1 2018-02-21 Martin Liska <mliska@suse.cz>
4 * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
5 Add "native" as a possible value.
6 * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
7 when native cpu detection is available.
9 2018-02-21 Jakub Jelinek <jakub@redhat.com>
10 Martin Sebor <msebor@redhat.com>
12 PR tree-optimization/84478
13 * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
15 * gimple-fold.c (get_range_strlen): Make minlen const and assume it
16 can't be NULL. Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
17 support which is conservatively correct, for 2 only stay conservative
18 for maxlen. Formatting and comment capitalization fixes. Add STRICT
19 argument to the 2 argument get_range_strlen, adjust 6 arg
20 get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
22 (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
23 (gimple_fold_builtin_strlen): Pass true as last argument to
26 2018-02-20 Martin Sebor <msebor@redhat.com>
29 * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
30 (builtin_memref::set_base_and_offset): Same. Handle inner references.
31 (builtin_memref::builtin_memref): Factor out parts into
32 set_base_and_offset and call it.
34 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
37 * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
38 is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
39 greater precision. If to_mode is a MODE_PARTIAL_INT, stop the
40 search at the associated MODE_INT.
42 2018-02-20 Jeff Law <law@redhat.com>
45 PR tree-optimization/81592
47 * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
48 for range data rather than using global data.
49 * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
50 range data rather than using global data.
51 * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
52 pass it to children as needed.
53 (struct directive::fmtresult): Similarly.
54 (struct directive::set_width): Similarly.
55 (struct directive::set_precision): Similarly.
56 (format_integer, format_directive, parse_directive): Similarly.
57 (format_none): Accept unnamed vr_values parameter.
58 (format_percent, format_floating, format_character): Similarly.
59 (format_string, format_plain): Similarly.
60 * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
61 the EVRP range analyzer for range data rather than using global data.
62 * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
63 gimple-ssa-evrp-analyze.h
64 (class sprintf_dom_walker): Add after_dom_children member function.
65 Add evrp_range_analyzer member.
66 (sprintf_dom_walker::before_dom_children): Call into the EVRP
67 range analyzer as needed.
68 (sprintf_dom_walker::after_dom_children): New member function.
69 * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
71 (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
72 (evrp_range_analyzer::pop_to_marker): Likewise.
74 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
76 PR tree-optimization/84419
77 * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
78 with the required type if its current type is compatible but
81 2018-02-20 Jakub Jelinek <jakub@redhat.com>
84 * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
87 2018-02-20 Martin Liska <mliska@suse.cz>
90 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
91 possible values if we don't have a hint.
93 2018-02-20 Martin Liska <mliska@suse.cz>
97 * final.c (shorten_branches): Build align_tab array with one
99 * opts.c (finish_options): Add alignment option limit check.
100 (MAX_CODE_ALIGN): Likewise.
101 (MAX_CODE_ALIGN_VALUE): Likewise.
102 * doc/invoke.texi: Document maximum allowed option value for
103 all -falign-* options.
105 2018-02-19 Jakub Jelinek <jakub@redhat.com>
108 * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
109 * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
110 * var-tracking.c (emit_note_insn_var_location): Remove all references
111 to NOTE_INSN_CALL_ARG_LOCATION.
112 (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
113 the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
114 Use copy_rtx_if_shared.
115 * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
116 NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
117 (dwarf2out_var_location): Remove handling of
118 NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
120 * final.c (final_scan_insn): Remove all references to
121 NOTE_INSN_CALL_ARG_LOCATION.
122 (rest_of_clean_state): Likewise. Remove REG_CALL_ARG_LOCATION notes
123 before dumping final insns.
124 * except.c (emit_note_eh_region_end): Remove all references to
125 NOTE_INSN_CALL_ARG_LOCATION.
126 * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
127 * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
128 * config/arc/arc.c (hwloop_optimize): Likewise.
129 * config/arm/arm.c (create_fix_barrier): Likewise.
130 * config/s390/s390.c (s390_chunkify_start): Likewise.
131 * config/sh/sh.c (find_barrier): Likewise.
132 * config/i386/i386.c (rest_of_insert_endbranch,
133 ix86_seh_fixup_eh_fallthru): Likewise.
134 * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
135 * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
136 * config/frv/frv.c (frv_function_prologue): Likewise.
137 * emit-rtl.c (try_split): Likewise. Copy over REG_CALL_ARG_LOCATION
139 (note_outside_basic_block_p): Remove all references to
140 NOTE_INSN_CALL_ARG_LOCATION.
141 * gengtype.c (adjust_field_rtx_def): Likewise.
142 * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
144 * jump.c (cleanup_barriers, delete_related_insns): Likewise.
145 * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
148 * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
151 PR tree-optimization/84452
152 * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
153 expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
156 2018-02-19 Martin Liska <mliska@suse.cz>
159 * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
161 2018-02-19 Martin Liska <mliska@suse.cz>
162 Richard Sandiford <richard.sandiford@linaro.org>
164 PR tree-optimization/82491
165 * gimple-fold.c (get_base_constructor): Make earlier bail out
168 2018-02-19 Carl Love <cel@us.ibm.com>
170 * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
171 BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
173 * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
176 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
178 * config/rl78/rl78.md (movdf): New define expand.
180 2018-02-19 Martin Liska <mliska@suse.cz>
183 * doc/invoke.texi: Fix typo.
184 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
186 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
188 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
189 handle rs6000_single_float and rs6000_double_float specially for
192 2018-02-16 Jeff Law <law@redhat.com>
194 * config/rx/rx.c (add_pop_cfi_notes): New function.;
197 2018-02-16 Jakub Jelinek <jakub@redhat.com>
200 * ipa-inline.c (inline_small_functions): Fix a typo.
202 2018-02-16 Nathan Sidwell <nathan@acm.org>
204 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
206 2018-02-16 Carl Love <cel@us.ibm.com>
208 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
209 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
210 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
211 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
212 expansion to P8V_BUILTIN_VEC_FLOAT2.
214 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
216 PR rtl-optimization/70023
217 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
218 src_regno into account.
220 2018-02-16 Carl Love <cel@us.ibm.com>
222 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
223 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
224 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
225 * config/rs6000/rs6000.c: Remove case statements for
226 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
227 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
228 and P9V_BUILTIN_VEC_VINSERT4B.
229 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
230 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
231 * config/rs6000/vsx.md:
232 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
235 2018-02-16 Carl Love <cel@us.ibm.com>
237 * config/rs6000/altivec.h: Add builtin names vec_extract4b
239 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
241 * config/rs6000/rs6000-c.c: Add the definitions for
242 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
243 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
244 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
245 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
246 definition for insert4b and define insn *insert3b_internal.
247 * doc/extend.texi: Add documentation for vec_extract4b.
249 2018-02-16 Nathan Sidwell <nathan@acm.org>
251 * doc/extend.texi (Backwards Compatibility): Mention friend
252 injection. Note for-scope is deprecated.
253 * doc/invoke.texi (-ffriend-injection): Deprecate.
255 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
257 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
258 that moved to I2, also allow destinations that are a paradoxical
259 subreg (instead of a normal reg).
261 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
264 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
267 2018-02-16 Richard Biener <rguenther@suse.de>
269 PR tree-optimization/84037
270 PR tree-optimization/84016
272 * config/i386/i386.c (ix86_builtin_vectorization_cost):
273 Adjust vec_construct for the fact we need additional higher latency
274 128bit inserts for AVX256 and AVX512 vector builds.
275 (ix86_add_stmt_cost): Scale vector construction cost for
278 2018-02-16 Richard Biener <rguenther@suse.de>
280 PR tree-optimization/84417
281 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
282 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
283 (non_rewritable_lvalue_p): Likewise, use poly-ints.
285 2018-02-16 Martin Liska <mliska@suse.cz>
288 * internal-fn.def (ASAN_CHECK): Set proper flags.
289 (ASAN_MARK): Likewise.
291 2018-02-16 Julia Koval <julia.koval@intel.com>
293 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
296 2018-02-16 Jakub Jelinek <jakub@redhat.com>
299 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
300 Use ++iter rather than iter++ for std::list iterators.
301 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
302 defer deleting them until all nodes in the forest are processed. Do
303 free even leaf nodes. Change to_process into auto_vec.
306 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
307 * vec.h (vec_default_construct): Use memset instead of placement new
308 if BROKEN_VALUE_INITIALIZATION is defined.
309 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
310 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
313 PR rtl-optimization/83723
314 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
315 * lra.c (lra_substitute_pseudo): Likewise. If true, use
316 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
318 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
320 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
322 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
324 PR rtl-optimization/81443
325 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
326 from inner REGs to paradoxical SUBREGs.
328 2018-02-16 Richard Biener <rguenther@suse.de>
330 PR tree-optimization/84399
331 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
332 For operands we can analyze at their definition make sure we can
333 analyze them at each use as well.
335 2018-02-16 Richard Biener <rguenther@suse.de>
337 PR tree-optimization/84190
338 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
339 volatile accesses if the decl isn't volatile.
341 2018-02-15 Jason Merrill <jason@redhat.com>
343 PR c++/84314 - ICE with templates and fastcall attribute.
344 * attribs.c (build_type_attribute_qual_variant): Don't clobber
345 TYPE_CANONICAL on an existing type.
347 2018-02-15 Jakub Jelinek <jakub@redhat.com>
349 PR tree-optimization/84383
350 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
351 dstoff nor call operand_equal_p if dstbase is NULL.
353 PR tree-optimization/84334
354 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
355 also a CONSTANT_CLASS_P, punt.
357 2018-02-14 Jim Wilson <jimw@sifive.com>
359 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
360 first SMALL_OPERAND check. New local min_second_step. Move assert
361 to where locals are set. Add TARGET_RVC support.
362 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
364 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
366 * doc/invoke.texi: Correct -Wformat-overflow code sample.
368 2018-02-14 Martin Sebor <msebor@redhat.com>
370 PR tree-optimization/83698
371 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
372 arrays constrain the offset range to their bounds.
373 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
374 (builtin_access::overlap): Avoid setting the size of overlap if it's
376 (maybe_diag_overlap): Also consider arrays when deciding what values
377 of offsets to include in diagnostics.
379 2018-02-14 Martin Sebor <msebor@redhat.com>
382 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
383 that correspond to the kind of a declaration.
385 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
388 * config/pa/pa.md: Load address of PIC label using the linkage table
389 if the label is nonlocal.
391 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
393 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
394 warning message if user requests -maltivec=be.
395 * doc/invoke.texi: Document deprecation of -maltivec=be.
397 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
400 * config/rs6000/rs6000-c.c: Update definitions for
401 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
402 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
404 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
407 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
408 add _get_ssp intrinsics. Remove argument from
409 __builtin_ia32_rdssp[d|q].
410 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
411 * config/i386/i386-builtin.def: Remove argument from
412 __builtin_ia32_rdssp[d|q].
413 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
414 ix86_expand_special_args_builtin for _rdssp[d|q].
415 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
416 Clear register before usage.
417 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
418 Add documentation for new _get_ssp and _inc_ssp intrinsics.
420 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
422 PR tree-optimization/84357
423 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
424 operand 1 of an ARRAY_REF too.
426 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
429 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
430 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
432 (set_of_reg): New struct.
433 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
434 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
435 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
437 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
438 Split into bitclr, bitset, bitinvert patterns if appropriate.
439 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
440 use rx_fuse_in_memory_bitop.
441 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
442 to named insn, correct maximum insn length.
444 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
447 * machmode.def: Define a complex mode for PARTIAL_INT.
448 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
450 * doc/rtl.texi: Document CSPImode.
451 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
453 (msp430_hard_regno_nregs_with_padding): Likewise.
