1 2018-03-02 Jakub Jelinek <jakub@redhat.com>
3 * substring-locations.h (format_warning_va): Formatting fix for
5 (format_warning_at_substring): Fix up ATTRIBUTE_GCC_DIAG second
7 (format_warning_n_va, format_warning_at_substring_n): New prototypes.
8 * substring-locations.c: Include intl.h.
9 (format_warning_va): Turned into small wrapper around
10 format_warning_n_va, renamed to ...
11 (format_warning_n_va): ... this, add N and PLURAL_GMSGID arguments,
12 rename GMSGID to SINGULAR_GMSGID, if SINGULAR_GMSGID != PLURAL_GMSGID,
14 (format_warning_at_substring_n): New function.
15 * gimple-ssa-sprintf.c: Remove GCC diagnostic ignored pragma.
16 (fmtwarn): Add ATTRIBUTE_GCC_DIAG. Turn into a copy of
17 format_warning_at_substring with just a shorter name instead of
18 const function pointer.
19 (fmtwarn_n): New function.
20 (maybe_warn, format_directive, parse_directive): Use fmtwarn_n where
21 appropriate, get rid of all the fmtstr temporaries, move conditionals
22 with G_() wrapped string literals directly into fmtwarn arguments,
23 cast dir.len to (int), formatting fixes.
25 2018-03-02 Thomas Schwinge <thomas@codesourcery.com>
27 * doc/invoke.texi: Remove "Cilk Plus" references.
29 2018-03-02 Jakub Jelinek <jakub@redhat.com>
30 Richard Biener <rguenther@suse.de>
33 * expr.c (expand_expr_real_1) <case CALL_EXPR>: Don't emit diagnostics
34 for error or warning attributes if CALL_FROM_THUNK_P is set.
37 2018-03-02 Jakub Jelinek <jakub@redhat.com>
40 * config/pa/pa.h (TARGET_CPU_CPP_BUILTINS): Predefine
41 __SIZEOF_128__ macro if HPUX_LONG_DOUBLE_LIBRARY.
44 * config/ia64/ia64.h (TARGET_CPU_CPP_BUILTINS): Predefine
45 __SIZEOF_{FPREG,FLOAT{80,128}}__ macros.
47 * predict.c (test_prediction_value_range): Use PROB_UNINITIALIZED
48 instead of -1U in last predictors element's probability member.
50 2018-03-02 Eric Botcazou <ebotcazou@adacore.com>
53 * ipa-devirt.c (odr_subtypes_equivalent_p): Get the ODR type of both
54 arguments if they are comparable.
56 2018-03-02 Richard Sandiford <richard.sandiford@linaro.org>
58 PR tree-optimization/84634
59 * tree-vect-stmts.c (vectorizable_store, vectorizable_load): Replace
60 masks and masked_loop_p with a single loop_masks, making sure it's
61 null for bb vectorization.
63 2018-03-02 Richard Sandiford <richard.sandiford@linaro.org>
65 * tree-vect-data-refs.c (vect_analyze_data_ref_dependence)
66 (vect_analyze_data_ref_access): Use loop->safe_len rather than
67 loop->force_vectorize to check whether there is no alias.
69 2018-03-02 Jakub Jelinek <jakub@redhat.com>
72 * rtl.h (prev_real_nondebug_insn, next_real_nondebug_insn): New
74 * emit-rtl.c (next_real_insn, prev_real_insn): Fix up function
76 (next_real_nondebug_insn, prev_real_nondebug_insn): New functions.
77 * cfgcleanup.c (try_head_merge_bb): Use prev_real_nondebug_insn
78 instead of a loop around prev_real_insn.
79 * combine.c (move_deaths): Use prev_real_nondebug_insn instead of
83 * config/i386/i386.c (ix86_print_operand): Use conditional
84 output_operand_lossage instead of gcc_assert if CONST_VECTOR is not
87 2018-03-02 Richard Biener <rguenther@suse.de>
89 PR tree-optimization/84427
90 * tree-ssa-pre.c (bitmap_remove_expr_from_set): Remove.
91 (bitmap_set_subtract_values): Rewrite to handle multiple
94 (prune_clobbered_mems): Likewise.
95 (phi_translate): Take edge instead of pred/phiblock.
96 (phi_translate_1): Likewise.
97 (phi_translate_set): Likewise. Insert all translated
98 exprs for a value into the set, keeping possibly multiple
99 expressions per value.
100 (compute_antic_aux): Adjust for phi_translate changes.
101 When intersecting union the expressions and prune those
102 not in the final value set, keeping possibly multiple
103 expressions per value. Do not use value-insertion
104 for unioning ANTIC_OUT U EXP_GEN - TMP_GEN but merge
105 all expressions. Add verification that the value-sets
106 only shrink during iteration.
107 (compute_partial_antic_aux): Adjust for the phi_translate changes.
108 (do_pre_regular_insertion): Likewise.
109 (do_pre_partial_partial_insertion): Likewise.
111 2018-03-02 Richard Biener <rguenther@suse.de>
114 * config/darwin.c (saved_debug_info_level): New static global.
115 (darwin_asm_lto_start): Disable debug info generation for LTO out.
116 (darwin_asm_lto_end): Restore debug info generation settings.
118 2018-03-01 Martin Liska <mliska@suse.cz>
121 * sanopt.c (sanitize_rewrite_addressable_params): Do not handle
124 2018-03-01 Richard Biener <rguenther@suse.de>
127 * dwarf2out.c (gen_variable_die): Properly handle late VLA
128 type annotation with LTO when debug was disabled at compile-time.
130 2018-03-01 Matthew Fortune <mfortune@gmail.com>
132 * config/mips/mips.c (mips_final_prescan_insn): Fix incorrect
134 (mips_final_postscan_insn): Likewise.
136 2018-03-01 Richard Sandiford <richard.sandiford@linaro.org>
138 PR rtl-optimization/84528
139 * alias.c (init_alias_target): Add commentary.
140 (init_alias_analysis): Only give HARD_FRAME_POINTER_REGNUM
141 a unique base value if the frame pointer is not eliminated
142 to the stack pointer.
144 2018-03-01 Tom de Vries <tom@codesourcery.com>
146 PR rtl-optimization/83327
147 * lra-int.h (hard_regs_spilled_into): Declare.
148 * lra.c (hard_regs_spilled_into): Define.
149 (init_reg_info): Init hard_regs_spilled_into.
150 * lra-spills.c (assign_spill_hard_regs): Update hard_regs_spilled_into.
151 * lra-lives.c (make_hard_regno_born, make_hard_regno_dead)
152 (process_bb_lives): Handle hard_regs_spilled_into.
153 (lra_create_live_ranges_1): Before doing liveness propagation, clear
154 regs in all_hard_regs_bitmap if set in hard_regs_spilled_into.
156 2018-02-28 David Edelsohn <dje.gcc@gmail.com>
158 * config.gcc (powerpc-ibm-aix7.1.*): New stanza.
159 (powerpc-ibm-aix[789]*): Default to AIX 7.2.
160 * config/rs6000/aix71.h (TARGET_DEFAULT): Revert to Power4 ISA.
161 * config/rs6000/aix72.h: New file.
163 2018-02-28 Jakub Jelinek <jakub@redhat.com>
165 * gimple-ssa-warn-restrict.c (maybe_diag_overlap): Use warning_n
166 instead of warning_at with conditional singular and plural messages
170 * stor-layout.c (update_alignment_for_field): For
171 targetm.ms_bitfield_layout_p (rli->t), if !is_bitfield
172 && !DECL_PACKED (field), do the alignment update, just use
173 only desired_align instead of MAX (type_align, desired_align)
175 (place_field): Don't do known_align < desired_align handling
176 early if targetm.ms_bitfield_layout_p (rli->t) and rli->prev_field
177 is non-NULL, instead do it after rli->prev_field handling and
178 only if not within a bitfield word. For DECL_PACKED (field)
179 use type_align of BITS_PER_UNIT.
181 2018-02-28 Eric Botcazou <ebotcazou@adacore.com>
183 * config/aarch64/aarch64.c (aarch64_emit_probe_stack_range): Remove
184 superfluous parentheses and trailing spaces.
186 2018-02-28 Richard Biener <rguenther@suse.de>
188 PR tree-optimization/84584
189 * graphite-scop-detection.c (scop_detection::add_scop): Discard
190 SCoPs with fake exit edge.
192 2018-02-28 Martin Liska <mliska@suse.cz>
195 * timevar.c (timer::print): Fix format to properly print 100%
198 2018-02-28 Richard Biener <rguenther@suse.de>
201 * genmatch.c (capture_info::walk_match): Do not mark
202 captured expressions without operands as expr_p given
203 they act more like predicates and should be subject to
204 "lost tail" side-effect preserving.
206 2018-02-28 Alexandre Oliva <aoliva@redhat.com>
208 PR rtl-optimization/81611
209 * auto-inc-dec.c (attempt_change): Move dead note from
210 mem_insn if it's the next use of regno
211 (find_address): Take address use of reg holding
212 non-incremented value. Add parm to limit search to the named
214 (merge_in_block): Attempt to use a mem insn that is the next
215 use of the original regno.
217 2018-02-27 Martin Sebor <msebor@redhat.com>
220 * gcc/doc/invoke.texi (-Wmissing-attributes): New option.
221 * gcc/print-tree.c (print_node): Handle DECL_UNINLINABLE.
223 2018-02-27 Martin Sebor <msebor@redhat.com>
226 * diagnostic-core.h (warning_n, error_n, inform_n): Change
227 n argument to unsigned HOST_WIDE_INT.
228 * diagnostic.c (warning_n, error_n, inform_n): Ditto.
229 (diagnostic_n_impl): Ditto. Handle arguments in excess of LONG_MAX.
230 * gimple-ssa-sprintf.c (format_directive): Simplify inform_n call.
231 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Use warning_n.
233 2018-02-27 Richard Biener <rguenther@suse.de>
235 PR tree-optimization/84512
236 * tree-vect-loop.c (vect_compute_single_scalar_iteration_cost):
237 Do not use the estimate returned from record_stmt_cost for
238 the scalar iteration cost but sum properly using add_stmt_cost.
240 2018-02-27 Richard Biener <rguenther@suse.de>
242 PR tree-optimization/84466
243 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
244 Adjust last change to less strictly validate use operands.
246 2018-02-27 Martin Liska <mliska@suse.cz>
248 PR gcov-profile/84548
249 * gcov.c (process_file): Allow partial overlap and consider it
250 also as group functions.
251 (output_lines): Properly calculate range of lines for a group.
253 2018-02-27 Martin Liska <mliska@suse.cz>
255 * timevar.c (timer::print_row): Remove 'usr', 'sys', 'wall' and
256 'ggc' suffixes. Change first column width.
257 (timer::print): Fix formatting of the column.
259 2018-02-27 Alexandre Oliva <aoliva@redhat.com>
261 * tree-ssa-live.c (remove_unused_scope_block_p): Do not
262 preserve inline entry blocks for the sake of debug inline
263 entry point markers alone.
264 (remove_unused_locals): Suggest in comments a better place to
265 force the preservation of inline entry blocks that are
266 otherwise unused, but do not preserve them.
268 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
270 * config/i386/i386.c (ix86_output_indirect_jmp): Update comments.
272 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
275 * config/i386/constraints.md (Bs): Replace
276 ix86_indirect_branch_register with
277 TARGET_INDIRECT_BRANCH_REGISTER.
279 * config/i386/i386.md (indirect_jump): Likewise.
280 (tablejump): Likewise.
281 (*sibcall_memory): Likewise.
282 (*sibcall_value_memory): Likewise.
283 Peepholes of indirect call and jump via memory: Likewise.
284 (*sibcall_GOT_32): Disallowed for TARGET_INDIRECT_BRANCH_REGISTER.
285 (*sibcall_value_GOT_32): Likewise.
286 * config/i386/predicates.md (indirect_branch_operand): Likewise.
287 (GOT_memory_operand): Likewise.
288 (call_insn_operand): Likewise.
289 (sibcall_insn_operand): Likewise.
290 (GOT32_symbol_operand): Likewise.
291 * config/i386/i386.h (TARGET_INDIRECT_BRANCH_REGISTER): New.
293 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
295 PR rtl-optimization/83496
296 * reorg.c (steal_delay_list_from_target): Change REDUNDANT array from
297 booleans to RTXes. Call fix_reg_dead_note on every non-null element.
298 (steal_delay_list_from_fallthrough): Call fix_reg_dead_note on a
299 redundant insn, if any.
300 (relax_delay_slots): Likewise.
301 (update_reg_unused_notes): Rename REDUNDANT_INSN to OTHER_INSN.
303 2018-02-26 Richard Sandiford <richard.sandiford@linaro.org>
305 PR tree-optimization/83965
306 * tree-vect-patterns.c (vect_reassociating_reduction_p): Assume
307 that grouped statements are part of a reduction chain. Return
308 true if the statement is not marked as a reduction itself but
310 (vect_recog_dot_prod_pattern): Don't check whether the statement
311 is part of a group here.
312 (vect_recog_sad_pattern): Likewise.
313 (vect_recog_widen_sum_pattern): Likewise.
315 2018-02-26 Eric Botcazou <ebotcazou@adacore.com>
318 * final.c (rest_of_clean_state): Also look for calls inside sequences.
320 2018-02-26 H.J. Lu <hongjiu.lu@intel.com>
323 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Remove
325 (ix86_output_indirect_function_return): New prototype.
326 (ix86_split_simple_return_pop_internal): Likewise.
327 * config/i386/i386.c (indirect_return_via_cx): New.
328 (indirect_return_via_cx_bnd): Likewise.
329 (indirect_thunk_name): Handle return va CX_REG.
330 (output_indirect_thunk_function): Create alias for
331 __x86_return_thunk_[re]cx and __x86_return_thunk_[re]cx_bnd.
332 (ix86_output_indirect_jmp): Remove the bool argument.
333 (ix86_output_indirect_function_return): New function.
334 (ix86_split_simple_return_pop_internal): Likewise.
335 * config/i386/i386.md (*indirect_jump): Don't pass false
336 to ix86_output_indirect_jmp.
337 (*tablejump_1): Likewise.
338 (simple_return_pop_internal): Change it to define_insn_and_split.
339 Call ix86_split_simple_return_pop_internal to split it for
341 (simple_return_indirect_internal): Call
342 ix86_output_indirect_function_return instead of
343 ix86_output_indirect_jmp.
345 2018-02-26 Jakub Jelinek <jakub@redhat.com>
348 * vec.h (vec_default_construct): For BROKEN_VALUE_INITIALIZATION use
349 memset and value initialization afterwards.
351 2018-02-26 Christophe Lyon <christophe.lyon@linaro.org>
353 * Makefile.in (lto-wrapper): Use ALL_LINKERFLAGS.
355 2018-02-26 Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
358 * common/config/aarch64/aarch64-common.c
359 (aarch_option_optimization_table[]): Switch
360 off fomit-frame-pointer
362 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
363 Chung-Ju Wu <jasonwucj@gmail.com>
365 * config/nds32/nds32-multiple.md (load_multiple): Disallow
367 (store_multiple): Ditto.
369 2018-02-26 Kito Cheng <kito.cheng@gmail.com>
371 * config.gcc: Add --with-cpu support for nds32 target.
372 * config/nds32/nds32-opts.h (nds32_cpu_type): New.
373 * config/nds32/nds32.opt: Add -mcpu= option.
375 2018-02-25 Segher Boessenkool <segher@kernel.crashing.org>
377 * config/rs6000/rs6000.opt (mvrsave=no, mvrsave=yes, isel=no,
378 isel=yes): Warn for these deprecated options.
380 2018-02-23 David Edelsohn <dje.gcc@gmail.com>
382 * config/rs6000/aix71.h (TARGET_DEFAULT): Change to
383 ISA_2_5_MASKS_EMBEDDED.
385 2018-02-23 Jakub Jelinek <jakub@redhat.com>
387 * ipa-prop.c (ipa_vr_ggc_hash_traits::hash): Hash p->min and
388 p->max as pointers rather than using iterative_hash_expr.
390 2018-02-23 Carl Love <cel@us.ibm.com>
392 * config/rs6000/rs6000-builtin.def: Change VSIGNED2 and VUNSIGNED2
393 macro expansions from BU_VSX_2 to BU_P8V_VSX_2 and BU_VSX_OVERLOAD_2 to
395 * config/rs6000/rs6000-c.c: Change VSX_BUILTIN_VEC_VSIGNED2 to
396 P8V_BUILTIN_VEC_VSIGNED2. Change VSX_BUILTIN_VEC_VUNSIGNED2 to
397 P8V_BUILTIN_VEC_VUNSIGNED2.
399 2018-02-22 Vladimir Makarov <vmakarov@redhat.com>
402 * lra-int.h (LRA_UNKNOWN_ALT, LRA_NON_CLOBBERED_ALT): New macros.
403 * lra.c (lra_set_insn_recog_data, lra_update_insn_recog_data): Use
405 * lra-constraints.c (curr_insn_transform): Set up
406 LRA_NON_CLOBBERED_ALT for moves processed on the fast path. Use
408 (remove_inheritance_pseudos): Use LRA_UNKNOWN_ALT.
409 * lra-eliminations.c (spill_pseudos): Ditto.
410 (process_insn_for_elimination): Ditto.
411 * lra-lives.c (reg_early_clobber_p): Use the new macros.
412 * lra-spills.c (spill_pseudos): Use LRA_UNKNOWN_ALT and
413 LRA_NON_CLOBBERED_ALT.
415 2018-02-22 Martin Sebor <msebor@redhat.com>
417 PR tree-optimization/84480
418 * gimple-fold.c (gimple_fold_builtin_strcpy): Move warnings
419 to maybe_diag_stxncpy_trunc. Call it.
420 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Integrate warnings
421 from gimple_fold_builtin_strcpy. Print inlining stack.
422 (handle_builtin_stxncpy): Print inlining stack.
423 * tree-ssa-strlen.h (maybe_diag_stxncpy_trunc): Declare.
425 2018-02-22 H.J. Lu <hongjiu.lu@intel.com>
428 * config/i386/i386.c (ix86_set_indirect_branch_type): Issue an
429 error when -mindirect-branch=thunk-extern, -fcf-protection=branch
430 and -fcheck-pointer-bounds are used together.
431 (indirect_thunk_prefix): New enum.
432 (indirect_thunk_need_prefix): New function.
433 (indirect_thunk_name): Replace need_bnd_p with need_prefix. Use
434 "_nt" instead of "_bnd" for NOTRACK prefix.
435 (output_indirect_thunk): Replace need_bnd_p with need_prefix.
436 (output_indirect_thunk_function): Likewise.
438 (ix86_code_end): Update output_indirect_thunk_function calls.
439 (ix86_output_indirect_branch_via_reg): Replace
440 ix86_bnd_prefixed_insn_p with indirect_thunk_need_prefix.
441 (ix86_output_indirect_branch_via_push): Likewise.
442 (ix86_output_function_return): Likewise.
443 * doc/invoke.texi: Document -mindirect-branch=thunk-extern is
444 incompatible with -fcf-protection=branch and
445 -fcheck-pointer-bounds.
447 2018-02-22 Steve Ellcey <sellcey@cavium.com>
450 * config/aarch64/aarch64.c (aarch64_print_address_internal):
451 Change gcc_assert call to output_operand_lossage.
453 2018-02-22 Steve Ellcey <sellcey@cavium.com>
455 * doc/extend.texi (__builtin_extend_pointer): Document builtin.
457 2018-02-22 DJ Delorie <dj@redhat.com>
458 Sebastian Perta <sebastian.perta@renesas.com>
459 Oleg Endo <olegendo@gcc.gnu.org>
461 * config/rx/rx.c (rx_rtx_costs): New function.
462 (TARGET_RTX_COSTS): Override to use rx_rtx_costs.
464 2018-02-22 Thomas Preud'homme <thomas.preudhomme@arm.com>
466 * config/arm/t-multilib: Map Armv8-R to Armv7 multilibs.
468 2018-02-22 Martin Liska <mliska@suse.cz>
471 * common/config/arm/arm-common.c (arm_print_hint_for_cpu_option):
472 Add "native" as a possible value.
474 2018-02-22 Martin Liska <mliska@suse.cz>
477 * config/i386/i386.c (ix86_option_override_internal):
478 Add "native" as a possible value for -march and -mtune.
480 2018-02-22 Jakub Jelinek <jakub@redhat.com>
483 * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
484 to all type variants.
486 PR tree-optimization/84503
487 * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
488 width as info->bitpos + info->bitsize - start.
489 (merged_store_group::merge_overlapping): Simplify width computation.
490 (check_no_overlap): New function.
491 (imm_store_chain_info::try_coalesce_bswap): Compute expected
492 start + width and last_order of the group, fail if check_no_overlap
494 (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
495 to group if check_no_overlap fails.
497 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
499 * config/rs6000/altivec.md: Delete contraint arguments to
500 define_expand, define_split, and define_peephole2, and in
501 define_insn_and_split if always unused.
502 * config/rs6000/darwin.md: Ditto.
503 * config/rs6000/dfp.md: Ditto.
504 * config/rs6000/rs6000.md: Ditto.
505 * config/rs6000/sync.md: Ditto.
506 * config/rs6000/vector.md: Ditto.
507 * config/rs6000/vsx.md: Ditto.
509 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
511 * config/rs6000/altivec.md: Write output control strings as braced
512 blocks instead of double-quoted strings.
513 * config/rs6000/darwin.md: Ditto.
514 * config/rs6000/rs6000.md: Ditto.
515 * config/rs6000/vector.md: Ditto.
516 * config/rs6000/vsx.md: Ditto.
518 2018-02-21 Jason Merrill <jason@redhat.com>
520 PR c++/84314 - ICE with templates and fastcall attribute.
521 * attribs.c (build_type_attribute_qual_variant): Remove assert.
523 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
525 * ipa-cp.c (determine_versionability): Fix comment typos.
527 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
530 * ipa-cp.c (determine_versionability): Do not version functions caling
533 2018-02-21 Martin Liska <mliska@suse.cz>
536 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
537 Add "native" as a possible value.
538 * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT): Define
539 the macro when native cpu detection is available.
541 2018-02-21 Martin Liska <mliska@suse.cz>
544 * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
545 Add "native" as a possible value.
546 * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
547 when native cpu detection is available.
549 2018-02-21 Jakub Jelinek <jakub@redhat.com>
550 Martin Sebor <msebor@redhat.com>
552 PR tree-optimization/84478
553 * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
555 * gimple-fold.c (get_range_strlen): Make minlen const and assume it
556 can't be NULL. Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
557 support which is conservatively correct, for 2 only stay conservative
558 for maxlen. Formatting and comment capitalization fixes. Add STRICT
559 argument to the 2 argument get_range_strlen, adjust 6 arg
560 get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
562 (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
563 (gimple_fold_builtin_strlen): Pass true as last argument to
566 2018-02-20 Martin Sebor <msebor@redhat.com>
569 * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
570 (builtin_memref::set_base_and_offset): Same. Handle inner references.
571 (builtin_memref::builtin_memref): Factor out parts into
572 set_base_and_offset and call it.
574 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
577 * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
578 is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
579 greater precision. If to_mode is a MODE_PARTIAL_INT, stop the
580 search at the associated MODE_INT.
582 2018-02-20 Jeff Law <law@redhat.com>
585 PR tree-optimization/81592
587 * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
588 for range data rather than using global data.
589 * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
590 range data rather than using global data.
591 * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
592 pass it to children as needed.
593 (struct directive::fmtresult): Similarly.
594 (struct directive::set_width): Similarly.
595 (struct directive::set_precision): Similarly.
596 (format_integer, format_directive, parse_directive): Similarly.
597 (format_none): Accept unnamed vr_values parameter.
598 (format_percent, format_floating, format_character): Similarly.
599 (format_string, format_plain): Similarly.
600 * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
601 the EVRP range analyzer for range data rather than using global data.
602 * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
603 gimple-ssa-evrp-analyze.h
604 (class sprintf_dom_walker): Add after_dom_children member function.
605 Add evrp_range_analyzer member.
606 (sprintf_dom_walker::before_dom_children): Call into the EVRP
607 range analyzer as needed.
608 (sprintf_dom_walker::after_dom_children): New member function.
609 * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
611 (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
612 (evrp_range_analyzer::pop_to_marker): Likewise.
614 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
616 PR tree-optimization/84419
617 * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
618 with the required type if its current type is compatible but
621 2018-02-20 Jakub Jelinek <jakub@redhat.com>
624 * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
627 2018-02-20 Martin Liska <mliska@suse.cz>
630 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
631 possible values if we don't have a hint.
633 2018-02-20 Martin Liska <mliska@suse.cz>
637 * final.c (shorten_branches): Build align_tab array with one
639 * opts.c (finish_options): Add alignment option limit check.
640 (MAX_CODE_ALIGN): Likewise.
641 (MAX_CODE_ALIGN_VALUE): Likewise.
642 * doc/invoke.texi: Document maximum allowed option value for
643 all -falign-* options.
645 2018-02-19 Jakub Jelinek <jakub@redhat.com>
648 * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
649 * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
650 * var-tracking.c (emit_note_insn_var_location): Remove all references
651 to NOTE_INSN_CALL_ARG_LOCATION.
652 (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
653 the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
654 Use copy_rtx_if_shared.
655 * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
656 NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
657 (dwarf2out_var_location): Remove handling of
658 NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
660 * final.c (final_scan_insn): Remove all references to
661 NOTE_INSN_CALL_ARG_LOCATION.
662 (rest_of_clean_state): Likewise. Remove REG_CALL_ARG_LOCATION notes
663 before dumping final insns.
664 * except.c (emit_note_eh_region_end): Remove all references to
665 NOTE_INSN_CALL_ARG_LOCATION.
666 * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
667 * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
668 * config/arc/arc.c (hwloop_optimize): Likewise.
669 * config/arm/arm.c (create_fix_barrier): Likewise.
670 * config/s390/s390.c (s390_chunkify_start): Likewise.
671 * config/sh/sh.c (find_barrier): Likewise.
672 * config/i386/i386.c (rest_of_insert_endbranch,
673 ix86_seh_fixup_eh_fallthru): Likewise.
674 * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
675 * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
676 * config/frv/frv.c (frv_function_prologue): Likewise.
677 * emit-rtl.c (try_split): Likewise. Copy over REG_CALL_ARG_LOCATION
679 (note_outside_basic_block_p): Remove all references to
680 NOTE_INSN_CALL_ARG_LOCATION.
681 * gengtype.c (adjust_field_rtx_def): Likewise.
682 * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
684 * jump.c (cleanup_barriers, delete_related_insns): Likewise.
685 * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
688 * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
691 PR tree-optimization/84452
692 * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
693 expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
696 2018-02-19 Martin Liska <mliska@suse.cz>
699 * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
701 2018-02-19 Martin Liska <mliska@suse.cz>
702 Richard Sandiford <richard.sandiford@linaro.org>
704 PR tree-optimization/82491
705 * gimple-fold.c (get_base_constructor): Make earlier bail out
708 2018-02-19 Carl Love <cel@us.ibm.com>
710 * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
711 BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
713 * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
716 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
718 * config/rl78/rl78.md (movdf): New define expand.
720 2018-02-19 Martin Liska <mliska@suse.cz>
723 * doc/invoke.texi: Fix typo.
724 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
726 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
728 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
729 handle rs6000_single_float and rs6000_double_float specially for
732 2018-02-16 Jeff Law <law@redhat.com>
734 * config/rx/rx.c (add_pop_cfi_notes): New function.;
737 2018-02-16 Jakub Jelinek <jakub@redhat.com>
740 * ipa-inline.c (inline_small_functions): Fix a typo.
742 2018-02-16 Nathan Sidwell <nathan@acm.org>
744 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
746 2018-02-16 Carl Love <cel@us.ibm.com>
748 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
749 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
750 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
751 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
752 expansion to P8V_BUILTIN_VEC_FLOAT2.
754 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
756 PR rtl-optimization/70023
757 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
758 src_regno into account.
760 2018-02-16 Carl Love <cel@us.ibm.com>
762 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
763 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
764 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
765 * config/rs6000/rs6000.c: Remove case statements for
766 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
767 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
768 and P9V_BUILTIN_VEC_VINSERT4B.
769 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
770 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
771 * config/rs6000/vsx.md:
772 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
775 2018-02-16 Carl Love <cel@us.ibm.com>
777 * config/rs6000/altivec.h: Add builtin names vec_extract4b
779 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
781 * config/rs6000/rs6000-c.c: Add the definitions for
782 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
783 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
784 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
785 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
786 definition for insert4b and define insn *insert3b_internal.
787 * doc/extend.texi: Add documentation for vec_extract4b.
789 2018-02-16 Nathan Sidwell <nathan@acm.org>
791 * doc/extend.texi (Backwards Compatibility): Mention friend
792 injection. Note for-scope is deprecated.
793 * doc/invoke.texi (-ffriend-injection): Deprecate.
795 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
797 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
798 that moved to I2, also allow destinations that are a paradoxical
799 subreg (instead of a normal reg).
801 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
804 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
807 2018-02-16 Richard Biener <rguenther@suse.de>
809 PR tree-optimization/84037
810 PR tree-optimization/84016
812 * config/i386/i386.c (ix86_builtin_vectorization_cost):
813 Adjust vec_construct for the fact we need additional higher latency
814 128bit inserts for AVX256 and AVX512 vector builds.
815 (ix86_add_stmt_cost): Scale vector construction cost for
818 2018-02-16 Richard Biener <rguenther@suse.de>
820 PR tree-optimization/84417
821 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
822 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
823 (non_rewritable_lvalue_p): Likewise, use poly-ints.
825 2018-02-16 Martin Liska <mliska@suse.cz>
828 * internal-fn.def (ASAN_CHECK): Set proper flags.
829 (ASAN_MARK): Likewise.
831 2018-02-16 Julia Koval <julia.koval@intel.com>
833 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
836 2018-02-16 Jakub Jelinek <jakub@redhat.com>
839 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
840 Use ++iter rather than iter++ for std::list iterators.
841 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
842 defer deleting them until all nodes in the forest are processed. Do
843 free even leaf nodes. Change to_process into auto_vec.
846 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
847 * vec.h (vec_default_construct): Use memset instead of placement new
848 if BROKEN_VALUE_INITIALIZATION is defined.
849 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
850 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
853 PR rtl-optimization/83723
854 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
855 * lra.c (lra_substitute_pseudo): Likewise. If true, use
856 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
858 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
860 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
862 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
864 PR rtl-optimization/81443
865 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
866 from inner REGs to paradoxical SUBREGs.
