR_ARCH8_4): New.
5373 (AARCH64_ISA_F16FML): New ISA flag.
5374 (TARGET_F16FML): New feature flag for fp16fml.
5375 (doc/invoke.texi): Document new fp16fml option.
5377 2018-01-10 Michael Collison <michael.collison@arm.com>
5379 * config/aarch64/aarch64-builtins.c:
5380 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
5381 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5382 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
5383 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
5384 (AARCH64_ISA_SHA3): New ISA flag.
5385 (TARGET_SHA3): New feature flag for sha3.
5386 * config/aarch64/iterators.md (sha512_op): New int attribute.
5387 (CRYPTO_SHA512): New int iterator.
5388 (UNSPEC_SHA512H): New unspec.
5389 (UNSPEC_SHA512H2): Ditto.
5390 (UNSPEC_SHA512SU0): Ditto.
5391 (UNSPEC_SHA512SU1): Ditto.
5392 * config/aarch64/aarch64-simd-builtins.def
5393 (aarch64_crypto_sha512hqv2di): New builtin.
5394 (aarch64_crypto_sha512h2qv2di): Ditto.
5395 (aarch64_crypto_sha512su0qv2di): Ditto.
5396 (aarch64_crypto_sha512su1qv2di): Ditto.
5397 (aarch64_eor3qv8hi): Ditto.
5398 (aarch64_rax1qv2di): Ditto.
5399 (aarch64_xarqv2di): Ditto.
5400 (aarch64_bcaxqv8hi): Ditto.
5401 * config/aarch64/aarch64-simd.md:
5402 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
5403 (aarch64_crypto_sha512su0qv2di): Ditto.
5404 (aarch64_crypto_sha512su1qv2di): Ditto.
5405 (aarch64_eor3qv8hi): Ditto.
5406 (aarch64_rax1qv2di): Ditto.
5407 (aarch64_xarqv2di): Ditto.
5408 (aarch64_bcaxqv8hi): Ditto.
5409 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
5410 (vsha512h2q_u64): Ditto.
5411 (vsha512su0q_u64): Ditto.
5412 (vsha512su1q_u64): Ditto.
5413 (veor3q_u16): Ditto.
5414 (vrax1q_u64): Ditto.
5416 (vbcaxq_u16): Ditto.
5417 * config/arm/types.md (crypto_sha512): New type attribute.
5418 (crypto_sha3): Ditto.
5419 (doc/invoke.texi): Document new sha3 option.
5421 2018-01-10 Michael Collison <michael.collison@arm.com>
5423 * config/aarch64/aarch64-builtins.c:
5424 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
5425 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5426 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
5427 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
5428 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
5429 (AARCH64_ISA_SM4): New ISA flag.
5430 (TARGET_SM4): New feature flag for sm4.
5431 * config/aarch64/aarch64-simd-builtins.def
5432 (aarch64_sm3ss1qv4si): Ditto.
5433 (aarch64_sm3tt1aq4si): Ditto.
5434 (aarch64_sm3tt1bq4si): Ditto.
5435 (aarch64_sm3tt2aq4si): Ditto.
5436 (aarch64_sm3tt2bq4si): Ditto.
5437 (aarch64_sm3partw1qv4si): Ditto.
5438 (aarch64_sm3partw2qv4si): Ditto.
5439 (aarch64_sm4eqv4si): Ditto.
5440 (aarch64_sm4ekeyqv4si): Ditto.
5441 * config/aarch64/aarch64-simd.md:
5442 (aarch64_sm3ss1qv4si): Ditto.
5443 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
5444 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
5445 (aarch64_sm4eqv4si): Ditto.
5446 (aarch64_sm4ekeyqv4si): Ditto.
5447 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
5448 (sm3part_op): Ditto.
5449 (CRYPTO_SM3TT): Ditto.
5450 (CRYPTO_SM3PART): Ditto.
5451 (UNSPEC_SM3SS1): New unspec.
5452 (UNSPEC_SM3TT1A): Ditto.
5453 (UNSPEC_SM3TT1B): Ditto.
5454 (UNSPEC_SM3TT2A): Ditto.
5455 (UNSPEC_SM3TT2B): Ditto.
5456 (UNSPEC_SM3PARTW1): Ditto.
5457 (UNSPEC_SM3PARTW2): Ditto.
5458 (UNSPEC_SM4E): Ditto.
5459 (UNSPEC_SM4EKEY): Ditto.
5460 * config/aarch64/constraints.md (Ui2): New constraint.
5461 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
5462 * config/arm/types.md (crypto_sm3): New type attribute.
5463 (crypto_sm4): Ditto.
5464 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
5465 (vsm3tt1aq_u32): Ditto.
5466 (vsm3tt1bq_u32): Ditto.
5467 (vsm3tt2aq_u32): Ditto.
5468 (vsm3tt2bq_u32): Ditto.
5469 (vsm3partw1q_u32): Ditto.
5470 (vsm3partw2q_u32): Ditto.
5471 (vsm4eq_u32): Ditto.
5472 (vsm4ekeyq_u32): Ditto.
5473 (doc/invoke.texi): Document new sm4 option.
5475 2018-01-10 Michael Collison <michael.collison@arm.com>
5477 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
5478 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
5479 (AARCH64_FL_FOR_ARCH8_4): New.
5480 (AARCH64_FL_V8_4): New flag.
5481 (doc/invoke.texi): Document new armv8.4-a option.
5483 2018-01-10 Michael Collison <michael.collison@arm.com>
5485 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5486 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
5487 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
5488 * config/aarch64/aarch64-option-extension.def: Add
5489 AARCH64_OPT_EXTENSION of 'sha2'.
5490 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
5491 (crypto): Disable sha2 and aes if crypto disabled.
5492 (crypto): Enable aes and sha2 if enabled.
5493 (simd): Disable sha2 and aes if simd disabled.
5494 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
5496 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
5497 (TARGET_SHA2): New feature flag for sha2.
5498 (TARGET_AES): New feature flag for aes.
5499 * config/aarch64/aarch64-simd.md:
5500 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
5501 conditional on TARGET_AES.
5502 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
5503 (aarch64_crypto_sha1hsi): Make pattern conditional
5505 (aarch64_crypto_sha1hv4si): Ditto.
5506 (aarch64_be_crypto_sha1hv4si): Ditto.
5507 (aarch64_crypto_sha1su1v4si): Ditto.
5508 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
5509 (aarch64_crypto_sha1su0v4si): Ditto.
5510 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
5511 (aarch64_crypto_sha256su0v4si): Ditto.
5512 (aarch64_crypto_sha256su1v4si): Ditto.
5513 (doc/invoke.texi): Document new aes and sha2 options.
5515 2018-01-10 Martin Sebor <msebor@redhat.com>
5517 PR tree-optimization/83781
5518 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
5521 2018-01-11 Martin Sebor <msebor@gmail.com>
5522 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5524 PR tree-optimization/83501
5525 PR tree-optimization/81703
5527 * tree-ssa-strlen.c (get_string_cst): Rename...
5528 (get_string_len): ...to this. Handle global constants.
5529 (handle_char_store): Adjust.
5531 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
5532 Jim Wilson <jimw@sifive.com>
5534 * config/riscv/riscv-protos.h (riscv_output_return): New.
5535 * config/riscv/riscv.c (struct machine_function): New naked_p field.
5536 (riscv_attribute_table, riscv_output_return),
5537 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
5538 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
5539 (riscv_compute_frame_info): Only compute frame->mask if not a naked
5541 (riscv_expand_prologue): Add early return for naked function.
5542 (riscv_expand_epilogue): Likewise.
5543 (riscv_function_ok_for_sibcall): Return false for naked function.
5544 (riscv_set_current_function): New.
5545 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
5546 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
5547 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
5548 * doc/extend.texi (RISC-V Function Attributes): New.
5550 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
5552 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
5553 check for 128-bit long double before checking TCmode.
5554 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
5555 128-bit long doubles before checking TFmode or TCmode.
5556 (FLOAT128_IBM_P): Likewise.
5558 2018-01-10 Martin Sebor <msebor@redhat.com>
5560 PR tree-optimization/83671
5561 * builtins.c (c_strlen): Unconditionally return zero for the empty
5563 Use -Warray-bounds for warnings.
5564 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
5565 for non-constant array indices with COMPONENT_REF, arrays of
5566 arrays, and pointers to arrays.
5567 (gimple_fold_builtin_strlen): Determine and set length range for
5568 non-constant character arrays.
5570 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
5573 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
5576 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
5578 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
5580 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5583 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
5584 VECTOR_MEM_ALTIVEC_OR_VSX_P.
5585 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
5586 indexed_or_indirect_operand predicate.
5587 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
5588 (*vsx_le_perm_load_v8hi): Likewise.
5589 (*vsx_le_perm_load_v16qi): Likewise.
5590 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
5591 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
5592 (*vsx_le_perm_store_v8hi): Likewise.
5593 (*vsx_le_perm_store_v16qi): Likewise.
5594 (eight unnamed splitters): Likewise.
5596 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5598 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
5599 * config/rs6000/emmintrin.h: Likewise.
5600 * config/rs6000/mmintrin.h: Likewise.
5601 * config/rs6000/xmmintrin.h: Likewise.
5603 2018-01-10 David Malcolm <dmalcolm@redhat.com>
5606 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
5608 * tree.c (tree_nop_conversion): Return true for location wrapper
5610 (maybe_wrap_with_location): New function.
5611 (selftest::check_strip_nops): New function.
5612 (selftest::test_location_wrappers): New function.