455 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
458 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
460 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
462 PR rtl-optimization/84169
463 * combine.c (try_combine): New variable split_i2i3. Set it to true if
464 we generated a parallel as new i3 and we split that to new i2 and i3
465 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
466 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
467 those to i2, not i1. Partially rewrite this scan code.
469 2018-02-13 Jakub Jelinek <jakub@redhat.com>
472 * stor-layout.c (place_field): For variable length fields, adjust
473 offset_align afterwards not just based on the field's alignment,
474 but also on the size.
477 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
478 of exps and logs in the use_exp2 case.
480 2018-02-13 Jeff Law <law@redhat.com>
482 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
485 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
488 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
492 * common.opt (gas-loc-support, gas-locview-support): New.
493 (ginline-points, ginternal-reset-location-views): New.
494 * doc/invoke.texi: Document them. Use @itemx where intended.
495 (gvariable-location-views): Adjust.
496 * target.def (reset_location_view): New.
497 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
498 (TARGET_RESET_LOCATION_VIEW): New.
499 * doc/tm.texi: Rebuilt.
500 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
501 (dwarf2out_default_as_locview_support): New.
502 (output_asm_line_debug_info): Use option variables.
503 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
504 (output_loc_list): Likewise.
505 (add_high_low_attributes): Check option variables.
506 Don't output entry view attribute in strict mode.
507 (gen_inlined_subroutine_die): Check option variables.
508 (dwarf2out_inline_entry): Likewise.
509 (init_sections_and_labels): Likewise.
510 (dwarf2out_early_finish): Likewise.
511 (maybe_reset_location_view): New, from...
512 (dwarf2out_var_location): ... here. Call it.
513 * debug.h (dwarf2out_default_as_loc_support): Declare.
514 (dwarf2out_default_as_locview_support): Declare.
515 * hooks.c (hook_int_rtx_insn_0): New.
516 * hooks.h (hook_int_rtx_insn_0): Declare.
517 * toplev.c (process_options): Take -gas-loc-support and
518 -gas-locview-support from dwarf2out. Enable
519 -gvariable-location-views by default only with locview
520 assembler support. Enable -ginternal-reset-location-views by
521 default only if the target defines the corresponding hook.
522 Enable -ginline-points by default if location views are
523 enabled; force it disabled if statement frontiers are
525 * tree-inline.c (expand_call_inline): Check option variables.
526 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
528 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
530 PR tree-optimization/84321
531 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
532 handling. Also check whether the anti-range contains any values
533 that satisfy the mask; switch to a VR_RANGE if not.
535 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
538 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
540 2018-02-13 Martin Jambor <mjambor@suse.cz>
543 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
544 of call statements, also set location of a load to a temporary.
546 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
548 * config/rl78/rl78.c (add_vector_labels): New function.
549 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
550 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
551 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
552 which checks that no arguments are passed.
553 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
554 * doc/extend.texi: Documentation for the new attribute.
556 2018-02-13 Andreas Schwab <schwab@suse.de>
558 * config/riscv/linux.h (CPP_SPEC): Define.
560 2018-02-13 Jakub Jelinek <jakub@redhat.com>
563 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
564 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
565 OPTION_MASK_ISA_AES as first argument to def_builtin_const
566 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
567 instead of OPTION_MASK_ISA_PCLMUL as first argument to
568 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
569 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
570 temporarily for AES and PCLMUL builtins.
572 PR tree-optimization/84339
573 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
574 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
578 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
579 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
580 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
582 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
584 * omp-simd-clone.h: New file.
585 * omp-simd-clone.c: Include omp-simd-clone.h.
586 (expand_simd_clones): No longer static.
587 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
588 cgraph.h and omp-simd-clone.h.
589 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
590 (vect_recog_widen_shift_pattern): Formatting fix.
591 (vect_pattern_recog_1): Don't check optab for calls.
594 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
595 operands[2] into a REG before using gen_lowpart on it.
597 2018-02-12 Jeff Law <law@redhat.com>
600 * config/sh/sh.c (find_barrier): Consider a sibling call
603 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
604 successfully back substituting a reg.
606 2018-02-12 Richard Biener <rguenther@suse.de>
608 PR tree-optimization/84037
609 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
610 parameter, move visited init to caller.
611 (vect_slp_analyze_operations): Separate cost from validity
612 check, initialize visited once for all instances.
613 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
615 * tree-vect-stmts.c (vect_model_simple_cost): Make early
617 (vect_model_promotion_demotion_cost): Likewise.
618 (vectorizable_bswap): Guard cost modeling with !slp_node
619 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
621 (vectorizable_call): Likewise.
622 (vectorizable_conversion): Likewise.
623 (vectorizable_assignment): Likewise.
624 (vectorizable_shift): Likewise.
625 (vectorizable_operation): Likewise.
626 (vectorizable_store): Likewise.
627 (vectorizable_load): Likewise.
628 (vectorizable_condition): Likewise.
629 (vectorizable_comparison): Likewise.
631 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
634 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
635 (ASAN_MARK): Fix fnspec to account for return value, change pointer
636 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
638 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
641 * params.def (inline-min-speedup): Increase from 8 to 15.
642 (max-inline-insns-auto): Decrease from 40 to 30.
643 * ipa-split.c (consider_split): Add some buffer for function to
644 be considered inlining candidate.
645 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
648 2018-02-12 Richard Biener <rguenther@suse.de>
650 PR tree-optimization/84037
651 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
652 matched stmts if we cannot swap the non-matched ones.
654 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
656 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
657 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
658 _mm_maskz_scalef_round_ss): New intrinsics.
659 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
660 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
661 __builtin_ia32_scalefss_round): Remove.
662 (__builtin_ia32_scalefsd_mask_round,
663 __builtin_ia32_scalefss_mask_round): New intrinsics.
664 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
665 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
666 ((match_operand:VF_128 2 "<round_nimm_predicate>"
667 "<round_constraint>")): Changed to ...
668 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
669 "<round_scalar_constraint>")): ... this.
670 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
671 %0, %1, %2<round_op3>}"): Changed to ...
672 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
673 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
674 %2<round_scalar_mask_op3>}"): ... this.
675 * config/i386/subst.md (round_scalar_nimm_predicate): New.
677 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
679 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
680 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
681 (_mm_maskz_sqrt_round_ss): New intrinsics.
682 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
683 (__builtin_ia32_sqrtsd_mask_round)
684 (__builtin_ia32_sqrtss_mask_round): New builtins.
685 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
686 (__builtin_ia32_sqrtss_round): Remove.
687 (__builtin_ia32_sqrtsd_mask_round)
688 (__builtin_ia32_sqrtss_mask_round): New builtins.
689 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
690 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
691 ((match_operand:VF_128 1 "vector_operand"
692 "xBm,<round_constraint>")): Changed to ...
693 ((match_operand:VF_128 1 "vector_operand"
694 "xBm,<round_scalar_constraint>")): ... this.
695 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
696 %0, %2, %<iptr>1<round_op3>}): Changed to ...
697 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
698 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
699 %<iptr>1<round_scalar_mask_op3>}): ... this.
700 ((set_attr "prefix" "<round_prefix>")): Changed to ...
701 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
703 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
706 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
707 Cast vec_cmpeq result to correct type.
708 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
709 Cast vec_cmpgt result to correct type.
711 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
713 * final.c (final_scan_insn_1): Renamed from...
714 (final_scan_insn): ... this. New wrapper, to recover
715 seen from the outermost call in recursive ones.
716 * config/sparc/sparc.c (output_return): Drop seen from call.
717 (output_sibcall): Likewise.
718 * config/visium/visium.c (output_branch): Likewise.
720 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
722 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
725 2018-02-10 Alan Modra <amodra@gmail.com>
728 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
729 Specify LR as an input.
731 2018-02-10 Jakub Jelinek <jakub@redhat.com>
734 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
735 remove_member_access_dummy_vars): New functions.
736 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
737 lower_omp_1, execute_lower_omp): Use them.
739 PR rtl-optimization/84308
740 * shrink-wrap.c (spread_components): Release todo vector.
742 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
744 PR rtl-optimization/57193
745 * ira-color.c (struct allocno_color_data): Add member
746 conflict_allocno_hard_prefs.
747 (update_conflict_allocno_hard_prefs): New.
748 (bucket_allocno_compare_func): Add a preference based on
749 conflict_allocno_hard_prefs.
750 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
751 (color_allocnos): Remove a dead code. Initiate
752 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
754 2018-02-09 Jakub Jelinek <jakub@redhat.com>
757 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
758 constraint from =wa to wa. Avoid a subreg on the output operand,
759 instead use a pseudo and subreg it in a move.
760 (p9_xxbrd_<mode>): Changed to ...
761 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
762 (p9_xxbrd_v2df): New expander.
763 (p9_xxbrw_<mode>): Changed to ...
764 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
765 (p9_xxbrw_v4sf): New expander.
767 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
769 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
771 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
774 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
775 multiply in 32-bit mode.
776 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
777 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
780 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
782 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
783 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
784 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
785 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
787 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
790 * dwarf2out.c (is_trivial_indirect_ref): New function.
791 (dwarf2out_late_global_decl): Do not generate a location
792 attribute for variables that have a non-trivial DECL_VALUE_EXPR
793 and that are not defined in the current unit.
795 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
797 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
798 instead of a libcall for UNORDERED.
800 2018-02-09 Tamar Christina <tamar.christina@arm.com>
803 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
804 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
806 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
809 * config/s390/s390.c (s390_set_current_function): Invoke
810 s390_indirect_branch_settings also if fndecl didn't change.
812 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
814 * config/rs6000/rs6000.md (blockage): Set length to zero.
816 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
818 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
820 2018-02-09 Jakub Jelinek <jakub@redhat.com>
823 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
824 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
828 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
829 PARALLEL incoming that failed vt_get_decl_and_offset check.
832 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
833 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
835 (get_variable_section): For decls in named .bss* sections pass true as
836 second argument to bss_initializer_p.
838 2018-02-09 Marek Polacek <polacek@redhat.com>
839 Jakub Jelinek <jakub@redhat.com>
842 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
843 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
844 Sync some changes from cxx_fold_indirect_ref.
846 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
848 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
850 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
851 (BLOCK_INLINE_ENTRY_LABEL): New.
852 (dwarf2out_var_location): Disregard inline entry markers.
853 (inline_entry_data): New struct.
854 (inline_entry_data_hasher): New hashtable type.
855 (inline_entry_data_hasher::hash): New.
856 (inline_entry_data_hasher::equal): New.
857 (inline_entry_data_table): New variable.
858 (add_high_low_attributes): Add DW_AT_entry_pc and
859 DW_AT_GNU_entry_view attributes if a pending entry is found
860 in inline_entry_data_table. Add old entry_pc attribute only
861 if debug nonbinding markers are disabled.
862 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
864 (block_within_block_p, dwarf2out_inline_entry): New.
865 (dwarf2out_finish): Check that no entries remained in
866 inline_entry_data_table.
867 * final.c (reemit_insn_block_notes): Handle inline entry notes.
868 (final_scan_insn, notice_source_line): Likewise.
869 (rest_of_clean_state): Skip inline entry markers.
870 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
872 * gimple.c (gimple_build_debug_inline_entry): New.
873 * gimple.h (enum gimple_debug_subcode): Add
874 GIMPLE_DEBUG_INLINE_ENTRY.
875 (gimple_build_debug_inline_entry): Declare.
876 (gimple_debug_inline_entry_p): New.
877 (gimple_debug_nonbind_marker_p): Adjust.
878 * insn-notes.def (INLINE_ENTRY): New.
879 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
880 inline entry marker notes.
881 (print_insn): Likewise.