868 2018-02-16 Richard Biener <rguenther@suse.de>
870 PR tree-optimization/84399
871 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
872 For operands we can analyze at their definition make sure we can
873 analyze them at each use as well.
875 2018-02-16 Richard Biener <rguenther@suse.de>
877 PR tree-optimization/84190
878 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
879 volatile accesses if the decl isn't volatile.
881 2018-02-15 Jason Merrill <jason@redhat.com>
883 PR c++/84314 - ICE with templates and fastcall attribute.
884 * attribs.c (build_type_attribute_qual_variant): Don't clobber
885 TYPE_CANONICAL on an existing type.
887 2018-02-15 Jakub Jelinek <jakub@redhat.com>
889 PR tree-optimization/84383
890 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
891 dstoff nor call operand_equal_p if dstbase is NULL.
893 PR tree-optimization/84334
894 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
895 also a CONSTANT_CLASS_P, punt.
897 2018-02-14 Jim Wilson <jimw@sifive.com>
899 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
900 first SMALL_OPERAND check. New local min_second_step. Move assert
901 to where locals are set. Add TARGET_RVC support.
902 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
904 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
906 * doc/invoke.texi: Correct -Wformat-overflow code sample.
908 2018-02-14 Martin Sebor <msebor@redhat.com>
910 PR tree-optimization/83698
911 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
912 arrays constrain the offset range to their bounds.
913 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
914 (builtin_access::overlap): Avoid setting the size of overlap if it's
916 (maybe_diag_overlap): Also consider arrays when deciding what values
917 of offsets to include in diagnostics.
919 2018-02-14 Martin Sebor <msebor@redhat.com>
922 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
923 that correspond to the kind of a declaration.
925 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
928 * config/pa/pa.md: Load address of PIC label using the linkage table
929 if the label is nonlocal.
931 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
933 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
934 warning message if user requests -maltivec=be.
935 * doc/invoke.texi: Document deprecation of -maltivec=be.
937 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
940 * config/rs6000/rs6000-c.c: Update definitions for
941 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
942 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
944 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
947 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
948 add _get_ssp intrinsics. Remove argument from
949 __builtin_ia32_rdssp[d|q].
950 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
951 * config/i386/i386-builtin.def: Remove argument from
952 __builtin_ia32_rdssp[d|q].
953 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
954 ix86_expand_special_args_builtin for _rdssp[d|q].
955 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
956 Clear register before usage.
957 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
958 Add documentation for new _get_ssp and _inc_ssp intrinsics.
960 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
962 PR tree-optimization/84357
963 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
964 operand 1 of an ARRAY_REF too.
966 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
969 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
970 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
972 (set_of_reg): New struct.
973 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
974 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
975 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
977 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
978 Split into bitclr, bitset, bitinvert patterns if appropriate.
979 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
980 use rx_fuse_in_memory_bitop.
981 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
982 to named insn, correct maximum insn length.
984 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
987 * machmode.def: Define a complex mode for PARTIAL_INT.
988 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
990 * doc/rtl.texi: Document CSPImode.
991 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
993 (msp430_hard_regno_nregs_with_padding): Likewise.
995 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
998 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
1000 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
1002 PR rtl-optimization/84169
1003 * combine.c (try_combine): New variable split_i2i3. Set it to true if
1004 we generated a parallel as new i3 and we split that to new i2 and i3
1005 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
1006 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
1007 those to i2, not i1. Partially rewrite this scan code.
1009 2018-02-13 Jakub Jelinek <jakub@redhat.com>
1012 * stor-layout.c (place_field): For variable length fields, adjust
1013 offset_align afterwards not just based on the field's alignment,
1014 but also on the size.
1017 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
1018 of exps and logs in the use_exp2 case.
1020 2018-02-13 Jeff Law <law@redhat.com>
1022 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
1025 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
1028 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
1032 * common.opt (gas-loc-support, gas-locview-support): New.
1033 (ginline-points, ginternal-reset-location-views): New.
1034 * doc/invoke.texi: Document them. Use @itemx where intended.
1035 (gvariable-location-views): Adjust.
1036 * target.def (reset_location_view): New.
1037 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
1038 (TARGET_RESET_LOCATION_VIEW): New.
1039 * doc/tm.texi: Rebuilt.
1040 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
1041 (dwarf2out_default_as_locview_support): New.
1042 (output_asm_line_debug_info): Use option variables.
1043 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
1044 (output_loc_list): Likewise.
1045 (add_high_low_attributes): Check option variables.
1046 Don't output entry view attribute in strict mode.
1047 (gen_inlined_subroutine_die): Check option variables.
1048 (dwarf2out_inline_entry): Likewise.
1049 (init_sections_and_labels): Likewise.
1050 (dwarf2out_early_finish): Likewise.
1051 (maybe_reset_location_view): New, from...
1052 (dwarf2out_var_location): ... here. Call it.
1053 * debug.h (dwarf2out_default_as_loc_support): Declare.
1054 (dwarf2out_default_as_locview_support): Declare.
1055 * hooks.c (hook_int_rtx_insn_0): New.
1056 * hooks.h (hook_int_rtx_insn_0): Declare.
1057 * toplev.c (process_options): Take -gas-loc-support and
1058 -gas-locview-support from dwarf2out. Enable
1059 -gvariable-location-views by default only with locview
1060 assembler support. Enable -ginternal-reset-location-views by
1061 default only if the target defines the corresponding hook.
1062 Enable -ginline-points by default if location views are
1063 enabled; force it disabled if statement frontiers are
1065 * tree-inline.c (expand_call_inline): Check option variables.
1066 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
1068 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
1070 PR tree-optimization/84321
1071 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
1072 handling. Also check whether the anti-range contains any values
1073 that satisfy the mask; switch to a VR_RANGE if not.
1075 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
1078 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
1080 2018-02-13 Martin Jambor <mjambor@suse.cz>
1083 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
1084 of call statements, also set location of a load to a temporary.
1086 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
1088 * config/rl78/rl78.c (add_vector_labels): New function.
1089 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
1090 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
1091 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
1092 which checks that no arguments are passed.
1093 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
1094 * doc/extend.texi: Documentation for the new attribute.
1096 2018-02-13 Andreas Schwab <schwab@suse.de>
1098 * config/riscv/linux.h (CPP_SPEC): Define.
1100 2018-02-13 Jakub Jelinek <jakub@redhat.com>
1103 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
1104 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
1105 OPTION_MASK_ISA_AES as first argument to def_builtin_const
1106 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
1107 instead of OPTION_MASK_ISA_PCLMUL as first argument to
1108 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
1109 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
1110 temporarily for AES and PCLMUL builtins.
1112 PR tree-optimization/84339
1113 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
1114 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
1118 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
1119 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
1120 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
1122 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
1124 * omp-simd-clone.h: New file.
1125 * omp-simd-clone.c: Include omp-simd-clone.h.
1126 (expand_simd_clones): No longer static.
1127 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
1128 cgraph.h and omp-simd-clone.h.
1129 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
1130 (vect_recog_widen_shift_pattern): Formatting fix.
1131 (vect_pattern_recog_1): Don't check optab for calls.
1134 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
1135 operands[2] into a REG before using gen_lowpart on it.
1137 2018-02-12 Jeff Law <law@redhat.com>
1140 * config/sh/sh.c (find_barrier): Consider a sibling call
1143 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
1144 successfully back substituting a reg.
1146 2018-02-12 Richard Biener <rguenther@suse.de>
1148 PR tree-optimization/84037
1149 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
1150 parameter, move visited init to caller.
1151 (vect_slp_analyze_operations): Separate cost from validity
1152 check, initialize visited once for all instances.
1153 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
1155 * tree-vect-stmts.c (vect_model_simple_cost): Make early
1157 (vect_model_promotion_demotion_cost): Likewise.
1158 (vectorizable_bswap): Guard cost modeling with !slp_node
1159 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
1161 (vectorizable_call): Likewise.
1162 (vectorizable_conversion): Likewise.
1163 (vectorizable_assignment): Likewise.
1164 (vectorizable_shift): Likewise.
1165 (vectorizable_operation): Likewise.
1166 (vectorizable_store): Likewise.
1167 (vectorizable_load): Likewise.
1168 (vectorizable_condition): Likewise.
1169 (vectorizable_comparison): Likewise.
1171 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
1174 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
1175 (ASAN_MARK): Fix fnspec to account for return value, change pointer
1176 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
1178 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1181 * params.def (inline-min-speedup): Increase from 8 to 15.
1182 (max-inline-insns-auto): Decrease from 40 to 30.
1183 * ipa-split.c (consider_split): Add some buffer for function to
1184 be considered inlining candidate.
1185 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
1188 2018-02-12 Richard Biener <rguenther@suse.de>
1190 PR tree-optimization/84037
1191 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
1192 matched stmts if we cannot swap the non-matched ones.
1194 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
1196 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
1197 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
1198 _mm_maskz_scalef_round_ss): New intrinsics.
1199 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
1200 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
1201 __builtin_ia32_scalefss_round): Remove.
1202 (__builtin_ia32_scalefsd_mask_round,
1203 __builtin_ia32_scalefss_mask_round): New intrinsics.
1204 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
1205 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
1206 ((match_operand:VF_128 2 "<round_nimm_predicate>"
1207 "<round_constraint>")): Changed to ...
1208 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
1209 "<round_scalar_constraint>")): ... this.
1210 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
1211 %0, %1, %2<round_op3>}"): Changed to ...
1212 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
1213 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
1214 %2<round_scalar_mask_op3>}"): ... this.
1215 * config/i386/subst.md (round_scalar_nimm_predicate): New.
1217 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
1219 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
1220 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
1221 (_mm_maskz_sqrt_round_ss): New intrinsics.
1222 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
1223 (__builtin_ia32_sqrtsd_mask_round)
1224 (__builtin_ia32_sqrtss_mask_round): New builtins.
1225 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
1226 (__builtin_ia32_sqrtss_round): Remove.
1227 (__builtin_ia32_sqrtsd_mask_round)
1228 (__builtin_ia32_sqrtss_mask_round): New builtins.
1229 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
1230 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
1231 ((match_operand:VF_128 1 "vector_operand"
1232 "xBm,<round_constraint>")): Changed to ...
1233 ((match_operand:VF_128 1 "vector_operand"
1234 "xBm,<round_scalar_constraint>")): ... this.
1235 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
1236 %0, %2, %<iptr>1<round_op3>}): Changed to ...
1237 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
1238 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
1239 %<iptr>1<round_scalar_mask_op3>}): ... this.
1240 ((set_attr "prefix" "<round_prefix>")): Changed to ...
1241 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
1243 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
1246 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
1247 Cast vec_cmpeq result to correct type.
1248 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
1249 Cast vec_cmpgt result to correct type.
1251 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
1253 * final.c (final_scan_insn_1): Renamed from...
1254 (final_scan_insn): ... this. New wrapper, to recover
1255 seen from the outermost call in recursive ones.
1256 * config/sparc/sparc.c (output_return): Drop seen from call.
1257 (output_sibcall): Likewise.
1258 * config/visium/visium.c (output_branch): Likewise.
1260 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
1262 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
1265 2018-02-10 Alan Modra <amodra@gmail.com>
1268 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
1269 Specify LR as an input.
1271 2018-02-10 Jakub Jelinek <jakub@redhat.com>
1274 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
1275 remove_member_access_dummy_vars): New functions.
1276 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
1277 lower_omp_1, execute_lower_omp): Use them.
1279 PR rtl-optimization/84308
1280 * shrink-wrap.c (spread_components): Release todo vector.
1282 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
1284 PR rtl-optimization/57193
1285 * ira-color.c (struct allocno_color_data): Add member
1286 conflict_allocno_hard_prefs.
1287 (update_conflict_allocno_hard_prefs): New.
1288 (bucket_allocno_compare_func): Add a preference based on
1289 conflict_allocno_hard_prefs.
1290 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
1291 (color_allocnos): Remove a dead code. Initiate
1292 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
1294 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1297 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
1298 constraint from =wa to wa. Avoid a subreg on the output operand,
1299 instead use a pseudo and subreg it in a move.
1300 (p9_xxbrd_<mode>): Changed to ...
1301 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
1302 (p9_xxbrd_v2df): New expander.
1303 (p9_xxbrw_<mode>): Changed to ...
1304 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
1305 (p9_xxbrw_v4sf): New expander.
1307 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1309 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
1311 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
1314 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
1315 multiply in 32-bit mode.
1316 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
1317 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
1320 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
1322 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
1323 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
1324 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
1325 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
1327 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
1330 * dwarf2out.c (is_trivial_indirect_ref): New function.
1331 (dwarf2out_late_global_decl): Do not generate a location
1332 attribute for variables that have a non-trivial DECL_VALUE_EXPR
1333 and that are not defined in the current unit.
1335 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1337 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
1338 instead of a libcall for UNORDERED.
1340 2018-02-09 Tamar Christina <tamar.christina@arm.com>
1343 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
1344 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
1346 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1349 * config/s390/s390.c (s390_set_current_function): Invoke
1350 s390_indirect_branch_settings also if fndecl didn't change.
1352 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1354 * config/rs6000/rs6000.md (blockage): Set length to zero.
1356 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
1358 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
1360 2018-02-09 Jakub Jelinek <jakub@redhat.com>
1363 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
1364 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
1368 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
1369 PARALLEL incoming that failed vt_get_decl_and_offset check.
1372 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
1373 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
1375 (get_variable_section): For decls in named .bss* sections pass true as
1376 second argument to bss_initializer_p.
1378 2018-02-09 Marek Polacek <polacek@redhat.com>
1379 Jakub Jelinek <jakub@redhat.com>
1382 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
1383 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
1384 Sync some changes from cxx_fold_indirect_ref.
1386 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
1388 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
1390 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
1391 (BLOCK_INLINE_ENTRY_LABEL): New.
1392 (dwarf2out_var_location): Disregard inline entry markers.
1393 (inline_entry_data): New struct.
1394 (inline_entry_data_hasher): New hashtable type.
1395 (inline_entry_data_hasher::hash): New.
1396 (inline_entry_data_hasher::equal): New.
1397 (inline_entry_data_table): New variable.
1398 (add_high_low_attributes): Add DW_AT_entry_pc and
1399 DW_AT_GNU_entry_view attributes if a pending entry is found
1400 in inline_entry_data_table. Add old entry_pc attribute only
1401 if debug nonbinding markers are disabled.
1402 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
1403 markers are enabled.
1404 (block_within_block_p, dwarf2out_inline_entry): New.
1405 (dwarf2out_finish): Check that no entries remained in
1406 inline_entry_data_table.
1407 * final.c (reemit_insn_block_notes): Handle inline entry notes.
1408 (final_scan_insn, notice_source_line): Likewise.
1409 (rest_of_clean_state): Skip inline entry markers.
1410 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
1412 * gimple.c (gimple_build_debug_inline_entry): New.
1413 * gimple.h (enum gimple_debug_subcode): Add
1414 GIMPLE_DEBUG_INLINE_ENTRY.
1415 (gimple_build_debug_inline_entry): Declare.
1416 (gimple_debug_inline_entry_p): New.
1417 (gimple_debug_nonbind_marker_p): Adjust.
1418 * insn-notes.def (INLINE_ENTRY): New.
1419 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
1420 inline entry marker notes.
1421 (print_insn): Likewise.
1422 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
1423 (INSN_DEBUG_MARKER_KIND): Likewise.
1424 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
1425 * tree-inline.c (expand_call_inline): Build and insert
1426 debug_inline_entry stmt.
1427 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
1428 inline entry blocks early, if nonbind markers are enabled.
1429 (dump_scope_block): Dump fragment info.
1430 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
1431 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
1432 (gimple_build_debug_inline_entry): New.
1433 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
1434 Enable/disable inline entry points too.
1435 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
1436 (DEBUG_INSN): Describe inline entry markers.
1438 * common.opt (gvariable-location-views): New.
1439 (gvariable-location-views=incompat5): New.
1440 * config.in: Rebuilt.
1441 * configure: Rebuilt.
1442 * configure.ac: Test assembler for view support.
1443 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
1444 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
1445 * dwarf2out.c (var_loc_view): New typedef.
1446 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
1447 (dwarf2out_locviews_in_attribute): New.
1448 (dwarf2out_locviews_in_loclist): New.
1449 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
1450 (enum dw_line_info_opcode): Add LI_adv_address.
1451 (struct dw_line_info_table): Add view.
1452 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
1453 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
1454 (zero_view_p): New variable.
1455 (ZERO_VIEW_P): New macro.
1456 (output_asm_line_debug_info): New.
1457 (struct var_loc_node): Add view.
1458 (add_AT_view_list, AT_loc_list): New.
1459 (add_var_loc_to_decl): Add view param. Test it against last.
1460 (new_loc_list): Add view params. Record them.
1461 (AT_loc_list_ptr): Handle loc and view lists.
1462 (view_list_to_loc_list_val_node): New.
1463 (print_dw_val): Handle dw_val_class_view_list.
1464 (size_of_die): Likewise.
1465 (value_format): Likewise.
1466 (loc_list_has_views): New.
1467 (gen_llsym): Set vl_symbol too.
1468 (maybe_gen_llsym, skip_loc_list_entry): New.
1469 (dwarf2out_maybe_output_loclist_view_pair): New.
1470 (output_loc_list): Output view list or entries too.
1471 (output_view_list_offset): New.
1472 (output_die): Handle dw_val_class_view_list.
1473 (output_dwarf_version): New.
1474 (output_compilation_unit_header): Use it.
1475 (output_skeleton_debug_sections): Likewise.
1476 (output_rnglists, output_line_info): Likewise.
1477 (output_pubnames, output_aranges): Update version comments.
1478 (output_one_line_info_table): Output view numbers in asm comments.
1479 (dw_loc_list): Determine current endview, pass it to new_loc_list.
1480 Call maybe_gen_llsym.
1481 (loc_list_from_tree_1): Adjust.
1482 (add_AT_location_description): Create view list attribute if
1483 needed, check it's absent otherwise.
1484 (convert_cfa_to_fb_loc_list): Adjust.
1485 (maybe_emit_file): Call output_asm_line_debug_info for test.
1486 (dwarf2out_var_location): Reset views as needed. Precompute
1487 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
1488 attribute. Set view.
1489 (new_line_info_table): Reset next view.
1490 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
1491 (dwarf2out_source_line): Likewise. Output view resets and labels to
1492 the assembler, or select appropriate line info opcodes.
1493 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
1494 (optimize_string_length): Catch it. Adjust.
1495 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
1496 dw_val_class_view_list, and remove it if no longer needed.
1497 (hash_loc_list): Hash view numbers.
1498 (loc_list_hasher::equal): Compare them.
1499 (optimize_location_lists): Check whether a view list symbol is
1500 needed, and whether the locview attribute is present, and
1501 whether they match. Remove the locview attribute if no longer
1503 (index_location_lists): Call skip_loc_list_entry for test.
1504 (dwarf2out_finish): Call output_asm_line_debug_info for test.
1505 Use output_dwarf_version.
1506 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
1507 (struct dw_val_node): Add val_view_list.
1508 * final.c (SEEN_NEXT_VIEW): New.
1509 (set_next_view_needed): New.
1510 (clear_next_view_needed): New.
1511 (maybe_output_next_view): New.
1512 (final_start_function): Rename to...
1513 (final_start_function_1): ... this. Take pointer to FIRST,
1514 add SEEN parameter. Emit param bindings in the initial view.
1515 (final_start_function): Reintroduce SEEN-less interface.
1516 (final): Rename to...
1517 (final_1): ... this. Take SEEN parameter. Output final pending
1518 next view at the end.
1519 (final): Reintroduce seen-less interface.
1520 (final_scan_insn): Output pending next view before switching
1521 sections or ending a block. Mark the next view as needed when
1522 outputting variable locations. Notify debug backend of section
1523 changes, and of location view changes.
1524 (rest_of_handle_final): Adjust.
1525 * toplev.c (process_options): Autodetect value for debug variable
1526 location views option. Warn on incompat5 without -gdwarf-5.
1527 * doc/invoke.texi (gvariable-location-views): New.
1528 (gvariable-location-views=incompat5): New.
1529 (gno-variable-location-views): New.
1531 2018-02-08 David Malcolm <dmalcolm@redhat.com>
1533 PR tree-optimization/84136
1534 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
1535 that the result of find_edge is non-NULL.
1537 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1540 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
1541 storing integer register in SImode. Fix cost of 256 and 512
1542 byte aligned SSE register store.
1544 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1546 * config/i386/i386.c (ix86_multiplication_cost): Fix
1547 multiplication cost for TARGET_AVX512DQ.
1549 2018-02-08 Marek Polacek <polacek@redhat.com>
1551 PR tree-optimization/84238
1552 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
1555 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1557 PR tree-optimization/84265
1558 * tree-vect-stmts.c (vectorizable_store): Don't treat
1559 VMAT_CONTIGUOUS accesses as grouped.
1560 (vectorizable_load): Likewise.
1562 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1564 PR tree-optimization/81635
1565 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
1566 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
1567 (test_round_for_mask): New functions.
1568 (wide_int_cc_tests): Call test_round_for_mask.
1569 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
1570 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
1571 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
1572 range returned by get_range_info.
1574 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1577 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
1578 * symtab.c: Include builtins.h
1579 (symtab_node::output_to_lto_symbol_table_p): Move here
1580 from lto-streamer-out.c:output_symbol_p.
1581 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
1582 (output_symbol_p): Move all logic to symtab.c
1583 (produce_symtab): Update.
1585 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1587 * config/s390/s390-opts.h (enum indirect_branch): Define.
1588 * config/s390/s390-protos.h (s390_return_addr_from_memory)
1589 (s390_indirect_branch_via_thunk)
1590 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
1591 (enum s390_indirect_branch_type): Define.
1592 * config/s390/s390.c (struct s390_frame_layout, struct
1593 machine_function): Remove.
1594 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
1595 (indirect_branch_table_label_no, indirect_branch_table_name):
1597 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
1598 (enum s390_indirect_branch_option): Define.
1599 (s390_return_addr_from_memory): New function.
1600 (s390_handle_string_attribute): New function.
1601 (s390_attribute_table): Add new attribute handler.
1602 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
1603 (s390_indirect_branch_via_thunk): New function.
1604 (s390_indirect_branch_via_inline_thunk): New function.
1605 (s390_function_ok_for_sibcall): When jumping via thunk disallow
1606 sibling call optimization for non z10 compiles.
1607 (s390_emit_call): Force indirect branch target to be a single
1608 register. Add r1 clobber for non-z10 compiles.
1609 (s390_emit_epilogue): Emit return jump via return_use expander.
1610 (s390_reorg): Handle JUMP_INSNs as execute targets.
1611 (s390_option_override_internal): Perform validity checks for the
1612 new command line options.
1613 (s390_indirect_branch_attrvalue): New function.
1614 (s390_indirect_branch_settings): New function.
1615 (s390_set_current_function): Invoke s390_indirect_branch_settings.
1616 (s390_output_indirect_thunk_function): New function.
1617 (s390_code_end): Implement target hook.
1618 (s390_case_values_threshold): Implement target hook.
1619 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
1621 * config/s390/s390.h (struct s390_frame_layout)
1622 (struct machine_function): Move here from s390.c.
1623 (TARGET_INDIRECT_BRANCH_NOBP_RET)
1624 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
1625 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
1626 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
1627 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
1628 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
1629 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
1630 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
1631 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
1632 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
1633 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
1634 (mnemonic attribute): Add values which aren't recognized
1636 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
1637 pattern for branch conversion. Fix mnemonic attribute.
1638 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
1639 indirect branch via thunk if requested.
1640 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
1641 ("*indirect_jump"): Disable for branch conversion using out of
1643 ("indirect_jump_via_thunk<mode>_z10")
1644 ("indirect_jump_via_thunk<mode>")
1645 ("indirect_jump_via_inlinethunk<mode>_z10")
1646 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
1647 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
1648 ("casesi_jump_via_inlinethunk<mode>_z10")
1649 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
1650 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
1651 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
1652 ("*indirect2_jump"): Disable for branch conversion.
1653 ("casesi_jump"): Turn into expander and expand patterns for branch
1655 ("return_use"): New expander.
1656 ("*return"): Emit return via thunk and rename it to ...
1657 ("*return<mode>"): ... this one.
1658 * config/s390/s390.opt: Add new options and and enum for the
1661 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1663 * lra-constraints.c (match_reload): Unconditionally use
1664 gen_lowpart_SUBREG, rather than selecting between that
1665 and equivalent gen_rtx_SUBREG code.
1667 2018-02-08 Richard Biener <rguenther@suse.de>
1669 PR tree-optimization/84233
1670 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
1671 changed flag instead of boguously re-using phi_inserted.
1673 2018-02-08 Martin Jambor <mjambor@suse.cz>
1675 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
1676 static local variables.
1678 2018-02-08 Richard Biener <rguenther@suse.de>
1680 PR tree-optimization/84278
1681 * tree-vect-stmts.c (vectorizable_store): When looking for
1682 smaller vector types to perform grouped strided loads/stores
1683 make sure the mode is supported by the target.
1684 (vectorizable_load): Likewise.
1686 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1688 * config/aarch64/aarch64.c (aarch64_components_for_bb):
1689 Increase LDP/STP opportunities by adding adjacent callee-saves.
1691 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1693 PR rtl-optimization/84068
1694 PR rtl-optimization/83459
1695 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
1697 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
1699 PR tree-optimization/84224
1700 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
1701 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
1704 2018-02-07 Iain Sandoe <iain@codesourcery.com>
1707 * config/rs6000/altivec.md (*restore_world): Remove LR use.
1708 * config/rs6000/predicates.md (restore_world_operation): Adjust op
1709 count, remove one USE.
1711 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
1713 * doc/install.texi (Configuration): Document the
1714 --with-long-double-format={ibm,ieee} PowerPC configuration
1718 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
1719 Convert from define_expand to be define_insn_and_split. Rework
1720 float/double/_Float128 conversions to QI/HI/SImode to work with
1721 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
1722 conversions to QI/HImode types did a store and then a load to
1723 truncate the value. For conversions to VSX registers, don't split
1724 the insn, instead emit the code directly. Use the code iterator
1725 any_fix to combine signed and unsigned conversions.
1726 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
1727 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
1728 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1729 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
1730 (fix_<mode>di2_hw): Likewise.
1731 (fixuns_<mode>di2_hw): Likewise.
1732 (fix_<mode>si2_hw): Likewise.
1733 (fixuns_<mode>si2_hw): Likewise.
1734 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
1735 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1736 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1737 fix<uns>_trunc<SFDF:mode>si2_p8.
1738 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1740 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1741 (fix<uns>_<mode>_mem): Likewise.
1742 (fctiw<u>z_<mode>_mem): Likewise.
1743 (fix<uns>_<mode>_mem): Likewise.
1744 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1745 the register allocator from doing a direct move to the GPRs to do
1746 a store, and instead use the ISA 3.0 store byte/half-word from
1747 vector register instruction. For IEEE 128-bit floating point,
1748 also optimize stores of 32-bit ints.
1749 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1751 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1753 * genextract.c (push_pathstr_operand): New function to support
1755 (walk_rtx): Call push_pathstr_operand.
1756 (print_path): Support [a-zA-Z].
1758 2018-02-07 Richard Biener <rguenther@suse.de>
1760 PR tree-optimization/84037
1761 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1762 (cse_and_gimplify_to_preheader): Declare.
1763 (vect_get_place_in_interleaving_chain): Likewise.
1764 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1766 (_loop_vec_info::~_loop_vec_info): Delete it.
1767 (cse_and_gimplify_to_preheader): New function.
1768 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1769 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1770 (vectorizable_load): Likewise. For grouped stores always base
1771 the IV on the first element.
1772 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1773 condition before gimplifying.
1775 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1777 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1778 *DIV_EXPR and *MOD_EXPR.
1780 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1783 * config/i386/i386.c (ix86_option_override_internal): Mask out
1784 the CF_SET bit when checking -fcf-protection.
1786 2018-02-07 Tom de Vries <tom@codesourcery.com>
1789 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1792 2018-02-07 Richard Biener <rguenther@suse.de>
1794 PR tree-optimization/84204
1795 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1798 PR tree-optimization/84205
1799 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1800 special-case isl_ast_op_zdiv_r.
1802 PR tree-optimization/84223
1803 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1804 Only add conditions from within the region.
1805 (gather_bbs::after_dom_children): Adjust.
1807 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1810 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1811 * config/avr/avr.md: Only post-reload split REG-REG moves if
1812 either register is REGERAL_REG_P.
1814 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1816 PR tree-optimization/84235
1817 * tree-ssa-scopedtables.c
1818 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1819 if the subtraction is performed in floating point type where NaNs are
1820 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1821 build 1. Formatting fix.
1823 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1826 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1827 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1828 and skip it regardless of bb boundaries. Use CALL_P macro,
1829 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1830 unnecessarily, formatting fix.
1832 2018-02-06 Michael Collison <michael.collison@arm.com>
1834 * config/arm/thumb2.md:
1835 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1836 (*thumb_mov_notscc): Ditto.
1838 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1841 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1842 unsigned_fix, not "s".
1844 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1846 * configure.ac (gcc_fn_eh_frame_ro): New function.
1847 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1848 correct .eh_frame permissions.
1849 * configure: Regenerate.
1851 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1853 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1856 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1858 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1859 Display warning message for -mno-speculate-indirect-jumps.
1861 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1863 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1865 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1867 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1869 PR tree-optimization/84225
1870 * tree-eh.c (find_trapping_overflow): Only call
1871 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1873 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1876 * config/i386/i386.c: Reimplement the check of possible options
1877 -mibt/-mshstk conbination. Change error messages.
1878 * doc/invoke.texi: Fix a typo: remove extra '='.
1880 2018-02-06 Marek Polacek <polacek@redhat.com>
1882 PR tree-optimization/84228
1883 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1885 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1888 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1889 emitted arch directives.
1890 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1891 __ARM_FEATURE_COPROC before changing architectures.
1893 2018-02-06 Richard Biener <rguenther@suse.de>
1895 * config/i386/i386.c (print_reg): Fix typo.
1896 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1898 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1900 * configure: Regenerate.
1902 2018-02-05 Martin Sebor <msebor@redhat.com>
1904 PR tree-optimization/83369
1905 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1908 2018-02-05 Martin Liska <mliska@suse.cz>
1910 * doc/invoke.texi: Cherry-pick upstream r323995.
1912 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1914 * ira.c (ira_init_register_move_cost): Adjust comment.