5613 (selftest::tree_c_tests): Call it.
5614 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
5615 (maybe_wrap_with_location): New decl.
5616 (EXPR_LOCATION_WRAPPER_P): New macro.
5617 (location_wrapper_p): New inline function.
5618 (tree_strip_any_location_wrapper): New inline function.
5620 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5623 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5624 stack_realign_offset for the largest alignment of stack slot
5626 (ix86_find_max_used_stack_alignment): New function.
5627 (ix86_finalize_stack_frame_flags): Use it. Set
5628 max_used_stack_alignment if we don't realign stack.
5629 * config/i386/i386.h (machine_function): Add
5630 max_used_stack_alignment.
5632 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5634 * config/arm/arm.opt (-mbranch-cost): New option.
5635 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5638 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5641 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5642 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5644 2018-01-10 Richard Biener <rguenther@suse.de>
5647 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5648 early out so it also covers the case where we have a non-NULL
5651 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5653 PR tree-optimization/83753
5654 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5655 for non-strided grouped accesses if the number of elements is 1.
5657 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5660 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5661 * i386.h (TARGET_USE_GATHER): Define.
5662 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5664 2018-01-10 Martin Liska <mliska@suse.cz>
5667 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5668 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5670 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5671 CLEANUP_NO_PARTITIONING is not set.
5673 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5675 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5676 for vectors, as a partial revert of r254296.
5677 * rtl.h (const_vec_p): Delete.
5678 (const_vec_duplicate_p): Don't test for vector CONSTs.
5679 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5680 * expmed.c (make_tree): Likewise.
5683 * common.md (E, F): Use CONSTANT_P instead of checking for
5685 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5686 checking for CONST_VECTOR.
5688 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5691 * predict.c (force_edge_cold): Handle in more sane way edges
5694 2018-01-09 Carl Love <cel@us.ibm.com>
5696 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5698 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5699 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5700 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5701 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5702 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5703 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5704 * config/rs6000/rs6000-protos.h: Add extern defition for
5705 rs6000_generate_float2_double_code.
5706 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5708 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5709 (float2_v2df): Add define_expand.
5711 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5714 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5715 op_mode in the force_to_mode call.
5717 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5719 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5720 instead of checking each element individually.
5721 (aarch64_evpc_uzp): Likewise.
5722 (aarch64_evpc_zip): Likewise.
5723 (aarch64_evpc_ext): Likewise.
5724 (aarch64_evpc_rev): Likewise.
5725 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5726 instead of checking each element individually. Return true without
5728 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5729 whether all selected elements come from the same input, instead of
5730 checking each element individually. Remove calls to gen_rtx_REG,
5731 start_sequence and end_sequence and instead assert that no rtl is
5734 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5736 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5737 order of HIGH and CONST checks.
5739 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5741 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5742 if the destination isn't an SSA_NAME.
5744 2018-01-09 Richard Biener <rguenther@suse.de>
5746 PR tree-optimization/83668
5747 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5749 (canonicalize_loop_form): ... here, renamed from ...
5750 (canonicalize_loop_closed_ssa_form): ... this and amended to
5751 swap successor edges for loop exit blocks to make us use
5752 the RPO order we need for initial schedule generation.
5754 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5756 PR tree-optimization/64811
5757 * match.pd: When optimizing comparisons with Inf, avoid
5758 introducing or losing exceptions from comparisons with NaN.
5760 2018-01-09 Martin Liska <mliska@suse.cz>
5763 * asan.c (shadow_mem_size): Add gcc_assert.
5765 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5767 Don't save registers in main().
5770 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5771 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5772 * config/avr/avr.c (avr_set_current_function): Don't error if
5773 naked, OS_task or OS_main are specified at the same time.
5774 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5776 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5778 * common/config/avr/avr-common.c (avr_option_optimization_table):
5779 Switch on -mmain-is-OS_task for optimizing compilations.
5781 2018-01-09 Richard Biener <rguenther@suse.de>
5783 PR tree-optimization/83572
5784 * graphite.c: Include cfganal.h.
5785 (graphite_transform_loops): Connect infinite loops to exit
5786 and remove fake edges at the end.
5788 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5790 * ipa-inline.c (edge_badness): Revert accidental checkin.
5792 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5795 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5796 symbols; not inline clones.
5798 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5801 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5802 hard registers. Formatting fixes.
5804 PR preprocessor/83722
5805 * gcc.c (try_generate_repro): Pass
5806 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5807 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5810 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5811 Kito Cheng <kito.cheng@gmail.com>
5813 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5814 (riscv_leaf_function_p): Delete.
5815 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5817 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5819 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5821 (do_ifelse): New function.
5822 (do_isel): New function.
5823 (do_sub3): New function.
5824 (do_add3): New function.
5825 (do_load_mask_compare): New function.
5826 (do_overlap_load_compare): New function.
5827 (expand_compare_loop): New function.
5828 (expand_block_compare): Call expand_compare_loop() when appropriate.
5829 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5831 (-mblock-compare-inline-loop-limit): New option.
5833 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5836 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5837 Reverse order of second and third operands in first alternative.
5838 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5839 of first and second elements in UNSPEC_VPERMR vector.
5840 (altivec_expand_vec_perm_le): Likewise.
5842 2018-01-08 Jeff Law <law@redhat.com>
5844 PR rtl-optimizatin/81308
5845 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5846 (process_switch): If group_case_labels makes a change, then set
5848 (pass_convert_switch::execute): If a switch is converted, then
5849 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5851 PR rtl-optimization/81308
5852 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5855 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5857 PR target/83663 - Revert r255946
5858 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5859 generation for cases where splatting a value is not useful.
5860 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5861 across a vec_duplicate and a paradoxical subreg forming a vector
5862 mode to a vec_concat.
5864 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5866 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5867 -march=armv8.3-a variants.
5868 * config/arm/t-multilib: Likewise.
5869 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5871 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5873 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5875 (cceq_ior_compare_complement): Give it a name so I can use it, and
5876 change boolean_or_operator predicate to boolean_operator so it can
5877 be used to generate a crand.
5878 (eqne): New code iterator.
5879 (bd/bd_neg): New code_attrs.
5880 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5881 a single define_insn.
5882 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5883 decrement (bdnzt/bdnzf/bdzt/bdzf).
5884 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5885 with the new names of the branch decrement patterns, and added the
5886 names of the branch decrement conditional patterns.
5888 2018-01-08 Richard Biener <rguenther@suse.de>
5890 PR tree-optimization/83563
5891 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5894 2018-01-08 Richard Biener <rguenther@suse.de>
5897 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5899 2018-01-08 Richard Biener <rguenther@suse.de>
5901 PR tree-optimization/83685
5902 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5903 references to abnormals.
5905 2018-01-08 Richard Biener <rguenther@suse.de>
5908 * dwarf2out.c (output_indirect_strings): Handle empty
5909 skeleton_debug_str_hash.
5910 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5912 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5914 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5915 (emit_store_direct): Likewise.
5916 (arc_trampoline_adjust_address): Likewise.
5917 (arc_asm_trampoline_template): New function.
5918 (arc_initialize_trampoline): Use asm_trampoline_template.
5919 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5920 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5921 * config/arc/arc.md (flush_icache): Delete pattern.
5923 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5925 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5926 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5929 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5932 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5933 by not USED_FOR_TARGET.
5934 (make_pass_resolve_sw_modes): Likewise.
5936 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5938 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5941 2018-01-08 Richard Biener <rguenther@suse.de>
5944 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5946 2018-01-08 Richard Biener <rguenther@suse.de>
5949 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5951 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5954 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5955 basic blocks with a small number of successors.
5956 (convert_control_dep_chain_into_preds): Improve handling of
5958 (dump_predicates): Split apart into...
5959 (dump_pred_chain): ...here...
5960 (dump_pred_info): ...and here.
5961 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5962 (can_chain_union_be_invalidated_p): Improve check for invalidation
5964 (uninit_uses_cannot_happen): Avoid unnecessary if
5965 convert_control_dep_chain_into_preds yielded nothing.
5967 2018-01-06 Martin Sebor <msebor@redhat.com>
5969 PR tree-optimization/83640
5970 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5971 subtracting negative offset from size.
5972 (builtin_access::overlap): Adjust offset bounds of the access to fall
5973 within the size of the object if possible.
5975 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5977 PR rtl-optimization/83699
5978 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5979 extract_bit_field_as_subreg to cases in which the extracted
5980 value is also a vector.
5982 * lra-constraints.c (process_alt_operands): Test for the equivalence
5983 substitutions when detecting a possible reload cycle.
5985 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5988 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5989 by default if flag_selective_schedling{,2}. Formatting fixes.
5991 PR rtl-optimization/83682
5992 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5993 if it has non-VECTOR_MODE element mode.
5994 (vec_duplicate_p): Likewise.
5997 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5998 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
6000 2018-01-05 Jakub Jelinek <jakub@redhat.com>
6003 * config/i386/i386-builtin.def
6004 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
6005 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
6006 Require also OPTION_MASK_ISA_AVX512F in addition to
6007 OPTION_MASK_ISA_GFNI.
6008 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
6009 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
6010 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
6011 to OPTION_MASK_ISA_GFNI.
6012 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
6013 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
6014 OPTION_MASK_ISA_AVX512BW.
6015 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
6016 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
6017 addition to OPTION_MASK_ISA_GFNI.
6018 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
6019 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
6020 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
6021 to OPTION_MASK_ISA_GFNI.
6022 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
6023 a requirement for all ISAs rather than any of them with a few
6025 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
6027 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
6028 bitmasks to be enabled with 3 exceptions, instead of requiring any
6029 enabled ISA with lots of exceptions.