882 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
883 (INSN_DEBUG_MARKER_KIND): Likewise.
884 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
885 * tree-inline.c (expand_call_inline): Build and insert
886 debug_inline_entry stmt.
887 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
888 inline entry blocks early, if nonbind markers are enabled.
889 (dump_scope_block): Dump fragment info.
890 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
891 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
892 (gimple_build_debug_inline_entry): New.
893 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
894 Enable/disable inline entry points too.
895 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
896 (DEBUG_INSN): Describe inline entry markers.
898 * common.opt (gvariable-location-views): New.
899 (gvariable-location-views=incompat5): New.
900 * config.in: Rebuilt.
901 * configure: Rebuilt.
902 * configure.ac: Test assembler for view support.
903 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
904 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
905 * dwarf2out.c (var_loc_view): New typedef.
906 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
907 (dwarf2out_locviews_in_attribute): New.
908 (dwarf2out_locviews_in_loclist): New.
909 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
910 (enum dw_line_info_opcode): Add LI_adv_address.
911 (struct dw_line_info_table): Add view.
912 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
913 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
914 (zero_view_p): New variable.
915 (ZERO_VIEW_P): New macro.
916 (output_asm_line_debug_info): New.
917 (struct var_loc_node): Add view.
918 (add_AT_view_list, AT_loc_list): New.
919 (add_var_loc_to_decl): Add view param. Test it against last.
920 (new_loc_list): Add view params. Record them.
921 (AT_loc_list_ptr): Handle loc and view lists.
922 (view_list_to_loc_list_val_node): New.
923 (print_dw_val): Handle dw_val_class_view_list.
924 (size_of_die): Likewise.
925 (value_format): Likewise.
926 (loc_list_has_views): New.
927 (gen_llsym): Set vl_symbol too.
928 (maybe_gen_llsym, skip_loc_list_entry): New.
929 (dwarf2out_maybe_output_loclist_view_pair): New.
930 (output_loc_list): Output view list or entries too.
931 (output_view_list_offset): New.
932 (output_die): Handle dw_val_class_view_list.
933 (output_dwarf_version): New.
934 (output_compilation_unit_header): Use it.
935 (output_skeleton_debug_sections): Likewise.
936 (output_rnglists, output_line_info): Likewise.
937 (output_pubnames, output_aranges): Update version comments.
938 (output_one_line_info_table): Output view numbers in asm comments.
939 (dw_loc_list): Determine current endview, pass it to new_loc_list.
940 Call maybe_gen_llsym.
941 (loc_list_from_tree_1): Adjust.
942 (add_AT_location_description): Create view list attribute if
943 needed, check it's absent otherwise.
944 (convert_cfa_to_fb_loc_list): Adjust.
945 (maybe_emit_file): Call output_asm_line_debug_info for test.
946 (dwarf2out_var_location): Reset views as needed. Precompute
947 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
949 (new_line_info_table): Reset next view.
950 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
951 (dwarf2out_source_line): Likewise. Output view resets and labels to
952 the assembler, or select appropriate line info opcodes.
953 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
954 (optimize_string_length): Catch it. Adjust.
955 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
956 dw_val_class_view_list, and remove it if no longer needed.
957 (hash_loc_list): Hash view numbers.
958 (loc_list_hasher::equal): Compare them.
959 (optimize_location_lists): Check whether a view list symbol is
960 needed, and whether the locview attribute is present, and
961 whether they match. Remove the locview attribute if no longer
963 (index_location_lists): Call skip_loc_list_entry for test.
964 (dwarf2out_finish): Call output_asm_line_debug_info for test.
965 Use output_dwarf_version.
966 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
967 (struct dw_val_node): Add val_view_list.
968 * final.c (SEEN_NEXT_VIEW): New.
969 (set_next_view_needed): New.
970 (clear_next_view_needed): New.
971 (maybe_output_next_view): New.
972 (final_start_function): Rename to...
973 (final_start_function_1): ... this. Take pointer to FIRST,
974 add SEEN parameter. Emit param bindings in the initial view.
975 (final_start_function): Reintroduce SEEN-less interface.
976 (final): Rename to...
977 (final_1): ... this. Take SEEN parameter. Output final pending
978 next view at the end.
979 (final): Reintroduce seen-less interface.
980 (final_scan_insn): Output pending next view before switching
981 sections or ending a block. Mark the next view as needed when
982 outputting variable locations. Notify debug backend of section
983 changes, and of location view changes.
984 (rest_of_handle_final): Adjust.
985 * toplev.c (process_options): Autodetect value for debug variable
986 location views option. Warn on incompat5 without -gdwarf-5.
987 * doc/invoke.texi (gvariable-location-views): New.
988 (gvariable-location-views=incompat5): New.
989 (gno-variable-location-views): New.
991 2018-02-08 David Malcolm <dmalcolm@redhat.com>
993 PR tree-optimization/84136
994 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
995 that the result of find_edge is non-NULL.
997 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1000 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
1001 storing integer register in SImode. Fix cost of 256 and 512
1002 byte aligned SSE register store.
1004 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1006 * config/i386/i386.c (ix86_multiplication_cost): Fix
1007 multiplication cost for TARGET_AVX512DQ.
1009 2018-02-08 Marek Polacek <polacek@redhat.com>
1011 PR tree-optimization/84238
1012 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
1015 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1017 PR tree-optimization/84265
1018 * tree-vect-stmts.c (vectorizable_store): Don't treat
1019 VMAT_CONTIGUOUS accesses as grouped.
1020 (vectorizable_load): Likewise.
1022 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1024 PR tree-optimization/81635
1025 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
1026 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
1027 (test_round_for_mask): New functions.
1028 (wide_int_cc_tests): Call test_round_for_mask.
1029 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
1030 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
1031 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
1032 range returned by get_range_info.
1034 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1037 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
1038 * symtab.c: Include builtins.h
1039 (symtab_node::output_to_lto_symbol_table_p): Move here
1040 from lto-streamer-out.c:output_symbol_p.
1041 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
1042 (output_symbol_p): Move all logic to symtab.c
1043 (produce_symtab): Update.
1045 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1047 * config/s390/s390-opts.h (enum indirect_branch): Define.
1048 * config/s390/s390-protos.h (s390_return_addr_from_memory)
1049 (s390_indirect_branch_via_thunk)
1050 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
1051 (enum s390_indirect_branch_type): Define.
1052 * config/s390/s390.c (struct s390_frame_layout, struct
1053 machine_function): Remove.
1054 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
1055 (indirect_branch_table_label_no, indirect_branch_table_name):
1057 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
1058 (enum s390_indirect_branch_option): Define.
1059 (s390_return_addr_from_memory): New function.
1060 (s390_handle_string_attribute): New function.
1061 (s390_attribute_table): Add new attribute handler.
1062 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
1063 (s390_indirect_branch_via_thunk): New function.
1064 (s390_indirect_branch_via_inline_thunk): New function.
1065 (s390_function_ok_for_sibcall): When jumping via thunk disallow
1066 sibling call optimization for non z10 compiles.
1067 (s390_emit_call): Force indirect branch target to be a single
1068 register. Add r1 clobber for non-z10 compiles.
1069 (s390_emit_epilogue): Emit return jump via return_use expander.
1070 (s390_reorg): Handle JUMP_INSNs as execute targets.
1071 (s390_option_override_internal): Perform validity checks for the
1072 new command line options.
1073 (s390_indirect_branch_attrvalue): New function.
1074 (s390_indirect_branch_settings): New function.
1075 (s390_set_current_function): Invoke s390_indirect_branch_settings.
1076 (s390_output_indirect_thunk_function): New function.
1077 (s390_code_end): Implement target hook.
1078 (s390_case_values_threshold): Implement target hook.
1079 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
1081 * config/s390/s390.h (struct s390_frame_layout)
1082 (struct machine_function): Move here from s390.c.
1083 (TARGET_INDIRECT_BRANCH_NOBP_RET)
1084 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
1085 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
1086 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
1087 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
1088 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
1089 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
1090 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
1091 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
1092 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
1093 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
1094 (mnemonic attribute): Add values which aren't recognized
1096 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
1097 pattern for branch conversion. Fix mnemonic attribute.
1098 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
1099 indirect branch via thunk if requested.
1100 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
1101 ("*indirect_jump"): Disable for branch conversion using out of
1103 ("indirect_jump_via_thunk<mode>_z10")
1104 ("indirect_jump_via_thunk<mode>")
1105 ("indirect_jump_via_inlinethunk<mode>_z10")
1106 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
1107 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
1108 ("casesi_jump_via_inlinethunk<mode>_z10")
1109 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
1110 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
1111 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
1112 ("*indirect2_jump"): Disable for branch conversion.
1113 ("casesi_jump"): Turn into expander and expand patterns for branch
1115 ("return_use"): New expander.
1116 ("*return"): Emit return via thunk and rename it to ...
1117 ("*return<mode>"): ... this one.
1118 * config/s390/s390.opt: Add new options and and enum for the
1121 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1123 * lra-constraints.c (match_reload): Unconditionally use
1124 gen_lowpart_SUBREG, rather than selecting between that
1125 and equivalent gen_rtx_SUBREG code.
1127 2018-02-08 Richard Biener <rguenther@suse.de>
1129 PR tree-optimization/84233
1130 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
1131 changed flag instead of boguously re-using phi_inserted.
1133 2018-02-08 Martin Jambor <mjambor@suse.cz>
1135 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
1136 static local variables.
1138 2018-02-08 Richard Biener <rguenther@suse.de>
1140 PR tree-optimization/84278
1141 * tree-vect-stmts.c (vectorizable_store): When looking for
1142 smaller vector types to perform grouped strided loads/stores
1143 make sure the mode is supported by the target.
1144 (vectorizable_load): Likewise.
1146 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1148 * config/aarch64/aarch64.c (aarch64_components_for_bb):
1149 Increase LDP/STP opportunities by adding adjacent callee-saves.
1151 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1153 PR rtl-optimization/84068
1154 PR rtl-optimization/83459
1155 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
1157 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
1159 PR tree-optimization/84224
1160 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
1161 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
1164 2018-02-07 Iain Sandoe <iain@codesourcery.com>
1167 * config/rs6000/altivec.md (*restore_world): Remove LR use.
1168 * config/rs6000/predicates.md (restore_world_operation): Adjust op
1169 count, remove one USE.
1171 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
1173 * doc/install.texi (Configuration): Document the
1174 --with-long-double-format={ibm,ieee} PowerPC configuration
1178 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
1179 Convert from define_expand to be define_insn_and_split. Rework
1180 float/double/_Float128 conversions to QI/HI/SImode to work with
1181 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
1182 conversions to QI/HImode types did a store and then a load to
1183 truncate the value. For conversions to VSX registers, don't split
1184 the insn, instead emit the code directly. Use the code iterator
1185 any_fix to combine signed and unsigned conversions.
1186 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
1187 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
1188 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1189 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
1190 (fix_<mode>di2_hw): Likewise.
1191 (fixuns_<mode>di2_hw): Likewise.
1192 (fix_<mode>si2_hw): Likewise.
1193 (fixuns_<mode>si2_hw): Likewise.
1194 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
1195 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1196 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1197 fix<uns>_trunc<SFDF:mode>si2_p8.
1198 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1200 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1201 (fix<uns>_<mode>_mem): Likewise.
1202 (fctiw<u>z_<mode>_mem): Likewise.
1203 (fix<uns>_<mode>_mem): Likewise.
1204 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1205 the register allocator from doing a direct move to the GPRs to do
1206 a store, and instead use the ISA 3.0 store byte/half-word from
1207 vector register instruction. For IEEE 128-bit floating point,
1208 also optimize stores of 32-bit ints.