1916 2018-02-05 Martin Liska <mliska@suse.cz>
1918 PR gcov-profile/84137
1919 * doc/gcov.texi: Fix typo in documentation.
1921 2018-02-05 Martin Liska <mliska@suse.cz>
1923 PR gcov-profile/83879
1924 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1925 using dlopen functionality.
1927 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1929 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1930 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1931 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1932 _mm_maskz_range_round_ss): New intrinsics.
1933 (__builtin_ia32_rangesd128_round)
1934 (__builtin_ia32_rangess128_round): Remove.
1935 (__builtin_ia32_rangesd128_mask_round,
1936 __builtin_ia32_rangess128_mask_round): New builtins.
1937 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1938 __builtin_ia32_rangess128_round): Remove.
1939 (__builtin_ia32_rangesd128_mask_round,
1940 __builtin_ia32_rangess128_mask_round): New builtins.
1941 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1942 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1943 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1944 "<round_saeonly_constraint>")): Changed to ...
1945 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1946 "<round_saeonly_scalar_constraint>")): ... this.
1947 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1948 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1949 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1950 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1951 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1953 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1955 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1957 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1958 Remove all values except native, 8540 and 8548.
1960 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1962 * config/i386/i386.c (ix86_output_function_return): Pass
1963 INVALID_REGNUM, instead of -1, as invalid register number to
1964 indirect_thunk_name and output_indirect_thunk.
1966 2018-02-02 Julia Koval <julia.koval@intel.com>
1968 * config.gcc: Add -march=icelake.
1969 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1970 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1971 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1972 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1973 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1974 (processor_target_table): Add icelake.
1975 (ix86_option_override_internal): Handle new PTAs.
1976 (get_builtin_code_for_version): Handle icelake.
1977 (M_INTEL_COREI7_ICELAKE): New.
1978 (fold_builtin_cpu): Handle icelake.
1979 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1980 * doc/invoke.texi: Add -march=icelake.
1982 2018-02-02 Julia Koval <julia.koval@intel.com>
1984 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1985 to wide_int_bitmask.
1986 * wide-int-bitmask.h: New.
1988 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1991 * config/i386/i386.md: Replace Pmode with word_mode in
1992 builtin_setjmp_setup and builtin_longjmp to support x32.
1994 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1998 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
2000 (rs6000_supported_cpu_names): New static variable.
2001 (linux_cpu_translation_table): Likewise.
2002 (elf_platform) <cpu>: Define new static variable and use it.
2003 Translate kernel AT_PLATFORM name to canonical name if needed.
2004 Error if platform name is unknown.
2006 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
2009 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
2011 2018-02-01 Jeff Law <law@redhat.com>
2014 * config/i386/i386.c (release_scratch_register_on_entry): Add new
2015 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
2016 the scratch if RELEASE_VIA_POP is false.
2017 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
2018 If we have to save a temporary register, decrement SIZE appropriately.
2019 Pass new arguments to release_scratch_register_on_entry.
2020 (ix86_adjust_stack_and_probe): Likewise.
2021 (ix86_emit_probe_stack_range): Pass new arguments to
2022 release_scratch_register_on_entry.
2024 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
2026 PR rtl-optimization/84157
2027 * combine.c (change_zero_ext): Use REG_P predicate in
2028 front of HARD_REGISTER_P predicate.
2030 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
2032 * config/avr/avr.c (avr_option_override): Move disabling of
2033 -fdelete-null-pointer-checks to...
2034 * common/config/avr/avr-common.c (avr_option_optimization_table):
2037 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2039 PR tree-optimization/81635
2040 * tree-data-ref.c (split_constant_offset_1): For types that
2041 wrap on overflow, try to use range info to prove that wrapping
2044 2018-02-01 Renlin Li <renlin.li@arm.com>
2047 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
2049 (aarch64_register_move_cost): Likewise.
2050 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
2052 (REG_CLASS_NAMES): Likewise.
2053 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
2054 TAILCALL_ADDR_REGS. Remove IP registers.
2055 * config/aarch64/aarch64.md (Ucs): Update register constraint.
2057 2018-02-01 Richard Biener <rguenther@suse.de>
2059 * domwalk.h (dom_walker::dom_walker): Add additional constructor
2060 for specifying RPO order and allow NULL for that.
2061 * domwalk.c (dom_walker::dom_walker): Likewise.
2062 (dom_walker::walk): Handle NULL RPO order.
2063 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
2065 (rewrite_update_dom_walker): Likewise.
2066 (mark_def_dom_walker): Likewise.
2068 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2070 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
2071 (aarch64_maybe_expand_sve_subreg_move): Declare.
2072 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
2073 * config/aarch64/predicates.md (aarch64_any_register_operand): New
2075 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
2076 that are semantically a reverse operation.
2077 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
2078 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
2079 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
2081 (aarch64_can_change_mode_class): For big-endian, forbid changes
2082 between two SVE modes if they have different element sizes.
2084 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2086 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
2087 the TImode handling for big-endian targets.
2089 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2091 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
2092 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
2094 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
2095 Remove BSWAP handing for big-endian targets and use the form of
2096 LD1RQ appropariate for the mode.
2098 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2100 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
2101 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
2104 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
2107 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
2108 check for operands that need to go through aarch64_sve_reload_be.
2110 2018-02-01 Jakub Jelinek <jakub@redhat.com>
2112 PR tree-optimization/81661
2113 PR tree-optimization/84117
2114 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
2115 * tree-eh.c: Include gimplify.h.
2116 (find_trapping_overflow, replace_trapping_overflow,
2117 rewrite_to_non_trapping_overflow): New functions.
2118 * tree-vect-loop.c: Include tree-eh.h.
2119 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
2120 * tree-data-ref.c: Include tree-eh.h.
2121 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
2123 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
2125 PR rtl-optimization/84123
2126 * combine.c (change_zero_ext): Check if hard register satisfies
2127 can_change_dest_mode before calling gen_lowpart_SUBREG.
2129 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
2132 * ira.c (ira_init_register_move_cost): Remove assert.
2134 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2136 PR rtl-optimization/84071
2137 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
2138 * doc/tm.texi: Regenerate.
2140 2018-01-31 Richard Biener <rguenther@suse.de>
2142 PR tree-optimization/84132
2143 * tree-data-ref.c (analyze_miv_subscript): Properly
2144 check whether evolution_function_is_affine_multivariate_p
2145 before calling gcd_of_steps_may_divide_p.
2147 2018-01-31 Julia Koval <julia.koval@intel.com>
2150 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
2151 * config/i386/i386.md (rdpid_rex64) New.
2152 (rdpid): Make 32bit only.
2154 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
2157 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
2158 an IDENTIFIER_NODE for FUNCTION_TYPE's.
2160 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2163 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2165 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2167 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
2169 PR rtl-optimization/84071
2170 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
2171 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
2173 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
2175 * config/arc/arc.c (arc_handle_aux_attribute): New function.
2176 (arc_attribute_table): Add 'aux' attribute.
2177 (arc_in_small_data_p): Consider aux like variables.
2178 (arc_is_aux_reg_p): New function.
2179 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
2180 (arc_get_aux_arg): New function.
2181 (prepare_move_operands): Handle aux-register access.
2182 (arc_handle_aux_attribute): New function.
2183 * doc/extend.texi (ARC Variable attributes): Add subsection.
2185 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
2187 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
2188 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
2189 (arc_attribute_table): Add 'uncached' attribute.
2190 (arc_print_operand): Print '.di' flag for uncached memory
2192 (arc_in_small_data_p): Do not consider for small data the uncached
2194 (arc_is_uncached_mem_p): New function.
2195 * config/arc/predicates.md (compact_store_memory_operand): Check
2196 for uncached memory accesses.
2197 (nonvol_nonimm_operand): Likewise.
2198 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
2200 2018-01-31 Jakub Jelinek <jakub@redhat.com>
2203 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
2204 falign-loops=): Add Optimization flag.
2206 2018-01-30 Jeff Law <law@redhat.com>
2209 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
2210 INT_REGISTERS_SAVED. Check it prior to calling
2211 get_scratch_register_on_entry.
2212 (ix86_adjust_stack_and_probe): Similarly.
2213 (ix86_emit_probe_stack_range): Similarly.
2214 (ix86_expand_prologue): Corresponding changes.
2216 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2219 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
2220 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
2222 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
2225 * lra-constraints.c (curr_insn_transform): Process AND in the
2228 2018-01-30 Jakub Jelinek <jakub@redhat.com>
2230 PR rtl-optimization/83986
2231 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
2232 dependence against last_pending_memory_flush in addition to
2235 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
2237 PR tree-optimization/81611
2238 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
2241 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
2244 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
2247 2018-01-30 Richard Biener <rguenther@suse.de>
2248 Jakub Jelinek <jakub@redhat.com>
2250 PR tree-optimization/84111
2251 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
2252 inner loops added during recursion, as they don't have up-to-date
2255 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2258 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
2259 (can_inline_edge_by_limits_p): ... here.
2260 (can_early_inline_edge_p, check_callers,
2261 update_caller_keys, update_callee_keys, recursive_inlining,
2262 add_new_edges_to_heap, speculation_useful_p,
2263 inline_small_functions,
2264 inline_small_functions, flatten_function,
2265 inline_to_all_callers_1): Update.
2267 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
2269 * profile-count.c (profile_count::combine_with_ipa_count): Handle
2272 2018-01-30 Richard Biener <rguenther@suse.de>
2274 PR tree-optimization/83008
2275 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
2276 invariant and constant vector uses in stmts when they need
2279 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2282 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
2283 * configure: Regenerate.
2285 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2287 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
2289 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
2290 Use gen_rtx_REG rather than gen_lowpart.
2292 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2294 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
2295 rather than 0 when creating partial subregs.
2297 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
2299 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
2302 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
2305 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
2306 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
2307 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
2308 flags. This restores the settings used before the 2017-07-24.
2309 Turning off pre increment/decrement/modify allows IVOPTS to
2310 optimize DF/SF loops where the index is an int.
2312 2018-01-29 Richard Biener <rguenther@suse.de>
2313 Kelvin Nilsen <kelvin@gcc.gnu.org>
2316 * tree-vect-stmts.c (vectorizable_call): Don't call
2317 targetm.vectorize_builtin_md_vectorized_function if callee is
2320 2018-01-22 Carl Love <cel@us.ibm.com>
2322 * doc/extend.tex: Fix typo in second arg in
2323 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
2325 2018-01-29 Richard Biener <rguenther@suse.de>
2327 PR tree-optimization/84086
2328 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
2329 (flush_ssaname_freelist): When SSA names were released reset
2330 the SCEV hash table.
2332 2018-01-29 Richard Biener <rguenther@suse.de>
2334 PR tree-optimization/84057
2335 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
2336 removed paths when removing edges.
2338 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
2340 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
2341 -mfunction-return=@var{choice}.
2343 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
2346 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
2348 (layout::print_source_line): Likewise.
2349 (test_get_line_width_without_trailing_whitespace): Add test cases.
2351 2018-01-27 Jakub Jelinek <jakub@redhat.com>
2354 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
2357 2018-01-26 Jim Wilson <jimw@sifive.com>
2359 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
2361 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
2364 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2366 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
2367 and CMP + SUB-immediate -> SUBS.
2369 2018-01-26 Martin Sebor <msebor@redhat.com>
2371 PR tree-optimization/83896
2372 * tree-ssa-strlen.c (get_string_len): Rename...
2373 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
2374 Avoid assuming length is constant.
2375 (handle_char_store): Use HOST_WIDE_INT for string length.
2377 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
2380 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
2381 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
2383 2018-01-26 Richard Biener <rguenther@suse.de>
2385 PR rtl-optimization/84003
2386 * dse.c (record_store): Only record redundant stores when
2387 the earlier store aliases at least all accesses the later one does.
2389 2018-01-26 Jakub Jelinek <jakub@redhat.com>
2391 PR rtl-optimization/83985
2392 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
2393 REG_CFA_RESTORE insns.
2394 (delete_unmarked_insns): Don't ignore separate shrink wrapping
2395 REG_CFA_RESTORE insns here.
2398 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
2399 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
2401 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2403 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
2404 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
2405 (arc_init): Likewise.
2406 (arc_override_options): Likewise.
2407 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
2409 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
2411 * config/arc/arc.h (TARGET_DBNZ): Define.
2412 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
2413 properly set the tune attribute.
2414 (dbnz): Use TARGET_DBNZ guard.
2415 * config/arc/arc.opt (mtune): Add core3 option.
2417 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2419 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
2420 recognize new pic like addresses.
2421 (arc_delegitimize_address): Clean up.
2423 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2425 * config/arc/arc-arches.def: Option mrf16 valid for all
2427 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
2428 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
2429 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
2430 * config/arc/arc-tables.opt: Regenerate.
2431 * config/arc/arc.c (arc_conditional_register_usage): Handle
2432 reduced register file case.
2433 (arc_file_start): Set must have build attributes.
2434 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
2436 * config/arc/arc.opt (mrf16): Add new option.
2437 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
2438 * config/arc/genmultilib.awk: Handle new mrf16 option.
2439 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
2440 * config/arc/t-multilib: Regenerate.
2441 * doc/invoke.texi (ARC Options): Document mrf16 option.
2443 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2445 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
2446 * config/arc/arc.c (arc_handle_secure_attribute): New function.
2447 (arc_attribute_table): Add 'secure_call' attribute.
2448 (arc_print_operand): Print secure call operand.
2449 (arc_function_ok_for_sibcall): Don't optimize tail calls when
2451 (arc_is_secure_call_p): New function. * config/arc/arc.md
2452 (call_i): Add support for sjli instruction.
2453 (call_value_i): Likewise.
2454 * config/arc/constraints.md (Csc): New constraint.
2456 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
2457 John Eric Martin <John.Martin@emmicro-us.com>
2459 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
2460 * config/arc/arc.c (_arc_jli_section): New struct.
2461 (arc_jli_section): New type.
2462 (rc_jli_sections): New static variable.
2463 (arc_handle_jli_attribute): New function.
2464 (arc_attribute_table): Add jli_always and jli_fixed attribute.
2465 (arc_file_end): New function.
2466 (TARGET_ASM_FILE_END): Define.
2467 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
2468 (arc_add_jli_section): New function.
2469 (jli_call_scan): Likewise.
2470 (arc_reorg): Call jli_call_scan.
2471 (arc_output_addsi): Remove 'S' from printing asm operand.
2472 (arc_is_jli_call_p): New function.
2473 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
2475 (movhi_insn): Likewise.
2476 (movsi_insn): Likewise.
2477 (movsi_set_cc_insn): Likewise.
2478 (loadqi_update): Likewise.
2479 (load_zeroextendqisi_update): Likewise.
2480 (load_signextendqisi_update): Likewise.
2481 (loadhi_update): Likewise.
2482 (load_zeroextendhisi_update): Likewise.
2483 (load_signextendhisi_update): Likewise.
2484 (loadsi_update): Likewise.
2485 (loadsf_update): Likewise.
2486 (movsicc_insn): Likewise.
2487 (bset_insn): Likewise.
2488 (bxor_insn): Likewise.
2489 (bclr_insn): Likewise.
2490 (bmsk_insn): Likewise.
2491 (bicsi3_insn): Likewise.
2492 (cmpsi_cc_c_insn): Likewise.
2493 (movsi_ne): Likewise.
2494 (movsi_cond_exec): Likewise.
2495 (clrsbsi2): Likewise.
2506 (call_i): Remove 'S' asm letter, add jli instruction.
2507 (call_value_i): Likewise.
2508 * config/arc/arc.op (mjli-always): New option.
2509 * config/arc/constraints.md (Cji): New constraint.
2510 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
2512 (subsf3_fpx): Likewise.
2513 (mulsf3_fpx): Likewise.
2514 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
2516 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
2518 * doc/invoke.texi (ARC): Document mjli-always option.
2520 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
2522 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
2523 avoid addition with 0 and use incw and decw where possible.
2525 2018-01-26 Richard Biener <rguenther@suse.de>
2527 PR tree-optimization/81082
2528 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
2529 association if it requires casting to unsigned.
2530 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
2531 from fold_plusminus_mult_expr to catch important cases late when
2532 range info is available.
2534 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2536 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
2537 * configure.ac (hidden_linkonce): New test.
2538 * configure: Regenerate.
2539 * config.in: Regenerate.
2541 2018-01-26 Julia Koval <julia.koval@intel.com>
2543 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
2544 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
2545 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
2546 _mm_mask_bitshuffle_epi64_mask): Fix type.
2547 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
2548 USI_FTYPE_V4DI_V4DI_USI): Remove.
2549 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
2550 __builtin_ia32_vpshufbitqmb256_mask,
2551 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
2552 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
2553 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
2555 2018-01-26 Alan Modra <amodra@gmail.com>
2558 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
2559 UNSPEC_VBPERMQ. Sort other unspecs.
2561 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
2563 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
2565 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
2568 * predict.c (drop_profile): Do not push/pop cfun; update also
2570 (handle_missing_profiles): Fix logic looking for zero profiles.
2572 2018-01-25 Jakub Jelinek <jakub@redhat.com>
2575 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
2576 on functions with #pragma omp declare simd or functions with simd
2578 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
2579 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
2580 Remove trailing \n from warning_at calls.
2582 2018-01-25 Tom de Vries <tom@codesourcery.com>
2585 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2586 for neutered workers.
2588 2018-01-24 Joseph Myers <joseph@codesourcery.com>
2591 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
2592 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
2594 2018-01-24 Jeff Law <law@redhat.com>
2597 * i386.c (get_probe_interval): Move to earlier point.
2598 (ix86_compute_frame_layout): If -fstack-clash-protection and
2599 the frame is larger than the probe interval, then use pushes
2600 to save registers rather than reg->mem moves.
2601 (ix86_expand_prologue): Remove conditional for int_registers_saved
2604 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
2607 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
2608 min/max for never referenced object.
2610 2018-01-24 Jakub Jelinek <jakub@redhat.com>
2613 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
2615 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
2616 attributes from DECL_ATTRIBUTES (decl) without affecting
2617 DECL_ATTRIBUTES (current_function_decl).
2618 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
2619 functions with non-NULL DECL_ABSTRACT_ORIGIN.
2621 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
2623 PR tree-optimization/83979
2624 * fold-const.c (fold_comparison): Use constant_boolean_node
2625 instead of boolean_{true,false}_node.
2627 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
2629 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
2632 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2634 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2635 Simplify the clause that sets the length attribute.
2636 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2637 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
2638 clause that sets the length attribute.
2639 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2641 2018-01-24 Tom de Vries <tom@codesourcery.com>
2644 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
2645 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
2646 Add strict parameter.
2647 (prevent_branch_around_nothing): Insert dummy insn between branch to
2648 label and label with no ptx insn inbetween.
2649 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
2651 2018-01-24 Tom de Vries <tom@codesourcery.com>
2654 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2655 for neutered threads in warp.
2656 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
2658 2018-01-24 Richard Biener <rguenther@suse.de>
2660 PR tree-optimization/83176
2661 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
2664 2018-01-24 Richard Biener <rguenther@suse.de>
2666 PR tree-optimization/82819
2667 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
2668 code generating pluses that are no-ops in the target precision.
2670 2018-01-24 Richard Biener <rguenther@suse.de>
2673 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
2675 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2677 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
2678 to merge probabilities.
2679 * predict.c (probably_never_executed): Also mark as cold functions
2680 with global 0 profile and guessed local profile.
2681 * profile-count.c (profile_probability::combine_with_count): New
2683 * profile-count.h (profile_probability::operator*,
2684 profile_probability::operator*=, profile_probability::operator/,
2685 profile_probability::operator/=): Reduce precision to adjusted
2686 and set value to guessed on contradictory divisions.
2687 (profile_probability::combine_with_freq): Remove.
2688 (profile_probability::combine_wiht_count): Declare.
2689 (profile_count::force_nonzero):: Set to adjusted.
2690 (profile_count::probability_in):: Set quality to adjusted.
2691 * tree-ssa-tail-merge.c (replace_block_by): Use
2694 2018-01-23 Andrew Waterman <andrew@sifive.com>
2695 Jim Wilson <jimw@sifive.com>
2697 * config/riscv/riscv.c (riscv_stack_boundary): New.
2698 (riscv_option_override): Set riscv_stack_boundary. Handle
2699 riscv_preferred_stack_boundary_arg.
2700 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
2701 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
2702 (STACK_BOUNDARY): Set to riscv_stack_boundary.
2703 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
2704 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
2705 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
2707 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
2710 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
2711 of struct ix86_frame.
2712 (ix86_expand_epilogue): Likewise. Add a local variable for
2713 the reg_save_offset field in struct ix86_frame.
2715 2018-01-23 Bin Cheng <bin.cheng@arm.com>
2717 PR tree-optimization/82604
2718 * tree-loop-distribution.c (enum partition_kind): New enum item
2719 PKIND_PARTIAL_MEMSET.
2720 (partition_builtin_p): Support above new enum item.
2721 (generate_code_for_partition): Ditto.
2722 (compute_access_range): Differentiate cases that equality can be
2723 proven at all loops, the innermost loops or no loops.
2724 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
2725 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
2726 (finalize_partitions, distribute_loop): Don't fuse partition of
2727 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
2728 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
2731 2018-01-23 Martin Liska <mliska@suse.cz>
2733 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
2734 order to ignore the predictor.
2735 (PRED_POLYMORPHIC_CALL): Likewise.
2736 (PRED_RECURSIVE_CALL): Likewise.
2738 2018-01-23 Martin Liska <mliska@suse.cz>
2740 * tree-profile.c (tree_profiling): Print function header to
2741 aware reader which function we are working on.
2742 * value-prof.c (gimple_find_values_to_profile): Do not print
2743 not interesting value histograms.
2745 2018-01-23 Martin Liska <mliska@suse.cz>
2747 * profile-count.h (enum profile_quality): Add
2748 profile_uninitialized as the first value. Do not number values
2749 as they are zero based.
2750 (profile_count::verify): Update sanity check.
2751 (profile_probability::verify): Likewise.
2753 2018-01-23 Nathan Sidwell <nathan@acm.org>
2755 * doc/invoke.texi (ffor-scope): Deprecate.
2757 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2759 PR tree-optimization/83510
2760 * domwalk.c (set_all_edges_as_executable): New function.
2761 (dom_walker::dom_walker): Convert bool param
2762 "skip_unreachable_blocks" to enum reachability. Move setup of
2763 edge flags to set_all_edges_as_executable and only do it when
2764 reachability is REACHABLE_BLOCKS.
2765 * domwalk.h (enum dom_walker::reachability): New enum.
2766 (dom_walker::dom_walker): Convert bool param
2767 "skip_unreachable_blocks" to enum reachability.
2768 (set_all_edges_as_executable): New decl.
2769 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2770 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2772 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2773 but converting true to REACHABLE_BLOCKS.
2774 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2776 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2777 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2778 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2780 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2781 if check_all_array_refs will be called.
2783 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2785 * tree.c (selftest::test_location_wrappers): Add more test
2788 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2790 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2791 (selftest::test_bit_in_range): Likewise.
2793 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2796 * doc/sourcebuild.texi (vect_float): Say that the selector
2797 only describes the situation when -funsafe-math-optimizations is on.
2798 (vect_float_strict): Document.
2800 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2802 PR tree-optimization/83965
2803 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2804 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2805 instead of checking only for a reduction.
2806 (vect_recog_widen_sum_pattern): Likewise.
2808 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2810 * predict.c (probably_never_executed): Only use precise profile info.
2811 (compute_function_frequency): Skip after inlining hack since we now
2812 have quality checking.
2814 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2816 * profile-count.h (profile_probability::very_unlikely,
2817 profile_probability::unlikely, profile_probability::even): Set
2818 precision to guessed.
2820 2018-01-23 Richard Biener <rguenther@suse.de>
2822 PR tree-optimization/83963
2823 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2824 Properly terminate dominator walk when crossing the exit edge not
2825 when visiting its source block.
2827 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2830 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2831 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2833 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2835 PR tree-optimization/83957
2836 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2837 semicolon after for body surrounded by braces.
2839 PR tree-optimization/83081
2840 * profile-count.h (profile_probability::split): New method.
2841 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2842 Use profile_probability::split.
2843 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2844 when splitting a single conditional jump into 2.
2846 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2848 PR tree-optimization/69452
2849 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2852 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2854 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2855 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2856 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2858 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2860 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2861 * config/rl78/rl78.md (movdi): New define_expand.
2862 * config/rl78/rl78.c (rl78_split_movdi): New function.
2864 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2867 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2869 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2870 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2871 128-bit to produce an UNSPEC move to get the double word with the
2872 signbit and then a shift directly to do signbit.
2873 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2874 implementation with a new version that just does either a direct
2875 move or a regular move. Move memory interface to separate insns.
2876 Move insns so they are next to the expander.
2877 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2878 with signbit move. Split big and little endian case.
2879 (signbit<mode>2_dm_mem_le): Likewise.
2880 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2881 (signbit<mode>2_dm2): Likewise.
2883 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2885 * config/rl78/rl78.md (anddi3): New define_expand.
2887 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2889 * config/rl78/rl78.md (umindi3): New define_expand.
2891 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2893 * config/rl78/rl78.md (smindi3): New define_expand.
2895 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2897 * config/rl78/rl78.md (smaxdi3): New define_expand.
2899 2018-01-22 Carl Love <cel@us.ibm.com>
2901 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2902 LVX_V1TI): Add macro expansion.
2903 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2904 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2905 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2906 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2907 Change check to determine if the instruction is a byte reversing
2908 entry. Fix typo in comment.
2909 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2910 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2911 Add def_builtin calls for new builtins.
2912 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2913 Add define_insn expansion.
2915 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2917 * config/rl78/rl78.md (umaxdi3): New define_expand.
2919 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2921 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2922 for non-QImode registers.
2924 2018-01-22 Richard Biener <rguenther@suse.de>
2926 PR tree-optimization/83963
2927 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2928 including the loop exit block.
2929 (scop_detection::merge_sese): Likewise.
2930 (scop_detection::add_scop): Do it here instead.
2932 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2934 * doc/sourcebuild.texi (arm_softfloat): Document.
2936 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2939 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2940 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2941 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2943 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2944 David Edelsohn <dje.gcc@gmail.com>
2947 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2948 Change "crset eq" to "crset 2".
2949 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2950 (*call_indirect_aix<mode>_nospec): Likewise.
2951 (*call_value_indirect_aix<mode>_nospec): Likewise.
2952 (*call_indirect_elfv2<mode>_nospec): Likewise.
2953 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2954 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2955 change assembly output from . to $.
2956 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2957 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2958 (*tablejump<mode>_internal1_nospec): Likewise.
2960 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2963 * config/sh/sh_optimize_sett_clrt.cc:
2964 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2966 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2968 PR tree-optimization/83940
2969 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2970 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2971 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2972 use it to pass back the definition type.
2973 (vect_check_store_rhs): Likewise rhs_dt_out.
2974 (vect_build_gather_load_calls): Add a mask_dt argument and use
2975 it instead of a call to vect_is_simple_use.
2976 (vectorizable_store): Update calls to vect_check_load_store_mask
2977 and vect_check_store_rhs. Use the dt returned by the latter instead
2978 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2979 instead of calls to vect_is_simple_use. Pass the scalar rather
2980 than the vector operand to vect_is_simple_use when handling
2981 second and subsequent copies of an rhs value.
2982 (vectorizable_load): Update calls to vect_check_load_store_mask
2983 and vect_build_gather_load_calls. Use the cached mask_dt and
2984 gs_info.offset_dt instead of calls to vect_is_simple_use.
2986 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2989 * tree-emutls.c: Include gimplify.h.
2990 (lower_emutls_2): New function.
2991 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2992 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2993 it before further processing.
2996 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2997 UINTVAL (trueop1) instead of INTVAL (op1).
2999 2018-01-19 Jakub Jelinek <jakub@redhat.com>
3003 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
3004 INCOMING_FRAME_SP_OFFSET if not defined.
3005 (scan_trace): Add ENTRY argument. If true and
3006 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
3007 emit a note to adjust the CFA offset.
3008 (create_cfi_notes): Adjust scan_trace callers.
3009 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
3010 INCOMING_FRAME_SP_OFFSET in the CIE.
3011 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
3012 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
3014 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
3015 * doc/tm.texi: Regenerated.
3017 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
3019 PR rtl-optimization/83147
3020 * lra-constraints.c (remove_inheritance_pseudos): Use
3021 lra_substitute_pseudo_within_insn.
3023 2018-01-19 Tom de Vries <tom@codesourcery.com>
3024 Cesar Philippidis <cesar@codesourcery.com>
3027 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
3029 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
3032 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
3033 spaces for function labels.
3035 2018-01-19 Martin Liska <mliska@suse.cz>
3037 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
3038 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
3039 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
3040 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
3041 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
3042 (PRED_CONST_RETURN): Change from 69 to 65.
3043 (PRED_NULL_RETURN): Change from 91 to 71.
3044 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
3045 (PRED_LOOP_GUARD): Change from 66 to 73.
3047 2018-01-19 Martin Liska <mliska@suse.cz>
3049 * predict.c (predict_insn_def): Add new assert.
3050 (struct branch_predictor): Change type to signed integer.
3051 (test_prediction_value_range): Amend test to cover
3053 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
3054 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
3055 (PRED_LOOP_ITERATIONS_MAX): Likewise.
3056 (PRED_LOOP_IV_COMPARE): Likewise.
3057 * predict.h (PROB_UNINITIALIZED): Define new constant.
3059 2018-01-19 Martin Liska <mliska@suse.cz>
3061 * predict.c (dump_prediction): Add new format for
3062 analyze_brprob.py script which is enabled with -details
3064 * profile-count.h (precise_p): New function.
3066 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
3068 PR tree-optimization/83922
3069 * tree-vect-loop.c (vect_verify_full_masking): Return false if
3070 there are no statements that need masking.
3071 (vect_active_double_reduction_p): New function.
3072 (vect_analyze_loop_operations): Use it when handling phis that
3073 are not in the loop header.
3075 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
3077 PR tree-optimization/83914
3078 * tree-vect-loop.c (vectorizable_induction): Don't convert
3079 init_expr or apply the peeling adjustment for inductions
3080 that are nested within the vectorized loop.
3082 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3084 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
3087 2018-01-18 Jakub Jelinek <jakub@redhat.com>
3091 * function.h (gimplify_parameters): Add gimple_seq * argument.
3092 * function.c: Include gimple.h and options.h.
3093 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
3094 for the added local temporaries if needed.