6030 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
6031 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
6032 Change avx512bw in isa attribute to avx512f.
6033 * config/i386/sgxintrin.h: Add license boilerplate.
6034 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
6035 to __AVX512F__ and __AVX512VL to __AVX512VL__.
6036 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
6037 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
6039 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
6040 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
6041 temporarily sse2 rather than sse if not enabled already.
6044 * config/i386/sse.md (VI248_VLBW): Rename to ...
6045 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
6046 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
6047 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
6048 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
6049 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
6050 mode iterator instead of VI248_VLBW.
6052 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
6054 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
6055 (record_modified): Skip clobbers; add debug output.
6056 (param_change_prob): Use sreal frequencies.
6058 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
6060 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
6061 punt for user-aligned variables.
6063 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
6065 * tree-chrec.c (chrec_contains_symbols): Return true for
6068 2018-01-05 Sudakshina Das <sudi.das@arm.com>
6071 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
6072 of (x|y) == x for BICS pattern.
6074 2018-01-05 Jakub Jelinek <jakub@redhat.com>
6076 PR tree-optimization/83605
6077 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
6078 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
6081 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
6083 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
6084 * config/epiphany/rtems.h: New file.
6086 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6087 Uros Bizjak <ubizjak@gmail.com>
6090 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
6091 QIreg_operand instead of register_operand predicate.
6092 * config/i386/i386.c (ix86_rop_should_change_byte_p,
6093 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
6094 comments instead of -fmitigate[-_]rop.
6096 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
6099 * cgraphunit.c (symbol_table::compile): Switch to text_section
6100 before calling assembly_start debug hook.
6101 * run-rtl-passes.c (run_rtl_passes): Likewise.
6104 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6106 * tree-vrp.c (extract_range_from_binary_expr_1): Check
6107 range_int_cst_p rather than !symbolic_range_p before calling
6108 extract_range_from_multiplicative_op_1.
6110 2018-01-04 Jeff Law <law@redhat.com>
6112 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
6113 redundant test in assertion.
6115 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6117 * doc/rtl.texi: Document machine_mode wrapper classes.
6119 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6121 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
6124 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6126 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
6127 the VEC_PERM_EXPR fold to fail.
6129 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6132 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
6133 to switched_sections.
6135 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6138 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
6141 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
6144 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
6145 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
6147 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6150 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
6151 is BLKmode and bitpos not zero or mode change is needed.
6153 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6156 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
6159 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
6162 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
6163 instead of MULT rtx. Update all corresponding splitters.
6165 (*ssub<modesuffix>): Ditto.
6167 (*cmp_sadd_di): Update split patterns.
6168 (*cmp_sadd_si): Ditto.
6169 (*cmp_sadd_sidi): Ditto.
6170 (*cmp_ssub_di): Ditto.
6171 (*cmp_ssub_si): Ditto.
6172 (*cmp_ssub_sidi): Ditto.
6173 * config/alpha/predicates.md (const23_operand): New predicate.
6174 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
6175 Look for ASHIFT, not MULT inner operand.
6176 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
6178 2018-01-04 Martin Liska <mliska@suse.cz>
6180 PR gcov-profile/83669
6181 * gcov.c (output_intermediate_file): Add version to intermediate
6183 * doc/gcov.texi: Document new field 'version' in intermediate
6184 file format. Fix location of '-k' option of gcov command.
6186 2018-01-04 Martin Liska <mliska@suse.cz>
6189 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
6191 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6193 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
6195 2018-01-03 Martin Sebor <msebor@redhat.com>
6197 PR tree-optimization/83655
6198 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
6199 checking calls with invalid arguments.
6201 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6203 * tree-vect-stmts.c (vect_get_store_rhs): New function.
6204 (vectorizable_mask_load_store): Delete.
6205 (vectorizable_call): Return false for masked loads and stores.
6206 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
6207 instead of gimple_assign_rhs1.
6208 (vectorizable_load): Handle IFN_MASK_LOAD.
6209 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
6211 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6213 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
6215 (vectorizable_mask_load_store): ...here.
6216 (vectorizable_load): ...and here.
6218 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6220 * tree-vect-stmts.c (vect_build_all_ones_mask)
6221 (vect_build_zero_merge_argument): New functions, split out from...
6222 (vectorizable_load): ...here.
6224 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6226 * tree-vect-stmts.c (vect_check_store_rhs): New function,
6228 (vectorizable_mask_load_store): ...here.
6229 (vectorizable_store): ...and here.
6231 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6233 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
6235 (vectorizable_mask_load_store): ...here.
6237 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6239 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
6240 (vect_model_store_cost): Take a vec_load_store_type instead of a
6242 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
6243 (vect_model_store_cost): Take a vec_load_store_type instead of a
6245 (vectorizable_mask_load_store): Update accordingly.
6246 (vectorizable_store): Likewise.
6247 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
6249 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6251 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
6252 IFN_MASK_LOAD calls here rather than...
6253 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
6255 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6256 Alan Hayward <alan.hayward@arm.com>
6257 David Sherwood <david.sherwood@arm.com>
6259 * expmed.c (extract_bit_field_1): For vector extracts,
6260 fall back to extract_bit_field_as_subreg if vec_extract
6263 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6264 Alan Hayward <alan.hayward@arm.com>
6265 David Sherwood <david.sherwood@arm.com>
6267 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
6268 they are variable or constant sized.
6269 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
6270 slots for constant-sized data.
6272 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6273 Alan Hayward <alan.hayward@arm.com>
6274 David Sherwood <david.sherwood@arm.com>
6276 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
6277 handling COND_EXPRs with boolean comparisons, try to find a better
6278 basis for the mask type than the boolean itself.
6280 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6282 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
6283 is calculated and how it can be overridden.
6284 * genmodes.c (max_bitsize_mode_any_mode): New variable.
6285 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
6287 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
6290 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6291 Alan Hayward <alan.hayward@arm.com>
6292 David Sherwood <david.sherwood@arm.com>
6294 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
6295 Remove the mode argument.
6296 (aarch64_simd_valid_immediate): Remove the mode and inverse
6298 * config/aarch64/iterators.md (bitsize): New iterator.
6299 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
6300 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
6301 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
6302 aarch64_simd_valid_immediate.
6303 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
6304 (aarch64_reg_or_bic_imm): Likewise.
6305 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
6306 with an insn_type enum and msl with a modifier_type enum.
6307 Replace element_width with a scalar_mode. Change the shift
6308 to unsigned int. Add constructors for scalar_float_mode and
6309 scalar_int_mode elements.
6310 (aarch64_vect_float_const_representable_p): Delete.
6311 (aarch64_can_const_movi_rtx_p)
6312 (aarch64_simd_scalar_immediate_valid_for_move)
6313 (aarch64_simd_make_constant): Update call to
6314 aarch64_simd_valid_immediate.
6315 (aarch64_advsimd_valid_immediate_hs): New function.
6316 (aarch64_advsimd_valid_immediate): Likewise.
6317 (aarch64_simd_valid_immediate): Remove mode and inverse
6318 arguments. Rewrite to use the above. Use const_vec_duplicate_p
6319 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
6320 and aarch64_float_const_representable_p on the result.
6321 (aarch64_output_simd_mov_immediate): Remove mode argument.
6322 Update call to aarch64_simd_valid_immediate and use of
6323 simd_immediate_info.
6324 (aarch64_output_scalar_simd_mov_immediate): Update call
6327 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6328 Alan Hayward <alan.hayward@arm.com>
6329 David Sherwood <david.sherwood@arm.com>
6331 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
6332 (mode_nunits): Likewise CONST_MODE_NUNITS.
6333 * machmode.def (ADJUST_NUNITS): Document.
6334 * genmodes.c (mode_data::need_nunits_adj): New field.
6335 (blank_mode): Update accordingly.
6336 (adj_nunits): New variable.
6337 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
6339 (emit_mode_size_inline): Set need_bytesize_adj for all modes
6340 listed in adj_nunits.
6341 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
6342 listed in adj_nunits. Don't emit case statements for such modes.
6343 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
6344 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
6345 nothing if adj_nunits is nonnull.
6346 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
6347 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
6348 (emit_mode_fbit): Update use of print_maybe_const_decl.
6349 (emit_move_size): Likewise. Treat the array as non-const
6351 (emit_mode_adjustments): Handle adj_nunits.
6353 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6355 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
6356 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
6357 (VECTOR_MODES): Use it.
6358 (make_vector_modes): Take the prefix as an argument.
6360 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6361 Alan Hayward <alan.hayward@arm.com>
6362 David Sherwood <david.sherwood@arm.com>
6364 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
6365 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
6366 for MODE_VECTOR_BOOL.
6367 * machmode.def (VECTOR_BOOL_MODE): Document.
6368 * genmodes.c (VECTOR_BOOL_MODE): New macro.
6369 (make_vector_bool_mode): New function.
6370 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
6372 * lto-streamer-in.c (lto_input_mode_table): Likewise.
6373 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
6375 * stor-layout.c (int_mode_for_mode): Likewise.
6376 * tree.c (build_vector_type_for_mode): Likewise.
6377 * varasm.c (output_constant_pool_2): Likewise.
6378 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
6379 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
6380 for MODE_VECTOR_BOOL.
6381 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
6382 of mode class checks.
6383 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
6384 instead of a list of mode class checks.
6385 (expand_vector_scalar_condition): Likewise.
6386 (type_for_widest_vector_mode): Handle BImode as an inner mode.