1209 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1211 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1213 * genextract.c (push_pathstr_operand): New function to support
1215 (walk_rtx): Call push_pathstr_operand.
1216 (print_path): Support [a-zA-Z].
1218 2018-02-07 Richard Biener <rguenther@suse.de>
1220 PR tree-optimization/84037
1221 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1222 (cse_and_gimplify_to_preheader): Declare.
1223 (vect_get_place_in_interleaving_chain): Likewise.
1224 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1226 (_loop_vec_info::~_loop_vec_info): Delete it.
1227 (cse_and_gimplify_to_preheader): New function.
1228 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1229 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1230 (vectorizable_load): Likewise. For grouped stores always base
1231 the IV on the first element.
1232 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1233 condition before gimplifying.
1235 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1237 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1238 *DIV_EXPR and *MOD_EXPR.
1240 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1243 * config/i386/i386.c (ix86_option_override_internal): Mask out
1244 the CF_SET bit when checking -fcf-protection.
1246 2018-02-07 Tom de Vries <tom@codesourcery.com>
1249 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1252 2018-02-07 Richard Biener <rguenther@suse.de>
1254 PR tree-optimization/84204
1255 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1258 PR tree-optimization/84205
1259 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1260 special-case isl_ast_op_zdiv_r.
1262 PR tree-optimization/84223
1263 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1264 Only add conditions from within the region.
1265 (gather_bbs::after_dom_children): Adjust.
1267 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1270 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1271 * config/avr/avr.md: Only post-reload split REG-REG moves if
1272 either register is REGERAL_REG_P.
1274 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1276 PR tree-optimization/84235
1277 * tree-ssa-scopedtables.c
1278 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1279 if the subtraction is performed in floating point type where NaNs are
1280 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1281 build 1. Formatting fix.
1283 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1286 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1287 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1288 and skip it regardless of bb boundaries. Use CALL_P macro,
1289 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1290 unnecessarily, formatting fix.
1292 2018-02-06 Michael Collison <michael.collison@arm.com>
1294 * config/arm/thumb2.md:
1295 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1296 (*thumb_mov_notscc): Ditto.
1298 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1301 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1302 unsigned_fix, not "s".
1304 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1306 * configure.ac (gcc_fn_eh_frame_ro): New function.
1307 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1308 correct .eh_frame permissions.
1309 * configure: Regenerate.
1311 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1313 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1316 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1318 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1319 Display warning message for -mno-speculate-indirect-jumps.
1321 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1323 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1325 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1327 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1329 PR tree-optimization/84225
1330 * tree-eh.c (find_trapping_overflow): Only call
1331 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1333 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1336 * config/i386/i386.c: Reimplement the check of possible options
1337 -mibt/-mshstk conbination. Change error messages.
1338 * doc/invoke.texi: Fix a typo: remove extra '='.
1340 2018-02-06 Marek Polacek <polacek@redhat.com>
1342 PR tree-optimization/84228
1343 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1345 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1348 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1349 emitted arch directives.
1350 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1351 __ARM_FEATURE_COPROC before changing architectures.
1353 2018-02-06 Richard Biener <rguenther@suse.de>
1355 * config/i386/i386.c (print_reg): Fix typo.
1356 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1358 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1360 * configure: Regenerate.
1362 2018-02-05 Martin Sebor <msebor@redhat.com>
1364 PR tree-optimization/83369
1365 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1368 2018-02-05 Martin Liska <mliska@suse.cz>
1370 * doc/invoke.texi: Cherry-pick upstream r323995.
1372 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1374 * ira.c (ira_init_register_move_cost): Adjust comment.
1376 2018-02-05 Martin Liska <mliska@suse.cz>
1378 PR gcov-profile/84137
1379 * doc/gcov.texi: Fix typo in documentation.
1381 2018-02-05 Martin Liska <mliska@suse.cz>
1383 PR gcov-profile/83879
1384 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1385 using dlopen functionality.
1387 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1389 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1390 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1391 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1392 _mm_maskz_range_round_ss): New intrinsics.
1393 (__builtin_ia32_rangesd128_round)
1394 (__builtin_ia32_rangess128_round): Remove.
1395 (__builtin_ia32_rangesd128_mask_round,
1396 __builtin_ia32_rangess128_mask_round): New builtins.
1397 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1398 __builtin_ia32_rangess128_round): Remove.
1399 (__builtin_ia32_rangesd128_mask_round,
1400 __builtin_ia32_rangess128_mask_round): New builtins.
1401 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1402 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1403 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1404 "<round_saeonly_constraint>")): Changed to ...
1405 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1406 "<round_saeonly_scalar_constraint>")): ... this.
1407 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1408 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1409 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1410 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1411 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1413 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1415 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1417 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1418 Remove all values except native, 8540 and 8548.
1420 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1422 * config/i386/i386.c (ix86_output_function_return): Pass
1423 INVALID_REGNUM, instead of -1, as invalid register number to
1424 indirect_thunk_name and output_indirect_thunk.
1426 2018-02-02 Julia Koval <julia.koval@intel.com>
1428 * config.gcc: Add -march=icelake.
1429 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1430 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1431 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1432 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1433 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1434 (processor_target_table): Add icelake.
1435 (ix86_option_override_internal): Handle new PTAs.
1436 (get_builtin_code_for_version): Handle icelake.
1437 (M_INTEL_COREI7_ICELAKE): New.
1438 (fold_builtin_cpu): Handle icelake.
1439 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1440 * doc/invoke.texi: Add -march=icelake.
1442 2018-02-02 Julia Koval <julia.koval@intel.com>
1444 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1445 to wide_int_bitmask.
1446 * wide-int-bitmask.h: New.
1448 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1451 * config/i386/i386.md: Replace Pmode with word_mode in
1452 builtin_setjmp_setup and builtin_longjmp to support x32.
1454 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1458 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1460 (rs6000_supported_cpu_names): New static variable.
1461 (linux_cpu_translation_table): Likewise.
1462 (elf_platform) <cpu>: Define new static variable and use it.
1463 Translate kernel AT_PLATFORM name to canonical name if needed.
1464 Error if platform name is unknown.
1466 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1469 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1471 2018-02-01 Jeff Law <law@redhat.com>
1474 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1475 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1476 the scratch if RELEASE_VIA_POP is false.
1477 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1478 If we have to save a temporary register, decrement SIZE appropriately.
1479 Pass new arguments to release_scratch_register_on_entry.
1480 (ix86_adjust_stack_and_probe): Likewise.
1481 (ix86_emit_probe_stack_range): Pass new arguments to
1482 release_scratch_register_on_entry.
1484 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1486 PR rtl-optimization/84157
1487 * combine.c (change_zero_ext): Use REG_P predicate in
1488 front of HARD_REGISTER_P predicate.
1490 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1492 * config/avr/avr.c (avr_option_override): Move disabling of
1493 -fdelete-null-pointer-checks to...
1494 * common/config/avr/avr-common.c (avr_option_optimization_table):
1497 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1499 PR tree-optimization/81635
1500 * tree-data-ref.c (split_constant_offset_1): For types that
1501 wrap on overflow, try to use range info to prove that wrapping
1504 2018-02-01 Renlin Li <renlin.li@arm.com>
1507 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1509 (aarch64_register_move_cost): Likewise.
1510 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1512 (REG_CLASS_NAMES): Likewise.
1513 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1514 TAILCALL_ADDR_REGS. Remove IP registers.
1515 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1517 2018-02-01 Richard Biener <rguenther@suse.de>
1519 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1520 for specifying RPO order and allow NULL for that.
1521 * domwalk.c (dom_walker::dom_walker): Likewise.
1522 (dom_walker::walk): Handle NULL RPO order.
1523 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1525 (rewrite_update_dom_walker): Likewise.
1526 (mark_def_dom_walker): Likewise.
1528 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1530 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1531 (aarch64_maybe_expand_sve_subreg_move): Declare.
1532 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1533 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1535 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1536 that are semantically a reverse operation.
1537 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1538 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1539 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1541 (aarch64_can_change_mode_class): For big-endian, forbid changes
1542 between two SVE modes if they have different element sizes.
1544 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1546 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1547 the TImode handling for big-endian targets.
1549 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1551 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1552 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1554 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1555 Remove BSWAP handing for big-endian targets and use the form of
1556 LD1RQ appropariate for the mode.
1558 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1560 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1561 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1564 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1567 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1568 check for operands that need to go through aarch64_sve_reload_be.
1570 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1572 PR tree-optimization/81661
1573 PR tree-optimization/84117
1574 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1575 * tree-eh.c: Include gimplify.h.
1576 (find_trapping_overflow, replace_trapping_overflow,
1577 rewrite_to_non_trapping_overflow): New functions.
1578 * tree-vect-loop.c: Include tree-eh.h.
1579 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1580 * tree-data-ref.c: Include tree-eh.h.
1581 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1583 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1585 PR rtl-optimization/84123
1586 * combine.c (change_zero_ext): Check if hard register satisfies
1587 can_change_dest_mode before calling gen_lowpart_SUBREG.
1589 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1592 * ira.c (ira_init_register_move_cost): Remove assert.
1594 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1596 PR rtl-optimization/84071
1597 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1598 * doc/tm.texi: Regenerate.
1600 2018-01-31 Richard Biener <rguenther@suse.de>
1602 PR tree-optimization/84132
1603 * tree-data-ref.c (analyze_miv_subscript): Properly
1604 check whether evolution_function_is_affine_multivariate_p
1605 before calling gcd_of_steps_may_divide_p.
1607 2018-01-31 Julia Koval <julia.koval@intel.com>
1610 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1611 * config/i386/i386.md (rdpid_rex64) New.
1612 (rdpid): Make 32bit only.
1614 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1617 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1618 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1620 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1623 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1625 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1627 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1629 PR rtl-optimization/84071
1630 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1631 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1633 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1635 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1636 (arc_attribute_table): Add 'aux' attribute.
1637 (arc_in_small_data_p): Consider aux like variables.
1638 (arc_is_aux_reg_p): New function.
1639 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1640 (arc_get_aux_arg): New function.
1641 (prepare_move_operands): Handle aux-register access.
1642 (arc_handle_aux_attribute): New function.
1643 * doc/extend.texi (ARC Variable attributes): Add subsection.
1645 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1647 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1648 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1649 (arc_attribute_table): Add 'uncached' attribute.
1650 (arc_print_operand): Print '.di' flag for uncached memory
1652 (arc_in_small_data_p): Do not consider for small data the uncached
1654 (arc_is_uncached_mem_p): New function.
1655 * config/arc/predicates.md (compact_store_memory_operand): Check
1656 for uncached memory accesses.
1657 (nonvol_nonimm_operand): Likewise.
1658 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1660 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1663 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1664 falign-loops=): Add Optimization flag.
1666 2018-01-30 Jeff Law <law@redhat.com>
1669 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1670 INT_REGISTERS_SAVED. Check it prior to calling
1671 get_scratch_register_on_entry.
1672 (ix86_adjust_stack_and_probe): Similarly.
1673 (ix86_emit_probe_stack_range): Similarly.
1674 (ix86_expand_prologue): Corresponding changes.
1676 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1679 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1680 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1682 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1685 * lra-constraints.c (curr_insn_transform): Process AND in the
1688 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1690 PR rtl-optimization/83986
1691 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1692 dependence against last_pending_memory_flush in addition to
1695 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1697 PR tree-optimization/81611
1698 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1701 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1704 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1707 2018-01-30 Richard Biener <rguenther@suse.de>
1708 Jakub Jelinek <jakub@redhat.com>
1710 PR tree-optimization/84111
1711 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1712 inner loops added during recursion, as they don't have up-to-date
1715 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1718 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1719 (can_inline_edge_by_limits_p): ... here.