3095 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
3096 if there are any parameter cleanups, wrap whole body into a
3097 try/finally with the cleanups.
3099 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
3102 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
3103 Use GET_MODE_CLASS for scalar floating point.
3105 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
3109 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
3110 Fix call of call_cgraph_insertion_hooks.
3112 2018-01-18 Martin Sebor <msebor@redhat.com>
3114 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
3116 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
3119 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
3122 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
3125 * common.opt: (-ffile-prefix-map): New option.
3126 * opts.c (common_handle_option): Defer it.
3127 * opts-global.c (handle_common_deferred_options): Handle it.
3128 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
3129 * file-prefix-map.h: New file.
3130 (remap_debug_filename, add_debug_prefix_map): ...here.
3131 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
3132 * final.c (debug_prefix_map, add_debug_prefix_map
3133 remap_debug_filename): Move to...
3134 * file-prefix-map.c: New file.
3135 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
3136 generalize, get rid of alloca(), use strrchr() instead of strchr().
3137 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
3138 Implement in terms of add_prefix_map().
3139 (remap_macro_filename, remap_debug_filename): Implement in term of
3141 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
3142 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
3143 * dbxout.c: Include file-prefix-map.h.
3144 * varasm.c: Likewise.
3145 * vmsdbgout.c: Likewise.
3146 * xcoffout.c: Likewise.
3147 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
3148 * doc/cppopts.texi (-fmacro-prefix-map): Document.
3149 * doc/invoke.texi (-ffile-prefix-map): Document.
3150 (-fdebug-prefix-map): Update description.
3152 2018-01-18 Martin Liska <mliska@suse.cz>
3154 * config/i386/i386.c (indirect_thunk_name): Document that also
3156 (output_indirect_thunk): Document why both instructions
3157 (pause and lfence) are generated.
3159 2018-01-18 Richard Biener <rguenther@suse.de>
3161 PR tree-optimization/83887
3162 * graphite-scop-detection.c
3163 (scop_detection::get_nearest_dom_with_single_entry): Remove.
3164 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
3165 (scop_detection::merge_sese): Re-implement with a flood-fill
3166 algorithm that properly finds a SESE region if it exists.
3168 2018-01-18 Jakub Jelinek <jakub@redhat.com>
3171 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
3172 pointer_diff optimizations use view_convert instead of convert.
3174 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3176 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
3177 Generate different code for -mno-speculate-indirect-jumps.
3178 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
3179 (*call_indirect_aix<mode>): Disable for
3180 -mno-speculate-indirect-jumps.
3181 (*call_indirect_aix<mode>_nospec): New define_insn.
3182 (*call_value_indirect_aix<mode>): Disable for
3183 -mno-speculate-indirect-jumps.
3184 (*call_value_indirect_aix<mode>_nospec): New define_insn.
3185 (*sibcall_nonlocal_sysv<mode>): Generate different code for
3186 -mno-speculate-indirect-jumps.
3187 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
3189 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
3191 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
3192 long double type, set the flags for noting the default long double
3193 type, even if we don't pass or return a long double type.
3195 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
3198 * ipa-inline.c (flatten_function): Do not overwrite final inlining
3201 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
3203 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
3204 support for merge[hl].
3205 (fold_mergehl_helper): New helper function.
3206 (tree-vector-builder.h): New #include for tree_vector_builder usage.
3207 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
3208 (altivec_vmrglw_direct): Add xxmrglw insn.
3210 2018-01-17 Andrew Waterman <andrew@sifive.com>
3212 * config/riscv/riscv.c (riscv_conditional_register_usage): If
3213 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
3215 2018-01-17 David Malcolm <dmalcolm@redhat.com>
3218 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
3219 call the lto_location_cache before reading the
3220 DECL_SOURCE_LOCATION of the types.
3222 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
3223 Richard Sandiford <richard.sandiford@linaro.org>
3225 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
3226 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
3227 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
3228 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
3229 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
3231 * config/aarch64/constraints.md (aarch64_movti_operand):
3233 * config/aarch64/predicates.md (Uti): Add new constraint.
3235 2018-01-17 Carl Love <cel@us.ibm.com>
3237 * config/rs6000/vsx.md (define_expand xl_len_r,
3238 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
3239 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
3241 (define_expand, define_insn): Move the shift left from the
3242 define_insn to the define_expand for lxvl and stxvl instructions.
3243 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
3244 and XL_LEN_R definitions to PURE.
3246 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
3248 * config/i386/i386.c (indirect_thunk_name): Declare regno
3249 as unsigned int. Compare regno with INVALID_REGNUM.
3250 (output_indirect_thunk): Ditto.
3251 (output_indirect_thunk_function): Ditto.
3252 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
3253 in the call to output_indirect_thunk_function.
3255 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
3258 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
3259 rather than the size of inner_type to determine the stack slot size
3260 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
3262 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
3265 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
3268 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
3270 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
3271 endian Linux systems to optionally enable multilibs for selecting
3272 the long double type if the user configured an explicit type.
3273 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
3274 have no long double multilibs if not defined.
3275 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
3276 warn if the user used -mabi={ieee,ibm}longdouble and we built
3277 multilibs for long double.
3278 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
3279 appropriate multilib option.
3280 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
3282 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
3283 for building long double multilibs.
3284 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
3286 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
3288 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
3291 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
3293 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
3296 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
3299 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
3302 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3304 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
3305 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
3307 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
3309 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
3310 different rtl trees depending on TARGET_64BIT.
3311 (rs6000_gen_lvx): Likewise.
3313 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
3315 * config/visium/visium.md (nop): Tweak comment.
3316 (hazard_nop): Likewise.
3318 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3320 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
3321 -mspeculate-indirect-jumps.
3322 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
3323 for -mno-speculate-indirect-jumps.
3324 (*call_indirect_elfv2<mode>_nospec): New define_insn.
3325 (*call_value_indirect_elfv2<mode>): Disable for
3326 -mno-speculate-indirect-jumps.
3327 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
3328 (indirect_jump): Emit different RTL for
3329 -mno-speculate-indirect-jumps.
3330 (*indirect_jump<mode>): Disable for
3331 -mno-speculate-indirect-jumps.
3332 (*indirect_jump<mode>_nospec): New define_insn.
3333 (tablejump): Emit different RTL for
3334 -mno-speculate-indirect-jumps.
3335 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
3336 (tablejumpsi_nospec): New define_expand.
3337 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
3338 (tablejumpdi_nospec): New define_expand.
3339 (*tablejump<mode>_internal1): Disable for
3340 -mno-speculate-indirect-jumps.
3341 (*tablejump<mode>_internal1_nospec): New define_insn.
3342 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
3345 2018-01-16 Artyom Skrobov tyomitch@gmail.com
3347 * caller-save.c (insert_save): Drop unnecessary parameter. All
3350 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3351 Richard Biener <rguenth@suse.de>
3354 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
3355 return early, inline manually is_gimple_sizepos. Make sure if we
3356 call gimplify_expr we don't end up with a gimple constant.
3357 * tree.c (variably_modified_type_p): Don't return true for
3358 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
3359 * gimplify.h (is_gimple_sizepos): Remove.
3361 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3363 PR tree-optimization/83857
3364 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
3365 vectorizable_live_operation for pure SLP statements.
3366 (vectorizable_live_operation): Handle PHIs.
3368 2018-01-16 Richard Biener <rguenther@suse.de>
3370 PR tree-optimization/83867
3371 * tree-vect-stmts.c (vect_transform_stmt): Precompute
3372 nested_in_vect_loop_p since the scalar stmt may get invalidated.
3374 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3377 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
3378 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
3379 If off is not INTEGER_CST, issue a may not be aligned warning
3380 rather than isn't aligned. Use isn%'t rather than isn't.
3381 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
3383 <case MULT_EXPR>: Improve the case when bottom and one of the
3384 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
3385 operand, in that case check if the other operand is multiple of
3386 bottom divided by the INTEGER_CST operand.
3388 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3391 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
3392 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
3393 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
3394 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
3395 * config/pa/pa.c (pa_function_arg_advance): Likewise.
3396 (pa_function_arg, pa_arg_partial_bytes): Likewise.
3397 (pa_function_arg_size): New function.
3399 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3401 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
3402 in a separate statement.
3404 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
3406 PR tree-optimization/83847
3407 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
3408 group gathers and scatters.
3410 2018-01-16 Jakub Jelinek <jakub@redhat.com>
3412 PR rtl-optimization/86620
3413 * params.def (max-sched-ready-insns): Bump minimum value to 1.
3415 PR rtl-optimization/83213
3416 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
3417 to last if both are JUMP_INSNs.
3419 PR tree-optimization/83843
3420 * gimple-ssa-store-merging.c
3421 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
3422 store_immediate_info for bswap/nop orig_stores.
3424 2018-01-15 Andrew Waterman <andrew@sifive.com>
3426 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
3428 <UDIV>: Increase cost if !TARGET_DIV.
3430 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
3432 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
3433 (define_attr "cr_logical_3op"): New.
3434 (cceq_ior_compare): Adjust.
3435 (cceq_ior_compare_complement): Adjust.
3436 (*cceq_rev_compare): Adjust.
3437 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
3438 (is_cracked_insn): Adjust.
3439 (insn_must_be_first_in_group): Adjust.
3440 * config/rs6000/40x.md: Adjust.
3441 * config/rs6000/440.md: Adjust.
3442 * config/rs6000/476.md: Adjust.
3443 * config/rs6000/601.md: Adjust.
3444 * config/rs6000/603.md: Adjust.
3445 * config/rs6000/6xx.md: Adjust.
3446 * config/rs6000/7450.md: Adjust.
3447 * config/rs6000/7xx.md: Adjust.
3448 * config/rs6000/8540.md: Adjust.
3449 * config/rs6000/cell.md: Adjust.
3450 * config/rs6000/e300c2c3.md: Adjust.
3451 * config/rs6000/e500mc.md: Adjust.
3452 * config/rs6000/e500mc64.md: Adjust.
3453 * config/rs6000/e5500.md: Adjust.
3454 * config/rs6000/e6500.md: Adjust.
3455 * config/rs6000/mpc.md: Adjust.
3456 * config/rs6000/power4.md: Adjust.
3457 * config/rs6000/power5.md: Adjust.
3458 * config/rs6000/power6.md: Adjust.
3459 * config/rs6000/power7.md: Adjust.
3460 * config/rs6000/power8.md: Adjust.
3461 * config/rs6000/power9.md: Adjust.
3462 * config/rs6000/rs64.md: Adjust.
3463 * config/rs6000/titan.md: Adjust.
3465 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3467 * config/i386/predicates.md (indirect_branch_operand): Rewrite
3468 ix86_indirect_branch_register logic.
3470 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3472 * config/i386/constraints.md (Bs): Update
3473 ix86_indirect_branch_register check. Don't check
3474 ix86_indirect_branch_register with GOT_memory_operand.
3476 * config/i386/predicates.md (GOT_memory_operand): Don't check
3477 ix86_indirect_branch_register here.
3478 (GOT32_symbol_operand): Likewise.
3480 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3482 * config/i386/predicates.md (constant_call_address_operand):
3483 Rewrite ix86_indirect_branch_register logic.
3484 (sibcall_insn_operand): Likewise.
3486 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3488 * config/i386/constraints.md (Bs): Replace
3489 ix86_indirect_branch_thunk_register with
3490 ix86_indirect_branch_register.
3492 * config/i386/i386.md (indirect_jump): Likewise.
3493 (tablejump): Likewise.
3494 (*sibcall_memory): Likewise.
3495 (*sibcall_value_memory): Likewise.
3496 Peepholes of indirect call and jump via memory: Likewise.
3497 * config/i386/i386.opt: Likewise.
3498 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3499 (GOT_memory_operand): Likewise.
3500 (call_insn_operand): Likewise.
3501 (sibcall_insn_operand): Likewise.
3502 (GOT32_symbol_operand): Likewise.
3504 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3507 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
3508 type rather than type addr's type points to.
3509 (expand_omp_atomic_mutex): Likewise.
3510 (expand_omp_atomic): Likewise.
3512 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3515 * config/i386/i386.c (output_indirect_thunk_function): Use
3516 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
3517 for __x86_return_thunk.
3519 2018-01-15 Richard Biener <rguenther@suse.de>
3522 * expmed.c (extract_bit_field_1): Fix typo.
3524 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3527 * config/arm/iterators.md (VF): New mode iterator.
3528 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
3529 Remove integer-related logic from pattern.
3530 (neon_vabd<mode>_3): Likewise.
3532 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3535 * common.opt (fstrict-overflow): No longer an alias.
3536 (fwrapv-pointer): New option.
3537 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
3538 also for pointer types based on flag_wrapv_pointer.
3539 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
3540 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
3541 opts->x_flag_wrapv got set.
3542 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
3543 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
3544 POINTER_TYPE_OVERFLOW_UNDEFINED.
3545 * match.pd: Likewise in address comparison pattern.
3546 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
3548 2018-01-15 Richard Biener <rguenther@suse.de>
3551 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
3552 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
3553 Reset type names to their identifier if their TYPE_DECL doesn't
3554 have linkage (and thus is used for ODR and devirt).
3555 (save_debug_info_for_decl): Remove.
3556 (save_debug_info_for_type): Likewise.
3557 (add_tree_to_fld_list): Adjust.
3558 * tree-pretty-print.c (dump_generic_node): Make dumping of
3559 type names more robust.
3561 2018-01-15 Richard Biener <rguenther@suse.de>
3563 * BASE-VER: Bump to 8.0.1.
3565 2018-01-14 Martin Sebor <msebor@redhat.com>
3568 * builtins.c (check_access): Avoid warning when the no-warning bit
3571 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
3573 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
3574 * ira-color (allocno_hard_regs_compare): Likewise.
3576 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
3579 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
3580 Use .pushsection/.popsection.
3582 2018-01-14 Martin Sebor <msebor@redhat.com>
3585 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
3587 2018-01-14 Jakub Jelinek <jakub@redhat.com>
3589 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
3590 entry from extra_headers.
3591 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
3592 extra_headers, make the list bitwise identical to the i?86-*-* one.
3594 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3596 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
3597 -mcmodel=large with -mindirect-branch=thunk,
3598 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
3599 -mfunction-return=thunk-extern.
3600 * doc/invoke.texi: Document -mcmodel=large is incompatible with
3601 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
3602 -mfunction-return=thunk and -mfunction-return=thunk-extern.
3604 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3606 * config/i386/i386.c (print_reg): Print the name of the full
3607 integer register without '%'.
3608 (ix86_print_operand): Handle 'V'.
3609 * doc/extend.texi: Document 'V' modifier.
3611 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3613 * config/i386/constraints.md (Bs): Disallow memory operand for
3614 -mindirect-branch-register.
3616 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3617 (GOT_memory_operand): Likewise.
3618 (call_insn_operand): Likewise.
3619 (sibcall_insn_operand): Likewise.
3620 (GOT32_symbol_operand): Likewise.
3621 * config/i386/i386.md (indirect_jump): Call convert_memory_address
3622 for -mindirect-branch-register.
3623 (tablejump): Likewise.
3624 (*sibcall_memory): Likewise.
3625 (*sibcall_value_memory): Likewise.
3626 Disallow peepholes of indirect call and jump via memory for
3627 -mindirect-branch-register.
3628 (*call_pop): Replace m with Bw.
3629 (*call_value_pop): Likewise.
3630 (*sibcall_pop_memory): Replace m with Bs.
3631 * config/i386/i386.opt (mindirect-branch-register): New option.
3632 * doc/invoke.texi: Document -mindirect-branch-register option.
3634 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3636 * config/i386/i386-protos.h (ix86_output_function_return): New.
3637 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
3638 set function_return_type.
3639 (indirect_thunk_name): Add ret_p to indicate thunk for function
3641 (output_indirect_thunk_function): Pass false to
3642 indirect_thunk_name.
3643 (ix86_output_indirect_branch_via_reg): Likewise.
3644 (ix86_output_indirect_branch_via_push): Likewise.
3645 (output_indirect_thunk_function): Create alias for function
3646 return thunk if regno < 0.
3647 (ix86_output_function_return): New function.
3648 (ix86_handle_fndecl_attribute): Handle function_return.
3649 (ix86_attribute_table): Add function_return.
3650 * config/i386/i386.h (machine_function): Add
3651 function_return_type.
3652 * config/i386/i386.md (simple_return_internal): Use
3653 ix86_output_function_return.
3654 (simple_return_internal_long): Likewise.
3655 * config/i386/i386.opt (mfunction-return=): New option.
3656 (indirect_branch): Mention -mfunction-return=.
3657 * doc/extend.texi: Document function_return function attribute.
3658 * doc/invoke.texi: Document -mfunction-return= option.
3660 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3662 * config/i386/i386-opts.h (indirect_branch): New.
3663 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
3664 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
3665 with local indirect jump when converting indirect call and jump.
3666 (ix86_set_indirect_branch_type): New.
3667 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
3668 (indirectlabelno): New.
3669 (indirect_thunk_needed): Likewise.
3670 (indirect_thunk_bnd_needed): Likewise.
3671 (indirect_thunks_used): Likewise.
3672 (indirect_thunks_bnd_used): Likewise.
3673 (INDIRECT_LABEL): Likewise.
3674 (indirect_thunk_name): Likewise.
3675 (output_indirect_thunk): Likewise.
3676 (output_indirect_thunk_function): Likewise.
3677 (ix86_output_indirect_branch_via_reg): Likewise.
3678 (ix86_output_indirect_branch_via_push): Likewise.
3679 (ix86_output_indirect_branch): Likewise.
3680 (ix86_output_indirect_jmp): Likewise.
3681 (ix86_code_end): Call output_indirect_thunk_function if needed.
3682 (ix86_output_call_insn): Call ix86_output_indirect_branch if
3684 (ix86_handle_fndecl_attribute): Handle indirect_branch.
3685 (ix86_attribute_table): Add indirect_branch.
3686 * config/i386/i386.h (machine_function): Add indirect_branch_type
3687 and has_local_indirect_jump.
3688 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
3690 (tablejump): Likewise.
3691 (*indirect_jump): Use ix86_output_indirect_jmp.
3692 (*tablejump_1): Likewise.
3693 (simple_return_indirect_internal): Likewise.
3694 * config/i386/i386.opt (mindirect-branch=): New option.
3695 (indirect_branch): New.
3698 (thunk-inline): Likewise.
3699 (thunk-extern): Likewise.
3700 * doc/extend.texi: Document indirect_branch function attribute.
3701 * doc/invoke.texi: Document -mindirect-branch= option.
3703 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
3706 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
3708 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
3710 * ipa-inline.c (want_inline_small_function_p): Return false if
3711 inlining has already failed with CIF_FINAL_ERROR.
3712 (update_caller_keys): Call want_inline_small_function_p before
3714 (update_callee_keys): Likewise.
3716 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
3718 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
3720 (rs6000_quadword_masked_address_p): Likewise.
3721 (quad_aligned_load_p): Likewise.
3722 (quad_aligned_store_p): Likewise.
3723 (const_load_sequence_p): Add comment to describe the outer-most loop.
3724 (mimic_memory_attributes_and_flags): New function.
3725 (rs6000_gen_stvx): Likewise.
3726 (replace_swapped_aligned_store): Likewise.
3727 (rs6000_gen_lvx): Likewise.
3728 (replace_swapped_aligned_load): Likewise.
3729 (replace_swapped_load_constant): Capitalize argument name in
3730 comment describing this function.
3731 (rs6000_analyze_swaps): Add a third pass to search for vector loads
3732 and stores that access quad-word aligned addresses and replace
3733 with stvx or lvx instructions when appropriate.
3734 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
3735 New function prototype.
3736 (rs6000_quadword_masked_address_p): Likewise.
3737 (rs6000_gen_lvx): Likewise.
3738 (rs6000_gen_stvx): Likewise.
3739 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3740 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3741 when memory address is aligned.
3742 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3743 this split to select lvx instruction when memory address is aligned.
3744 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3745 instruction when memory address is aligned.
3746 (*vsx_le_perm_load_v16qi): Likewise.
3747 (four unnamed splitters): Modify to select the stvx instruction
3748 when memory is aligned.
3750 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3752 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3753 which appears in the queue multiple times.
3755 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3756 Alan Hayward <alan.hayward@arm.com>
3757 David Sherwood <david.sherwood@arm.com>
3759 * tree-vectorizer.h (vec_lower_bound): New structure.
3760 (_loop_vec_info): Add check_nonzero and lower_bounds.
3761 (LOOP_VINFO_CHECK_NONZERO): New macro.
3762 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3763 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3764 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3765 fields. Make seg_len the distance travelled, not including the
3767 (dr_direction_indicator): Declare.
3768 (dr_zero_step_indicator): Likewise.
3769 (dr_known_forward_stride_p): Likewise.
3770 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3772 (runtime_alias_check_p): Allow runtime alias checks with
3774 (operator ==): Compare access_size and align.
3775 (prune_runtime_alias_test_list): Rework for new distinction between
3776 the access_size and seg_len.
3777 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3779 (get_segment_min_max): New function.
3780 (create_intersect_range_checks): Use it.
3781 (dr_step_indicator): New function.
3782 (dr_direction_indicator): Likewise.
3783 (dr_zero_step_indicator): Likewise.
3784 (dr_known_forward_stride_p): Likewise.
3785 * tree-loop-distribution.c (data_ref_segment_size): Return
3786 DR_STEP * (niters - 1).
3787 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3789 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3790 (vect_preserves_scalar_order_p): New function, split out from...
3791 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3792 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3793 (vect_vfa_access_size): New function.
3794 (vect_vfa_align): Likewise.
3795 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3796 (dump_lower_bound): New function.
3797 (vect_check_lower_bound): Likewise.
3798 (vect_small_gap_p): Likewise.
3799 (vectorizable_with_step_bound_p): Likewise.
3800 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3801 depencies if the vectorization factor is 1. Convert the checks
3802 for nonzero steps into checks on the bounds of DR_STEP. Try using
3803 a bunds check for variable steps if the minimum required step is
3804 relatively small. Update calls to the dr_with_seg_len
3805 constructor and to vect_compile_time_alias.
3806 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3808 (vect_loop_versioning): Call it.
3809 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3811 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3813 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3814 Alan Hayward <alan.hayward@arm.com>
3815 David Sherwood <david.sherwood@arm.com>
3817 * doc/sourcebuild.texi (vect_scatter_store): Document.
3818 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3820 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3822 * genopinit.c (main): Add supports_vec_scatter_store and
3823 supports_vec_scatter_store_cached to target_optabs.
3824 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3825 IFN_MASK_SCATTER_STORE.
3826 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3828 * internal-fn.h (internal_store_fn_p): Declare.
3829 (internal_fn_stored_value_index): Likewise.
3830 * internal-fn.c (scatter_store_direct): New macro.
3831 (expand_scatter_store_optab_fn): New function.
3832 (direct_scatter_store_optab_supported_p): New macro.
3833 (internal_store_fn_p): New function.
3834 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3835 IFN_MASK_SCATTER_STORE.
3836 (internal_fn_mask_index): Likewise.
3837 (internal_fn_stored_value_index): New function.
3838 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3840 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3841 * optabs-query.c (supports_vec_scatter_store_p): New function.
3842 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3843 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3844 true for scatter stores.
3845 (vect_gather_scatter_fn_p): Handle scatter stores too.
3846 (vect_check_gather_scatter): Consider using scatter stores if
3847 supports_vec_scatter_store_p.
3848 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3850 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3851 internal_fn_stored_value_index.
3852 (check_load_store_masking): Handle scatter stores too.
3853 (vect_get_store_rhs): Make public.
3854 (vectorizable_call): Use internal_store_fn_p.
3855 (vectorizable_store): Handle scatter store internal functions.
3856 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3857 when deciding whether the end of the group has been reached.
3858 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3859 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3860 (mask_scatter_store<mode>): New insns.
3862 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3863 Alan Hayward <alan.hayward@arm.com>
3864 David Sherwood <david.sherwood@arm.com>
3866 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3867 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3868 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3870 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3871 Use vect_truncate_gather_scatter_offset if we can't treat the
3872 operation as a normal gather load or scatter store.
3873 (get_group_load_store_type): Take the gather_scatter_info
3874 as argument. Try using a gather load or scatter store for
3875 single-element groups.
3876 (get_load_store_type): Update calls to get_group_load_store_type
3877 and vect_use_strided_gather_scatters_p.
3879 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3880 Alan Hayward <alan.hayward@arm.com>
3881 David Sherwood <david.sherwood@arm.com>
3883 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3884 optional tree argument.
3885 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3887 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3888 but continue to use the current value as a fallback.
3889 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3890 to compare the updates.
3891 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3892 (get_load_store_type): Use it when handling a strided access.
3893 (vect_get_strided_load_store_ops): New function.
3894 (vect_get_data_ptr_increment): Likewise.
3895 (vectorizable_load): Handle strided gather loads. Always pass
3896 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3898 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3899 Alan Hayward <alan.hayward@arm.com>
3900 David Sherwood <david.sherwood@arm.com>
3902 * doc/md.texi (gather_load@var{m}): Document.
3903 (mask_gather_load@var{m}): Likewise.
3904 * genopinit.c (main): Add supports_vec_gather_load and
3905 supports_vec_gather_load_cached to target_optabs.
3906 * optabs-tree.c (init_tree_optimization_optabs): Use
3907 ggc_cleared_alloc to allocate target_optabs.
3908 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3909 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3911 * internal-fn.h (internal_load_fn_p): Declare.
3912 (internal_gather_scatter_fn_p): Likewise.
3913 (internal_fn_mask_index): Likewise.
3914 (internal_gather_scatter_fn_supported_p): Likewise.
3915 * internal-fn.c (gather_load_direct): New macro.
3916 (expand_gather_load_optab_fn): New function.
3917 (direct_gather_load_optab_supported_p): New macro.
3918 (direct_internal_fn_optab): New function.
3919 (internal_load_fn_p): Likewise.
3920 (internal_gather_scatter_fn_p): Likewise.
3921 (internal_fn_mask_index): Likewise.
3922 (internal_gather_scatter_fn_supported_p): Likewise.
3923 * optabs-query.c (supports_at_least_one_mode_p): New function.
3924 (supports_vec_gather_load_p): Likewise.
3925 * optabs-query.h (supports_vec_gather_load_p): Declare.
3926 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3927 and memory_type field.
3928 (NUM_PATTERNS): Bump to 15.
3929 * tree-vect-data-refs.c: Include internal-fn.h.
3930 (vect_gather_scatter_fn_p): New function.
3931 (vect_describe_gather_scatter_call): Likewise.
3932 (vect_check_gather_scatter): Try using internal functions for
3933 gather loads. Recognize existing calls to a gather load function.
3934 (vect_analyze_data_refs): Consider using gather loads if
3935 supports_vec_gather_load_p.
3936 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3937 (vect_get_gather_scatter_offset_type): Likewise.
3938 (vect_convert_mask_for_vectype): Likewise.
3939 (vect_add_conversion_to_patterm): Likewise.
3940 (vect_try_gather_scatter_pattern): Likewise.
3941 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3942 (vect_vect_recog_func_ptrs): Add it.
3943 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3944 internal_fn_mask_index and internal_gather_scatter_fn_p.
3945 (check_load_store_masking): Take the gather_scatter_info as an
3946 argument and handle gather loads.
3947 (vect_get_gather_scatter_ops): New function.
3948 (vectorizable_call): Check internal_load_fn_p.
3949 (vectorizable_load): Likewise. Handle gather load internal
3951 (vectorizable_store): Update call to check_load_store_masking.
3952 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3953 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3954 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3955 (aarch64_gather_scale_operand_d): New predicates.
3956 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3957 (mask_gather_load<mode>): New insns.
3959 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3960 Alan Hayward <alan.hayward@arm.com>
3961 David Sherwood <david.sherwood@arm.com>
3963 * optabs.def (fold_left_plus_optab): New optab.
3964 * doc/md.texi (fold_left_plus_@var{m}): Document.
3965 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3966 * internal-fn.c (fold_left_direct): Define.
3967 (expand_fold_left_optab_fn): Likewise.
3968 (direct_fold_left_optab_supported_p): Likewise.
3969 * fold-const-call.c (fold_const_fold_left): New function.
3970 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3971 * tree-parloops.c (valid_reduction_p): New function.
3972 (gather_scalar_reductions): Use it.
3973 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3974 (vect_finish_replace_stmt): Declare.
3975 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3976 (needs_fold_left_reduction_p): New function, split out from...
3977 (vect_is_simple_reduction): ...here. Accept reductions that
3978 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3979 (vect_force_simple_reduction): Also store the reduction type in
3980 the assignment's STMT_VINFO_REDUC_TYPE.
3981 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3982 (merge_with_identity): New function.
3983 (vect_expand_fold_left): Likewise.
3984 (vectorize_fold_left_reduction): Likewise.
3985 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3986 scalar phi in place for it. Check for target support and reject
3987 cases that would reassociate the operation. Defer the transform
3988 phase to vectorize_fold_left_reduction.
3989 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3990 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3991 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3993 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3995 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3996 call to ifc_temp_var.
3998 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3999 Alan Hayward <alan.hayward@arm.com>
4000 David Sherwood <david.sherwood@arm.com>
4002 * target.def (legitimize_address_displacement): Take the original
4003 offset as a poly_int.
4004 * targhooks.h (default_legitimize_address_displacement): Update
4006 * targhooks.c (default_legitimize_address_displacement): Likewise.
4007 * doc/tm.texi: Regenerate.
4008 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
4009 as an argument, moving assert of ad->disp == ad->disp_term to...
4010 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
4011 Try calling targetm.legitimize_address_displacement before expanding
4012 the address rather than afterwards, and adjust for the new interface.
4013 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
4014 Match the new hook interface. Handle SVE addresses.
4015 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
4018 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4020 * Makefile.in (OBJS): Add early-remat.o.
4021 * target.def (select_early_remat_modes): New hook.
4022 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
4023 * doc/tm.texi: Regenerate.
4024 * targhooks.h (default_select_early_remat_modes): Declare.
4025 * targhooks.c (default_select_early_remat_modes): New function.
4026 * timevar.def (TV_EARLY_REMAT): New timevar.
4027 * passes.def (pass_early_remat): New pass.
4028 * tree-pass.h (make_pass_early_remat): Declare.
4029 * early-remat.c: New file.
4030 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
4032 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
4034 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4035 Alan Hayward <alan.hayward@arm.com>
4036 David Sherwood <david.sherwood@arm.com>
4038 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
4039 vfm1 with a bound_epilog parameter.