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 * machmode.h (mode_size): Change from unsigned short to
6394 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
6395 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6396 or if measurement_type is not polynomial.
6397 (fixed_size_mode::includes_p): Check for constant-sized modes.
6398 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
6399 return a poly_uint16 rather than an unsigned short.
6400 (emit_mode_size): Change the type of mode_size from unsigned short
6401 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
6402 (emit_mode_adjustments): Cope with polynomial vector sizes.
6403 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6405 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6407 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
6408 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
6409 * caller-save.c (setup_save_areas): Likewise.
6410 (replace_reg_with_saved_mem): Likewise.
6411 * calls.c (emit_library_call_value_1): Likewise.
6412 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
6413 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
6414 (gen_lowpart_for_combine): Likewise.
6415 * convert.c (convert_to_integer_1): Likewise.
6416 * cse.c (equiv_constant, cse_insn): Likewise.
6417 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
6418 (cselib_subst_to_values): Likewise.
6419 * dce.c (word_dce_process_block): Likewise.
6420 * df-problems.c (df_word_lr_mark_ref): Likewise.
6421 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
6422 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
6423 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
6424 (rtl_for_decl_location): Likewise.
6425 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
6426 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
6427 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
6428 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
6429 (expand_expr_real_1): Likewise.
6430 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
6431 (pad_below): Likewise.
6432 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6433 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
6434 * ira.c (get_subreg_tracking_sizes): Likewise.
6435 * ira-build.c (ira_create_allocno_objects): Likewise.
6436 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
6437 (ira_sort_regnos_for_alter_reg): Likewise.
6438 * ira-costs.c (record_operand_costs): Likewise.
6439 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
6440 (resolve_simple_move): Likewise.
6441 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
6442 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
6443 (lra_constraints): Likewise.
6444 (CONST_POOL_OK_P): Reject variable-sized modes.
6445 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
6446 (add_pseudo_to_slot, lra_spill): Likewise.
6447 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6448 * optabs-query.c (get_best_extraction_insn): Likewise.
6449 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6450 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
6451 (expand_mult_highpart, valid_multiword_target_p): Likewise.
6452 * recog.c (offsettable_address_addr_space_p): Likewise.
6453 * regcprop.c (maybe_mode_change): Likewise.
6454 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
6455 * regrename.c (build_def_use): Likewise.
6456 * regstat.c (dump_reg_info): Likewise.
6457 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
6458 (find_reloads, find_reloads_subreg_address): Likewise.
6459 * reload1.c (eliminate_regs_1): Likewise.
6460 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
6461 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
6462 (simplify_binary_operation_1, simplify_subreg): Likewise.
6463 * targhooks.c (default_function_arg_padding): Likewise.
6464 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
6465 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
6466 (verify_gimple_assign_ternary): Likewise.
6467 * tree-inline.c (estimate_move_cost): Likewise.
6468 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6469 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
6470 (get_address_cost_ainc): Likewise.
6471 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
6472 (vect_supportable_dr_alignment): Likewise.
6473 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6474 (vectorizable_reduction): Likewise.
6475 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
6476 (vectorizable_operation, vectorizable_load): Likewise.
6477 * tree.c (build_same_sized_truth_vector_type): Likewise.
6478 * valtrack.c (cleanup_auto_inc_dec): Likewise.
6479 * var-tracking.c (emit_note_insn_var_location): Likewise.
6480 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
6481 (ADDR_VEC_ALIGN): Likewise.
6483 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6484 Alan Hayward <alan.hayward@arm.com>
6485 David Sherwood <david.sherwood@arm.com>
6487 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
6489 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6490 or if measurement_type is polynomial.
6491 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
6492 * combine.c (make_extraction): Likewise.
6493 * dse.c (find_shift_sequence): Likewise.
6494 * dwarf2out.c (mem_loc_descriptor): Likewise.
6495 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
6496 (extract_bit_field, extract_low_bits): Likewise.
6497 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
6498 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
6499 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
6500 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
6501 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6502 * reload.c (find_reloads): Likewise.
6503 * reload1.c (alter_reg): Likewise.
6504 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
6505 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
6506 * tree-if-conv.c (predicate_mem_writes): Likewise.
6507 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
6508 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
6509 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
6510 * valtrack.c (dead_debug_insert_temp): Likewise.
6511 * varasm.c (mergeable_constant_section): Likewise.
6512 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
6514 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6515 Alan Hayward <alan.hayward@arm.com>
6516 David Sherwood <david.sherwood@arm.com>
6518 * expr.c (expand_assignment): Cope with polynomial mode sizes
6519 when assigning to a CONCAT.
6521 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6522 Alan Hayward <alan.hayward@arm.com>
6523 David Sherwood <david.sherwood@arm.com>
6525 * machmode.h (mode_precision): Change from unsigned short to
6527 (mode_to_precision): Return a poly_uint16 rather than an unsigned
6529 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
6530 or if measurement_type is not polynomial.
6531 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
6532 in which the mode is already known to be a scalar_int_mode.
6533 * genmodes.c (emit_mode_precision): Change the type of mode_precision
6534 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
6536 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6537 for GET_MODE_PRECISION.
6538 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6539 for GET_MODE_PRECISION.
6540 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
6542 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
6543 (expand_field_assignment, make_extraction): Likewise.
6544 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
6545 (get_last_value): Likewise.
6546 * convert.c (convert_to_integer_1): Likewise.
6547 * cse.c (cse_insn): Likewise.
6548 * expr.c (expand_expr_real_1): Likewise.
6549 * lra-constraints.c (simplify_operand_subreg): Likewise.
6550 * optabs-query.c (can_atomic_load_p): Likewise.
6551 * optabs.c (expand_atomic_load): Likewise.
6552 (expand_atomic_store): Likewise.
6553 * ree.c (combine_reaching_defs): Likewise.
6554 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
6555 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
6556 * tree.h (type_has_mode_precision_p): Likewise.
6557 * ubsan.c (instrument_si_overflow): Likewise.
6559 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6560 Alan Hayward <alan.hayward@arm.com>
6561 David Sherwood <david.sherwood@arm.com>
6563 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
6564 polynomial numbers of units.
6565 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
6566 (valid_vector_subparts_p): New function.
6567 (build_vector_type): Remove temporary shim and take the number
6568 of units as a poly_uint64 rather than an int.
6569 (build_opaque_vector_type): Take the number of units as a
6570 poly_uint64 rather than an int.
6571 * tree.c (build_vector_from_ctor): Handle polynomial
6572 TYPE_VECTOR_SUBPARTS.
6573 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
6574 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
6575 (build_vector_from_val): If the number of units is variable,
6576 use build_vec_duplicate_cst for constant operands and
6577 VEC_DUPLICATE_EXPR otherwise.
6578 (make_vector_type): Remove temporary is_constant ().
6579 (build_vector_type, build_opaque_vector_type): Take the number of
6580 units as a poly_uint64 rather than an int.
6581 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
6583 * cfgexpand.c (expand_debug_expr): Likewise.
6584 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
6585 (store_constructor, expand_expr_real_1): Likewise.
6586 (const_scalar_mask_from_tree): Likewise.
6587 * fold-const-call.c (fold_const_reduction): Likewise.
6588 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
6589 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
6590 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
6591 (fold_relational_const): Likewise.
6592 (native_interpret_vector): Likewise. Change the size from an
6593 int to an unsigned int.
6594 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
6595 TYPE_VECTOR_SUBPARTS.
6596 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
6597 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
6598 duplicating a non-constant operand into a variable-length vector.
6599 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
6600 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
6601 * ipa-icf.c (sem_variable::equals): Likewise.
6602 * match.pd: Likewise.
6603 * omp-simd-clone.c (simd_clone_subparts): Likewise.
6604 * print-tree.c (print_node): Likewise.
6605 * stor-layout.c (layout_type): Likewise.
6606 * targhooks.c (default_builtin_vectorization_cost): Likewise.
6607 * tree-cfg.c (verify_gimple_comparison): Likewise.
6608 (verify_gimple_assign_binary): Likewise.
6609 (verify_gimple_assign_ternary): Likewise.
6610 (verify_gimple_assign_single): Likewise.
6611 * tree-pretty-print.c (dump_generic_node): Likewise.
6612 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6613 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
6614 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
6615 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
6616 (vect_shift_permute_load_chain): Likewise.
6617 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
6618 (expand_vector_condition, optimize_vector_constructor): Likewise.
6619 (lower_vec_perm, get_compute_type): Likewise.
6620 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6621 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6622 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6623 (vect_recog_mask_conversion_pattern): Likewise.
6624 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6625 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6626 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6627 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6628 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6629 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6630 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6631 (supportable_widening_operation): Likewise.
6632 (supportable_narrowing_operation): Likewise.
6633 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6635 * varasm.c (output_constant): Likewise.
6637 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6638 Alan Hayward <alan.hayward@arm.com>
6639 David Sherwood <david.sherwood@arm.com>
6641 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6642 so that both the length == 3 and length != 3 cases set up their
6643 own permute vectors. Add comments explaining why we know the
6644 number of elements is constant.
6645 (vect_permute_load_chain): Likewise.
6647 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6648 Alan Hayward <alan.hayward@arm.com>
6649 David Sherwood <david.sherwood@arm.com>
6651 * machmode.h (mode_nunits): Change from unsigned char to
6653 (ONLY_FIXED_SIZE_MODES): New macro.
6654 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6655 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6656 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6658 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6659 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6660 or if measurement_type is not polynomial.
6661 * genmodes.c (ZERO_COEFFS): New macro.