1720 (can_early_inline_edge_p, check_callers,
1721 update_caller_keys, update_callee_keys, recursive_inlining,
1722 add_new_edges_to_heap, speculation_useful_p,
1723 inline_small_functions,
1724 inline_small_functions, flatten_function,
1725 inline_to_all_callers_1): Update.
1727 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1729 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1732 2018-01-30 Richard Biener <rguenther@suse.de>
1734 PR tree-optimization/83008
1735 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1736 invariant and constant vector uses in stmts when they need
1739 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1742 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1743 * configure: Regenerate.
1745 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1747 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1749 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1750 Use gen_rtx_REG rather than gen_lowpart.
1752 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1754 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1755 rather than 0 when creating partial subregs.
1757 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1759 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1762 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1765 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1766 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1767 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1768 flags. This restores the settings used before the 2017-07-24.
1769 Turning off pre increment/decrement/modify allows IVOPTS to
1770 optimize DF/SF loops where the index is an int.
1772 2018-01-29 Richard Biener <rguenther@suse.de>
1773 Kelvin Nilsen <kelvin@gcc.gnu.org>
1776 * tree-vect-stmts.c (vectorizable_call): Don't call
1777 targetm.vectorize_builtin_md_vectorized_function if callee is
1780 2018-01-22 Carl Love <cel@us.ibm.com>
1782 * doc/extend.tex: Fix typo in second arg in
1783 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1785 2018-01-29 Richard Biener <rguenther@suse.de>
1787 PR tree-optimization/84086
1788 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1789 (flush_ssaname_freelist): When SSA names were released reset
1790 the SCEV hash table.
1792 2018-01-29 Richard Biener <rguenther@suse.de>
1794 PR tree-optimization/84057
1795 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1796 removed paths when removing edges.
1798 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1800 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1801 -mfunction-return=@var{choice}.
1803 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1806 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1808 (layout::print_source_line): Likewise.
1809 (test_get_line_width_without_trailing_whitespace): Add test cases.
1811 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1814 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1817 2018-01-26 Jim Wilson <jimw@sifive.com>
1819 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1821 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1824 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1826 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1827 and CMP + SUB-immediate -> SUBS.
1829 2018-01-26 Martin Sebor <msebor@redhat.com>
1831 PR tree-optimization/83896
1832 * tree-ssa-strlen.c (get_string_len): Rename...
1833 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1834 Avoid assuming length is constant.
1835 (handle_char_store): Use HOST_WIDE_INT for string length.
1837 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1840 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1841 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1843 2018-01-26 Richard Biener <rguenther@suse.de>
1845 PR rtl-optimization/84003
1846 * dse.c (record_store): Only record redundant stores when
1847 the earlier store aliases at least all accesses the later one does.
1849 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1851 PR rtl-optimization/83985
1852 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1853 REG_CFA_RESTORE insns.
1854 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1855 REG_CFA_RESTORE insns here.
1858 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1859 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1861 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1863 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1864 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1865 (arc_init): Likewise.
1866 (arc_override_options): Likewise.
1867 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1869 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1871 * config/arc/arc.h (TARGET_DBNZ): Define.
1872 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1873 properly set the tune attribute.
1874 (dbnz): Use TARGET_DBNZ guard.
1875 * config/arc/arc.opt (mtune): Add core3 option.
1877 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1879 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1880 recognize new pic like addresses.
1881 (arc_delegitimize_address): Clean up.
1883 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1885 * config/arc/arc-arches.def: Option mrf16 valid for all
1887 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1888 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1889 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1890 * config/arc/arc-tables.opt: Regenerate.
1891 * config/arc/arc.c (arc_conditional_register_usage): Handle
1892 reduced register file case.
1893 (arc_file_start): Set must have build attributes.
1894 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1896 * config/arc/arc.opt (mrf16): Add new option.
1897 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1898 * config/arc/genmultilib.awk: Handle new mrf16 option.
1899 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1900 * config/arc/t-multilib: Regenerate.
1901 * doc/invoke.texi (ARC Options): Document mrf16 option.
1903 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1905 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1906 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1907 (arc_attribute_table): Add 'secure_call' attribute.
1908 (arc_print_operand): Print secure call operand.
1909 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1911 (arc_is_secure_call_p): New function. * config/arc/arc.md
1912 (call_i): Add support for sjli instruction.
1913 (call_value_i): Likewise.
1914 * config/arc/constraints.md (Csc): New constraint.
1916 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1917 John Eric Martin <John.Martin@emmicro-us.com>
1919 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1920 * config/arc/arc.c (_arc_jli_section): New struct.
1921 (arc_jli_section): New type.
1922 (rc_jli_sections): New static variable.
1923 (arc_handle_jli_attribute): New function.
1924 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1925 (arc_file_end): New function.
1926 (TARGET_ASM_FILE_END): Define.
1927 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1928 (arc_add_jli_section): New function.
1929 (jli_call_scan): Likewise.
1930 (arc_reorg): Call jli_call_scan.
1931 (arc_output_addsi): Remove 'S' from printing asm operand.
1932 (arc_is_jli_call_p): New function.
1933 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1935 (movhi_insn): Likewise.
1936 (movsi_insn): Likewise.
1937 (movsi_set_cc_insn): Likewise.
1938 (loadqi_update): Likewise.
1939 (load_zeroextendqisi_update): Likewise.
1940 (load_signextendqisi_update): Likewise.
1941 (loadhi_update): Likewise.
1942 (load_zeroextendhisi_update): Likewise.
1943 (load_signextendhisi_update): Likewise.
1944 (loadsi_update): Likewise.
1945 (loadsf_update): Likewise.
1946 (movsicc_insn): Likewise.
1947 (bset_insn): Likewise.
1948 (bxor_insn): Likewise.
1949 (bclr_insn): Likewise.
1950 (bmsk_insn): Likewise.
1951 (bicsi3_insn): Likewise.
1952 (cmpsi_cc_c_insn): Likewise.
1953 (movsi_ne): Likewise.
1954 (movsi_cond_exec): Likewise.
1955 (clrsbsi2): Likewise.
1966 (call_i): Remove 'S' asm letter, add jli instruction.
1967 (call_value_i): Likewise.
1968 * config/arc/arc.op (mjli-always): New option.
1969 * config/arc/constraints.md (Cji): New constraint.
1970 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
1972 (subsf3_fpx): Likewise.
1973 (mulsf3_fpx): Likewise.
1974 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
1976 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
1978 * doc/invoke.texi (ARC): Document mjli-always option.
1980 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
1982 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
1983 avoid addition with 0 and use incw and decw where possible.
1985 2018-01-26 Richard Biener <rguenther@suse.de>
1987 PR tree-optimization/81082
1988 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
1989 association if it requires casting to unsigned.
1990 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
1991 from fold_plusminus_mult_expr to catch important cases late when
1992 range info is available.
1994 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1996 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
1997 * configure.ac (hidden_linkonce): New test.
1998 * configure: Regenerate.
1999 * config.in: Regenerate.
2001 2018-01-26 Julia Koval <julia.koval@intel.com>
2003 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
2004 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
2005 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
2006 _mm_mask_bitshuffle_epi64_mask): Fix type.
2007 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
2008 USI_FTYPE_V4DI_V4DI_USI): Remove.
2009 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
2010 __builtin_ia32_vpshufbitqmb256_mask,
2011 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
2012 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
2013 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
2015 2018-01-26 Alan Modra <amodra@gmail.com>
2018 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
2019 UNSPEC_VBPERMQ. Sort other unspecs.
2021 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
2023 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
2025 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
2028 * predict.c (drop_profile): Do not push/pop cfun; update also
2030 (handle_missing_profiles): Fix logic looking for zero profiles.
2032 2018-01-25 Jakub Jelinek <jakub@redhat.com>
2035 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
2036 on functions with #pragma omp declare simd or functions with simd
2038 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
2039 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
2040 Remove trailing \n from warning_at calls.
2042 2018-01-25 Tom de Vries <tom@codesourcery.com>
2045 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2046 for neutered workers.
2048 2018-01-24 Joseph Myers <joseph@codesourcery.com>
2051 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
2052 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
2054 2018-01-24 Jeff Law <law@redhat.com>
2057 * i386.c (get_probe_interval): Move to earlier point.
2058 (ix86_compute_frame_layout): If -fstack-clash-protection and
2059 the frame is larger than the probe interval, then use pushes
2060 to save registers rather than reg->mem moves.
2061 (ix86_expand_prologue): Remove conditional for int_registers_saved
2064 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
2067 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
2068 min/max for never referenced object.
2070 2018-01-24 Jakub Jelinek <jakub@redhat.com>
2073 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
2075 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
2076 attributes from DECL_ATTRIBUTES (decl) without affecting
2077 DECL_ATTRIBUTES (current_function_decl).
2078 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
2079 functions with non-NULL DECL_ABSTRACT_ORIGIN.
2081 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
2083 PR tree-optimization/83979
2084 * fold-const.c (fold_comparison): Use constant_boolean_node
2085 instead of boolean_{true,false}_node.
2087 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
2089 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
2092 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2094 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2095 Simplify the clause that sets the length attribute.
2096 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2097 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
2098 clause that sets the length attribute.
2099 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2101 2018-01-24 Tom de Vries <tom@codesourcery.com>
2104 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
2105 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
2106 Add strict parameter.
2107 (prevent_branch_around_nothing): Insert dummy insn between branch to
2108 label and label with no ptx insn inbetween.
2109 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
2111 2018-01-24 Tom de Vries <tom@codesourcery.com>
2114 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2115 for neutered threads in warp.
2116 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
2118 2018-01-24 Richard Biener <rguenther@suse.de>
2120 PR tree-optimization/83176
2121 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
2124 2018-01-24 Richard Biener <rguenther@suse.de>
2126 PR tree-optimization/82819
2127 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
2128 code generating pluses that are no-ops in the target precision.
2130 2018-01-24 Richard Biener <rguenther@suse.de>
2133 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
2135 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2137 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
2138 to merge probabilities.
2139 * predict.c (probably_never_executed): Also mark as cold functions
2140 with global 0 profile and guessed local profile.
2141 * profile-count.c (profile_probability::combine_with_count): New
2143 * profile-count.h (profile_probability::operator*,
2144 profile_probability::operator*=, profile_probability::operator/,
2145 profile_probability::operator/=): Reduce precision to adjusted
2146 and set value to guessed on contradictory divisions.
2147 (profile_probability::combine_with_freq): Remove.
2148 (profile_probability::combine_wiht_count): Declare.
2149 (profile_count::force_nonzero):: Set to adjusted.
2150 (profile_count::probability_in):: Set quality to adjusted.
2151 * tree-ssa-tail-merge.c (replace_block_by): Use
2154 2018-01-23 Andrew Waterman <andrew@sifive.com>
2155 Jim Wilson <jimw@sifive.com>
2157 * config/riscv/riscv.c (riscv_stack_boundary): New.
2158 (riscv_option_override): Set riscv_stack_boundary. Handle
2159 riscv_preferred_stack_boundary_arg.
2160 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
2161 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
2162 (STACK_BOUNDARY): Set to riscv_stack_boundary.
2163 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
2164 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
2165 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
2167 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
2170 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
2171 of struct ix86_frame.
2172 (ix86_expand_epilogue): Likewise. Add a local variable for
2173 the reg_save_offset field in struct ix86_frame.
2175 2018-01-23 Bin Cheng <bin.cheng@arm.com>
2177 PR tree-optimization/82604
2178 * tree-loop-distribution.c (enum partition_kind): New enum item
2179 PKIND_PARTIAL_MEMSET.
2180 (partition_builtin_p): Support above new enum item.