4040 (vect_do_peeling): Update calls accordingly, and move the prologue
4041 call earlier in the function. Treat the base bound_epilog as 0 for
4042 fully-masked loops and retain vf - 1 for other loops. Add 1 to
4043 this base when peeling for gaps.
4044 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
4045 with fully-masked loops.
4046 (vect_estimate_min_profitable_iters): Handle the single peeled
4047 iteration in that case.
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-vect-data-refs.c (vect_analyze_group_access_1): Allow
4054 single-element interleaving even if the size is not a power of 2.
4055 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
4056 accesses for single-element interleaving if the group size is
4059 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4060 Alan Hayward <alan.hayward@arm.com>
4061 David Sherwood <david.sherwood@arm.com>
4063 * doc/md.texi (fold_extract_last_@var{m}): Document.
4064 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
4065 * optabs.def (fold_extract_last_optab): New optab.
4066 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
4067 * internal-fn.c (fold_extract_direct): New macro.
4068 (expand_fold_extract_optab_fn): Likewise.
4069 (direct_fold_extract_optab_supported_p): Likewise.
4070 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
4071 * tree-vect-loop.c (vect_model_reduction_cost): Handle
4072 EXTRACT_LAST_REDUCTION.
4073 (get_initial_def_for_reduction): Do not create an initial vector
4074 for EXTRACT_LAST_REDUCTION reductions.
4075 (vectorizable_reduction): Leave the scalar phi in place for
4076 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
4077 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
4078 epilogue code for EXTRACT_LAST_REDUCTION and defer the
4079 transform phase to vectorizable_condition.
4080 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
4082 (vect_finish_stmt_generation): ...here.
4083 (vect_finish_replace_stmt): New function.
4084 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
4085 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
4087 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
4089 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4090 Alan Hayward <alan.hayward@arm.com>
4091 David Sherwood <david.sherwood@arm.com>
4093 * doc/md.texi (extract_last_@var{m}): Document.
4094 * optabs.def (extract_last_optab): New optab.
4095 * internal-fn.def (EXTRACT_LAST): New internal function.
4096 * internal-fn.c (cond_unary_direct): New macro.
4097 (expand_cond_unary_optab_fn): Likewise.
4098 (direct_cond_unary_optab_supported_p): Likewise.
4099 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
4100 loops using EXTRACT_LAST.
4101 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
4102 (extract_last_<mode>): ...this optab.
4103 (vec_extract<mode><Vel>): Update accordingly.
4105 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4106 Alan Hayward <alan.hayward@arm.com>
4107 David Sherwood <david.sherwood@arm.com>
4109 * target.def (empty_mask_is_expensive): New hook.
4110 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
4111 * doc/tm.texi: Regenerate.
4112 * targhooks.h (default_empty_mask_is_expensive): Declare.
4113 * targhooks.c (default_empty_mask_is_expensive): New function.
4114 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
4115 if the target says that empty masks are expensive.
4116 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
4118 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
4120 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4121 Alan Hayward <alan.hayward@arm.com>
4122 David Sherwood <david.sherwood@arm.com>
4124 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
4125 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
4126 (vect_use_loop_mask_for_alignment_p): New function.
4127 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
4128 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
4129 niters_skip argument. Make sure that the first niters_skip elements
4130 of the first iteration are inactive.
4131 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
4132 Update call to vect_set_loop_masks_directly.
4133 (get_misalign_in_elems): New function, split out from...
4134 (vect_gen_prolog_loop_niters): ...here.
4135 (vect_update_init_of_dr): Take a code argument that specifies whether
4136 the adjustment should be added or subtracted.
4137 (vect_update_init_of_drs): Likewise.
4138 (vect_prepare_for_masked_peels): New function.
4139 (vect_do_peeling): Skip prologue peeling if we're using a mask
4140 instead. Update call to vect_update_inits_of_drs.
4141 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4143 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
4144 alignment. Do not include the number of peeled iterations in
4145 the minimum threshold in that case.
4146 (vectorizable_induction): Adjust the start value down by
4147 LOOP_VINFO_MASK_SKIP_NITERS iterations.
4148 (vect_transform_loop): Call vect_prepare_for_masked_peels.
4149 Take the number of skipped iterations into account when calculating
4151 * tree-vect-stmts.c (vect_gen_while_not): New function.
4153 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4154 Alan Hayward <alan.hayward@arm.com>
4155 David Sherwood <david.sherwood@arm.com>
4157 * doc/sourcebuild.texi (vect_fully_masked): Document.
4158 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
4160 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
4162 (vect_analyze_loop_2): ...here. Don't check the vectorization
4163 factor against the number of loop iterations if the loop is
4166 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4167 Alan Hayward <alan.hayward@arm.com>
4168 David Sherwood <david.sherwood@arm.com>
4170 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
4171 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
4172 (dump_groups): Update accordingly.
4173 (iv_use::mem_type): New member variable.
4174 (address_p): New function.
4175 (record_use): Add a mem_type argument and initialize the new
4177 (record_group_use): Add a mem_type argument. Use address_p.
4178 Remove obsolete null checks of base_object. Update call to record_use.
4179 (find_interesting_uses_op): Update call to record_group_use.
4180 (find_interesting_uses_cond): Likewise.
4181 (find_interesting_uses_address): Likewise.
4182 (get_mem_type_for_internal_fn): New function.
4183 (find_address_like_use): Likewise.
4184 (find_interesting_uses_stmt): Try find_address_like_use before
4185 calling find_interesting_uses_op.
4186 (addr_offset_valid_p): Use the iv mem_type field as the type
4187 of the addressed memory.
4188 (add_autoinc_candidates): Likewise.
4189 (get_address_cost): Likewise.
4190 (split_small_address_groups_p): Use address_p.
4191 (split_address_groups): Likewise.
4192 (add_iv_candidate_for_use): Likewise.
4193 (autoinc_possible_for_pair): Likewise.
4194 (rewrite_groups): Likewise.
4195 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
4196 (determine_group_iv_cost): Update after split of USE_ADDRESS.
4197 (get_alias_ptr_type_for_ptr_address): New function.
4198 (rewrite_use_address): Rewrite address uses in calls that were
4199 identified by find_address_like_use.
4201 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4202 Alan Hayward <alan.hayward@arm.com>
4203 David Sherwood <david.sherwood@arm.com>
4205 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
4207 * gimple-expr.h (is_gimple_addressable: Likewise.
4208 * gimple-expr.c (is_gimple_address): Likewise.
4209 * internal-fn.c (expand_call_mem_ref): New function.
4210 (expand_mask_load_optab_fn): Use it.
4211 (expand_mask_store_optab_fn): Likewise.
4213 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4214 Alan Hayward <alan.hayward@arm.com>
4215 David Sherwood <david.sherwood@arm.com>
4217 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
4218 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
4219 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
4220 (cond_umax@var{mode}): Document.
4221 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
4222 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
4223 (cond_umin_optab, cond_umax_optab): New optabs.
4224 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
4225 (COND_IOR, COND_XOR): New internal functions.
4226 * internal-fn.h (get_conditional_internal_fn): Declare.
4227 * internal-fn.c (cond_binary_direct): New macro.
4228 (expand_cond_binary_optab_fn): Likewise.
4229 (direct_cond_binary_optab_supported_p): Likewise.
4230 (get_conditional_internal_fn): New function.
4231 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
4232 Cope with reduction statements that are vectorized as calls rather
4234 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
4235 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
4236 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
4237 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
4238 (UNSPEC_COND_EOR): New unspecs.
4239 (optab): Add mappings for them.
4240 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
4241 (sve_int_op, sve_fp_op): New int attributes.
4243 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4244 Alan Hayward <alan.hayward@arm.com>
4245 David Sherwood <david.sherwood@arm.com>
4247 * optabs.def (while_ult_optab): New optab.
4248 * doc/md.texi (while_ult@var{m}@var{n}): Document.
4249 * internal-fn.def (WHILE_ULT): New internal function.
4250 * internal-fn.h (direct_internal_fn_supported_p): New override
4251 that takes two types as argument.
4252 * internal-fn.c (while_direct): New macro.
4253 (expand_while_optab_fn): New function.
4254 (convert_optab_supported_p): Likewise.
4255 (direct_while_optab_supported_p): New macro.
4256 * wide-int.h (wi::udiv_ceil): New function.
4257 * tree-vectorizer.h (rgroup_masks): New structure.
4258 (vec_loop_masks): New typedef.
4259 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
4261 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
4262 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
4263 (vect_max_vf): New function.
4264 (slpeel_make_loop_iterate_ntimes): Delete.
4265 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
4266 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
4267 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
4268 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
4269 internal-fn.h, stor-layout.h and optabs-query.h.
4270 (vect_set_loop_mask): New function.
4271 (add_preheader_seq): Likewise.
4272 (add_header_seq): Likewise.
4273 (interleave_supported_p): Likewise.
4274 (vect_maybe_permute_loop_masks): Likewise.
4275 (vect_set_loop_masks_directly): Likewise.
4276 (vect_set_loop_condition_masked): Likewise.
4277 (vect_set_loop_condition_unmasked): New function, split out from
4278 slpeel_make_loop_iterate_ntimes.
4279 (slpeel_make_loop_iterate_ntimes): Rename to..
4280 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
4281 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
4282 (vect_do_peeling): Update call accordingly.
4283 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
4285 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
4286 mask_compare_type, can_fully_mask_p and fully_masked_p.
4287 (release_vec_loop_masks): New function.
4288 (_loop_vec_info): Use it to free the loop masks.
4289 (can_produce_all_loop_masks_p): New function.
4290 (vect_get_max_nscalars_per_iter): Likewise.
4291 (vect_verify_full_masking): Likewise.
4292 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
4293 retries, and free the mask rgroups before retrying. Check loop-wide
4294 reasons for disallowing fully-masked loops. Make the final decision
4295 about whether use a fully-masked loop or not.
4296 (vect_estimate_min_profitable_iters): Do not assume that peeling
4297 for the number of iterations will be needed for fully-masked loops.
4298 (vectorizable_reduction): Disable fully-masked loops.
4299 (vectorizable_live_operation): Likewise.
4300 (vect_halve_mask_nunits): New function.
4301 (vect_double_mask_nunits): Likewise.
4302 (vect_record_loop_mask): Likewise.
4303 (vect_get_loop_mask): Likewise.
4304 (vect_transform_loop): Handle the case in which the final loop
4305 iteration might handle a partial vector. Call vect_set_loop_condition
4306 instead of slpeel_make_loop_iterate_ntimes.
4307 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
4308 (check_load_store_masking): New function.
4309 (prepare_load_store_mask): Likewise.
4310 (vectorizable_store): Handle fully-masked loops.
4311 (vectorizable_load): Likewise.
4312 (supportable_widening_operation): Use vect_halve_mask_nunits for
4314 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
4315 (vect_gen_while): New function.
4316 * config/aarch64/aarch64.md (umax<mode>3): New expander.
4317 (aarch64_uqdec<mode>): New insn.
4319 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4320 Alan Hayward <alan.hayward@arm.com>
4321 David Sherwood <david.sherwood@arm.com>
4323 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
4324 (reduc_xor_scal_optab): New optabs.
4325 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
4326 (reduc_xor_scal_@var{m}): Document.
4327 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
4328 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
4330 * fold-const-call.c (fold_const_call): Handle them.
4331 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
4332 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
4333 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
4334 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
4335 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
4336 (UNSPEC_XORV): New unspecs.
4337 (optab): Add entries for them.
4338 (BITWISEV): New int iterator.
4339 (bit_reduc_op): New int attributes.
4341 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4342 Alan Hayward <alan.hayward@arm.com>
4343 David Sherwood <david.sherwood@arm.com>
4345 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
4346 * internal-fn.def (VEC_SHL_INSERT): New internal function.
4347 * optabs.def (vec_shl_insert_optab): New optab.
4348 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
4349 (duplicate_and_interleave): Likewise.
4350 * tree-vect-loop.c: Include internal-fn.h.
4351 (neutral_op_for_slp_reduction): New function, split out from
4352 get_initial_defs_for_reduction.
4353 (get_initial_def_for_reduction): Handle option 2 for variable-length
4354 vectors by loading the neutral value into a vector and then shifting
4355 the initial value into element 0.
4356 (get_initial_defs_for_reduction): Replace the code argument with
4357 the neutral value calculated by neutral_op_for_slp_reduction.
4358 Use gimple_build_vector for constant-length vectors.
4359 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
4360 but the first group_size elements have a neutral value.
4361 Use duplicate_and_interleave otherwise.
4362 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
4363 Update call to get_initial_defs_for_reduction. Handle SLP
4364 reductions for variable-length vectors by creating one vector
4365 result for each scalar result, with the elements associated
4366 with other scalar results stubbed out with the neutral value.
4367 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
4368 Require IFN_VEC_SHL_INSERT for double reductions on
4369 variable-length vectors, or SLP reductions that have
4370 a neutral value. Require can_duplicate_and_interleave_p
4371 support for variable-length unchained SLP reductions if there
4372 is no neutral value, such as for MIN/MAX reductions. Also require
4373 the number of vector elements to be a multiple of the number of
4374 SLP statements when doing variable-length unchained SLP reductions.
4375 Update call to vect_create_epilog_for_reduction.
4376 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
4377 and remove initial values.
4378 (duplicate_and_interleave): Make public.
4379 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
4380 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
4382 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4383 Alan Hayward <alan.hayward@arm.com>
4384 David Sherwood <david.sherwood@arm.com>
4386 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
4387 (can_duplicate_and_interleave_p): New function.
4388 (vect_get_and_check_slp_defs): Take the vector of statements
4389 rather than just the current one. Remove excess parentheses.
4390 Restriction rejectinon of vect_constant_def and vect_external_def
4391 for variable-length vectors to boolean types, or types for which
4392 can_duplicate_and_interleave_p is false.
4393 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
4394 (duplicate_and_interleave): New function.
4395 (vect_get_constant_vectors): Use gimple_build_vector for
4396 constant-length vectors and suitable variable-length constant
4397 vectors. Use duplicate_and_interleave for other variable-length
4398 vectors. Don't defer the update when inserting new statements.
4400 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4401 Alan Hayward <alan.hayward@arm.com>
4402 David Sherwood <david.sherwood@arm.com>
4404 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
4405 min_profitable_iters doesn't go negative.
4407 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4408 Alan Hayward <alan.hayward@arm.com>
4409 David Sherwood <david.sherwood@arm.com>
4411 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
4412 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
4413 * optabs.def (vec_mask_load_lanes_optab): New optab.
4414 (vec_mask_store_lanes_optab): Likewise.
4415 * internal-fn.def (MASK_LOAD_LANES): New internal function.
4416 (MASK_STORE_LANES): Likewise.
4417 * internal-fn.c (mask_load_lanes_direct): New macro.
4418 (mask_store_lanes_direct): Likewise.
4419 (expand_mask_load_optab_fn): Handle masked operations.
4420 (expand_mask_load_lanes_optab_fn): New macro.
4421 (expand_mask_store_optab_fn): Handle masked operations.
4422 (expand_mask_store_lanes_optab_fn): New macro.
4423 (direct_mask_load_lanes_optab_supported_p): Likewise.
4424 (direct_mask_store_lanes_optab_supported_p): Likewise.
4425 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
4427 (vect_load_lanes_supported): Likewise.
4428 * tree-vect-data-refs.c (strip_conversion): New function.
4429 (can_group_stmts_p): Likewise.
4430 (vect_analyze_data_ref_accesses): Use it instead of checking
4431 for a pair of assignments.
4432 (vect_store_lanes_supported): Take a masked_p parameter.
4433 (vect_load_lanes_supported): Likewise.
4434 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
4435 vect_store_lanes_supported and vect_load_lanes_supported.
4436 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
4437 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
4438 parameter. Don't allow gaps for masked accesses.
4439 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
4440 and vect_load_lanes_supported.
4441 (get_load_store_type): Take a masked_p parameter and update
4442 call to get_group_load_store_type.
4443 (vectorizable_store): Update call to get_load_store_type.
4444 Handle IFN_MASK_STORE_LANES.
4445 (vectorizable_load): Update call to get_load_store_type.
4446 Handle IFN_MASK_LOAD_LANES.
4448 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4449 Alan Hayward <alan.hayward@arm.com>
4450 David Sherwood <david.sherwood@arm.com>
4452 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
4454 * config/aarch64/aarch64-protos.h
4455 (aarch64_sve_struct_memory_operand_p): Declare.
4456 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
4457 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
4458 (VPRED, vpred): Handle SVE structure modes.
4459 * config/aarch64/constraints.md (Utx): New constraint.
4460 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
4461 (aarch64_sve_struct_nonimmediate_operand): New predicates.
4462 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
4463 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
4464 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
4465 structure modes. Split into pieces after RA.
4466 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
4467 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
4469 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
4470 SVE structure modes.
4471 (aarch64_classify_address): Likewise.
4472 (sizetochar): Move earlier in file.
4473 (aarch64_print_operand): Handle SVE register lists.
4474 (aarch64_array_mode): New function.
4475 (aarch64_sve_struct_memory_operand_p): Likewise.
4476 (TARGET_ARRAY_MODE): Redefine.
4478 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4479 Alan Hayward <alan.hayward@arm.com>
4480 David Sherwood <david.sherwood@arm.com>
4482 * target.def (array_mode): New target hook.
4483 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
4484 * doc/tm.texi: Regenerate.
4485 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
4486 * hooks.c (hook_optmode_mode_uhwi_none): New function.
4487 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
4489 * stor-layout.c (mode_for_array): Likewise. Support polynomial
4492 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4493 Alan Hayward <alan.hayward@arm.com>
4494 David Sherwood <david.sherwood@arm.com>
4496 * fold-const.c (fold_binary_loc): Check the argument types
4497 rather than the result type when testing for a vector operation.
4499 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4501 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
4502 * doc/tm.texi: Regenerate.
4504 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4505 Alan Hayward <alan.hayward@arm.com>
4506 David Sherwood <david.sherwood@arm.com>
4508 * doc/invoke.texi (-msve-vector-bits=): Document new option.
4509 (sve): Document new AArch64 extension.
4510 * doc/md.texi (w): Extend the description of the AArch64
4511 constraint to include SVE vectors.
4512 (Upl, Upa): Document new AArch64 predicate constraints.
4513 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
4515 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
4516 (msve-vector-bits=): New option.
4517 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
4518 SVE when these are disabled.
4519 (sve): New extension.
4520 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
4521 modes. Adjust their number of units based on aarch64_sve_vg.
4522 (MAX_BITSIZE_MODE_ANY_MODE): Define.
4523 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
4524 aarch64_addr_query_type.
4525 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
4526 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
4527 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
4528 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
4529 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
4530 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
4531 (aarch64_simd_imm_zero_p): Delete.
4532 (aarch64_check_zero_based_sve_index_immediate): Declare.
4533 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4534 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4535 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4536 (aarch64_sve_float_mul_immediate_p): Likewise.
4537 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4539 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
4540 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
4541 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
4542 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
4543 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
4544 (aarch64_regmode_natural_size): Likewise.
4545 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
4546 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
4548 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
4549 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
4550 for VG and the SVE predicate registers.
4551 (V_ALIASES): Add a "z"-prefixed alias.
4552 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
4553 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
4554 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
4555 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
4556 (REG_CLASS_NAMES): Add entries for them.
4557 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
4558 and the predicate registers.
4559 (aarch64_sve_vg): Declare.
4560 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
4561 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
4562 (REGMODE_NATURAL_SIZE): Define.
4563 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
4565 * config/aarch64/aarch64.c: Include cfgrtl.h.
4566 (simd_immediate_info): Add a constructor for series vectors,
4567 and an associated step field.
4568 (aarch64_sve_vg): New variable.
4569 (aarch64_dbx_register_number): Handle VG and the predicate registers.
4570 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
4571 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
4572 (VEC_ANY_DATA, VEC_STRUCT): New constants.
4573 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
4574 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
4575 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
4576 (aarch64_get_mask_mode): New functions.
4577 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
4578 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
4579 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
4580 predicate modes and predicate registers. Explicitly restrict
4581 GPRs to modes of 16 bytes or smaller. Only allow FP registers
4582 to store a vector mode if it is recognized by
4583 aarch64_classify_vector_mode.
4584 (aarch64_regmode_natural_size): New function.
4585 (aarch64_hard_regno_caller_save_mode): Return the original mode
4587 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
4588 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
4589 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
4590 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
4592 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
4593 does not overlap dest if the function is frame-related. Handle
4595 (aarch64_split_add_offset): New function.
4596 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
4597 them aarch64_add_offset.
4598 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
4599 and update call to aarch64_sub_sp.
4600 (aarch64_add_cfa_expression): New function.
4601 (aarch64_expand_prologue): Pass extra temporary registers to the
4602 functions above. Handle the case in which we need to emit new
4603 DW_CFA_expressions for registers that were originally saved
4604 relative to the stack pointer, but now have to be expressed
4605 relative to the frame pointer.
4606 (aarch64_output_mi_thunk): Pass extra temporary registers to the
4608 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
4609 IP0 and IP1 values for SVE frames.
4610 (aarch64_expand_vec_series): New function.
4611 (aarch64_expand_sve_widened_duplicate): Likewise.
4612 (aarch64_expand_sve_const_vector): Likewise.
4613 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
4614 Handle SVE constants. Use emit_move_insn to move a force_const_mem
4615 into the register, rather than emitting a SET directly.
4616 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
4617 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
4618 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
4619 (offset_9bit_signed_scaled_p): New functions.
4620 (aarch64_replicate_bitmask_imm): New function.
4621 (aarch64_bitmask_imm): Use it.
4622 (aarch64_cannot_force_const_mem): Reject expressions involving
4623 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
4624 (aarch64_classify_index): Handle SVE indices, by requiring
4625 a plain register index with a scale that matches the element size.
4626 (aarch64_classify_address): Handle SVE addresses. Assert that
4627 the mode of the address is VOIDmode or an integer mode.
4628 Update call to aarch64_classify_symbol.
4629 (aarch64_classify_symbolic_expression): Update call to
4630 aarch64_classify_symbol.
4631 (aarch64_const_vec_all_in_range_p): New function.
4632 (aarch64_print_vector_float_operand): Likewise.
4633 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
4634 "vN" for FP registers with SVE modes. Handle (const ...) vectors
4635 and the FP immediates 1.0 and 0.5.
4636 (aarch64_print_address_internal): Handle SVE addresses.
4637 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
4638 (aarch64_regno_regclass): Handle predicate registers.
4639 (aarch64_secondary_reload): Handle big-endian reloads of SVE
4641 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
4642 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
4643 (aarch64_convert_sve_vector_bits): New function.
4644 (aarch64_override_options): Use it to handle -msve-vector-bits=.
4645 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4647 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
4648 Handle SVE vector and predicate modes. Accept VL-based constants
4649 that need only one temporary register, and VL offsets that require
4650 no temporary registers.
4651 (aarch64_conditional_register_usage): Mark the predicate registers
4652 as fixed if SVE isn't available.
4653 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
4654 Return true for SVE vector and predicate modes.
4655 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
4656 rather than an unsigned int. Handle SVE modes.
4657 (aarch64_preferred_simd_mode): Update call accordingly. Handle
4659 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
4661 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4662 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4663 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4664 (aarch64_sve_float_mul_immediate_p): New functions.
4665 (aarch64_sve_valid_immediate): New function.
4666 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
4667 Explicitly reject structure modes. Check for INDEX constants.
4668 Handle PTRUE and PFALSE constants.
4669 (aarch64_check_zero_based_sve_index_immediate): New function.
4670 (aarch64_simd_imm_zero_p): Delete.
4671 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
4672 vector modes. Accept constants in the range of CNT[BHWD].
4673 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
4674 ask for an Advanced SIMD mode.
4675 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
4676 (aarch64_simd_vector_alignment): Handle SVE predicates.
4677 (aarch64_vectorize_preferred_vector_alignment): New function.
4678 (aarch64_simd_vector_alignment_reachable): Use it instead of
4680 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
4681 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
4683 (MAX_VECT_LEN): Delete.
4684 (expand_vec_perm_d): Add a vec_flags field.
4685 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
4686 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
4687 (aarch64_evpc_ext): Don't apply a big-endian lane correction
4689 (aarch64_evpc_rev): Rename to...
4690 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
4691 (aarch64_evpc_rev_global): New function.
4692 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
4693 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
4695 (aarch64_evpc_sve_tbl): New function.
4696 (aarch64_expand_vec_perm_const_1): Update after rename of
4697 aarch64_evpc_rev. Handle SVE permutes too, trying
4698 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
4699 than aarch64_evpc_tbl.
4700 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
4701 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
4702 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
4703 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
4704 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
4705 (aarch64_expand_sve_vcond): New functions.
4706 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
4707 of aarch64_vector_mode_p.
4708 (aarch64_dwarf_poly_indeterminate_value): New function.
4709 (aarch64_compute_pressure_classes): Likewise.
4710 (aarch64_can_change_mode_class): Likewise.
4711 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
4712 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
4713 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
4714 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
4715 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
4716 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
4717 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
4718 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
4720 (Dn, Dl, Dr): Accept const as well as const_vector.
4721 (Dz): Likewise. Compare against CONST0_RTX.
4722 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
4723 of "vector" where appropriate.
4724 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
4725 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
4726 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
4727 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
4728 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
4729 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
4730 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
4731 (v_int_equiv): Extend to SVE modes.
4732 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
4734 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
4735 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4736 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4737 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4738 (SVE_COND_FP_CMP): New int iterators.
4739 (perm_hilo): Handle the new unpack unspecs.
4740 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4742 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4743 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4744 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4745 (aarch64_equality_operator, aarch64_constant_vector_operand)
4746 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4747 (aarch64_sve_nonimmediate_operand): Likewise.
4748 (aarch64_sve_general_operand): Likewise.
4749 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4750 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4751 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4752 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4753 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4754 (aarch64_sve_float_arith_immediate): Likewise.
4755 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4756 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4757 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4758 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4759 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4760 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4761 (aarch64_sve_float_arith_operand): Likewise.
4762 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4763 (aarch64_sve_float_mul_operand): Likewise.
4764 (aarch64_sve_vec_perm_operand): Likewise.
4765 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4766 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4767 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4768 as well as const_vector.
4769 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4770 in file. Use CONST0_RTX and CONSTM1_RTX.
4771 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4772 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4773 Use aarch64_simd_imm_zero.
4774 * config/aarch64/aarch64-sve.md: New file.
4775 * config/aarch64/aarch64.md: Include it.
4776 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4777 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4778 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4779 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4780 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4781 (sve): New attribute.
4782 (enabled): Disable instructions with the sve attribute unless
4784 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4785 aarch64_expand_mov_immediate.
4786 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4787 CNT[BHSD] immediates.
4788 (movti): Split CONST_POLY_INT moves into two halves.
4789 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4790 Split additions that need a temporary here if the destination
4791 is the stack pointer.
4792 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4793 (*add<mode>3_poly_1): New instruction.
4794 (set_clobber_cc): New expander.
4796 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4798 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4799 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4800 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4801 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4802 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4803 Change innermode from fixed_mode_size to machine_mode.
4804 (simplify_subreg): Update call accordingly. Handle a constant-sized
4805 subreg of a variable-length CONST_VECTOR.
4807 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4808 Alan Hayward <alan.hayward@arm.com>
4809 David Sherwood <david.sherwood@arm.com>
4811 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4812 (add_offset_to_base): New function, split out from...
4813 (create_mem_ref): ...here. When handling a scale other than 1,
4814 check first whether the address is valid without the offset.
4815 Add it into the base if so, leaving the index and scale as-is.
4817 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4820 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4821 fold_for_warn before checking if arg2 is INTEGER_CST.
4823 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4825 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4826 (store_multiple_operation): Delete.
4827 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4828 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4829 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4830 guarded by TARGET_STRING.
4831 (rs6000_output_load_multiple): Delete.
4832 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4833 OPTION_MASK_STRING / TARGET_STRING handling.
4834 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4835 (const rs6000_opt_masks) <"string">: Change mask to 0.
4836 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4837 (MASK_STRING): Delete.
4838 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4840 (load_multiple): Delete.
4847 (store_multiple): Delete.
4854 (movmemsi_8reg): Delete.
4855 (corresponding unnamed define_insn): Delete.
4856 (movmemsi_6reg): Delete.
4857 (corresponding unnamed define_insn): Delete.
4858 (movmemsi_4reg): Delete.
4859 (corresponding unnamed define_insn): Delete.
4860 (movmemsi_2reg): Delete.
4861 (corresponding unnamed define_insn): Delete.
4862 (movmemsi_1reg): Delete.
4863 (corresponding unnamed define_insn): Delete.
4864 * config/rs6000/rs6000.opt (mno-string): New.
4865 (mstring): Replace by deprecation warning stub.
4866 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4868 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4870 * regrename.c (regrename_do_replace): If replacing the same
4871 reg multiple times, try to reuse last created gen_raw_REG.
4874 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4875 main to workaround a bug in GDB.
4877 2018-01-12 Tom de Vries <tom@codesourcery.com>
4880 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4882 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4884 PR rtl-optimization/80481
4885 * ira-color.c (get_cap_member): New function.
4886 (allocnos_conflict_by_live_ranges_p): Use it.
4887 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4888 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4890 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4893 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4894 (*saddl_se_1): Ditto.
4896 (*ssubl_se_1): Ditto.
4898 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4900 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4901 rather than wi::to_widest for DR_INITs.
4902 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4903 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4904 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4906 (vect_analyze_group_access_1): Note that here.
4908 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4910 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4911 polynomial type sizes.
4913 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4915 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4916 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4917 (gimple_add_tmp_var): Likewise.
4919 2018-01-12 Martin Liska <mliska@suse.cz>
4921 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4922 (gimple_alloc_sizes): Likewise.
4923 (dump_gimple_statistics): Use PRIu64 in printf format.
4924 * gimple.h: Change uint64_t to int.
4926 2018-01-12 Martin Liska <mliska@suse.cz>
4928 * tree-core.h: Use uint64_t instead of int.
4929 * tree.c (tree_node_counts): Likewise.
4930 (tree_node_sizes): Likewise.
4931 (dump_tree_statistics): Use PRIu64 in printf format.
4933 2018-01-12 Martin Liska <mliska@suse.cz>
4935 * Makefile.in: As qsort_chk is implemented in vec.c, add
4936 vec.o to linkage of gencfn-macros.
4937 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4938 passing the info to record_node_allocation_statistics.