6662 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6664 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6665 Use ZERO_COEFFS when emitting initializers.
6666 * data-streamer.h (bp_pack_poly_value): New function.
6667 (bp_unpack_poly_value): Likewise.
6668 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6669 for GET_MODE_NUNITS.
6670 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6671 for GET_MODE_NUNITS.
6672 * tree.c (make_vector_type): Remove temporary shim and make
6673 the real function take the number of units as a poly_uint64
6675 (build_vector_type_for_mode): Handle polynomial nunits.
6676 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6677 * emit-rtl.c (const_vec_series_p_1): Likewise.
6678 (gen_rtx_CONST_VECTOR): Likewise.
6679 * fold-const.c (test_vec_duplicate_folding): Likewise.
6680 * genrecog.c (validate_pattern): Likewise.
6681 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6682 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6683 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6684 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6685 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6686 * rtlanal.c (subreg_get_info): Likewise.
6687 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6688 (vect_grouped_load_supported): Likewise.
6689 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6690 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6691 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6692 (simplify_const_unary_operation, simplify_binary_operation_1)
6693 (simplify_const_binary_operation, simplify_ternary_operation)
6694 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6695 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6696 instead of CONST_VECTOR_NUNITS.
6697 * varasm.c (output_constant_pool_2): Likewise.
6698 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6699 explicit-encoded elements in the XVEC for variable-length vectors.
6701 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6703 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6705 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6706 Alan Hayward <alan.hayward@arm.com>
6707 David Sherwood <david.sherwood@arm.com>
6709 * coretypes.h (fixed_size_mode): Declare.
6710 (fixed_size_mode_pod): New typedef.
6711 * builtins.h (target_builtins::x_apply_args_mode)
6712 (target_builtins::x_apply_result_mode): Change type to
6713 fixed_size_mode_pod.
6714 * builtins.c (apply_args_size, apply_result_size, result_vector)
6715 (expand_builtin_apply_args_1, expand_builtin_apply)
6716 (expand_builtin_return): Update accordingly.
6718 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6720 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6721 * cselib.c (cselib_hash_rtx): Likewise.
6722 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6723 CONST_VECTOR encoding.
6725 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6726 Jeff Law <law@redhat.com>
6729 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6730 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6731 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6732 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6735 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6736 explicitly probe *sp in a noreturn function if there were any callee
6737 register saves or frame pointer is needed.
6739 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6742 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6743 BLKmode for ternary, binary or unary expressions.
6746 * var-tracking.c (delete_vta_debug_insn): New inline function.
6747 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6748 insns from get_insns () to NULL instead of each bb separately.
6749 Use delete_vta_debug_insn. No longer static.
6750 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6751 delete_vta_debug_insns callers.
6752 * rtl.h (delete_vta_debug_insns): Declare.
6753 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6754 instead of variable_tracking_main.
6756 2018-01-03 Martin Sebor <msebor@redhat.com>
6758 PR tree-optimization/83603
6759 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6760 arguments past the endof the argument list in functions declared
6761 without a prototype.
6762 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6763 Avoid checking when arguments are null.
6765 2018-01-03 Martin Sebor <msebor@redhat.com>
6768 * doc/extend.texi (attribute const): Fix a typo.
6769 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6770 issuing -Wsuggest-attribute for void functions.
6772 2018-01-03 Martin Sebor <msebor@redhat.com>
6774 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6775 offset_int::from instead of wide_int::to_shwi.
6776 (maybe_diag_overlap): Remove assertion.
6777 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6778 * gimple-ssa-sprintf.c (format_directive): Same.
6779 (parse_directive): Same.
6780 (sprintf_dom_walker::compute_format_length): Same.
6781 (try_substitute_return_value): Same.
6783 2018-01-03 Jeff Law <law@redhat.com>
6786 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6787 non-constant residual for zero at runtime and avoid probing in
6788 that case. Reorganize code for trailing problem to mirror handling
6791 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6793 PR tree-optimization/83501
6794 * tree-ssa-strlen.c (get_string_cst): New.
6795 (handle_char_store): Call get_string_cst.
6797 2018-01-03 Martin Liska <mliska@suse.cz>
6799 PR tree-optimization/83593
6800 * tree-ssa-strlen.c: Include tree-cfg.h.
6801 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6802 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6803 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6805 (strlen_dom_walker::before_dom_children): Call
6806 gimple_purge_dead_eh_edges. Dump tranformation with details
6808 (strlen_dom_walker::before_dom_children): Update call by adding
6809 new argument cleanup_eh.
6810 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6812 2018-01-03 Martin Liska <mliska@suse.cz>
6815 * cif-code.def (VARIADIC_THUNK): New enum value.
6816 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6819 2018-01-03 Jan Beulich <jbeulich@suse.com>
6821 * sse.md (mov<mode>_internal): Tighten condition for when to use
6822 vmovdqu<ssescalarsize> for TI and OI modes.
6824 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6826 Update copyright years.
6828 2018-01-03 Martin Liska <mliska@suse.cz>
6831 * ipa-visibility.c (function_and_variable_visibility): Skip
6832 functions with noipa attribure.
6834 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6836 * gcc.c (process_command): Update copyright notice dates.
6837 * gcov-dump.c (print_version): Ditto.
6838 * gcov.c (print_version): Ditto.
6839 * gcov-tool.c (print_version): Ditto.
6840 * gengtype.c (create_file): Ditto.
6841 * doc/cpp.texi: Bump @copying's copyright year.
6842 * doc/cppinternals.texi: Ditto.
6843 * doc/gcc.texi: Ditto.
6844 * doc/gccint.texi: Ditto.
6845 * doc/gcov.texi: Ditto.
6846 * doc/install.texi: Ditto.
6847 * doc/invoke.texi: Ditto.
6849 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6851 * vector-builder.h (vector_builder::m_full_nelts): Change from
6852 unsigned int to poly_uint64.
6853 (vector_builder::full_nelts): Update prototype accordingly.
6854 (vector_builder::new_vector): Likewise.
6855 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6856 (vector_builder::operator ==): Likewise.
6857 (vector_builder::finalize): Likewise.
6858 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6859 Take the number of elements as a poly_uint64 rather than an
6861 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6862 from unsigned int to poly_uint64.
6863 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6864 (vec_perm_indices::new_vector): Likewise.
6865 (vec_perm_indices::length): Likewise.
6866 (vec_perm_indices::nelts_per_input): Likewise.
6867 (vec_perm_indices::input_nelts): Likewise.
6868 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6869 number of elements per input as a poly_uint64 rather than an
6870 unsigned int. Use the original encoding for variable-length
6871 vectors, rather than clamping each individual element.
6872 For the second and subsequent elements in each pattern,
6873 clamp the step and base before clamping their sum.
6874 (vec_perm_indices::series_p): Handle polynomial element counts.
6875 (vec_perm_indices::all_in_range_p): Likewise.
6876 (vec_perm_indices_to_tree): Likewise.
6877 (vec_perm_indices_to_rtx): Likewise.
6878 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6879 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6880 (tree_vector_builder::new_binary_operation): Handle polynomial
6881 element counts. Return false if we need to know the number
6882 of elements at compile time.
6883 * fold-const.c (fold_vec_perm): Punt if the number of elements
6884 isn't known at compile time.
6886 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6888 * vec-perm-indices.h (vec_perm_builder): Change element type
6889 from HOST_WIDE_INT to poly_int64.
6890 (vec_perm_indices::element_type): Update accordingly.
6891 (vec_perm_indices::clamp): Handle polynomial element_types.
6892 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6893 (vec_perm_indices::all_in_range_p): Likewise.
6894 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6896 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6897 polynomial vec_perm_indices element types.
6898 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6899 * fold-const.c (fold_vec_perm): Likewise.
6900 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6901 * tree-vect-generic.c (lower_vec_perm): Likewise.
6902 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6903 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6904 element type to HOST_WIDE_INT.
6906 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6907 Alan Hayward <alan.hayward@arm.com>
6908 David Sherwood <david.sherwood@arm.com>
6910 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6911 rather than an int. Use plus_constant.
6912 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6913 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6915 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6916 Alan Hayward <alan.hayward@arm.com>
6917 David Sherwood <david.sherwood@arm.com>
6919 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6920 a HOST_WIDE_INT to a poly_int64.
6922 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6923 Alan Hayward <alan.hayward@arm.com>
6924 David Sherwood <david.sherwood@arm.com>
6926 * calls.c (load_register_parameters): Cope with polynomial
6927 mode sizes. Require a constant size for BLKmode parameters
6928 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6929 forces a parameter to be padded at the lsb end in order to
6930 fill a complete number of words, require the parameter size
6931 to be ordered wrt UNITS_PER_WORD.
6933 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6934 Alan Hayward <alan.hayward@arm.com>
6935 David Sherwood <david.sherwood@arm.com>
6937 * reload1.c (spill_stack_slot_width): Change element type
6938 from unsigned int to poly_uint64_pod.
6939 (alter_reg): Treat mode sizes as polynomial.
6941 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6942 Alan Hayward <alan.hayward@arm.com>
6943 David Sherwood <david.sherwood@arm.com>
6945 * reload.c (complex_word_subreg_p): New function.
6946 (reload_inner_reg_of_subreg, push_reload): Use it.
6948 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6949 Alan Hayward <alan.hayward@arm.com>
6950 David Sherwood <david.sherwood@arm.com>
6952 * lra-constraints.c (process_alt_operands): Reject matched
6953 operands whose sizes aren't ordered.
6954 (match_reload): Refer to this check here.