2181 (generate_code_for_partition): Ditto.
2182 (compute_access_range): Differentiate cases that equality can be
2183 proven at all loops, the innermost loops or no loops.
2184 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
2185 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
2186 (finalize_partitions, distribute_loop): Don't fuse partition of
2187 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
2188 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
2191 2018-01-23 Martin Liska <mliska@suse.cz>
2193 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
2194 order to ignore the predictor.
2195 (PRED_POLYMORPHIC_CALL): Likewise.
2196 (PRED_RECURSIVE_CALL): Likewise.
2198 2018-01-23 Martin Liska <mliska@suse.cz>
2200 * tree-profile.c (tree_profiling): Print function header to
2201 aware reader which function we are working on.
2202 * value-prof.c (gimple_find_values_to_profile): Do not print
2203 not interesting value histograms.
2205 2018-01-23 Martin Liska <mliska@suse.cz>
2207 * profile-count.h (enum profile_quality): Add
2208 profile_uninitialized as the first value. Do not number values
2209 as they are zero based.
2210 (profile_count::verify): Update sanity check.
2211 (profile_probability::verify): Likewise.
2213 2018-01-23 Nathan Sidwell <nathan@acm.org>
2215 * doc/invoke.texi (ffor-scope): Deprecate.
2217 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2219 PR tree-optimization/83510
2220 * domwalk.c (set_all_edges_as_executable): New function.
2221 (dom_walker::dom_walker): Convert bool param
2222 "skip_unreachable_blocks" to enum reachability. Move setup of
2223 edge flags to set_all_edges_as_executable and only do it when
2224 reachability is REACHABLE_BLOCKS.
2225 * domwalk.h (enum dom_walker::reachability): New enum.
2226 (dom_walker::dom_walker): Convert bool param
2227 "skip_unreachable_blocks" to enum reachability.
2228 (set_all_edges_as_executable): New decl.
2229 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2230 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2232 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2233 but converting true to REACHABLE_BLOCKS.
2234 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2236 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2237 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2238 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2240 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2241 if check_all_array_refs will be called.
2243 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2245 * tree.c (selftest::test_location_wrappers): Add more test
2248 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2250 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2251 (selftest::test_bit_in_range): Likewise.
2253 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2256 * doc/sourcebuild.texi (vect_float): Say that the selector
2257 only describes the situation when -funsafe-math-optimizations is on.
2258 (vect_float_strict): Document.
2260 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2262 PR tree-optimization/83965
2263 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2264 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2265 instead of checking only for a reduction.
2266 (vect_recog_widen_sum_pattern): Likewise.
2268 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2270 * predict.c (probably_never_executed): Only use precise profile info.
2271 (compute_function_frequency): Skip after inlining hack since we now
2272 have quality checking.
2274 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2276 * profile-count.h (profile_probability::very_unlikely,
2277 profile_probability::unlikely, profile_probability::even): Set
2278 precision to guessed.
2280 2018-01-23 Richard Biener <rguenther@suse.de>
2282 PR tree-optimization/83963
2283 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2284 Properly terminate dominator walk when crossing the exit edge not
2285 when visiting its source block.
2287 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2290 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2291 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2293 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2295 PR tree-optimization/83957
2296 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2297 semicolon after for body surrounded by braces.
2299 PR tree-optimization/83081
2300 * profile-count.h (profile_probability::split): New method.
2301 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2302 Use profile_probability::split.
2303 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2304 when splitting a single conditional jump into 2.
2306 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2308 PR tree-optimization/69452
2309 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2312 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2314 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2315 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2316 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2318 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2320 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2321 * config/rl78/rl78.md (movdi): New define_expand.
2322 * config/rl78/rl78.c (rl78_split_movdi): New function.
2324 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2327 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2329 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2330 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2331 128-bit to produce an UNSPEC move to get the double word with the
2332 signbit and then a shift directly to do signbit.
2333 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2334 implementation with a new version that just does either a direct
2335 move or a regular move. Move memory interface to separate insns.
2336 Move insns so they are next to the expander.
2337 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2338 with signbit move. Split big and little endian case.
2339 (signbit<mode>2_dm_mem_le): Likewise.
2340 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2341 (signbit<mode>2_dm2): Likewise.
2343 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2345 * config/rl78/rl78.md (anddi3): New define_expand.
2347 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2349 * config/rl78/rl78.md (umindi3): New define_expand.
2351 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2353 * config/rl78/rl78.md (smindi3): New define_expand.
2355 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2357 * config/rl78/rl78.md (smaxdi3): New define_expand.
2359 2018-01-22 Carl Love <cel@us.ibm.com>
2361 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2362 LVX_V1TI): Add macro expansion.
2363 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2364 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2365 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2366 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2367 Change check to determine if the instruction is a byte reversing
2368 entry. Fix typo in comment.
2369 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2370 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2371 Add def_builtin calls for new builtins.
2372 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2373 Add define_insn expansion.
2375 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2377 * config/rl78/rl78.md (umaxdi3): New define_expand.
2379 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2381 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2382 for non-QImode registers.
2384 2018-01-22 Richard Biener <rguenther@suse.de>
2386 PR tree-optimization/83963
2387 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2388 including the loop exit block.
2389 (scop_detection::merge_sese): Likewise.
2390 (scop_detection::add_scop): Do it here instead.
2392 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2394 * doc/sourcebuild.texi (arm_softfloat): Document.
2396 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2399 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2400 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2401 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2403 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2404 David Edelsohn <dje.gcc@gmail.com>
2407 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2408 Change "crset eq" to "crset 2".
2409 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2410 (*call_indirect_aix<mode>_nospec): Likewise.
2411 (*call_value_indirect_aix<mode>_nospec): Likewise.
2412 (*call_indirect_elfv2<mode>_nospec): Likewise.
2413 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2414 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2415 change assembly output from . to $.
2416 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2417 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2418 (*tablejump<mode>_internal1_nospec): Likewise.
2420 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2423 * config/sh/sh_optimize_sett_clrt.cc:
2424 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2426 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2428 PR tree-optimization/83940
2429 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2430 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2431 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2432 use it to pass back the definition type.
2433 (vect_check_store_rhs): Likewise rhs_dt_out.
2434 (vect_build_gather_load_calls): Add a mask_dt argument and use
2435 it instead of a call to vect_is_simple_use.
2436 (vectorizable_store): Update calls to vect_check_load_store_mask
2437 and vect_check_store_rhs. Use the dt returned by the latter instead
2438 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2439 instead of calls to vect_is_simple_use. Pass the scalar rather
2440 than the vector operand to vect_is_simple_use when handling
2441 second and subsequent copies of an rhs value.
2442 (vectorizable_load): Update calls to vect_check_load_store_mask
2443 and vect_build_gather_load_calls. Use the cached mask_dt and
2444 gs_info.offset_dt instead of calls to vect_is_simple_use.
2446 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2449 * tree-emutls.c: Include gimplify.h.
2450 (lower_emutls_2): New function.
2451 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2452 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2453 it before further processing.
2456 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2457 UINTVAL (trueop1) instead of INTVAL (op1).
2459 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2463 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2464 INCOMING_FRAME_SP_OFFSET if not defined.
2465 (scan_trace): Add ENTRY argument. If true and
2466 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2467 emit a note to adjust the CFA offset.
2468 (create_cfi_notes): Adjust scan_trace callers.
2469 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2470 INCOMING_FRAME_SP_OFFSET in the CIE.
2471 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2472 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2474 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2475 * doc/tm.texi: Regenerated.
2477 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2479 PR rtl-optimization/83147
2480 * lra-constraints.c (remove_inheritance_pseudos): Use
2481 lra_substitute_pseudo_within_insn.
2483 2018-01-19 Tom de Vries <tom@codesourcery.com>
2484 Cesar Philippidis <cesar@codesourcery.com>
2487 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2489 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2492 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2493 spaces for function labels.
2495 2018-01-19 Martin Liska <mliska@suse.cz>
2497 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2498 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2499 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2500 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2501 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2502 (PRED_CONST_RETURN): Change from 69 to 65.
2503 (PRED_NULL_RETURN): Change from 91 to 71.
2504 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2505 (PRED_LOOP_GUARD): Change from 66 to 73.
2507 2018-01-19 Martin Liska <mliska@suse.cz>
2509 * predict.c (predict_insn_def): Add new assert.
2510 (struct branch_predictor): Change type to signed integer.
2511 (test_prediction_value_range): Amend test to cover
2513 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2514 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2515 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2516 (PRED_LOOP_IV_COMPARE): Likewise.
2517 * predict.h (PROB_UNINITIALIZED): Define new constant.
2519 2018-01-19 Martin Liska <mliska@suse.cz>
2521 * predict.c (dump_prediction): Add new format for
2522 analyze_brprob.py script which is enabled with -details
2524 * profile-count.h (precise_p): New function.
2526 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2528 PR tree-optimization/83922
2529 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2530 there are no statements that need masking.
2531 (vect_active_double_reduction_p): New function.
2532 (vect_analyze_loop_operations): Use it when handling phis that
2533 are not in the loop header.
2535 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2537 PR tree-optimization/83914
2538 * tree-vect-loop.c (vectorizable_induction): Don't convert
2539 init_expr or apply the peeling adjustment for inductions
2540 that are nested within the vectorized loop.
2542 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2544 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2547 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2551 * function.h (gimplify_parameters): Add gimple_seq * argument.
2552 * function.c: Include gimple.h and options.h.
2553 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2554 for the added local temporaries if needed.
2555 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2556 if there are any parameter cleanups, wrap whole body into a
2557 try/finally with the cleanups.
2559 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2562 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2563 Use GET_MODE_CLASS for scalar floating point.
2565 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2569 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2570 Fix call of call_cgraph_insertion_hooks.
2572 2018-01-18 Martin Sebor <msebor@redhat.com>
2574 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2576 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2579 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2582 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2585 * common.opt: (-ffile-prefix-map): New option.
2586 * opts.c (common_handle_option): Defer it.
2587 * opts-global.c (handle_common_deferred_options): Handle it.
2588 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2589 * file-prefix-map.h: New file.
2590 (remap_debug_filename, add_debug_prefix_map): ...here.
2591 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2592 * final.c (debug_prefix_map, add_debug_prefix_map
2593 remap_debug_filename): Move to...
2594 * file-prefix-map.c: New file.
2595 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2596 generalize, get rid of alloca(), use strrchr() instead of strchr().
2597 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2598 Implement in terms of add_prefix_map().
2599 (remap_macro_filename, remap_debug_filename): Implement in term of
2601 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2602 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2603 * dbxout.c: Include file-prefix-map.h.
2604 * varasm.c: Likewise.
2605 * vmsdbgout.c: Likewise.
2606 * xcoffout.c: Likewise.
2607 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2608 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2609 * doc/invoke.texi (-ffile-prefix-map): Document.
2610 (-fdebug-prefix-map): Update description.
2612 2018-01-18 Martin Liska <mliska@suse.cz>
2614 * config/i386/i386.c (indirect_thunk_name): Document that also
2616 (output_indirect_thunk): Document why both instructions
2617 (pause and lfence) are generated.
2619 2018-01-18 Richard Biener <rguenther@suse.de>
2621 PR tree-optimization/83887
2622 * graphite-scop-detection.c
2623 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2624 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2625 (scop_detection::merge_sese): Re-implement with a flood-fill
2626 algorithm that properly finds a SESE region if it exists.
2628 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2631 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2632 pointer_diff optimizations use view_convert instead of convert.
2634 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2636 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2637 Generate different code for -mno-speculate-indirect-jumps.