4939 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4941 * ggc-common.c (struct ggc_usage): Add operator== and use
4942 it in operator< and compare function.
4943 * mem-stats.h (struct mem_usage): Likewise.
4944 * vec.c (struct vec_usage): Remove operator< and compare
4945 function. Can be simply inherited.
4947 2018-01-12 Martin Jambor <mjambor@suse.cz>
4950 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4951 * tree-ssa-math-opts.c: Include domwalk.h.
4952 (convert_mult_to_fma_1): New function.
4953 (fma_transformation_info): New type.
4954 (fma_deferring_state): Likewise.
4955 (cancel_fma_deferring): New function.
4956 (result_of_phi): Likewise.
4957 (last_fma_candidate_feeds_initial_phi): Likewise.
4958 (convert_mult_to_fma): Added deferring logic, split actual
4959 transformation to convert_mult_to_fma_1.
4960 (math_opts_dom_walker): New type.
4961 (math_opts_dom_walker::after_dom_children): New method, body moved
4962 here from pass_optimize_widening_mul::execute, added deferring logic
4964 (pass_optimize_widening_mul::execute): Moved most of code to
4965 math_opts_dom_walker::after_dom_children.
4966 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4967 * config/i386/i386.c (ix86_option_override_internal): Added
4968 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4970 2018-01-12 Richard Biener <rguenther@suse.de>
4973 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4974 inline instance vars.
4976 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4979 * config/rx/rx.c (rx_is_restricted_memory_address):
4982 2018-01-12 Richard Biener <rguenther@suse.de>
4984 PR tree-optimization/80846
4985 * target.def (split_reduction): New target hook.
4986 * targhooks.c (default_split_reduction): New function.
4987 * targhooks.h (default_split_reduction): Declare.
4988 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4989 target requests first reduce vectors by combining low and high
4991 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4992 (get_vectype_for_scalar_type_and_size): Export.
4993 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4994 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4995 * doc/tm.texi: Regenerate.
4996 * config/i386/i386.c (ix86_split_reduction): Implement
4997 TARGET_VECTORIZE_SPLIT_REDUCTION.
4999 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
5002 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
5003 in PIC mode except for TARGET_VXWORKS_RTP.
5004 * config/sparc/sparc.c: Include cfgrtl.h.
5005 (TARGET_INIT_PIC_REG): Define.
5006 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
5007 (sparc_pic_register_p): New predicate.
5008 (sparc_legitimate_address_p): Use it.
5009 (sparc_legitimize_pic_address): Likewise.
5010 (sparc_delegitimize_address): Likewise.
5011 (sparc_mode_dependent_address_p): Likewise.
5012 (gen_load_pcrel_sym): Remove 4th parameter.
5013 (load_got_register): Adjust call to above. Remove obsolete stuff.
5014 (sparc_expand_prologue): Do not call load_got_register here.
5015 (sparc_flat_expand_prologue): Likewise.
5016 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
5017 (sparc_use_pseudo_pic_reg): New function.
5018 (sparc_init_pic_reg): Likewise.
5019 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
5020 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
5022 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
5024 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
5025 Add item for branch_cost.
5027 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
5029 PR rtl-optimization/83565
5030 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
5031 not extend the result to a larger mode for rotate operations.
5032 (num_sign_bit_copies1): Likewise.
5034 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5037 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
5039 Use values-Xc.o for -pedantic.
5040 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
5042 2018-01-12 Martin Liska <mliska@suse.cz>
5045 * ipa-devirt.c (final_warning_record::grow_type_warnings):
5047 (possible_polymorphic_call_targets): Use it.
5048 (ipa_devirt): Likewise.
5050 2018-01-12 Martin Liska <mliska@suse.cz>
5052 * profile-count.h (enum profile_quality): Use 0 as invalid
5053 enum value of profile_quality.
5055 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
5057 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
5058 -mext-string options.
5060 2018-01-12 Richard Biener <rguenther@suse.de>
5062 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
5063 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
5064 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
5066 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
5068 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
5070 * configure.ac (--with-long-double-format): Add support for the
5071 configuration option to change the default long double format on
5073 * config.gcc (powerpc*-linux*-*): Likewise.
5074 * configure: Regenerate.
5075 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
5076 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
5077 used without modification.
5079 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5081 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
5082 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
5083 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
5084 MISC_BUILTIN_SPEC_BARRIER.
5085 (rs6000_init_builtins): Likewise.
5086 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
5088 (speculation_barrier): New define_insn.
5089 * doc/extend.texi: Document __builtin_speculation_barrier.
5091 2018-01-11 Jakub Jelinek <jakub@redhat.com>
5094 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
5095 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
5096 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
5098 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
5099 integral modes instead of "ss" and "sd".
5100 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
5101 vectors with 32-bit and 64-bit elements.
5102 (vecdupssescalarmodesuffix): New mode attribute.
5103 (vec_dup<mode>): Use it.
5105 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
5108 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
5109 frame if argument is passed on stack.
5111 2018-01-11 Jakub Jelinek <jakub@redhat.com>
5114 * ree.c (combine_reaching_defs): Optimize also
5115 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
5116 reg2=any_extend(exp); reg1=reg2;, formatting fix.
5118 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
5121 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
5123 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
5126 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
5127 after they are computed.
5129 2018-01-11 Bin Cheng <bin.cheng@arm.com>
5131 PR tree-optimization/83695
5132 * gimple-loop-linterchange.cc
5133 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
5134 reset cached scev information after interchange.
5135 (pass_linterchange::execute): Remove call to scev_reset_htab.
5137 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5139 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
5140 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
5141 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
5142 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
5143 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
5144 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
5145 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
5146 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
5147 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
5148 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
5149 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
5150 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
5151 (V_lane_reg): Likewise.
5152 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
5154 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
5155 (vfmal_lane_low<mode>_intrinsic,
5156 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
5157 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
5158 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
5159 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
5160 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
5161 vfmsl_lane_high<mode>_intrinsic): New define_insns.
5163 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5165 * config/arm/arm-cpus.in (fp16fml): New feature.
5166 (ALL_SIMD): Add fp16fml.
5167 (armv8.2-a): Add fp16fml as an option.
5168 (armv8.3-a): Likewise.
5169 (armv8.4-a): Add fp16fml as part of fp16.
5170 * config/arm/arm.h (TARGET_FP16FML): Define.
5171 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
5173 * config/arm/arm-modes.def (V2HF): Define.
5174 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
5175 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
5176 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
5177 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
5178 vfmsl_low, vfmsl_high): New set of builtins.
5179 * config/arm/iterators.md (PLUSMINUS): New code iterator.
5180 (vfml_op): New code attribute.
5181 (VFMLHALVES): New int iterator.
5182 (VFML, VFMLSEL): New mode attributes.
5183 (V_reg): Define mapping for V2HF.
5184 (V_hi, V_lo): New mode attributes.
5185 (VF_constraint): Likewise.
5186 (vfml_half, vfml_half_selector): New int attributes.
5187 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
5189 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
5190 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
5192 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
5193 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
5194 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
5195 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
5197 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
5198 Document new effective target and option set.
5200 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5202 * config/arm/arm-cpus.in (armv8_4): New feature.
5203 (ARMv8_4a): New fgroup.
5204 (armv8.4-a): New arch.
5205 * config/arm/arm-tables.opt: Regenerate.
5206 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
5207 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
5208 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
5209 Add matching rules for -march=armv8.4-a and extensions.
5210 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
5212 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
5215 * config/rx/rx.md (BW): New mode attribute.
5216 (sync_lock_test_and_setsi): Add mode suffix to insn output.
5218 2018-01-11 Richard Biener <rguenther@suse.de>
5220 PR tree-optimization/83435
5221 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
5222 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
5223 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
5225 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5226 Alan Hayward <alan.hayward@arm.com>
5227 David Sherwood <david.sherwood@arm.com>
5229 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
5231 (aarch64_classify_address): Initialize it. Track polynomial offsets.
5232 (aarch64_print_address_internal): Use it to check for a zero offset.
5234 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5235 Alan Hayward <alan.hayward@arm.com>
5236 David Sherwood <david.sherwood@arm.com>
5238 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
5239 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
5240 Return a poly_int64 rather than a HOST_WIDE_INT.
5241 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
5242 rather than a HOST_WIDE_INT.
5243 * config/aarch64/aarch64.h (aarch64_frame): Protect with
5244 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
5245 hard_fp_offset, frame_size, initial_adjust, callee_offset and
5246 final_offset from HOST_WIDE_INT to poly_int64.
5247 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
5248 to_constant when getting the number of units in an Advanced SIMD
5250 (aarch64_builtin_vectorized_function): Check for a constant number
5252 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
5254 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
5255 attribute instead of GET_MODE_NUNITS.
5256 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
5257 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
5258 GET_MODE_SIZE for fixed-size registers.
5259 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
5260 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
5261 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
5262 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
5263 (aarch64_print_operand, aarch64_print_address_internal)
5264 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
5265 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
5266 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
5267 Handle polynomial GET_MODE_SIZE.
5268 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
5269 wider than SImode without modification.
5270 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
5271 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
5272 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
5273 passing and returning SVE modes.
5274 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
5275 rather than GEN_INT.
5276 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
5277 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
5278 (aarch64_allocate_and_probe_stack_space): Likewise.
5279 (aarch64_layout_frame): Cope with polynomial offsets.
5280 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
5281 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
5283 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
5284 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
5285 poly_int64 rather than a HOST_WIDE_INT.
5286 (aarch64_get_separate_components, aarch64_process_components)
5287 (aarch64_expand_prologue, aarch64_expand_epilogue)
5288 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
5289 (aarch64_anchor_offset): New function, split out from...
5290 (aarch64_legitimize_address): ...here.
5291 (aarch64_builtin_vectorization_cost): Handle polynomial
5292 TYPE_VECTOR_SUBPARTS.
5293 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
5295 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
5296 number of elements from the PARALLEL rather than the mode.
5297 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
5298 rather than GET_MODE_BITSIZE.
5299 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
5300 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
5301 (aarch64_expand_vec_perm_const_1): Handle polynomial
5302 d->perm.length () and d->perm elements.
5303 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
5304 Apply to_constant to d->perm elements.
5305 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
5306 polynomial CONST_VECTOR_NUNITS.
5307 (aarch64_move_pointer): Take amount as a poly_int64 rather
5309 (aarch64_progress_pointer): Avoid temporary variable.
5310 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
5311 the mode attribute instead of GET_MODE.
5313 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5314 Alan Hayward <alan.hayward@arm.com>
5315 David Sherwood <david.sherwood@arm.com>
5317 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
5318 x exists before using it.
5319 (aarch64_add_constant_internal): Rename to...
5320 (aarch64_add_offset_1): ...this. Replace regnum with separate
5321 src and dest rtxes. Handle the case in which they're different,
5322 including when the offset is zero. Replace scratchreg with an rtx.
5323 Use 2 additions if there is no spare register into which we can
5324 move a 16-bit constant.
5325 (aarch64_add_constant): Delete.
5326 (aarch64_add_offset): Replace reg with separate src and dest
5327 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
5328 Use aarch64_add_offset_1.
5329 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
5330 an rtx rather than an int. Take the delta as a poly_int64
5331 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
5332 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
5333 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
5334 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
5335 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
5337 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
5338 aarch64_add_constant.
5340 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5342 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
5343 Use scalar_float_mode.
5345 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5347 * config/aarch64/aarch64-simd.md
5348 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
5349 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
5350 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
5351 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
5352 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
5353 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
5354 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
5355 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
5356 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
5357 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
5359 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5362 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
5363 targ_options->x_arm_arch_string is non NULL.
5365 2018-01-11 Tamar Christina <tamar.christina@arm.com>
5367 * config/aarch64/aarch64.h
5368 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
5370 2018-01-11 Sudakshina Das <sudi.das@arm.com>
5373 * expmed.c (emit_store_flag_force): Swap if const op0
5374 and change VOIDmode to mode of op0.
5376 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
5378 PR rtl-optimization/83761
5379 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
5380 than bytes to mode_for_size.
5382 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5385 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
5386 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
5389 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5392 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
5393 when in layout mode.
5394 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
5395 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
5398 2018-01-10 Michael Collison <michael.collison@arm.com>
5400 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
5401 * config/aarch64/aarch64-option-extension.def: Add
5402 AARCH64_OPT_EXTENSION of 'fp16fml'.
5403 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5404 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
5405 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
5406 * config/aarch64/constraints.md (Ui7): New constraint.
5407 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
5408 (VFMLA_SEL_W): Ditto.
5411 (VFMLA16_LOW): New int iterator.
5412 (VFMLA16_HIGH): Ditto.
5413 (UNSPEC_FMLAL): New unspec.
5414 (UNSPEC_FMLSL): Ditto.
5415 (UNSPEC_FMLAL2): Ditto.
5416 (UNSPEC_FMLSL2): Ditto.
5417 (f16mac): New code attribute.
5418 * config/aarch64/aarch64-simd-builtins.def
5419 (aarch64_fmlal_lowv2sf): Ditto.
5420 (aarch64_fmlsl_lowv2sf): Ditto.
5421 (aarch64_fmlalq_lowv4sf): Ditto.
5422 (aarch64_fmlslq_lowv4sf): Ditto.
5423 (aarch64_fmlal_highv2sf): Ditto.
5424 (aarch64_fmlsl_highv2sf): Ditto.
5425 (aarch64_fmlalq_highv4sf): Ditto.
5426 (aarch64_fmlslq_highv4sf): Ditto.
5427 (aarch64_fmlal_lane_lowv2sf): Ditto.
5428 (aarch64_fmlsl_lane_lowv2sf): Ditto.
5429 (aarch64_fmlal_laneq_lowv2sf): Ditto.
5430 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
5431 (aarch64_fmlalq_lane_lowv4sf): Ditto.
5432 (aarch64_fmlsl_lane_lowv4sf): Ditto.
5433 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
5434 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
5435 (aarch64_fmlal_lane_highv2sf): Ditto.
5436 (aarch64_fmlsl_lane_highv2sf): Ditto.
5437 (aarch64_fmlal_laneq_highv2sf): Ditto.
5438 (aarch64_fmlsl_laneq_highv2sf): Ditto.
5439 (aarch64_fmlalq_lane_highv4sf): Ditto.
5440 (aarch64_fmlsl_lane_highv4sf): Ditto.
5441 (aarch64_fmlalq_laneq_highv4sf): Ditto.
5442 (aarch64_fmlsl_laneq_highv4sf): Ditto.
5443 * config/aarch64/aarch64-simd.md:
5444 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
5445 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5446 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
5447 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
5448 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
5449 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
5450 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
5451 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
5452 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
5453 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
5454 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
5455 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
5456 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
5457 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
5458 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
5459 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
5460 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
5461 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
5462 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
5463 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
5464 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
5465 (vfmlsl_low_u32): Ditto.
5466 (vfmlalq_low_u32): Ditto.
5467 (vfmlslq_low_u32): Ditto.
5468 (vfmlal_high_u32): Ditto.
5469 (vfmlsl_high_u32): Ditto.
5470 (vfmlalq_high_u32): Ditto.
5471 (vfmlslq_high_u32): Ditto.
5472 (vfmlal_lane_low_u32): Ditto.
5473 (vfmlsl_lane_low_u32): Ditto.
5474 (vfmlal_laneq_low_u32): Ditto.
5475 (vfmlsl_laneq_low_u32): Ditto.
5476 (vfmlalq_lane_low_u32): Ditto.
5477 (vfmlslq_lane_low_u32): Ditto.
5478 (vfmlalq_laneq_low_u32): Ditto.
5479 (vfmlslq_laneq_low_u32): Ditto.
5480 (vfmlal_lane_high_u32): Ditto.
5481 (vfmlsl_lane_high_u32): Ditto.
5482 (vfmlal_laneq_high_u32): Ditto.
5483 (vfmlsl_laneq_high_u32): Ditto.
5484 (vfmlalq_lane_high_u32): Ditto.
5485 (vfmlslq_lane_high_u32): Ditto.
5486 (vfmlalq_laneq_high_u32): Ditto.
5487 (vfmlslq_laneq_high_u32): Ditto.
5488 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
5489 (AARCH64_FL_FOR_ARCH8_4): New.
5490 (AARCH64_ISA_F16FML): New ISA flag.
5491 (TARGET_F16FML): New feature flag for fp16fml.
5492 (doc/invoke.texi): Document new fp16fml option.
5494 2018-01-10 Michael Collison <michael.collison@arm.com>
5496 * config/aarch64/aarch64-builtins.c:
5497 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
5498 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5499 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
5500 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
5501 (AARCH64_ISA_SHA3): New ISA flag.
5502 (TARGET_SHA3): New feature flag for sha3.
5503 * config/aarch64/iterators.md (sha512_op): New int attribute.
5504 (CRYPTO_SHA512): New int iterator.
5505 (UNSPEC_SHA512H): New unspec.
5506 (UNSPEC_SHA512H2): Ditto.
5507 (UNSPEC_SHA512SU0): Ditto.
5508 (UNSPEC_SHA512SU1): Ditto.
5509 * config/aarch64/aarch64-simd-builtins.def
5510 (aarch64_crypto_sha512hqv2di): New builtin.
5511 (aarch64_crypto_sha512h2qv2di): Ditto.
5512 (aarch64_crypto_sha512su0qv2di): Ditto.
5513 (aarch64_crypto_sha512su1qv2di): Ditto.
5514 (aarch64_eor3qv8hi): Ditto.
5515 (aarch64_rax1qv2di): Ditto.
5516 (aarch64_xarqv2di): Ditto.
5517 (aarch64_bcaxqv8hi): Ditto.
5518 * config/aarch64/aarch64-simd.md:
5519 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
5520 (aarch64_crypto_sha512su0qv2di): Ditto.
5521 (aarch64_crypto_sha512su1qv2di): Ditto.
5522 (aarch64_eor3qv8hi): Ditto.
5523 (aarch64_rax1qv2di): Ditto.
5524 (aarch64_xarqv2di): Ditto.
5525 (aarch64_bcaxqv8hi): Ditto.
5526 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
5527 (vsha512h2q_u64): Ditto.
5528 (vsha512su0q_u64): Ditto.
5529 (vsha512su1q_u64): Ditto.
5530 (veor3q_u16): Ditto.
5531 (vrax1q_u64): Ditto.
5533 (vbcaxq_u16): Ditto.
5534 * config/arm/types.md (crypto_sha512): New type attribute.
5535 (crypto_sha3): Ditto.
5536 (doc/invoke.texi): Document new sha3 option.
5538 2018-01-10 Michael Collison <michael.collison@arm.com>
5540 * config/aarch64/aarch64-builtins.c:
5541 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
5542 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5543 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
5544 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
5545 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
5546 (AARCH64_ISA_SM4): New ISA flag.
5547 (TARGET_SM4): New feature flag for sm4.
5548 * config/aarch64/aarch64-simd-builtins.def
5549 (aarch64_sm3ss1qv4si): Ditto.
5550 (aarch64_sm3tt1aq4si): Ditto.
5551 (aarch64_sm3tt1bq4si): Ditto.
5552 (aarch64_sm3tt2aq4si): Ditto.
5553 (aarch64_sm3tt2bq4si): Ditto.
5554 (aarch64_sm3partw1qv4si): Ditto.
5555 (aarch64_sm3partw2qv4si): Ditto.
5556 (aarch64_sm4eqv4si): Ditto.
5557 (aarch64_sm4ekeyqv4si): Ditto.
5558 * config/aarch64/aarch64-simd.md:
5559 (aarch64_sm3ss1qv4si): Ditto.
5560 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
5561 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
5562 (aarch64_sm4eqv4si): Ditto.
5563 (aarch64_sm4ekeyqv4si): Ditto.
5564 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
5565 (sm3part_op): Ditto.
5566 (CRYPTO_SM3TT): Ditto.
5567 (CRYPTO_SM3PART): Ditto.
5568 (UNSPEC_SM3SS1): New unspec.
5569 (UNSPEC_SM3TT1A): Ditto.
5570 (UNSPEC_SM3TT1B): Ditto.
5571 (UNSPEC_SM3TT2A): Ditto.
5572 (UNSPEC_SM3TT2B): Ditto.
5573 (UNSPEC_SM3PARTW1): Ditto.
5574 (UNSPEC_SM3PARTW2): Ditto.
5575 (UNSPEC_SM4E): Ditto.
5576 (UNSPEC_SM4EKEY): Ditto.
5577 * config/aarch64/constraints.md (Ui2): New constraint.
5578 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
5579 * config/arm/types.md (crypto_sm3): New type attribute.
5580 (crypto_sm4): Ditto.
5581 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
5582 (vsm3tt1aq_u32): Ditto.
5583 (vsm3tt1bq_u32): Ditto.
5584 (vsm3tt2aq_u32): Ditto.
5585 (vsm3tt2bq_u32): Ditto.
5586 (vsm3partw1q_u32): Ditto.
5587 (vsm3partw2q_u32): Ditto.
5588 (vsm4eq_u32): Ditto.
5589 (vsm4ekeyq_u32): Ditto.
5590 (doc/invoke.texi): Document new sm4 option.
5592 2018-01-10 Michael Collison <michael.collison@arm.com>
5594 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
5595 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
5596 (AARCH64_FL_FOR_ARCH8_4): New.
5597 (AARCH64_FL_V8_4): New flag.
5598 (doc/invoke.texi): Document new armv8.4-a option.
5600 2018-01-10 Michael Collison <michael.collison@arm.com>
5602 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5603 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
5604 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
5605 * config/aarch64/aarch64-option-extension.def: Add
5606 AARCH64_OPT_EXTENSION of 'sha2'.
5607 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
5608 (crypto): Disable sha2 and aes if crypto disabled.
5609 (crypto): Enable aes and sha2 if enabled.
5610 (simd): Disable sha2 and aes if simd disabled.
5611 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
5613 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
5614 (TARGET_SHA2): New feature flag for sha2.
5615 (TARGET_AES): New feature flag for aes.
5616 * config/aarch64/aarch64-simd.md:
5617 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
5618 conditional on TARGET_AES.
5619 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
5620 (aarch64_crypto_sha1hsi): Make pattern conditional
5622 (aarch64_crypto_sha1hv4si): Ditto.
5623 (aarch64_be_crypto_sha1hv4si): Ditto.
5624 (aarch64_crypto_sha1su1v4si): Ditto.
5625 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
5626 (aarch64_crypto_sha1su0v4si): Ditto.
5627 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
5628 (aarch64_crypto_sha256su0v4si): Ditto.
5629 (aarch64_crypto_sha256su1v4si): Ditto.
5630 (doc/invoke.texi): Document new aes and sha2 options.
5632 2018-01-10 Martin Sebor <msebor@redhat.com>
5634 PR tree-optimization/83781
5635 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
5638 2018-01-11 Martin Sebor <msebor@gmail.com>
5639 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5641 PR tree-optimization/83501
5642 PR tree-optimization/81703
5644 * tree-ssa-strlen.c (get_string_cst): Rename...
5645 (get_string_len): ...to this. Handle global constants.
5646 (handle_char_store): Adjust.
5648 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
5649 Jim Wilson <jimw@sifive.com>
5651 * config/riscv/riscv-protos.h (riscv_output_return): New.
5652 * config/riscv/riscv.c (struct machine_function): New naked_p field.
5653 (riscv_attribute_table, riscv_output_return),
5654 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
5655 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
5656 (riscv_compute_frame_info): Only compute frame->mask if not a naked
5658 (riscv_expand_prologue): Add early return for naked function.
5659 (riscv_expand_epilogue): Likewise.
5660 (riscv_function_ok_for_sibcall): Return false for naked function.
5661 (riscv_set_current_function): New.
5662 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
5663 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
5664 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
5665 * doc/extend.texi (RISC-V Function Attributes): New.
5667 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
5669 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
5670 check for 128-bit long double before checking TCmode.
5671 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
5672 128-bit long doubles before checking TFmode or TCmode.
5673 (FLOAT128_IBM_P): Likewise.
5675 2018-01-10 Martin Sebor <msebor@redhat.com>
5677 PR tree-optimization/83671
5678 * builtins.c (c_strlen): Unconditionally return zero for the empty
5680 Use -Warray-bounds for warnings.
5681 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
5682 for non-constant array indices with COMPONENT_REF, arrays of
5683 arrays, and pointers to arrays.
5684 (gimple_fold_builtin_strlen): Determine and set length range for
5685 non-constant character arrays.
5687 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
5690 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
5693 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
5695 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
5697 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5700 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
5701 VECTOR_MEM_ALTIVEC_OR_VSX_P.
5702 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
5703 indexed_or_indirect_operand predicate.
5704 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
5705 (*vsx_le_perm_load_v8hi): Likewise.
5706 (*vsx_le_perm_load_v16qi): Likewise.
5707 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
5708 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
5709 (*vsx_le_perm_store_v8hi): Likewise.
5710 (*vsx_le_perm_store_v16qi): Likewise.
5711 (eight unnamed splitters): Likewise.
5713 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5715 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
5716 * config/rs6000/emmintrin.h: Likewise.
5717 * config/rs6000/mmintrin.h: Likewise.
5718 * config/rs6000/xmmintrin.h: Likewise.
5720 2018-01-10 David Malcolm <dmalcolm@redhat.com>
5723 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
5725 * tree.c (tree_nop_conversion): Return true for location wrapper
5727 (maybe_wrap_with_location): New function.
5728 (selftest::check_strip_nops): New function.
5729 (selftest::test_location_wrappers): New function.
5730 (selftest::tree_c_tests): Call it.
5731 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
5732 (maybe_wrap_with_location): New decl.
5733 (EXPR_LOCATION_WRAPPER_P): New macro.
5734 (location_wrapper_p): New inline function.
5735 (tree_strip_any_location_wrapper): New inline function.
5737 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5740 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5741 stack_realign_offset for the largest alignment of stack slot
5743 (ix86_find_max_used_stack_alignment): New function.
5744 (ix86_finalize_stack_frame_flags): Use it. Set
5745 max_used_stack_alignment if we don't realign stack.
5746 * config/i386/i386.h (machine_function): Add
5747 max_used_stack_alignment.
5749 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5751 * config/arm/arm.opt (-mbranch-cost): New option.
5752 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5755 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5758 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5759 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5761 2018-01-10 Richard Biener <rguenther@suse.de>
5764 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5765 early out so it also covers the case where we have a non-NULL
5768 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5770 PR tree-optimization/83753
5771 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5772 for non-strided grouped accesses if the number of elements is 1.
5774 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5777 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5778 * i386.h (TARGET_USE_GATHER): Define.
5779 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5781 2018-01-10 Martin Liska <mliska@suse.cz>
5784 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5785 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5787 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5788 CLEANUP_NO_PARTITIONING is not set.
5790 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5792 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5793 for vectors, as a partial revert of r254296.
5794 * rtl.h (const_vec_p): Delete.
5795 (const_vec_duplicate_p): Don't test for vector CONSTs.
5796 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5797 * expmed.c (make_tree): Likewise.
5800 * common.md (E, F): Use CONSTANT_P instead of checking for
5802 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5803 checking for CONST_VECTOR.
5805 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5808 * predict.c (force_edge_cold): Handle in more sane way edges
5811 2018-01-09 Carl Love <cel@us.ibm.com>
5813 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5815 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5816 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5817 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5818 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5819 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5820 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5821 * config/rs6000/rs6000-protos.h: Add extern defition for
5822 rs6000_generate_float2_double_code.
5823 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5825 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5826 (float2_v2df): Add define_expand.
5828 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5831 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5832 op_mode in the force_to_mode call.
5834 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5836 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5837 instead of checking each element individually.
5838 (aarch64_evpc_uzp): Likewise.
5839 (aarch64_evpc_zip): Likewise.
5840 (aarch64_evpc_ext): Likewise.
5841 (aarch64_evpc_rev): Likewise.
5842 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5843 instead of checking each element individually. Return true without
5845 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5846 whether all selected elements come from the same input, instead of
5847 checking each element individually. Remove calls to gen_rtx_REG,
5848 start_sequence and end_sequence and instead assert that no rtl is
5851 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5853 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5854 order of HIGH and CONST checks.
5856 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5858 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5859 if the destination isn't an SSA_NAME.
5861 2018-01-09 Richard Biener <rguenther@suse.de>
5863 PR tree-optimization/83668
5864 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5866 (canonicalize_loop_form): ... here, renamed from ...
5867 (canonicalize_loop_closed_ssa_form): ... this and amended to
5868 swap successor edges for loop exit blocks to make us use
5869 the RPO order we need for initial schedule generation.
5871 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5873 PR tree-optimization/64811
5874 * match.pd: When optimizing comparisons with Inf, avoid
5875 introducing or losing exceptions from comparisons with NaN.
5877 2018-01-09 Martin Liska <mliska@suse.cz>
5880 * asan.c (shadow_mem_size): Add gcc_assert.
5882 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5884 Don't save registers in main().
5887 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5888 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5889 * config/avr/avr.c (avr_set_current_function): Don't error if
5890 naked, OS_task or OS_main are specified at the same time.
5891 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5893 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5895 * common/config/avr/avr-common.c (avr_option_optimization_table):
5896 Switch on -mmain-is-OS_task for optimizing compilations.
5898 2018-01-09 Richard Biener <rguenther@suse.de>
5900 PR tree-optimization/83572
5901 * graphite.c: Include cfganal.h.
5902 (graphite_transform_loops): Connect infinite loops to exit
5903 and remove fake edges at the end.
5905 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5907 * ipa-inline.c (edge_badness): Revert accidental checkin.
5909 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5912 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5913 symbols; not inline clones.
5915 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5918 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5919 hard registers. Formatting fixes.
5921 PR preprocessor/83722
5922 * gcc.c (try_generate_repro): Pass
5923 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5924 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5927 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5928 Kito Cheng <kito.cheng@gmail.com>
5930 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5931 (riscv_leaf_function_p): Delete.
5932 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5934 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5936 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5938 (do_ifelse): New function.
5939 (do_isel): New function.
5940 (do_sub3): New function.
5941 (do_add3): New function.
5942 (do_load_mask_compare): New function.
5943 (do_overlap_load_compare): New function.
5944 (expand_compare_loop): New function.
5945 (expand_block_compare): Call expand_compare_loop() when appropriate.
5946 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5948 (-mblock-compare-inline-loop-limit): New option.
5950 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5953 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5954 Reverse order of second and third operands in first alternative.
5955 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5956 of first and second elements in UNSPEC_VPERMR vector.
5957 (altivec_expand_vec_perm_le): Likewise.