6956 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6957 Alan Hayward <alan.hayward@arm.com>
6958 David Sherwood <david.sherwood@arm.com>
6960 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6961 that the mode size is in the set {1, 2, 4, 8, 16}.
6963 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6964 Alan Hayward <alan.hayward@arm.com>
6965 David Sherwood <david.sherwood@arm.com>
6967 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6968 Use plus_constant instead of gen_rtx_PLUS.
6970 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6971 Alan Hayward <alan.hayward@arm.com>
6972 David Sherwood <david.sherwood@arm.com>
6974 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6975 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6976 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6977 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6978 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6979 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6980 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6981 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6982 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6983 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6985 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6986 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6987 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6988 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6989 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6990 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6991 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6992 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6993 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6994 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6996 * expr.c (emit_move_resolve_push): Treat the input and result
6997 of PUSH_ROUNDING as a poly_int64.
6998 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6999 (emit_push_insn): Likewise.
7000 * lra-eliminations.c (mark_not_eliminable): Likewise.
7001 * recog.c (push_operand): Likewise.
7002 * reload1.c (elimination_effects): Likewise.
7003 * rtlanal.c (nonzero_bits1): Likewise.
7004 * calls.c (store_one_arg): Likewise. Require the padding to be
7005 known at compile time.
7007 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7008 Alan Hayward <alan.hayward@arm.com>
7009 David Sherwood <david.sherwood@arm.com>
7011 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
7012 Use plus_constant instead of gen_rtx_PLUS.
7014 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7015 Alan Hayward <alan.hayward@arm.com>
7016 David Sherwood <david.sherwood@arm.com>
7018 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
7021 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7022 Alan Hayward <alan.hayward@arm.com>
7023 David Sherwood <david.sherwood@arm.com>
7025 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
7026 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
7027 via stack temporaries. Treat the mode size as polynomial too.
7029 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7030 Alan Hayward <alan.hayward@arm.com>
7031 David Sherwood <david.sherwood@arm.com>
7033 * expr.c (expand_expr_real_2): When handling conversions involving
7034 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
7035 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
7036 as a poly_uint64 too.
7038 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7039 Alan Hayward <alan.hayward@arm.com>
7040 David Sherwood <david.sherwood@arm.com>
7042 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
7044 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7045 Alan Hayward <alan.hayward@arm.com>
7046 David Sherwood <david.sherwood@arm.com>
7048 * combine.c (can_change_dest_mode): Handle polynomial
7049 REGMODE_NATURAL_SIZE.
7050 * expmed.c (store_bit_field_1): Likewise.
7051 * expr.c (store_constructor): Likewise.
7052 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
7053 and polynomial REGMODE_NATURAL_SIZE.
7054 (gen_lowpart_common): Likewise.
7055 * reginfo.c (record_subregs_of_mode): Likewise.
7056 * rtlanal.c (read_modify_subreg_p): Likewise.
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 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
7063 numbers of elements.
7065 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7066 Alan Hayward <alan.hayward@arm.com>
7067 David Sherwood <david.sherwood@arm.com>
7069 * match.pd: Cope with polynomial numbers of vector elements.
7071 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7072 Alan Hayward <alan.hayward@arm.com>
7073 David Sherwood <david.sherwood@arm.com>
7075 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
7076 in a POINTER_PLUS_EXPR.
7078 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7079 Alan Hayward <alan.hayward@arm.com>
7080 David Sherwood <david.sherwood@arm.com>
7082 * omp-simd-clone.c (simd_clone_subparts): New function.
7083 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
7084 (ipa_simd_modify_function_body): Likewise.
7086 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7087 Alan Hayward <alan.hayward@arm.com>
7088 David Sherwood <david.sherwood@arm.com>
7090 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
7091 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
7092 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
7093 (expand_vector_condition, vector_element): Likewise.
7094 (subparts_gt): New function.
7095 (get_compute_type): Use subparts_gt.
7096 (count_type_subparts): Delete.
7097 (expand_vector_operations_1): Use subparts_gt instead of
7098 count_type_subparts.
7100 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7101 Alan Hayward <alan.hayward@arm.com>
7102 David Sherwood <david.sherwood@arm.com>
7104 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
7105 (vect_compile_time_alias): ...this new function. Do the calculation
7106 on poly_ints rather than trees.
7107 (vect_prune_runtime_alias_test_list): Update call accordingly.
7109 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7110 Alan Hayward <alan.hayward@arm.com>
7111 David Sherwood <david.sherwood@arm.com>
7113 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
7115 (vect_schedule_slp_instance): Likewise.
7117 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7118 Alan Hayward <alan.hayward@arm.com>
7119 David Sherwood <david.sherwood@arm.com>
7121 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
7122 constant and extern definitions for variable-length vectors.
7123 (vect_get_constant_vectors): Note that the number of units
7124 is known to be constant.
7126 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7127 Alan Hayward <alan.hayward@arm.com>
7128 David Sherwood <david.sherwood@arm.com>
7130 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
7131 of units as polynomial. Choose between WIDE and NARROW based
7134 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7135 Alan Hayward <alan.hayward@arm.com>
7136 David Sherwood <david.sherwood@arm.com>
7138 * tree-vect-stmts.c (simd_clone_subparts): New function.
7139 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
7141 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7142 Alan Hayward <alan.hayward@arm.com>
7143 David Sherwood <david.sherwood@arm.com>
7145 * tree-vect-stmts.c (vectorizable_call): Treat the number of
7146 vectors as polynomial. Use build_index_vector for
7149 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7150 Alan Hayward <alan.hayward@arm.com>
7151 David Sherwood <david.sherwood@arm.com>
7153 * tree-vect-stmts.c (get_load_store_type): Treat the number of
7154 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
7155 for variable-length vectors.
7156 (vectorizable_mask_load_store): Treat the number of units as
7157 polynomial, asserting that it is constant if the condition has
7158 already been enforced.
7159 (vectorizable_store, vectorizable_load): Likewise.
7161 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7162 Alan Hayward <alan.hayward@arm.com>
7163 David Sherwood <david.sherwood@arm.com>
7165 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
7166 of units as polynomial. Punt if we can't tell at compile time
7167 which vector contains the final result.
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 * tree-vect-loop.c (vectorizable_induction): Treat the number
7174 of units as polynomial. Punt on SLP inductions. Use an integer
7175 VEC_SERIES_EXPR for variable-length integer reductions. Use a
7176 cast of such a series for variable-length floating-point
7179 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7180 Alan Hayward <alan.hayward@arm.com>
7181 David Sherwood <david.sherwood@arm.com>
7183 * tree.h (build_index_vector): Declare.
7184 * tree.c (build_index_vector): New function.
7185 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
7186 of units as polynomial, forcibly converting it to a constant if
7187 vectorizable_reduction has already enforced the condition.
7188 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
7189 to create a {1,2,3,...} vector.
7190 (vectorizable_reduction): Treat the number of units as polynomial.
7191 Choose vectype_in based on the largest scalar element size rather
7192 than the smallest number of units. Enforce the restrictions
7195 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7196 Alan Hayward <alan.hayward@arm.com>
7197 David Sherwood <david.sherwood@arm.com>
7199 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
7200 number of units as polynomial.
7202 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7203 Alan Hayward <alan.hayward@arm.com>
7204 David Sherwood <david.sherwood@arm.com>
7206 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
7207 * target.def (autovectorize_vector_sizes): Return the vector sizes
7208 by pointer, using vector_sizes rather than a bitmask.
7209 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
7210 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
7211 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
7213 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
7214 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
7215 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
7216 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
7217 * omp-general.c (omp_max_vf): Likewise.
7218 * omp-low.c (omp_clause_aligned_alignment): Likewise.
7219 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
7220 * tree-vect-loop.c (vect_analyze_loop): Likewise.
7221 * tree-vect-slp.c (vect_slp_bb): Likewise.
7222 * doc/tm.texi: Regenerate.
7223 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
7225 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
7226 the vector size as a poly_uint64 rather than an unsigned int.
7227 (current_vector_size): Change from an unsigned int to a poly_uint64.
7228 (get_vectype_for_scalar_type): Update accordingly.
7229 * tree.h (build_truth_vector_type): Take the size and number of
7230 units as a poly_uint64 rather than an unsigned int.
7231 (build_vector_type): Add a temporary overload that takes
7232 the number of units as a poly_uint64 rather than an unsigned int.
7233 * tree.c (make_vector_type): Likewise.
7234 (build_truth_vector_type): Take the number of units as a poly_uint64
7235 rather than an unsigned int.
7237 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7238 Alan Hayward <alan.hayward@arm.com>
7239 David Sherwood <david.sherwood@arm.com>
7241 * target.def (get_mask_mode): Take the number of units and length
7242 as poly_uint64s rather than unsigned ints.
7243 * targhooks.h (default_get_mask_mode): Update accordingly.
7244 * targhooks.c (default_get_mask_mode): Likewise.
7245 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
7246 * doc/tm.texi: Regenerate.
7248 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7249 Alan Hayward <alan.hayward@arm.com>
7250 David Sherwood <david.sherwood@arm.com>
7252 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
7253 * omp-general.c (omp_max_vf): Likewise.
7254 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
7255 (expand_omp_simd): Handle polynomial safelen.
7256 * omp-low.c (omplow_simd_context): Add a default constructor.
7257 (omplow_simd_context::max_vf): Change from int to poly_uint64.
7258 (lower_rec_simd_input_clauses): Update accordingly.
7259 (lower_rec_input_clauses): Likewise.
7261 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7262 Alan Hayward <alan.hayward@arm.com>
7263 David Sherwood <david.sherwood@arm.com>
7265 * tree-vectorizer.h (vect_nunits_for_cost): New function.