2638 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2639 (*call_indirect_aix<mode>): Disable for
2640 -mno-speculate-indirect-jumps.
2641 (*call_indirect_aix<mode>_nospec): New define_insn.
2642 (*call_value_indirect_aix<mode>): Disable for
2643 -mno-speculate-indirect-jumps.
2644 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2645 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2646 -mno-speculate-indirect-jumps.
2647 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2649 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2651 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2652 long double type, set the flags for noting the default long double
2653 type, even if we don't pass or return a long double type.
2655 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2658 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2661 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2663 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2664 support for merge[hl].
2665 (fold_mergehl_helper): New helper function.
2666 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2667 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2668 (altivec_vmrglw_direct): Add xxmrglw insn.
2670 2018-01-17 Andrew Waterman <andrew@sifive.com>
2672 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2673 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2675 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2678 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2679 call the lto_location_cache before reading the
2680 DECL_SOURCE_LOCATION of the types.
2682 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2683 Richard Sandiford <richard.sandiford@linaro.org>
2685 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2686 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2687 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2688 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2689 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2691 * config/aarch64/constraints.md (aarch64_movti_operand):
2693 * config/aarch64/predicates.md (Uti): Add new constraint.
2695 2018-01-17 Carl Love <cel@us.ibm.com>
2697 * config/rs6000/vsx.md (define_expand xl_len_r,
2698 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2699 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2701 (define_expand, define_insn): Move the shift left from the
2702 define_insn to the define_expand for lxvl and stxvl instructions.
2703 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2704 and XL_LEN_R definitions to PURE.
2706 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2708 * config/i386/i386.c (indirect_thunk_name): Declare regno
2709 as unsigned int. Compare regno with INVALID_REGNUM.
2710 (output_indirect_thunk): Ditto.
2711 (output_indirect_thunk_function): Ditto.
2712 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2713 in the call to output_indirect_thunk_function.
2715 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2718 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2719 rather than the size of inner_type to determine the stack slot size
2720 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2722 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2725 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2728 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2730 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2731 endian Linux systems to optionally enable multilibs for selecting
2732 the long double type if the user configured an explicit type.
2733 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2734 have no long double multilibs if not defined.
2735 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2736 warn if the user used -mabi={ieee,ibm}longdouble and we built
2737 multilibs for long double.
2738 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2739 appropriate multilib option.
2740 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2742 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2743 for building long double multilibs.
2744 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2746 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2748 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2751 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2753 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2756 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2759 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2762 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2764 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2765 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2767 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2769 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2770 different rtl trees depending on TARGET_64BIT.
2771 (rs6000_gen_lvx): Likewise.
2773 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2775 * config/visium/visium.md (nop): Tweak comment.
2776 (hazard_nop): Likewise.
2778 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2780 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2781 -mspeculate-indirect-jumps.
2782 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2783 for -mno-speculate-indirect-jumps.
2784 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2785 (*call_value_indirect_elfv2<mode>): Disable for
2786 -mno-speculate-indirect-jumps.
2787 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2788 (indirect_jump): Emit different RTL for
2789 -mno-speculate-indirect-jumps.
2790 (*indirect_jump<mode>): Disable for
2791 -mno-speculate-indirect-jumps.
2792 (*indirect_jump<mode>_nospec): New define_insn.
2793 (tablejump): Emit different RTL for
2794 -mno-speculate-indirect-jumps.
2795 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2796 (tablejumpsi_nospec): New define_expand.
2797 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2798 (tablejumpdi_nospec): New define_expand.
2799 (*tablejump<mode>_internal1): Disable for
2800 -mno-speculate-indirect-jumps.
2801 (*tablejump<mode>_internal1_nospec): New define_insn.
2802 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2805 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2807 * caller-save.c (insert_save): Drop unnecessary parameter. All
2810 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2811 Richard Biener <rguenth@suse.de>
2814 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2815 return early, inline manually is_gimple_sizepos. Make sure if we
2816 call gimplify_expr we don't end up with a gimple constant.
2817 * tree.c (variably_modified_type_p): Don't return true for
2818 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2819 * gimplify.h (is_gimple_sizepos): Remove.
2821 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2823 PR tree-optimization/83857
2824 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2825 vectorizable_live_operation for pure SLP statements.
2826 (vectorizable_live_operation): Handle PHIs.
2828 2018-01-16 Richard Biener <rguenther@suse.de>
2830 PR tree-optimization/83867
2831 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2832 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2834 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2837 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2838 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2839 If off is not INTEGER_CST, issue a may not be aligned warning
2840 rather than isn't aligned. Use isn%'t rather than isn't.
2841 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2843 <case MULT_EXPR>: Improve the case when bottom and one of the
2844 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2845 operand, in that case check if the other operand is multiple of
2846 bottom divided by the INTEGER_CST operand.
2848 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2851 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2852 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2853 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2854 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2855 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2856 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2857 (pa_function_arg_size): New function.
2859 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2861 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2862 in a separate statement.
2864 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2866 PR tree-optimization/83847
2867 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2868 group gathers and scatters.
2870 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2872 PR rtl-optimization/86620
2873 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2875 PR rtl-optimization/83213
2876 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2877 to last if both are JUMP_INSNs.
2879 PR tree-optimization/83843
2880 * gimple-ssa-store-merging.c
2881 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2882 store_immediate_info for bswap/nop orig_stores.
2884 2018-01-15 Andrew Waterman <andrew@sifive.com>
2886 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2888 <UDIV>: Increase cost if !TARGET_DIV.
2890 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2892 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2893 (define_attr "cr_logical_3op"): New.
2894 (cceq_ior_compare): Adjust.
2895 (cceq_ior_compare_complement): Adjust.
2896 (*cceq_rev_compare): Adjust.
2897 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2898 (is_cracked_insn): Adjust.
2899 (insn_must_be_first_in_group): Adjust.
2900 * config/rs6000/40x.md: Adjust.
2901 * config/rs6000/440.md: Adjust.
2902 * config/rs6000/476.md: Adjust.
2903 * config/rs6000/601.md: Adjust.
2904 * config/rs6000/603.md: Adjust.
2905 * config/rs6000/6xx.md: Adjust.
2906 * config/rs6000/7450.md: Adjust.
2907 * config/rs6000/7xx.md: Adjust.
2908 * config/rs6000/8540.md: Adjust.
2909 * config/rs6000/cell.md: Adjust.
2910 * config/rs6000/e300c2c3.md: Adjust.
2911 * config/rs6000/e500mc.md: Adjust.
2912 * config/rs6000/e500mc64.md: Adjust.
2913 * config/rs6000/e5500.md: Adjust.
2914 * config/rs6000/e6500.md: Adjust.
2915 * config/rs6000/mpc.md: Adjust.
2916 * config/rs6000/power4.md: Adjust.
2917 * config/rs6000/power5.md: Adjust.
2918 * config/rs6000/power6.md: Adjust.
2919 * config/rs6000/power7.md: Adjust.
2920 * config/rs6000/power8.md: Adjust.
2921 * config/rs6000/power9.md: Adjust.
2922 * config/rs6000/rs64.md: Adjust.
2923 * config/rs6000/titan.md: Adjust.
2925 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2927 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2928 ix86_indirect_branch_register logic.
2930 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2932 * config/i386/constraints.md (Bs): Update
2933 ix86_indirect_branch_register check. Don't check
2934 ix86_indirect_branch_register with GOT_memory_operand.
2936 * config/i386/predicates.md (GOT_memory_operand): Don't check
2937 ix86_indirect_branch_register here.
2938 (GOT32_symbol_operand): Likewise.
2940 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2942 * config/i386/predicates.md (constant_call_address_operand):
2943 Rewrite ix86_indirect_branch_register logic.
2944 (sibcall_insn_operand): Likewise.
2946 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2948 * config/i386/constraints.md (Bs): Replace
2949 ix86_indirect_branch_thunk_register with
2950 ix86_indirect_branch_register.
2952 * config/i386/i386.md (indirect_jump): Likewise.
2953 (tablejump): Likewise.
2954 (*sibcall_memory): Likewise.
2955 (*sibcall_value_memory): Likewise.
2956 Peepholes of indirect call and jump via memory: Likewise.
2957 * config/i386/i386.opt: Likewise.
2958 * config/i386/predicates.md (indirect_branch_operand): Likewise.
2959 (GOT_memory_operand): Likewise.
2960 (call_insn_operand): Likewise.
2961 (sibcall_insn_operand): Likewise.
2962 (GOT32_symbol_operand): Likewise.
2964 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2967 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
2968 type rather than type addr's type points to.
2969 (expand_omp_atomic_mutex): Likewise.
2970 (expand_omp_atomic): Likewise.
2972 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2975 * config/i386/i386.c (output_indirect_thunk_function): Use
2976 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
2977 for __x86_return_thunk.
2979 2018-01-15 Richard Biener <rguenther@suse.de>
2982 * expmed.c (extract_bit_field_1): Fix typo.
2984 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2987 * config/arm/iterators.md (VF): New mode iterator.
2988 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
2989 Remove integer-related logic from pattern.
2990 (neon_vabd<mode>_3): Likewise.
2992 2018-01-15 Jakub Jelinek <jakub@redhat.com>
2995 * common.opt (fstrict-overflow): No longer an alias.
2996 (fwrapv-pointer): New option.
2997 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
2998 also for pointer types based on flag_wrapv_pointer.
2999 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
3000 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
3001 opts->x_flag_wrapv got set.
3002 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
3003 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
3004 POINTER_TYPE_OVERFLOW_UNDEFINED.
3005 * match.pd: Likewise in address comparison pattern.
3006 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
3008 2018-01-15 Richard Biener <rguenther@suse.de>
3011 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
3012 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
3013 Reset type names to their identifier if their TYPE_DECL doesn't
3014 have linkage (and thus is used for ODR and devirt).
3015 (save_debug_info_for_decl): Remove.
3016 (save_debug_info_for_type): Likewise.
3017 (add_tree_to_fld_list): Adjust.
3018 * tree-pretty-print.c (dump_generic_node): Make dumping of
3019 type names more robust.
3021 2018-01-15 Richard Biener <rguenther@suse.de>
3023 * BASE-VER: Bump to 8.0.1.
3025 2018-01-14 Martin Sebor <msebor@redhat.com>
3028 * builtins.c (check_access): Avoid warning when the no-warning bit
3031 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
3033 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
3034 * ira-color (allocno_hard_regs_compare): Likewise.
3036 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
3039 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
3040 Use .pushsection/.popsection.
3042 2018-01-14 Martin Sebor <msebor@redhat.com>
3045 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
3047 2018-01-14 Jakub Jelinek <jakub@redhat.com>
3049 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
3050 entry from extra_headers.
3051 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
3052 extra_headers, make the list bitwise identical to the i?86-*-* one.
3054 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3056 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
3057 -mcmodel=large with -mindirect-branch=thunk,
3058 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
3059 -mfunction-return=thunk-extern.
3060 * doc/invoke.texi: Document -mcmodel=large is incompatible with
3061 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
3062 -mfunction-return=thunk and -mfunction-return=thunk-extern.
3064 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3066 * config/i386/i386.c (print_reg): Print the name of the full
3067 integer register without '%'.
3068 (ix86_print_operand): Handle 'V'.
3069 * doc/extend.texi: Document 'V' modifier.
3071 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3073 * config/i386/constraints.md (Bs): Disallow memory operand for
3074 -mindirect-branch-register.
3076 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3077 (GOT_memory_operand): Likewise.
3078 (call_insn_operand): Likewise.
3079 (sibcall_insn_operand): Likewise.
3080 (GOT32_symbol_operand): Likewise.
3081 * config/i386/i386.md (indirect_jump): Call convert_memory_address
3082 for -mindirect-branch-register.