5959 2018-01-08 Jeff Law <law@redhat.com>
5961 PR rtl-optimizatin/81308
5962 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5963 (process_switch): If group_case_labels makes a change, then set
5965 (pass_convert_switch::execute): If a switch is converted, then
5966 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5968 PR rtl-optimization/81308
5969 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5972 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5974 PR target/83663 - Revert r255946
5975 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5976 generation for cases where splatting a value is not useful.
5977 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5978 across a vec_duplicate and a paradoxical subreg forming a vector
5979 mode to a vec_concat.
5981 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5983 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5984 -march=armv8.3-a variants.
5985 * config/arm/t-multilib: Likewise.
5986 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5988 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5990 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5992 (cceq_ior_compare_complement): Give it a name so I can use it, and
5993 change boolean_or_operator predicate to boolean_operator so it can
5994 be used to generate a crand.
5995 (eqne): New code iterator.
5996 (bd/bd_neg): New code_attrs.
5997 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5998 a single define_insn.
5999 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
6000 decrement (bdnzt/bdnzf/bdzt/bdzf).
6001 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
6002 with the new names of the branch decrement patterns, and added the
6003 names of the branch decrement conditional patterns.
6005 2018-01-08 Richard Biener <rguenther@suse.de>
6007 PR tree-optimization/83563
6008 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
6011 2018-01-08 Richard Biener <rguenther@suse.de>
6014 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
6016 2018-01-08 Richard Biener <rguenther@suse.de>
6018 PR tree-optimization/83685
6019 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
6020 references to abnormals.
6022 2018-01-08 Richard Biener <rguenther@suse.de>
6025 * dwarf2out.c (output_indirect_strings): Handle empty
6026 skeleton_debug_str_hash.
6027 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
6029 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
6031 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
6032 (emit_store_direct): Likewise.
6033 (arc_trampoline_adjust_address): Likewise.
6034 (arc_asm_trampoline_template): New function.
6035 (arc_initialize_trampoline): Use asm_trampoline_template.
6036 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
6037 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
6038 * config/arc/arc.md (flush_icache): Delete pattern.
6040 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
6042 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
6043 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
6046 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
6049 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
6050 by not USED_FOR_TARGET.
6051 (make_pass_resolve_sw_modes): Likewise.
6053 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
6055 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
6058 2018-01-08 Richard Biener <rguenther@suse.de>
6061 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
6063 2018-01-08 Richard Biener <rguenther@suse.de>
6066 * match.pd ((t * 2) / 2) -> t): Add missing :c.
6068 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
6071 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
6072 basic blocks with a small number of successors.
6073 (convert_control_dep_chain_into_preds): Improve handling of
6075 (dump_predicates): Split apart into...
6076 (dump_pred_chain): ...here...
6077 (dump_pred_info): ...and here.
6078 (can_one_predicate_be_invalidated_p): Add debugging printfs.
6079 (can_chain_union_be_invalidated_p): Improve check for invalidation
6081 (uninit_uses_cannot_happen): Avoid unnecessary if
6082 convert_control_dep_chain_into_preds yielded nothing.
6084 2018-01-06 Martin Sebor <msebor@redhat.com>
6086 PR tree-optimization/83640
6087 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
6088 subtracting negative offset from size.
6089 (builtin_access::overlap): Adjust offset bounds of the access to fall
6090 within the size of the object if possible.
6092 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
6094 PR rtl-optimization/83699
6095 * expmed.c (extract_bit_field_1): Restrict the vector usage of
6096 extract_bit_field_as_subreg to cases in which the extracted
6097 value is also a vector.
6099 * lra-constraints.c (process_alt_operands): Test for the equivalence
6100 substitutions when detecting a possible reload cycle.
6102 2018-01-06 Jakub Jelinek <jakub@redhat.com>
6105 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
6106 by default if flag_selective_schedling{,2}. Formatting fixes.
6108 PR rtl-optimization/83682
6109 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
6110 if it has non-VECTOR_MODE element mode.
6111 (vec_duplicate_p): Likewise.
6114 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
6115 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
6117 2018-01-05 Jakub Jelinek <jakub@redhat.com>
6120 * config/i386/i386-builtin.def
6121 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
6122 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
6123 Require also OPTION_MASK_ISA_AVX512F in addition to
6124 OPTION_MASK_ISA_GFNI.
6125 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
6126 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
6127 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
6128 to OPTION_MASK_ISA_GFNI.
6129 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
6130 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
6131 OPTION_MASK_ISA_AVX512BW.
6132 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
6133 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
6134 addition to OPTION_MASK_ISA_GFNI.
6135 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
6136 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
6137 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
6138 to OPTION_MASK_ISA_GFNI.
6139 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
6140 a requirement for all ISAs rather than any of them with a few
6142 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
6144 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
6145 bitmasks to be enabled with 3 exceptions, instead of requiring any
6146 enabled ISA with lots of exceptions.
6147 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
6148 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
6149 Change avx512bw in isa attribute to avx512f.
6150 * config/i386/sgxintrin.h: Add license boilerplate.
6151 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
6152 to __AVX512F__ and __AVX512VL to __AVX512VL__.
6153 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
6154 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
6156 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
6157 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
6158 temporarily sse2 rather than sse if not enabled already.
6161 * config/i386/sse.md (VI248_VLBW): Rename to ...
6162 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
6163 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
6164 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
6165 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
6166 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
6167 mode iterator instead of VI248_VLBW.
6169 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
6171 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
6172 (record_modified): Skip clobbers; add debug output.
6173 (param_change_prob): Use sreal frequencies.
6175 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
6177 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
6178 punt for user-aligned variables.
6180 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
6182 * tree-chrec.c (chrec_contains_symbols): Return true for
6185 2018-01-05 Sudakshina Das <sudi.das@arm.com>
6188 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
6189 of (x|y) == x for BICS pattern.
6191 2018-01-05 Jakub Jelinek <jakub@redhat.com>
6193 PR tree-optimization/83605
6194 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
6195 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
6198 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
6200 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
6201 * config/epiphany/rtems.h: New file.
6203 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6204 Uros Bizjak <ubizjak@gmail.com>
6207 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
6208 QIreg_operand instead of register_operand predicate.
6209 * config/i386/i386.c (ix86_rop_should_change_byte_p,
6210 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
6211 comments instead of -fmitigate[-_]rop.
6213 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
6216 * cgraphunit.c (symbol_table::compile): Switch to text_section
6217 before calling assembly_start debug hook.
6218 * run-rtl-passes.c (run_rtl_passes): Likewise.
6221 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6223 * tree-vrp.c (extract_range_from_binary_expr_1): Check
6224 range_int_cst_p rather than !symbolic_range_p before calling
6225 extract_range_from_multiplicative_op_1.
6227 2018-01-04 Jeff Law <law@redhat.com>
6229 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
6230 redundant test in assertion.
6232 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6234 * doc/rtl.texi: Document machine_mode wrapper classes.
6236 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6238 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
6241 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6243 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
6244 the VEC_PERM_EXPR fold to fail.
6246 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6249 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
6250 to switched_sections.
6252 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6255 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
6258 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
6261 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
6262 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
6264 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6267 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
6268 is BLKmode and bitpos not zero or mode change is needed.
6270 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
6273 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
6276 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
6279 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
6280 instead of MULT rtx. Update all corresponding splitters.
6282 (*ssub<modesuffix>): Ditto.
6284 (*cmp_sadd_di): Update split patterns.
6285 (*cmp_sadd_si): Ditto.
6286 (*cmp_sadd_sidi): Ditto.
6287 (*cmp_ssub_di): Ditto.
6288 (*cmp_ssub_si): Ditto.
6289 (*cmp_ssub_sidi): Ditto.
6290 * config/alpha/predicates.md (const23_operand): New predicate.
6291 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
6292 Look for ASHIFT, not MULT inner operand.
6293 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
6295 2018-01-04 Martin Liska <mliska@suse.cz>
6297 PR gcov-profile/83669
6298 * gcov.c (output_intermediate_file): Add version to intermediate
6300 * doc/gcov.texi: Document new field 'version' in intermediate
6301 file format. Fix location of '-k' option of gcov command.
6303 2018-01-04 Martin Liska <mliska@suse.cz>
6306 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
6308 2018-01-04 Jakub Jelinek <jakub@redhat.com>
6310 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
6312 2018-01-03 Martin Sebor <msebor@redhat.com>
6314 PR tree-optimization/83655
6315 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
6316 checking calls with invalid arguments.
6318 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6320 * tree-vect-stmts.c (vect_get_store_rhs): New function.
6321 (vectorizable_mask_load_store): Delete.
6322 (vectorizable_call): Return false for masked loads and stores.
6323 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
6324 instead of gimple_assign_rhs1.
6325 (vectorizable_load): Handle IFN_MASK_LOAD.
6326 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
6328 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6330 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
6332 (vectorizable_mask_load_store): ...here.
6333 (vectorizable_load): ...and here.
6335 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6337 * tree-vect-stmts.c (vect_build_all_ones_mask)
6338 (vect_build_zero_merge_argument): New functions, split out from...
6339 (vectorizable_load): ...here.
6341 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6343 * tree-vect-stmts.c (vect_check_store_rhs): New function,
6345 (vectorizable_mask_load_store): ...here.
6346 (vectorizable_store): ...and here.
6348 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6350 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
6352 (vectorizable_mask_load_store): ...here.
6354 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6356 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
6357 (vect_model_store_cost): Take a vec_load_store_type instead of a
6359 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
6360 (vect_model_store_cost): Take a vec_load_store_type instead of a
6362 (vectorizable_mask_load_store): Update accordingly.
6363 (vectorizable_store): Likewise.
6364 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
6366 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6368 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
6369 IFN_MASK_LOAD calls here rather than...
6370 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
6372 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6373 Alan Hayward <alan.hayward@arm.com>
6374 David Sherwood <david.sherwood@arm.com>
6376 * expmed.c (extract_bit_field_1): For vector extracts,
6377 fall back to extract_bit_field_as_subreg if vec_extract
6380 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6381 Alan Hayward <alan.hayward@arm.com>
6382 David Sherwood <david.sherwood@arm.com>
6384 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
6385 they are variable or constant sized.
6386 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
6387 slots for constant-sized data.
6389 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6390 Alan Hayward <alan.hayward@arm.com>
6391 David Sherwood <david.sherwood@arm.com>
6393 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
6394 handling COND_EXPRs with boolean comparisons, try to find a better
6395 basis for the mask type than the boolean itself.
6397 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6399 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
6400 is calculated and how it can be overridden.
6401 * genmodes.c (max_bitsize_mode_any_mode): New variable.
6402 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
6404 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
6407 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6408 Alan Hayward <alan.hayward@arm.com>
6409 David Sherwood <david.sherwood@arm.com>
6411 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
6412 Remove the mode argument.
6413 (aarch64_simd_valid_immediate): Remove the mode and inverse
6415 * config/aarch64/iterators.md (bitsize): New iterator.
6416 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
6417 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
6418 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
6419 aarch64_simd_valid_immediate.
6420 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
6421 (aarch64_reg_or_bic_imm): Likewise.
6422 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
6423 with an insn_type enum and msl with a modifier_type enum.
6424 Replace element_width with a scalar_mode. Change the shift
6425 to unsigned int. Add constructors for scalar_float_mode and
6426 scalar_int_mode elements.
6427 (aarch64_vect_float_const_representable_p): Delete.
6428 (aarch64_can_const_movi_rtx_p)
6429 (aarch64_simd_scalar_immediate_valid_for_move)
6430 (aarch64_simd_make_constant): Update call to
6431 aarch64_simd_valid_immediate.
6432 (aarch64_advsimd_valid_immediate_hs): New function.
6433 (aarch64_advsimd_valid_immediate): Likewise.
6434 (aarch64_simd_valid_immediate): Remove mode and inverse
6435 arguments. Rewrite to use the above. Use const_vec_duplicate_p
6436 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
6437 and aarch64_float_const_representable_p on the result.
6438 (aarch64_output_simd_mov_immediate): Remove mode argument.
6439 Update call to aarch64_simd_valid_immediate and use of
6440 simd_immediate_info.
6441 (aarch64_output_scalar_simd_mov_immediate): Update call
6444 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6445 Alan Hayward <alan.hayward@arm.com>
6446 David Sherwood <david.sherwood@arm.com>
6448 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
6449 (mode_nunits): Likewise CONST_MODE_NUNITS.
6450 * machmode.def (ADJUST_NUNITS): Document.
6451 * genmodes.c (mode_data::need_nunits_adj): New field.
6452 (blank_mode): Update accordingly.
6453 (adj_nunits): New variable.
6454 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
6456 (emit_mode_size_inline): Set need_bytesize_adj for all modes
6457 listed in adj_nunits.
6458 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
6459 listed in adj_nunits. Don't emit case statements for such modes.
6460 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
6461 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
6462 nothing if adj_nunits is nonnull.
6463 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
6464 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
6465 (emit_mode_fbit): Update use of print_maybe_const_decl.
6466 (emit_move_size): Likewise. Treat the array as non-const
6468 (emit_mode_adjustments): Handle adj_nunits.
6470 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6472 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
6473 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
6474 (VECTOR_MODES): Use it.
6475 (make_vector_modes): Take the prefix as an argument.
6477 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6478 Alan Hayward <alan.hayward@arm.com>
6479 David Sherwood <david.sherwood@arm.com>
6481 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
6482 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
6483 for MODE_VECTOR_BOOL.
6484 * machmode.def (VECTOR_BOOL_MODE): Document.
6485 * genmodes.c (VECTOR_BOOL_MODE): New macro.
6486 (make_vector_bool_mode): New function.
6487 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
6489 * lto-streamer-in.c (lto_input_mode_table): Likewise.
6490 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
6492 * stor-layout.c (int_mode_for_mode): Likewise.
6493 * tree.c (build_vector_type_for_mode): Likewise.
6494 * varasm.c (output_constant_pool_2): Likewise.
6495 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
6496 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
6497 for MODE_VECTOR_BOOL.
6498 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
6499 of mode class checks.
6500 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
6501 instead of a list of mode class checks.
6502 (expand_vector_scalar_condition): Likewise.
6503 (type_for_widest_vector_mode): Handle BImode as an inner mode.
6505 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6506 Alan Hayward <alan.hayward@arm.com>
6507 David Sherwood <david.sherwood@arm.com>
6509 * machmode.h (mode_size): Change from unsigned short to
6511 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
6512 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6513 or if measurement_type is not polynomial.
6514 (fixed_size_mode::includes_p): Check for constant-sized modes.
6515 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
6516 return a poly_uint16 rather than an unsigned short.
6517 (emit_mode_size): Change the type of mode_size from unsigned short
6518 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
6519 (emit_mode_adjustments): Cope with polynomial vector sizes.
6520 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6522 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6524 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
6525 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
6526 * caller-save.c (setup_save_areas): Likewise.
6527 (replace_reg_with_saved_mem): Likewise.
6528 * calls.c (emit_library_call_value_1): Likewise.
6529 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
6530 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
6531 (gen_lowpart_for_combine): Likewise.
6532 * convert.c (convert_to_integer_1): Likewise.
6533 * cse.c (equiv_constant, cse_insn): Likewise.
6534 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
6535 (cselib_subst_to_values): Likewise.
6536 * dce.c (word_dce_process_block): Likewise.
6537 * df-problems.c (df_word_lr_mark_ref): Likewise.
6538 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
6539 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
6540 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
6541 (rtl_for_decl_location): Likewise.
6542 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
6543 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
6544 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
6545 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
6546 (expand_expr_real_1): Likewise.
6547 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
6548 (pad_below): Likewise.
6549 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6550 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
6551 * ira.c (get_subreg_tracking_sizes): Likewise.
6552 * ira-build.c (ira_create_allocno_objects): Likewise.
6553 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
6554 (ira_sort_regnos_for_alter_reg): Likewise.
6555 * ira-costs.c (record_operand_costs): Likewise.
6556 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
6557 (resolve_simple_move): Likewise.
6558 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
6559 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
6560 (lra_constraints): Likewise.
6561 (CONST_POOL_OK_P): Reject variable-sized modes.
6562 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
6563 (add_pseudo_to_slot, lra_spill): Likewise.
6564 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6565 * optabs-query.c (get_best_extraction_insn): Likewise.
6566 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6567 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
6568 (expand_mult_highpart, valid_multiword_target_p): Likewise.
6569 * recog.c (offsettable_address_addr_space_p): Likewise.
6570 * regcprop.c (maybe_mode_change): Likewise.
6571 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
6572 * regrename.c (build_def_use): Likewise.
6573 * regstat.c (dump_reg_info): Likewise.
6574 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
6575 (find_reloads, find_reloads_subreg_address): Likewise.
6576 * reload1.c (eliminate_regs_1): Likewise.
6577 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
6578 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
6579 (simplify_binary_operation_1, simplify_subreg): Likewise.
6580 * targhooks.c (default_function_arg_padding): Likewise.
6581 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
6582 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
6583 (verify_gimple_assign_ternary): Likewise.
6584 * tree-inline.c (estimate_move_cost): Likewise.
6585 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6586 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
6587 (get_address_cost_ainc): Likewise.
6588 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
6589 (vect_supportable_dr_alignment): Likewise.
6590 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6591 (vectorizable_reduction): Likewise.
6592 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
6593 (vectorizable_operation, vectorizable_load): Likewise.
6594 * tree.c (build_same_sized_truth_vector_type): Likewise.
6595 * valtrack.c (cleanup_auto_inc_dec): Likewise.
6596 * var-tracking.c (emit_note_insn_var_location): Likewise.
6597 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
6598 (ADDR_VEC_ALIGN): Likewise.
6600 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6601 Alan Hayward <alan.hayward@arm.com>
6602 David Sherwood <david.sherwood@arm.com>
6604 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
6606 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6607 or if measurement_type is polynomial.
6608 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
6609 * combine.c (make_extraction): Likewise.
6610 * dse.c (find_shift_sequence): Likewise.
6611 * dwarf2out.c (mem_loc_descriptor): Likewise.
6612 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
6613 (extract_bit_field, extract_low_bits): Likewise.
6614 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
6615 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
6616 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
6617 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
6618 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6619 * reload.c (find_reloads): Likewise.
6620 * reload1.c (alter_reg): Likewise.
6621 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
6622 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
6623 * tree-if-conv.c (predicate_mem_writes): Likewise.
6624 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
6625 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
6626 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
6627 * valtrack.c (dead_debug_insert_temp): Likewise.
6628 * varasm.c (mergeable_constant_section): Likewise.
6629 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
6631 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6632 Alan Hayward <alan.hayward@arm.com>
6633 David Sherwood <david.sherwood@arm.com>
6635 * expr.c (expand_assignment): Cope with polynomial mode sizes
6636 when assigning to a CONCAT.
6638 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6639 Alan Hayward <alan.hayward@arm.com>
6640 David Sherwood <david.sherwood@arm.com>
6642 * machmode.h (mode_precision): Change from unsigned short to
6644 (mode_to_precision): Return a poly_uint16 rather than an unsigned
6646 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
6647 or if measurement_type is not polynomial.
6648 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
6649 in which the mode is already known to be a scalar_int_mode.
6650 * genmodes.c (emit_mode_precision): Change the type of mode_precision
6651 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
6653 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6654 for GET_MODE_PRECISION.
6655 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6656 for GET_MODE_PRECISION.
6657 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
6659 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
6660 (expand_field_assignment, make_extraction): Likewise.
6661 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
6662 (get_last_value): Likewise.
6663 * convert.c (convert_to_integer_1): Likewise.
6664 * cse.c (cse_insn): Likewise.
6665 * expr.c (expand_expr_real_1): Likewise.
6666 * lra-constraints.c (simplify_operand_subreg): Likewise.
6667 * optabs-query.c (can_atomic_load_p): Likewise.
6668 * optabs.c (expand_atomic_load): Likewise.
6669 (expand_atomic_store): Likewise.
6670 * ree.c (combine_reaching_defs): Likewise.
6671 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
6672 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
6673 * tree.h (type_has_mode_precision_p): Likewise.
6674 * ubsan.c (instrument_si_overflow): Likewise.
6676 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6677 Alan Hayward <alan.hayward@arm.com>
6678 David Sherwood <david.sherwood@arm.com>
6680 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
6681 polynomial numbers of units.
6682 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
6683 (valid_vector_subparts_p): New function.
6684 (build_vector_type): Remove temporary shim and take the number
6685 of units as a poly_uint64 rather than an int.
6686 (build_opaque_vector_type): Take the number of units as a
6687 poly_uint64 rather than an int.
6688 * tree.c (build_vector_from_ctor): Handle polynomial
6689 TYPE_VECTOR_SUBPARTS.
6690 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
6691 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
6692 (build_vector_from_val): If the number of units is variable,
6693 use build_vec_duplicate_cst for constant operands and
6694 VEC_DUPLICATE_EXPR otherwise.
6695 (make_vector_type): Remove temporary is_constant ().
6696 (build_vector_type, build_opaque_vector_type): Take the number of
6697 units as a poly_uint64 rather than an int.
6698 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
6700 * cfgexpand.c (expand_debug_expr): Likewise.
6701 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
6702 (store_constructor, expand_expr_real_1): Likewise.
6703 (const_scalar_mask_from_tree): Likewise.
6704 * fold-const-call.c (fold_const_reduction): Likewise.
6705 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
6706 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
6707 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
6708 (fold_relational_const): Likewise.
6709 (native_interpret_vector): Likewise. Change the size from an
6710 int to an unsigned int.
6711 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
6712 TYPE_VECTOR_SUBPARTS.
6713 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
6714 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
6715 duplicating a non-constant operand into a variable-length vector.
6716 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
6717 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
6718 * ipa-icf.c (sem_variable::equals): Likewise.
6719 * match.pd: Likewise.
6720 * omp-simd-clone.c (simd_clone_subparts): Likewise.
6721 * print-tree.c (print_node): Likewise.
6722 * stor-layout.c (layout_type): Likewise.
6723 * targhooks.c (default_builtin_vectorization_cost): Likewise.
6724 * tree-cfg.c (verify_gimple_comparison): Likewise.
6725 (verify_gimple_assign_binary): Likewise.
6726 (verify_gimple_assign_ternary): Likewise.
6727 (verify_gimple_assign_single): Likewise.
6728 * tree-pretty-print.c (dump_generic_node): Likewise.
6729 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6730 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
6731 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
6732 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
6733 (vect_shift_permute_load_chain): Likewise.
6734 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
6735 (expand_vector_condition, optimize_vector_constructor): Likewise.
6736 (lower_vec_perm, get_compute_type): Likewise.
6737 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6738 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6739 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6740 (vect_recog_mask_conversion_pattern): Likewise.
6741 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6742 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6743 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6744 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6745 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6746 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6747 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6748 (supportable_widening_operation): Likewise.
6749 (supportable_narrowing_operation): Likewise.
6750 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6752 * varasm.c (output_constant): Likewise.
6754 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6755 Alan Hayward <alan.hayward@arm.com>
6756 David Sherwood <david.sherwood@arm.com>
6758 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6759 so that both the length == 3 and length != 3 cases set up their
6760 own permute vectors. Add comments explaining why we know the
6761 number of elements is constant.
6762 (vect_permute_load_chain): Likewise.
6764 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6765 Alan Hayward <alan.hayward@arm.com>
6766 David Sherwood <david.sherwood@arm.com>
6768 * machmode.h (mode_nunits): Change from unsigned char to
6770 (ONLY_FIXED_SIZE_MODES): New macro.
6771 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6772 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6773 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6775 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6776 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6777 or if measurement_type is not polynomial.
6778 * genmodes.c (ZERO_COEFFS): New macro.
6779 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6781 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6782 Use ZERO_COEFFS when emitting initializers.
6783 * data-streamer.h (bp_pack_poly_value): New function.
6784 (bp_unpack_poly_value): Likewise.
6785 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6786 for GET_MODE_NUNITS.
6787 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6788 for GET_MODE_NUNITS.
6789 * tree.c (make_vector_type): Remove temporary shim and make
6790 the real function take the number of units as a poly_uint64
6792 (build_vector_type_for_mode): Handle polynomial nunits.
6793 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6794 * emit-rtl.c (const_vec_series_p_1): Likewise.
6795 (gen_rtx_CONST_VECTOR): Likewise.
6796 * fold-const.c (test_vec_duplicate_folding): Likewise.
6797 * genrecog.c (validate_pattern): Likewise.
6798 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6799 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6800 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6801 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6802 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6803 * rtlanal.c (subreg_get_info): Likewise.
6804 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6805 (vect_grouped_load_supported): Likewise.
6806 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6807 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6808 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6809 (simplify_const_unary_operation, simplify_binary_operation_1)
6810 (simplify_const_binary_operation, simplify_ternary_operation)
6811 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6812 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6813 instead of CONST_VECTOR_NUNITS.
6814 * varasm.c (output_constant_pool_2): Likewise.
6815 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6816 explicit-encoded elements in the XVEC for variable-length vectors.
6818 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6820 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6822 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6823 Alan Hayward <alan.hayward@arm.com>
6824 David Sherwood <david.sherwood@arm.com>
6826 * coretypes.h (fixed_size_mode): Declare.
6827 (fixed_size_mode_pod): New typedef.
6828 * builtins.h (target_builtins::x_apply_args_mode)
6829 (target_builtins::x_apply_result_mode): Change type to
6830 fixed_size_mode_pod.
6831 * builtins.c (apply_args_size, apply_result_size, result_vector)
6832 (expand_builtin_apply_args_1, expand_builtin_apply)
6833 (expand_builtin_return): Update accordingly.
6835 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6837 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6838 * cselib.c (cselib_hash_rtx): Likewise.
6839 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6840 CONST_VECTOR encoding.
6842 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6843 Jeff Law <law@redhat.com>
6846 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6847 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6848 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6849 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6852 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6853 explicitly probe *sp in a noreturn function if there were any callee
6854 register saves or frame pointer is needed.
6856 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6859 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6860 BLKmode for ternary, binary or unary expressions.
6863 * var-tracking.c (delete_vta_debug_insn): New inline function.
6864 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6865 insns from get_insns () to NULL instead of each bb separately.
6866 Use delete_vta_debug_insn. No longer static.
6867 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6868 delete_vta_debug_insns callers.
6869 * rtl.h (delete_vta_debug_insns): Declare.
6870 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6871 instead of variable_tracking_main.
6873 2018-01-03 Martin Sebor <msebor@redhat.com>
6875 PR tree-optimization/83603
6876 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6877 arguments past the endof the argument list in functions declared
6878 without a prototype.
6879 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6880 Avoid checking when arguments are null.
6882 2018-01-03 Martin Sebor <msebor@redhat.com>
6885 * doc/extend.texi (attribute const): Fix a typo.
6886 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6887 issuing -Wsuggest-attribute for void functions.
6889 2018-01-03 Martin Sebor <msebor@redhat.com>
6891 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6892 offset_int::from instead of wide_int::to_shwi.
6893 (maybe_diag_overlap): Remove assertion.
6894 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6895 * gimple-ssa-sprintf.c (format_directive): Same.
6896 (parse_directive): Same.
6897 (sprintf_dom_walker::compute_format_length): Same.
6898 (try_substitute_return_value): Same.
6900 2018-01-03 Jeff Law <law@redhat.com>
6903 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6904 non-constant residual for zero at runtime and avoid probing in
6905 that case. Reorganize code for trailing problem to mirror handling
6908 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6910 PR tree-optimization/83501
6911 * tree-ssa-strlen.c (get_string_cst): New.
6912 (handle_char_store): Call get_string_cst.
6914 2018-01-03 Martin Liska <mliska@suse.cz>
6916 PR tree-optimization/83593
6917 * tree-ssa-strlen.c: Include tree-cfg.h.
6918 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6919 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6920 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6922 (strlen_dom_walker::before_dom_children): Call
6923 gimple_purge_dead_eh_edges. Dump tranformation with details
6925 (strlen_dom_walker::before_dom_children): Update call by adding
6926 new argument cleanup_eh.
6927 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6929 2018-01-03 Martin Liska <mliska@suse.cz>
6932 * cif-code.def (VARIADIC_THUNK): New enum value.
6933 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6936 2018-01-03 Jan Beulich <jbeulich@suse.com>
6938 * sse.md (mov<mode>_internal): Tighten condition for when to use
6939 vmovdqu<ssescalarsize> for TI and OI modes.
6941 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6943 Update copyright years.
6945 2018-01-03 Martin Liska <mliska@suse.cz>
6948 * ipa-visibility.c (function_and_variable_visibility): Skip
6949 functions with noipa attribure.
6951 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6953 * gcc.c (process_command): Update copyright notice dates.
6954 * gcov-dump.c (print_version): Ditto.
6955 * gcov.c (print_version): Ditto.
6956 * gcov-tool.c (print_version): Ditto.
6957 * gengtype.c (create_file): Ditto.
6958 * doc/cpp.texi: Bump @copying's copyright year.
6959 * doc/cppinternals.texi: Ditto.
6960 * doc/gcc.texi: Ditto.
6961 * doc/gccint.texi: Ditto.
6962 * doc/gcov.texi: Ditto.
6963 * doc/install.texi: Ditto.
6964 * doc/invoke.texi: Ditto.
6966 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6968 * vector-builder.h (vector_builder::m_full_nelts): Change from
6969 unsigned int to poly_uint64.
6970 (vector_builder::full_nelts): Update prototype accordingly.
6971 (vector_builder::new_vector): Likewise.
6972 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6973 (vector_builder::operator ==): Likewise.
6974 (vector_builder::finalize): Likewise.
6975 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6976 Take the number of elements as a poly_uint64 rather than an
6978 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6979 from unsigned int to poly_uint64.
6980 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6981 (vec_perm_indices::new_vector): Likewise.
6982 (vec_perm_indices::length): Likewise.
6983 (vec_perm_indices::nelts_per_input): Likewise.
6984 (vec_perm_indices::input_nelts): Likewise.
6985 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6986 number of elements per input as a poly_uint64 rather than an
6987 unsigned int. Use the original encoding for variable-length
6988 vectors, rather than clamping each individual element.
6989 For the second and subsequent elements in each pattern,
6990 clamp the step and base before clamping their sum.
6991 (vec_perm_indices::series_p): Handle polynomial element counts.
6992 (vec_perm_indices::all_in_range_p): Likewise.
6993 (vec_perm_indices_to_tree): Likewise.
6994 (vec_perm_indices_to_rtx): Likewise.
6995 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6996 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6997 (tree_vector_builder::new_binary_operation): Handle polynomial
6998 element counts. Return false if we need to know the number
6999 of elements at compile time.