7266 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
7267 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
7268 (vect_analyze_slp_cost): Likewise.
7269 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
7270 (vect_model_load_cost): Likewise.
7272 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7273 Alan Hayward <alan.hayward@arm.com>
7274 David Sherwood <david.sherwood@arm.com>
7276 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
7277 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
7278 from an unsigned int * to a poly_uint64_pod *.
7279 (calculate_unrolling_factor): New function.
7280 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
7282 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7283 Alan Hayward <alan.hayward@arm.com>
7284 David Sherwood <david.sherwood@arm.com>
7286 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
7287 from an unsigned int to a poly_uint64.
7288 (_loop_vec_info::slp_unrolling_factor): Likewise.
7289 (_loop_vec_info::vectorization_factor): Change from an int
7291 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
7292 (vect_get_num_vectors): New function.
7293 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
7294 (vect_get_num_copies): Use vect_get_num_vectors.
7295 (vect_analyze_data_ref_dependences): Change max_vf from an int *
7296 to an unsigned int *.
7297 (vect_analyze_data_refs): Change min_vf from an int * to a
7299 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7300 than an unsigned HOST_WIDE_INT.
7301 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
7302 (vect_analyze_data_ref_dependence): Change max_vf from an int *
7303 to an unsigned int *.
7304 (vect_analyze_data_ref_dependences): Likewise.
7305 (vect_compute_data_ref_alignment): Handle polynomial vf.
7306 (vect_enhance_data_refs_alignment): Likewise.
7307 (vect_prune_runtime_alias_test_list): Likewise.
7308 (vect_shift_permute_load_chain): Likewise.
7309 (vect_supportable_dr_alignment): Likewise.
7310 (dependence_distance_ge_vf): Take the vectorization factor as a
7311 poly_uint64 rather than an unsigned HOST_WIDE_INT.
7312 (vect_analyze_data_refs): Change min_vf from an int * to a
7314 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
7315 vfm1 as a poly_uint64 rather than an int. Make the same change
7316 for the returned bound_scalar.
7317 (vect_gen_vector_loop_niters): Handle polynomial vf.
7318 (vect_do_peeling): Likewise. Update call to
7319 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
7320 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
7322 * tree-vect-loop.c (vect_determine_vectorization_factor)
7323 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
7324 (vect_get_known_peeling_cost): Likewise.
7325 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
7326 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
7327 (vect_transform_loop): Likewise. Use the lowest possible VF when
7328 updating the upper bounds of the loop.
7329 (vect_min_worthwhile_factor): Make static. Return an unsigned int
7331 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
7332 polynomial unroll factors.
7333 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
7334 (vect_make_slp_decision): Likewise.
7335 (vect_supported_load_permutation_p): Likewise, and polynomial
7337 (vect_analyze_slp_cost): Handle polynomial vf.
7338 (vect_slp_analyze_node_operations): Likewise.
7339 (vect_slp_analyze_bb_1): Likewise.
7340 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7341 than an unsigned HOST_WIDE_INT.
7342 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
7343 (vectorizable_load): Handle polynomial vf.
7344 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
7346 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
7348 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7349 Alan Hayward <alan.hayward@arm.com>
7350 David Sherwood <david.sherwood@arm.com>
7352 * match.pd: Handle bit operations involving three constants
7353 and try to fold one pair.
7355 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7357 * tree-vect-loop-manip.c: Include gimple-fold.h.
7358 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
7359 niters_maybe_zero parameters. Handle other cases besides a step of 1.
7360 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
7361 Add a path that uses a step of VF instead of 1, but disable it
7363 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
7364 and niters_no_overflow parameters. Update calls to
7365 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
7366 Create a new SSA name if the latter choses to use a ste other
7367 than zero, and return it via niters_vector_mult_vf_var.
7368 * tree-vect-loop.c (vect_transform_loop): Update calls to
7369 vect_do_peeling, vect_gen_vector_loop_niters and
7370 slpeel_make_loop_iterate_ntimes.
7371 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
7372 (vect_gen_vector_loop_niters): Update declarations after above changes.
7374 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
7376 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
7377 128-bit round to integer instructions.
7378 (ceil<mode>2): Likewise.
7379 (btrunc<mode>2): Likewise.
7380 (round<mode>2): Likewise.
7382 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7384 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
7385 unaligned VSX load/store on P8/P9.
7386 (expand_block_clear): Allow the use of unaligned VSX
7387 load/store on P8/P9.
7389 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
7391 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
7393 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
7394 swap associated with both a load and a store.
7396 2018-01-02 Andrew Waterman <andrew@sifive.com>
7398 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
7399 * config/riscv/riscv.md (clear_cache): Use it.
7401 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
7403 * web.c: Remove out-of-date comment.
7405 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7407 * expr.c (fixup_args_size_notes): Check that any existing
7408 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
7409 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
7410 (emit_single_push_insn): ...here.
7412 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7414 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
7415 (const_vector_encoded_nelts): New function.
7416 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
7417 (const_vector_int_elt, const_vector_elt): Declare.
7418 * emit-rtl.c (const_vector_int_elt_1): New function.
7419 (const_vector_elt): Likewise.
7420 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
7421 of CONST_VECTOR_ELT.
7423 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7425 * expr.c: Include rtx-vector-builder.h.
7426 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
7427 directly on the tree encoding.
7428 (const_vector_from_tree): Likewise.
7429 * optabs.c: Include rtx-vector-builder.h.
7430 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
7431 sequence of "u" values.
7432 * vec-perm-indices.c: Include rtx-vector-builder.h.
7433 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
7434 directly on the vec_perm_indices encoding.
7436 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7438 * doc/rtl.texi (const_vector): Describe new encoding scheme.
7439 * Makefile.in (OBJS): Add rtx-vector-builder.o.
7440 * rtx-vector-builder.h: New file.
7441 * rtx-vector-builder.c: Likewise.
7442 * rtl.h (rtx_def::u2): Add a const_vector field.
7443 (CONST_VECTOR_NPATTERNS): New macro.
7444 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
7445 (CONST_VECTOR_DUPLICATE_P): Likewise.
7446 (CONST_VECTOR_STEPPED_P): Likewise.
7447 (CONST_VECTOR_ENCODED_ELT): Likewise.
7448 (const_vec_duplicate_p): Check for a duplicated vector encoding.
7449 (unwrap_const_vec_duplicate): Likewise.
7450 (const_vec_series_p): Check for a non-duplicated vector encoding.
7451 Say that the function only returns true for integer vectors.
7452 * emit-rtl.c: Include rtx-vector-builder.h.
7453 (gen_const_vec_duplicate_1): Delete.
7454 (gen_const_vector): Call gen_const_vec_duplicate instead of
7455 gen_const_vec_duplicate_1.
7456 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
7457 (gen_const_vec_duplicate): Use rtx_vector_builder.
7458 (gen_const_vec_series): Likewise.
7459 (gen_rtx_CONST_VECTOR): Likewise.
7460 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
7461 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7462 Build a new vector rather than modifying a CONST_VECTOR in-place.
7463 (handle_special_swappables): Update call accordingly.
7464 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
7465 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7466 Build a new vector rather than modifying a CONST_VECTOR in-place.
7467 (handle_special_swappables): Update call accordingly.
7469 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7471 * simplify-rtx.c (simplify_const_binary_operation): Use
7472 CONST_VECTOR_ELT instead of XVECEXP.
7474 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7476 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
7477 the selector elements to be different from the data elements
7478 if the selector is a VECTOR_CST.
7479 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
7480 ssizetype for the selector.
7482 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7484 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
7485 before testing each element individually.
7486 * tree-vect-generic.c (lower_vec_perm): Likewise.
7488 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7490 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
7491 * selftest-run-tests.c (selftest::run_tests): Call it.
7492 * vector-builder.h (vector_builder::operator ==): New function.
7493 (vector_builder::operator !=): Likewise.
7494 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
7495 (vec_perm_indices::all_from_input_p): New function.
7496 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7497 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
7498 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
7499 instead of reading the VECTOR_CST directly. Detect whether both
7500 vector inputs are the same before constructing the vec_perm_indices,
7501 and update the number of inputs argument accordingly. Use the
7502 utility functions added above. Only construct sel2 if we need to.
7504 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7506 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
7507 the broadcast of the low byte.
7508 (expand_mult_highpart): Use an explicit encoding for the permutes.
7509 * optabs-query.c (can_mult_highpart_p): Likewise.
7510 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
7511 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7512 (vectorizable_bswap): Likewise.
7513 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
7514 explicit encoding for the power-of-2 permutes.
7515 (vect_permute_store_chain): Likewise.
7516 (vect_grouped_load_supported): Likewise.
7517 (vect_permute_load_chain): Likewise.
7519 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7521 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
7522 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
7523 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
7524 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7525 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
7526 (vect_gen_perm_mask_any): Likewise.
7528 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7530 * int-vector-builder.h: New file.
7531 * vec-perm-indices.h: Include int-vector-builder.h.
7532 (vec_perm_indices): Redefine as an int_vector_builder.
7533 (auto_vec_perm_indices): Delete.
7534 (vec_perm_builder): Redefine as a stand-alone class.
7535 (vec_perm_indices::vec_perm_indices): New function.
7536 (vec_perm_indices::clamp): Likewise.
7537 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
7538 (vec_perm_indices::new_vector): New function.
7539 (vec_perm_indices::new_expanded_vector): Update for new
7540 vec_perm_indices class.
7541 (vec_perm_indices::rotate_inputs): New function.