7000 * fold-const.c (fold_vec_perm): Punt if the number of elements
7001 isn't known at compile time.
7003 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7005 * vec-perm-indices.h (vec_perm_builder): Change element type
7006 from HOST_WIDE_INT to poly_int64.
7007 (vec_perm_indices::element_type): Update accordingly.
7008 (vec_perm_indices::clamp): Handle polynomial element_types.
7009 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7010 (vec_perm_indices::all_in_range_p): Likewise.
7011 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
7013 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
7014 polynomial vec_perm_indices element types.
7015 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
7016 * fold-const.c (fold_vec_perm): Likewise.
7017 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
7018 * tree-vect-generic.c (lower_vec_perm): Likewise.
7019 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7020 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
7021 element type to HOST_WIDE_INT.
7023 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7024 Alan Hayward <alan.hayward@arm.com>
7025 David Sherwood <david.sherwood@arm.com>
7027 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
7028 rather than an int. Use plus_constant.
7029 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
7030 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
7032 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7033 Alan Hayward <alan.hayward@arm.com>
7034 David Sherwood <david.sherwood@arm.com>
7036 * calls.c (emit_call_1, expand_call): Change struct_value_size from
7037 a HOST_WIDE_INT to a poly_int64.
7039 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7040 Alan Hayward <alan.hayward@arm.com>
7041 David Sherwood <david.sherwood@arm.com>
7043 * calls.c (load_register_parameters): Cope with polynomial
7044 mode sizes. Require a constant size for BLKmode parameters
7045 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
7046 forces a parameter to be padded at the lsb end in order to
7047 fill a complete number of words, require the parameter size
7048 to be ordered wrt UNITS_PER_WORD.
7050 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7051 Alan Hayward <alan.hayward@arm.com>
7052 David Sherwood <david.sherwood@arm.com>
7054 * reload1.c (spill_stack_slot_width): Change element type
7055 from unsigned int to poly_uint64_pod.
7056 (alter_reg): Treat mode sizes as polynomial.
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 * reload.c (complex_word_subreg_p): New function.
7063 (reload_inner_reg_of_subreg, push_reload): Use it.
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 * lra-constraints.c (process_alt_operands): Reject matched
7070 operands whose sizes aren't ordered.
7071 (match_reload): Refer to this check here.
7073 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7074 Alan Hayward <alan.hayward@arm.com>
7075 David Sherwood <david.sherwood@arm.com>
7077 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
7078 that the mode size is in the set {1, 2, 4, 8, 16}.
7080 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7081 Alan Hayward <alan.hayward@arm.com>
7082 David Sherwood <david.sherwood@arm.com>
7084 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
7085 Use plus_constant instead of gen_rtx_PLUS.
7087 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7088 Alan Hayward <alan.hayward@arm.com>
7089 David Sherwood <david.sherwood@arm.com>
7091 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
7092 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
7093 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
7094 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
7095 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
7096 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
7097 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
7098 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
7099 * config/i386/i386.c (ix86_push_rounding): ...this new function.
7100 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
7102 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
7103 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
7104 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
7105 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
7106 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
7107 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
7108 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
7109 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
7110 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
7111 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
7113 * expr.c (emit_move_resolve_push): Treat the input and result
7114 of PUSH_ROUNDING as a poly_int64.
7115 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
7116 (emit_push_insn): Likewise.
7117 * lra-eliminations.c (mark_not_eliminable): Likewise.
7118 * recog.c (push_operand): Likewise.
7119 * reload1.c (elimination_effects): Likewise.
7120 * rtlanal.c (nonzero_bits1): Likewise.
7121 * calls.c (store_one_arg): Likewise. Require the padding to be
7122 known at compile time.
7124 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7125 Alan Hayward <alan.hayward@arm.com>
7126 David Sherwood <david.sherwood@arm.com>
7128 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
7129 Use plus_constant instead of gen_rtx_PLUS.
7131 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7132 Alan Hayward <alan.hayward@arm.com>
7133 David Sherwood <david.sherwood@arm.com>
7135 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
7138 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7139 Alan Hayward <alan.hayward@arm.com>
7140 David Sherwood <david.sherwood@arm.com>
7142 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
7143 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
7144 via stack temporaries. Treat the mode size as polynomial too.
7146 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7147 Alan Hayward <alan.hayward@arm.com>
7148 David Sherwood <david.sherwood@arm.com>
7150 * expr.c (expand_expr_real_2): When handling conversions involving
7151 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
7152 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
7153 as a poly_uint64 too.
7155 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7156 Alan Hayward <alan.hayward@arm.com>
7157 David Sherwood <david.sherwood@arm.com>
7159 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
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 * combine.c (can_change_dest_mode): Handle polynomial
7166 REGMODE_NATURAL_SIZE.
7167 * expmed.c (store_bit_field_1): Likewise.
7168 * expr.c (store_constructor): Likewise.
7169 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
7170 and polynomial REGMODE_NATURAL_SIZE.
7171 (gen_lowpart_common): Likewise.
7172 * reginfo.c (record_subregs_of_mode): Likewise.
7173 * rtlanal.c (read_modify_subreg_p): Likewise.
7175 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7176 Alan Hayward <alan.hayward@arm.com>
7177 David Sherwood <david.sherwood@arm.com>
7179 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
7180 numbers of elements.
7182 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7183 Alan Hayward <alan.hayward@arm.com>
7184 David Sherwood <david.sherwood@arm.com>
7186 * match.pd: Cope with polynomial numbers of vector elements.
7188 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7189 Alan Hayward <alan.hayward@arm.com>
7190 David Sherwood <david.sherwood@arm.com>
7192 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
7193 in a POINTER_PLUS_EXPR.
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 * omp-simd-clone.c (simd_clone_subparts): New function.
7200 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
7201 (ipa_simd_modify_function_body): Likewise.
7203 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7204 Alan Hayward <alan.hayward@arm.com>
7205 David Sherwood <david.sherwood@arm.com>
7207 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
7208 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
7209 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
7210 (expand_vector_condition, vector_element): Likewise.
7211 (subparts_gt): New function.
7212 (get_compute_type): Use subparts_gt.
7213 (count_type_subparts): Delete.
7214 (expand_vector_operations_1): Use subparts_gt instead of
7215 count_type_subparts.
7217 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7218 Alan Hayward <alan.hayward@arm.com>
7219 David Sherwood <david.sherwood@arm.com>
7221 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
7222 (vect_compile_time_alias): ...this new function. Do the calculation
7223 on poly_ints rather than trees.
7224 (vect_prune_runtime_alias_test_list): Update call accordingly.
7226 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7227 Alan Hayward <alan.hayward@arm.com>
7228 David Sherwood <david.sherwood@arm.com>
7230 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
7232 (vect_schedule_slp_instance): Likewise.
7234 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7235 Alan Hayward <alan.hayward@arm.com>
7236 David Sherwood <david.sherwood@arm.com>
7238 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
7239 constant and extern definitions for variable-length vectors.
7240 (vect_get_constant_vectors): Note that the number of units
7241 is known to be constant.
7243 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7244 Alan Hayward <alan.hayward@arm.com>
7245 David Sherwood <david.sherwood@arm.com>
7247 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
7248 of units as polynomial. Choose between WIDE and NARROW based
7251 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7252 Alan Hayward <alan.hayward@arm.com>
7253 David Sherwood <david.sherwood@arm.com>
7255 * tree-vect-stmts.c (simd_clone_subparts): New function.
7256 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
7258 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7259 Alan Hayward <alan.hayward@arm.com>
7260 David Sherwood <david.sherwood@arm.com>
7262 * tree-vect-stmts.c (vectorizable_call): Treat the number of
7263 vectors as polynomial. Use build_index_vector for
7266 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7267 Alan Hayward <alan.hayward@arm.com>
7268 David Sherwood <david.sherwood@arm.com>
7270 * tree-vect-stmts.c (get_load_store_type): Treat the number of
7271 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
7272 for variable-length vectors.
7273 (vectorizable_mask_load_store): Treat the number of units as
7274 polynomial, asserting that it is constant if the condition has
7275 already been enforced.
7276 (vectorizable_store, vectorizable_load): Likewise.
7278 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7279 Alan Hayward <alan.hayward@arm.com>
7280 David Sherwood <david.sherwood@arm.com>
7282 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
7283 of units as polynomial. Punt if we can't tell at compile time
7284 which vector contains the final result.
7286 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7287 Alan Hayward <alan.hayward@arm.com>
7288 David Sherwood <david.sherwood@arm.com>
7290 * tree-vect-loop.c (vectorizable_induction): Treat the number
7291 of units as polynomial. Punt on SLP inductions. Use an integer
7292 VEC_SERIES_EXPR for variable-length integer reductions. Use a
7293 cast of such a series for variable-length floating-point
7296 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7297 Alan Hayward <alan.hayward@arm.com>
7298 David Sherwood <david.sherwood@arm.com>
7300 * tree.h (build_index_vector): Declare.
7301 * tree.c (build_index_vector): New function.
7302 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
7303 of units as polynomial, forcibly converting it to a constant if
7304 vectorizable_reduction has already enforced the condition.
7305 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
7306 to create a {1,2,3,...} vector.
7307 (vectorizable_reduction): Treat the number of units as polynomial.
7308 Choose vectype_in based on the largest scalar element size rather
7309 than the smallest number of units. Enforce the restrictions
7312 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7313 Alan Hayward <alan.hayward@arm.com>
7314 David Sherwood <david.sherwood@arm.com>
7316 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
7317 number of units as polynomial.
7319 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7320 Alan Hayward <alan.hayward@arm.com>
7321 David Sherwood <david.sherwood@arm.com>
7323 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
7324 * target.def (autovectorize_vector_sizes): Return the vector sizes
7325 by pointer, using vector_sizes rather than a bitmask.
7326 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
7327 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
7328 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
7330 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
7331 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
7332 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
7333 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
7334 * omp-general.c (omp_max_vf): Likewise.
7335 * omp-low.c (omp_clause_aligned_alignment): Likewise.
7336 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
7337 * tree-vect-loop.c (vect_analyze_loop): Likewise.
7338 * tree-vect-slp.c (vect_slp_bb): Likewise.
7339 * doc/tm.texi: Regenerate.
7340 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
7342 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
7343 the vector size as a poly_uint64 rather than an unsigned int.
7344 (current_vector_size): Change from an unsigned int to a poly_uint64.
7345 (get_vectype_for_scalar_type): Update accordingly.
7346 * tree.h (build_truth_vector_type): Take the size and number of
7347 units as a poly_uint64 rather than an unsigned int.
7348 (build_vector_type): Add a temporary overload that takes
7349 the number of units as a poly_uint64 rather than an unsigned int.
7350 * tree.c (make_vector_type): Likewise.
7351 (build_truth_vector_type): Take the number of units as a poly_uint64
7352 rather than an unsigned int.
7354 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7355 Alan Hayward <alan.hayward@arm.com>
7356 David Sherwood <david.sherwood@arm.com>
7358 * target.def (get_mask_mode): Take the number of units and length
7359 as poly_uint64s rather than unsigned ints.
7360 * targhooks.h (default_get_mask_mode): Update accordingly.
7361 * targhooks.c (default_get_mask_mode): Likewise.
7362 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
7363 * doc/tm.texi: Regenerate.
7365 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7366 Alan Hayward <alan.hayward@arm.com>
7367 David Sherwood <david.sherwood@arm.com>
7369 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
7370 * omp-general.c (omp_max_vf): Likewise.
7371 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
7372 (expand_omp_simd): Handle polynomial safelen.
7373 * omp-low.c (omplow_simd_context): Add a default constructor.
7374 (omplow_simd_context::max_vf): Change from int to poly_uint64.
7375 (lower_rec_simd_input_clauses): Update accordingly.
7376 (lower_rec_input_clauses): Likewise.
7378 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7379 Alan Hayward <alan.hayward@arm.com>
7380 David Sherwood <david.sherwood@arm.com>
7382 * tree-vectorizer.h (vect_nunits_for_cost): New function.
7383 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
7384 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
7385 (vect_analyze_slp_cost): Likewise.
7386 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
7387 (vect_model_load_cost): Likewise.
7389 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7390 Alan Hayward <alan.hayward@arm.com>
7391 David Sherwood <david.sherwood@arm.com>
7393 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
7394 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
7395 from an unsigned int * to a poly_uint64_pod *.
7396 (calculate_unrolling_factor): New function.
7397 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
7399 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7400 Alan Hayward <alan.hayward@arm.com>
7401 David Sherwood <david.sherwood@arm.com>
7403 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
7404 from an unsigned int to a poly_uint64.
7405 (_loop_vec_info::slp_unrolling_factor): Likewise.
7406 (_loop_vec_info::vectorization_factor): Change from an int
7408 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
7409 (vect_get_num_vectors): New function.
7410 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
7411 (vect_get_num_copies): Use vect_get_num_vectors.
7412 (vect_analyze_data_ref_dependences): Change max_vf from an int *
7413 to an unsigned int *.
7414 (vect_analyze_data_refs): Change min_vf from an int * to a
7416 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7417 than an unsigned HOST_WIDE_INT.
7418 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
7419 (vect_analyze_data_ref_dependence): Change max_vf from an int *
7420 to an unsigned int *.
7421 (vect_analyze_data_ref_dependences): Likewise.
7422 (vect_compute_data_ref_alignment): Handle polynomial vf.
7423 (vect_enhance_data_refs_alignment): Likewise.
7424 (vect_prune_runtime_alias_test_list): Likewise.
7425 (vect_shift_permute_load_chain): Likewise.
7426 (vect_supportable_dr_alignment): Likewise.
7427 (dependence_distance_ge_vf): Take the vectorization factor as a
7428 poly_uint64 rather than an unsigned HOST_WIDE_INT.
7429 (vect_analyze_data_refs): Change min_vf from an int * to a
7431 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
7432 vfm1 as a poly_uint64 rather than an int. Make the same change
7433 for the returned bound_scalar.
7434 (vect_gen_vector_loop_niters): Handle polynomial vf.
7435 (vect_do_peeling): Likewise. Update call to
7436 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
7437 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
7439 * tree-vect-loop.c (vect_determine_vectorization_factor)
7440 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
7441 (vect_get_known_peeling_cost): Likewise.
7442 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
7443 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
7444 (vect_transform_loop): Likewise. Use the lowest possible VF when
7445 updating the upper bounds of the loop.
7446 (vect_min_worthwhile_factor): Make static. Return an unsigned int
7448 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
7449 polynomial unroll factors.
7450 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
7451 (vect_make_slp_decision): Likewise.
7452 (vect_supported_load_permutation_p): Likewise, and polynomial
7454 (vect_analyze_slp_cost): Handle polynomial vf.
7455 (vect_slp_analyze_node_operations): Likewise.
7456 (vect_slp_analyze_bb_1): Likewise.
7457 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
7458 than an unsigned HOST_WIDE_INT.
7459 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
7460 (vectorizable_load): Handle polynomial vf.
7461 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
7463 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
7465 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7466 Alan Hayward <alan.hayward@arm.com>
7467 David Sherwood <david.sherwood@arm.com>
7469 * match.pd: Handle bit operations involving three constants
7470 and try to fold one pair.
7472 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
7474 * tree-vect-loop-manip.c: Include gimple-fold.h.
7475 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
7476 niters_maybe_zero parameters. Handle other cases besides a step of 1.
7477 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
7478 Add a path that uses a step of VF instead of 1, but disable it
7480 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
7481 and niters_no_overflow parameters. Update calls to
7482 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
7483 Create a new SSA name if the latter choses to use a ste other
7484 than zero, and return it via niters_vector_mult_vf_var.
7485 * tree-vect-loop.c (vect_transform_loop): Update calls to
7486 vect_do_peeling, vect_gen_vector_loop_niters and
7487 slpeel_make_loop_iterate_ntimes.
7488 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
7489 (vect_gen_vector_loop_niters): Update declarations after above changes.
7491 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
7493 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
7494 128-bit round to integer instructions.
7495 (ceil<mode>2): Likewise.
7496 (btrunc<mode>2): Likewise.
7497 (round<mode>2): Likewise.
7499 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7501 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
7502 unaligned VSX load/store on P8/P9.
7503 (expand_block_clear): Allow the use of unaligned VSX
7504 load/store on P8/P9.
7506 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
7508 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
7510 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
7511 swap associated with both a load and a store.
7513 2018-01-02 Andrew Waterman <andrew@sifive.com>
7515 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
7516 * config/riscv/riscv.md (clear_cache): Use it.
7518 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
7520 * web.c: Remove out-of-date comment.
7522 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7524 * expr.c (fixup_args_size_notes): Check that any existing
7525 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
7526 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
7527 (emit_single_push_insn): ...here.
7529 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7531 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
7532 (const_vector_encoded_nelts): New function.
7533 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
7534 (const_vector_int_elt, const_vector_elt): Declare.
7535 * emit-rtl.c (const_vector_int_elt_1): New function.
7536 (const_vector_elt): Likewise.
7537 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
7538 of CONST_VECTOR_ELT.
7540 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7542 * expr.c: Include rtx-vector-builder.h.
7543 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
7544 directly on the tree encoding.
7545 (const_vector_from_tree): Likewise.
7546 * optabs.c: Include rtx-vector-builder.h.
7547 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
7548 sequence of "u" values.
7549 * vec-perm-indices.c: Include rtx-vector-builder.h.
7550 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
7551 directly on the vec_perm_indices encoding.
7553 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7555 * doc/rtl.texi (const_vector): Describe new encoding scheme.
7556 * Makefile.in (OBJS): Add rtx-vector-builder.o.
7557 * rtx-vector-builder.h: New file.
7558 * rtx-vector-builder.c: Likewise.
7559 * rtl.h (rtx_def::u2): Add a const_vector field.
7560 (CONST_VECTOR_NPATTERNS): New macro.
7561 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
7562 (CONST_VECTOR_DUPLICATE_P): Likewise.
7563 (CONST_VECTOR_STEPPED_P): Likewise.
7564 (CONST_VECTOR_ENCODED_ELT): Likewise.
7565 (const_vec_duplicate_p): Check for a duplicated vector encoding.
7566 (unwrap_const_vec_duplicate): Likewise.
7567 (const_vec_series_p): Check for a non-duplicated vector encoding.
7568 Say that the function only returns true for integer vectors.
7569 * emit-rtl.c: Include rtx-vector-builder.h.
7570 (gen_const_vec_duplicate_1): Delete.
7571 (gen_const_vector): Call gen_const_vec_duplicate instead of
7572 gen_const_vec_duplicate_1.
7573 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
7574 (gen_const_vec_duplicate): Use rtx_vector_builder.
7575 (gen_const_vec_series): Likewise.
7576 (gen_rtx_CONST_VECTOR): Likewise.
7577 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
7578 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7579 Build a new vector rather than modifying a CONST_VECTOR in-place.
7580 (handle_special_swappables): Update call accordingly.
7581 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
7582 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7583 Build a new vector rather than modifying a CONST_VECTOR in-place.
7584 (handle_special_swappables): Update call accordingly.
7586 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7588 * simplify-rtx.c (simplify_const_binary_operation): Use
7589 CONST_VECTOR_ELT instead of XVECEXP.
7591 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7593 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
7594 the selector elements to be different from the data elements
7595 if the selector is a VECTOR_CST.
7596 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
7597 ssizetype for the selector.
7599 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7601 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
7602 before testing each element individually.
7603 * tree-vect-generic.c (lower_vec_perm): Likewise.
7605 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7607 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
7608 * selftest-run-tests.c (selftest::run_tests): Call it.
7609 * vector-builder.h (vector_builder::operator ==): New function.
7610 (vector_builder::operator !=): Likewise.
7611 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
7612 (vec_perm_indices::all_from_input_p): New function.
7613 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7614 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
7615 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
7616 instead of reading the VECTOR_CST directly. Detect whether both
7617 vector inputs are the same before constructing the vec_perm_indices,
7618 and update the number of inputs argument accordingly. Use the
7619 utility functions added above. Only construct sel2 if we need to.
7621 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7623 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
7624 the broadcast of the low byte.
7625 (expand_mult_highpart): Use an explicit encoding for the permutes.
7626 * optabs-query.c (can_mult_highpart_p): Likewise.
7627 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
7628 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7629 (vectorizable_bswap): Likewise.
7630 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
7631 explicit encoding for the power-of-2 permutes.
7632 (vect_permute_store_chain): Likewise.
7633 (vect_grouped_load_supported): Likewise.
7634 (vect_permute_load_chain): Likewise.
7636 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7638 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
7639 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
7640 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
7641 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7642 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
7643 (vect_gen_perm_mask_any): Likewise.
7645 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7647 * int-vector-builder.h: New file.
7648 * vec-perm-indices.h: Include int-vector-builder.h.
7649 (vec_perm_indices): Redefine as an int_vector_builder.
7650 (auto_vec_perm_indices): Delete.
7651 (vec_perm_builder): Redefine as a stand-alone class.
7652 (vec_perm_indices::vec_perm_indices): New function.
7653 (vec_perm_indices::clamp): Likewise.
7654 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
7655 (vec_perm_indices::new_vector): New function.
7656 (vec_perm_indices::new_expanded_vector): Update for new
7657 vec_perm_indices class.
7658 (vec_perm_indices::rotate_inputs): New function.
7659 (vec_perm_indices::all_in_range_p): Operate directly on the
7660 encoded form, without computing elided elements.
7661 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
7662 encoding. Update for new vec_perm_indices class.
7663 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
7664 the given vec_perm_builder.
7665 (expand_vec_perm_var): Update vec_perm_builder constructor.
7666 (expand_mult_highpart): Use vec_perm_builder instead of
7667 auto_vec_perm_indices.
7668 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
7669 vec_perm_indices instead of auto_vec_perm_indices. Use a single
7670 or double series encoding as appropriate.
7671 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
7672 vec_perm_indices instead of auto_vec_perm_indices.
7673 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7674 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7675 (vect_permute_store_chain): Likewise.
7676 (vect_grouped_load_supported): Likewise.
7677 (vect_permute_load_chain): Likewise.
7678 (vect_shift_permute_load_chain): Likewise.
7679 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7680 (vect_transform_slp_perm_load): Likewise.
7681 (vect_schedule_slp_instance): Likewise.
7682 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7683 (vectorizable_mask_load_store): Likewise.
7684 (vectorizable_bswap): Likewise.
7685 (vectorizable_store): Likewise.
7686 (vectorizable_load): Likewise.
7687 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
7688 vec_perm_indices instead of auto_vec_perm_indices. Use
7689 tree_to_vec_perm_builder to read the vector from a tree.
7690 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
7691 vec_perm_builder instead of a vec_perm_indices.
7692 (have_whole_vector_shift): Use vec_perm_builder and
7693 vec_perm_indices instead of auto_vec_perm_indices. Leave the
7694 truncation to calc_vec_perm_mask_for_shift.
7695 (vect_create_epilog_for_reduction): Likewise.
7696 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
7697 from auto_vec_perm_indices to vec_perm_indices.
7698 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7699 instead of changing individual elements.
7700 (aarch64_vectorize_vec_perm_const): Use new_vector to install
7701 the vector in d.perm.
7702 * config/arm/arm.c (expand_vec_perm_d::perm): Change
7703 from auto_vec_perm_indices to vec_perm_indices.
7704 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7705 instead of changing individual elements.
7706 (arm_vectorize_vec_perm_const): Use new_vector to install
7707 the vector in d.perm.
7708 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
7709 Update vec_perm_builder constructor.
7710 (rs6000_expand_interleave): Likewise.
7711 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
7712 (rs6000_expand_interleave): Likewise.
7714 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7716 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
7717 to qimode could truncate the indices.
7718 * optabs.c (expand_vec_perm_var): Likewise.
7720 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7722 * Makefile.in (OBJS): Add vec-perm-indices.o.
7723 * vec-perm-indices.h: New file.
7724 * vec-perm-indices.c: Likewise.
7725 * target.h (vec_perm_indices): Replace with a forward class
7727 (auto_vec_perm_indices): Move to vec-perm-indices.h.
7728 * optabs.h: Include vec-perm-indices.h.
7729 (expand_vec_perm): Delete.
7730 (selector_fits_mode_p, expand_vec_perm_var): Declare.
7731 (expand_vec_perm_const): Declare.
7732 * target.def (vec_perm_const_ok): Replace with...
7733 (vec_perm_const): ...this new hook.
7734 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
7735 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7736 * doc/tm.texi: Regenerate.
7737 * optabs.def (vec_perm_const): Delete.
7738 * doc/md.texi (vec_perm_const): Likewise.
7739 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7740 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7741 expand_vec_perm for constant permutation vectors. Assert that
7742 the mode of variable permutation vectors is the integer equivalent
7743 of the mode that is being permuted.
7744 * optabs-query.h (selector_fits_mode_p): Declare.
7745 * optabs-query.c: Include vec-perm-indices.h.
7746 (selector_fits_mode_p): New function.
7747 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7748 is defined, instead of checking whether the vec_perm_const_optab
7749 exists. Use targetm.vectorize.vec_perm_const instead of
7750 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7751 fit in the vector mode before using a variable permute.
7752 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7753 vec_perm_indices instead of an rtx.
7754 (expand_vec_perm): Replace with...
7755 (expand_vec_perm_const): ...this new function. Take the selector
7756 as a vec_perm_indices rather than an rtx. Also take the mode of
7757 the selector. Update call to shift_amt_for_vec_perm_mask.
7758 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7759 Use vec_perm_indices::new_expanded_vector to expand the original
7760 selector into bytes. Check whether the indices fit in the vector
7761 mode before using a variable permute.
7762 (expand_vec_perm_var): Make global.
7763 (expand_mult_highpart): Use expand_vec_perm_const.
7764 * fold-const.c: Includes vec-perm-indices.h.
7765 * tree-ssa-forwprop.c: Likewise.
7766 * tree-vect-data-refs.c: Likewise.
7767 * tree-vect-generic.c: Likewise.
7768 * tree-vect-loop.c: Likewise.
7769 * tree-vect-slp.c: Likewise.
7770 * tree-vect-stmts.c: Likewise.
7771 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7773 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7774 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7775 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7776 (aarch64_vectorize_vec_perm_const): ...this new function.
7777 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7778 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7779 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7780 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7781 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7782 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7783 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7785 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7786 check for NEON modes.
7787 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7788 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7789 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7790 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7792 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7793 the old VEC_PERM_CONST conditions.
7794 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7795 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7796 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7797 (ia64_vectorize_vec_perm_const_ok): Merge into...
7798 (ia64_vectorize_vec_perm_const): ...this new function.
7799 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7800 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7801 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7802 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7803 * config/mips/mips.c (mips_expand_vec_perm_const)
7804 (mips_vectorize_vec_perm_const_ok): Merge into...
7805 (mips_vectorize_vec_perm_const): ...this new function.
7806 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7807 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7808 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7809 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7810 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7811 (rs6000_expand_vec_perm_const): Delete.
7812 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7814 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7815 (altivec_expand_vec_perm_const_le): Take each operand individually.
7816 Operate on constant selectors rather than rtxes.
7817 (altivec_expand_vec_perm_const): Likewise. Update call to
7818 altivec_expand_vec_perm_const_le.
7819 (rs6000_expand_vec_perm_const): Delete.
7820 (rs6000_vectorize_vec_perm_const_ok): Delete.
7821 (rs6000_vectorize_vec_perm_const): New function.
7822 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7823 an element count and rtx array.
7824 (rs6000_expand_extract_even): Update call accordingly.
7825 (rs6000_expand_interleave): Likewise.
7826 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7827 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7828 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7829 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7830 (rs6000_expand_vec_perm_const): Delete.
7831 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7832 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7833 (altivec_expand_vec_perm_const_le): Take each operand individually.
7834 Operate on constant selectors rather than rtxes.
7835 (altivec_expand_vec_perm_const): Likewise. Update call to
7836 altivec_expand_vec_perm_const_le.
7837 (rs6000_expand_vec_perm_const): Delete.
7838 (rs6000_vectorize_vec_perm_const_ok): Delete.
7839 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7840 reference to the SPE evmerge intructions.
7841 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7842 an element count and rtx array.
7843 (rs6000_expand_extract_even): Update call accordingly.
7844 (rs6000_expand_interleave): Likewise.
7845 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7846 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7848 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7850 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7852 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7853 vector mode and that that mode matches the mode of the data
7855 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7856 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7857 directly using expand_vec_perm_1 when forcing selectors into
7859 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7861 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7863 * optabs-query.h (can_vec_perm_p): Delete.
7864 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7865 * optabs-query.c (can_vec_perm_p): Split into...
7866 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7867 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7868 particular selector is valid.
7869 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7870 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7871 (vect_grouped_load_supported): Likewise.
7872 (vect_shift_permute_load_chain): Likewise.
7873 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7874 (vect_transform_slp_perm_load): Likewise.
7875 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7876 (vectorizable_bswap): Likewise.
7877 (vect_gen_perm_mask_checked): Likewise.
7878 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7879 implementations of variable permutation vectors into account
7880 when deciding which selector to use.
7881 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7882 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7883 with a false third argument.
7884 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7885 to test whether the constant selector is valid and can_vec_perm_var_p
7886 to test whether a variable selector is valid.
7888 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7890 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7891 * optabs-query.c (can_vec_perm_p): Likewise.
7892 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7893 instead of vec_perm_indices.
7894 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7895 (vect_gen_perm_mask_checked): Likewise,
7896 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7897 (vect_gen_perm_mask_checked): Likewise,
7899 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7901 * optabs-query.h (qimode_for_vec_perm): Declare.
7902 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7903 (qimode_for_vec_perm): ...this new function.
7904 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7906 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7908 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7909 does not have a conditional at the top.
7911 2018-01-02 Richard Biener <rguenther@suse.de>
7913 * ipa-inline.c (big_speedup_p): Fix expression.
7915 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7918 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7921 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7925 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7926 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7927 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7928 cond_taken_branch_cost 3->4.
7930 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7932 PR tree-optimization/83581
7933 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7934 TODO_cleanup_cfg if any changes have been made.
7937 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7938 convert_modes if target mode has the right side, but different mode
7942 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7943 last argument when extracting from CONCAT. If either from_real or
7944 from_imag is NULL, use expansion through memory. If result is not
7945 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7946 the parts directly to inner mode, if even that fails, use expansion
7950 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7951 check for bswap in mode rather than HImode and use that in expand_unop
7954 Copyright (C) 2018 Free Software Foundation, Inc.
7956 Copying and distribution of this file, with or without modification,
7957 are permitted in any medium without royalty provided the copyright
7958 notice and this notice are preserved.