7542 (vec_perm_indices::all_in_range_p): Operate directly on the
7543 encoded form, without computing elided elements.
7544 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
7545 encoding. Update for new vec_perm_indices class.
7546 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
7547 the given vec_perm_builder.
7548 (expand_vec_perm_var): Update vec_perm_builder constructor.
7549 (expand_mult_highpart): Use vec_perm_builder instead of
7550 auto_vec_perm_indices.
7551 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
7552 vec_perm_indices instead of auto_vec_perm_indices. Use a single
7553 or double series encoding as appropriate.
7554 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
7555 vec_perm_indices instead of auto_vec_perm_indices.
7556 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7557 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7558 (vect_permute_store_chain): Likewise.
7559 (vect_grouped_load_supported): Likewise.
7560 (vect_permute_load_chain): Likewise.
7561 (vect_shift_permute_load_chain): Likewise.
7562 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7563 (vect_transform_slp_perm_load): Likewise.
7564 (vect_schedule_slp_instance): Likewise.
7565 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7566 (vectorizable_mask_load_store): Likewise.
7567 (vectorizable_bswap): Likewise.
7568 (vectorizable_store): Likewise.
7569 (vectorizable_load): Likewise.
7570 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
7571 vec_perm_indices instead of auto_vec_perm_indices. Use
7572 tree_to_vec_perm_builder to read the vector from a tree.
7573 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
7574 vec_perm_builder instead of a vec_perm_indices.
7575 (have_whole_vector_shift): Use vec_perm_builder and
7576 vec_perm_indices instead of auto_vec_perm_indices. Leave the
7577 truncation to calc_vec_perm_mask_for_shift.
7578 (vect_create_epilog_for_reduction): Likewise.
7579 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
7580 from auto_vec_perm_indices to vec_perm_indices.
7581 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7582 instead of changing individual elements.
7583 (aarch64_vectorize_vec_perm_const): Use new_vector to install
7584 the vector in d.perm.
7585 * config/arm/arm.c (expand_vec_perm_d::perm): Change
7586 from auto_vec_perm_indices to vec_perm_indices.
7587 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7588 instead of changing individual elements.
7589 (arm_vectorize_vec_perm_const): Use new_vector to install
7590 the vector in d.perm.
7591 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
7592 Update vec_perm_builder constructor.
7593 (rs6000_expand_interleave): Likewise.
7594 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
7595 (rs6000_expand_interleave): Likewise.
7597 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7599 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
7600 to qimode could truncate the indices.
7601 * optabs.c (expand_vec_perm_var): Likewise.
7603 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7605 * Makefile.in (OBJS): Add vec-perm-indices.o.
7606 * vec-perm-indices.h: New file.
7607 * vec-perm-indices.c: Likewise.
7608 * target.h (vec_perm_indices): Replace with a forward class
7610 (auto_vec_perm_indices): Move to vec-perm-indices.h.
7611 * optabs.h: Include vec-perm-indices.h.
7612 (expand_vec_perm): Delete.
7613 (selector_fits_mode_p, expand_vec_perm_var): Declare.
7614 (expand_vec_perm_const): Declare.
7615 * target.def (vec_perm_const_ok): Replace with...
7616 (vec_perm_const): ...this new hook.
7617 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
7618 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7619 * doc/tm.texi: Regenerate.
7620 * optabs.def (vec_perm_const): Delete.
7621 * doc/md.texi (vec_perm_const): Likewise.
7622 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7623 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7624 expand_vec_perm for constant permutation vectors. Assert that
7625 the mode of variable permutation vectors is the integer equivalent
7626 of the mode that is being permuted.
7627 * optabs-query.h (selector_fits_mode_p): Declare.
7628 * optabs-query.c: Include vec-perm-indices.h.
7629 (selector_fits_mode_p): New function.
7630 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7631 is defined, instead of checking whether the vec_perm_const_optab
7632 exists. Use targetm.vectorize.vec_perm_const instead of
7633 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7634 fit in the vector mode before using a variable permute.
7635 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7636 vec_perm_indices instead of an rtx.
7637 (expand_vec_perm): Replace with...
7638 (expand_vec_perm_const): ...this new function. Take the selector
7639 as a vec_perm_indices rather than an rtx. Also take the mode of
7640 the selector. Update call to shift_amt_for_vec_perm_mask.
7641 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7642 Use vec_perm_indices::new_expanded_vector to expand the original
7643 selector into bytes. Check whether the indices fit in the vector
7644 mode before using a variable permute.
7645 (expand_vec_perm_var): Make global.
7646 (expand_mult_highpart): Use expand_vec_perm_const.
7647 * fold-const.c: Includes vec-perm-indices.h.
7648 * tree-ssa-forwprop.c: Likewise.
7649 * tree-vect-data-refs.c: Likewise.
7650 * tree-vect-generic.c: Likewise.
7651 * tree-vect-loop.c: Likewise.
7652 * tree-vect-slp.c: Likewise.
7653 * tree-vect-stmts.c: Likewise.
7654 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7656 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7657 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7658 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7659 (aarch64_vectorize_vec_perm_const): ...this new function.
7660 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7661 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7662 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7663 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7664 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7665 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7666 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7668 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7669 check for NEON modes.
7670 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7671 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7672 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7673 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7675 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7676 the old VEC_PERM_CONST conditions.
7677 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7678 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7679 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7680 (ia64_vectorize_vec_perm_const_ok): Merge into...
7681 (ia64_vectorize_vec_perm_const): ...this new function.
7682 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7683 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7684 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7685 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7686 * config/mips/mips.c (mips_expand_vec_perm_const)
7687 (mips_vectorize_vec_perm_const_ok): Merge into...
7688 (mips_vectorize_vec_perm_const): ...this new function.
7689 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7690 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7691 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7692 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7693 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7694 (rs6000_expand_vec_perm_const): Delete.
7695 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7697 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7698 (altivec_expand_vec_perm_const_le): Take each operand individually.
7699 Operate on constant selectors rather than rtxes.
7700 (altivec_expand_vec_perm_const): Likewise. Update call to
7701 altivec_expand_vec_perm_const_le.
7702 (rs6000_expand_vec_perm_const): Delete.
7703 (rs6000_vectorize_vec_perm_const_ok): Delete.
7704 (rs6000_vectorize_vec_perm_const): New function.
7705 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7706 an element count and rtx array.
7707 (rs6000_expand_extract_even): Update call accordingly.
7708 (rs6000_expand_interleave): Likewise.
7709 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7710 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7711 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7712 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7713 (rs6000_expand_vec_perm_const): Delete.
7714 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7715 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7716 (altivec_expand_vec_perm_const_le): Take each operand individually.
7717 Operate on constant selectors rather than rtxes.
7718 (altivec_expand_vec_perm_const): Likewise. Update call to
7719 altivec_expand_vec_perm_const_le.
7720 (rs6000_expand_vec_perm_const): Delete.
7721 (rs6000_vectorize_vec_perm_const_ok): Delete.
7722 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7723 reference to the SPE evmerge intructions.
7724 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7725 an element count and rtx array.
7726 (rs6000_expand_extract_even): Update call accordingly.
7727 (rs6000_expand_interleave): Likewise.
7728 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7729 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7731 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7733 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7735 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7736 vector mode and that that mode matches the mode of the data
7738 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7739 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7740 directly using expand_vec_perm_1 when forcing selectors into
7742 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7744 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7746 * optabs-query.h (can_vec_perm_p): Delete.
7747 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7748 * optabs-query.c (can_vec_perm_p): Split into...
7749 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7750 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7751 particular selector is valid.
7752 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7753 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7754 (vect_grouped_load_supported): Likewise.
7755 (vect_shift_permute_load_chain): Likewise.
7756 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7757 (vect_transform_slp_perm_load): Likewise.
7758 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7759 (vectorizable_bswap): Likewise.
7760 (vect_gen_perm_mask_checked): Likewise.
7761 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7762 implementations of variable permutation vectors into account
7763 when deciding which selector to use.
7764 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7765 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7766 with a false third argument.
7767 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7768 to test whether the constant selector is valid and can_vec_perm_var_p
7769 to test whether a variable selector is valid.
7771 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7773 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7774 * optabs-query.c (can_vec_perm_p): Likewise.
7775 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7776 instead of vec_perm_indices.
7777 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7778 (vect_gen_perm_mask_checked): Likewise,
7779 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7780 (vect_gen_perm_mask_checked): Likewise,
7782 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7784 * optabs-query.h (qimode_for_vec_perm): Declare.
7785 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7786 (qimode_for_vec_perm): ...this new function.
7787 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7789 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7791 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7792 does not have a conditional at the top.
7794 2018-01-02 Richard Biener <rguenther@suse.de>
7796 * ipa-inline.c (big_speedup_p): Fix expression.
7798 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7801 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7804 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7808 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7809 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7810 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7811 cond_taken_branch_cost 3->4.
7813 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7815 PR tree-optimization/83581
7816 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7817 TODO_cleanup_cfg if any changes have been made.
7820 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7821 convert_modes if target mode has the right side, but different mode
7825 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7826 last argument when extracting from CONCAT. If either from_real or
7827 from_imag is NULL, use expansion through memory. If result is not
7828 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7829 the parts directly to inner mode, if even that fails, use expansion
7833 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7834 check for bswap in mode rather than HImode and use that in expand_unop
7837 Copyright (C) 2018 Free Software Foundation, Inc.
7839 Copying and distribution of this file, with or without modification,
7840 are permitted in any medium without royalty provided the copyright
7841 notice and this notice are preserved.