3083 (tablejump): Likewise.
3084 (*sibcall_memory): Likewise.
3085 (*sibcall_value_memory): Likewise.
3086 Disallow peepholes of indirect call and jump via memory for
3087 -mindirect-branch-register.
3088 (*call_pop): Replace m with Bw.
3089 (*call_value_pop): Likewise.
3090 (*sibcall_pop_memory): Replace m with Bs.
3091 * config/i386/i386.opt (mindirect-branch-register): New option.
3092 * doc/invoke.texi: Document -mindirect-branch-register option.
3094 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3096 * config/i386/i386-protos.h (ix86_output_function_return): New.
3097 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
3098 set function_return_type.
3099 (indirect_thunk_name): Add ret_p to indicate thunk for function
3101 (output_indirect_thunk_function): Pass false to
3102 indirect_thunk_name.
3103 (ix86_output_indirect_branch_via_reg): Likewise.
3104 (ix86_output_indirect_branch_via_push): Likewise.
3105 (output_indirect_thunk_function): Create alias for function
3106 return thunk if regno < 0.
3107 (ix86_output_function_return): New function.
3108 (ix86_handle_fndecl_attribute): Handle function_return.
3109 (ix86_attribute_table): Add function_return.
3110 * config/i386/i386.h (machine_function): Add
3111 function_return_type.
3112 * config/i386/i386.md (simple_return_internal): Use
3113 ix86_output_function_return.
3114 (simple_return_internal_long): Likewise.
3115 * config/i386/i386.opt (mfunction-return=): New option.
3116 (indirect_branch): Mention -mfunction-return=.
3117 * doc/extend.texi: Document function_return function attribute.
3118 * doc/invoke.texi: Document -mfunction-return= option.
3120 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3122 * config/i386/i386-opts.h (indirect_branch): New.
3123 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
3124 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
3125 with local indirect jump when converting indirect call and jump.
3126 (ix86_set_indirect_branch_type): New.
3127 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
3128 (indirectlabelno): New.
3129 (indirect_thunk_needed): Likewise.
3130 (indirect_thunk_bnd_needed): Likewise.
3131 (indirect_thunks_used): Likewise.
3132 (indirect_thunks_bnd_used): Likewise.
3133 (INDIRECT_LABEL): Likewise.
3134 (indirect_thunk_name): Likewise.
3135 (output_indirect_thunk): Likewise.
3136 (output_indirect_thunk_function): Likewise.
3137 (ix86_output_indirect_branch_via_reg): Likewise.
3138 (ix86_output_indirect_branch_via_push): Likewise.
3139 (ix86_output_indirect_branch): Likewise.
3140 (ix86_output_indirect_jmp): Likewise.
3141 (ix86_code_end): Call output_indirect_thunk_function if needed.
3142 (ix86_output_call_insn): Call ix86_output_indirect_branch if
3144 (ix86_handle_fndecl_attribute): Handle indirect_branch.
3145 (ix86_attribute_table): Add indirect_branch.
3146 * config/i386/i386.h (machine_function): Add indirect_branch_type
3147 and has_local_indirect_jump.
3148 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
3150 (tablejump): Likewise.
3151 (*indirect_jump): Use ix86_output_indirect_jmp.
3152 (*tablejump_1): Likewise.
3153 (simple_return_indirect_internal): Likewise.
3154 * config/i386/i386.opt (mindirect-branch=): New option.
3155 (indirect_branch): New.
3158 (thunk-inline): Likewise.
3159 (thunk-extern): Likewise.
3160 * doc/extend.texi: Document indirect_branch function attribute.
3161 * doc/invoke.texi: Document -mindirect-branch= option.
3163 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
3166 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
3168 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
3170 * ipa-inline.c (want_inline_small_function_p): Return false if
3171 inlining has already failed with CIF_FINAL_ERROR.
3172 (update_caller_keys): Call want_inline_small_function_p before
3174 (update_callee_keys): Likewise.
3176 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
3178 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
3180 (rs6000_quadword_masked_address_p): Likewise.
3181 (quad_aligned_load_p): Likewise.
3182 (quad_aligned_store_p): Likewise.
3183 (const_load_sequence_p): Add comment to describe the outer-most loop.
3184 (mimic_memory_attributes_and_flags): New function.
3185 (rs6000_gen_stvx): Likewise.
3186 (replace_swapped_aligned_store): Likewise.
3187 (rs6000_gen_lvx): Likewise.
3188 (replace_swapped_aligned_load): Likewise.
3189 (replace_swapped_load_constant): Capitalize argument name in
3190 comment describing this function.
3191 (rs6000_analyze_swaps): Add a third pass to search for vector loads
3192 and stores that access quad-word aligned addresses and replace
3193 with stvx or lvx instructions when appropriate.
3194 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
3195 New function prototype.
3196 (rs6000_quadword_masked_address_p): Likewise.
3197 (rs6000_gen_lvx): Likewise.
3198 (rs6000_gen_stvx): Likewise.
3199 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3200 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3201 when memory address is aligned.
3202 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3203 this split to select lvx instruction when memory address is aligned.
3204 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3205 instruction when memory address is aligned.
3206 (*vsx_le_perm_load_v16qi): Likewise.
3207 (four unnamed splitters): Modify to select the stvx instruction
3208 when memory is aligned.
3210 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3212 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3213 which appears in the queue multiple times.
3215 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3216 Alan Hayward <alan.hayward@arm.com>
3217 David Sherwood <david.sherwood@arm.com>
3219 * tree-vectorizer.h (vec_lower_bound): New structure.
3220 (_loop_vec_info): Add check_nonzero and lower_bounds.
3221 (LOOP_VINFO_CHECK_NONZERO): New macro.
3222 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3223 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3224 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3225 fields. Make seg_len the distance travelled, not including the
3227 (dr_direction_indicator): Declare.
3228 (dr_zero_step_indicator): Likewise.
3229 (dr_known_forward_stride_p): Likewise.
3230 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3232 (runtime_alias_check_p): Allow runtime alias checks with
3234 (operator ==): Compare access_size and align.
3235 (prune_runtime_alias_test_list): Rework for new distinction between
3236 the access_size and seg_len.
3237 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3239 (get_segment_min_max): New function.
3240 (create_intersect_range_checks): Use it.
3241 (dr_step_indicator): New function.
3242 (dr_direction_indicator): Likewise.
3243 (dr_zero_step_indicator): Likewise.
3244 (dr_known_forward_stride_p): Likewise.
3245 * tree-loop-distribution.c (data_ref_segment_size): Return
3246 DR_STEP * (niters - 1).
3247 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3249 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3250 (vect_preserves_scalar_order_p): New function, split out from...
3251 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3252 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3253 (vect_vfa_access_size): New function.
3254 (vect_vfa_align): Likewise.
3255 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3256 (dump_lower_bound): New function.
3257 (vect_check_lower_bound): Likewise.
3258 (vect_small_gap_p): Likewise.
3259 (vectorizable_with_step_bound_p): Likewise.
3260 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3261 depencies if the vectorization factor is 1. Convert the checks
3262 for nonzero steps into checks on the bounds of DR_STEP. Try using
3263 a bunds check for variable steps if the minimum required step is
3264 relatively small. Update calls to the dr_with_seg_len
3265 constructor and to vect_compile_time_alias.
3266 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3268 (vect_loop_versioning): Call it.
3269 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3271 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3273 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3274 Alan Hayward <alan.hayward@arm.com>
3275 David Sherwood <david.sherwood@arm.com>
3277 * doc/sourcebuild.texi (vect_scatter_store): Document.
3278 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3280 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3282 * genopinit.c (main): Add supports_vec_scatter_store and
3283 supports_vec_scatter_store_cached to target_optabs.
3284 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3285 IFN_MASK_SCATTER_STORE.
3286 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3288 * internal-fn.h (internal_store_fn_p): Declare.
3289 (internal_fn_stored_value_index): Likewise.
3290 * internal-fn.c (scatter_store_direct): New macro.
3291 (expand_scatter_store_optab_fn): New function.
3292 (direct_scatter_store_optab_supported_p): New macro.
3293 (internal_store_fn_p): New function.
3294 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3295 IFN_MASK_SCATTER_STORE.
3296 (internal_fn_mask_index): Likewise.
3297 (internal_fn_stored_value_index): New function.
3298 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3300 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3301 * optabs-query.c (supports_vec_scatter_store_p): New function.
3302 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3303 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3304 true for scatter stores.
3305 (vect_gather_scatter_fn_p): Handle scatter stores too.
3306 (vect_check_gather_scatter): Consider using scatter stores if
3307 supports_vec_scatter_store_p.
3308 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3310 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3311 internal_fn_stored_value_index.
3312 (check_load_store_masking): Handle scatter stores too.
3313 (vect_get_store_rhs): Make public.
3314 (vectorizable_call): Use internal_store_fn_p.
3315 (vectorizable_store): Handle scatter store internal functions.
3316 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3317 when deciding whether the end of the group has been reached.
3318 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3319 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3320 (mask_scatter_store<mode>): New insns.
3322 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3323 Alan Hayward <alan.hayward@arm.com>
3324 David Sherwood <david.sherwood@arm.com>
3326 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3327 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3328 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3330 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3331 Use vect_truncate_gather_scatter_offset if we can't treat the
3332 operation as a normal gather load or scatter store.
3333 (get_group_load_store_type): Take the gather_scatter_info
3334 as argument. Try using a gather load or scatter store for
3335 single-element groups.
3336 (get_load_store_type): Update calls to get_group_load_store_type
3337 and vect_use_strided_gather_scatters_p.
3339 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3340 Alan Hayward <alan.hayward@arm.com>
3341 David Sherwood <david.sherwood@arm.com>
3343 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3344 optional tree argument.
3345 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3347 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3348 but continue to use the current value as a fallback.
3349 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3350 to compare the updates.
3351 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3352 (get_load_store_type): Use it when handling a strided access.
3353 (vect_get_strided_load_store_ops): New function.
3354 (vect_get_data_ptr_increment): Likewise.
3355 (vectorizable_load): Handle strided gather loads. Always pass
3356 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3358 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3359 Alan Hayward <alan.hayward@arm.com>
3360 David Sherwood <david.sherwood@arm.com>
3362 * doc/md.texi (gather_load@var{m}): Document.
3363 (mask_gather_load@var{m}): Likewise.
3364 * genopinit.c (main): Add supports_vec_gather_load and
3365 supports_vec_gather_load_cached to target_optabs.
3366 * optabs-tree.c (init_tree_optimization_optabs): Use
3367 ggc_cleared_alloc to allocate target_optabs.
3368 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3369 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3371 * internal-fn.h (internal_load_fn_p): Declare.
3372 (internal_gather_scatter_fn_p): Likewise.
3373 (internal_fn_mask_index): Likewise.
3374 (internal_gather_scatter_fn_supported_p): Likewise.
3375 * internal-fn.c (gather_load_direct): New macro.
3376 (expand_gather_load_optab_fn): New function.
3377 (direct_gather_load_optab_supported_p): New macro.
3378 (direct_internal_fn_optab): New function.
3379 (internal_load_fn_p): Likewise.
3380 (internal_gather_scatter_fn_p): Likewise.
3381 (internal_fn_mask_index): Likewise.
3382 (internal_gather_scatter_fn_supported_p): Likewise.
3383 * optabs-query.c (supports_at_least_one_mode_p): New function.
3384 (supports_vec_gather_load_p): Likewise.
3385 * optabs-query.h (supports_vec_gather_load_p): Declare.
3386 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3387 and memory_type field.
3388 (NUM_PATTERNS): Bump to 15.
3389 * tree-vect-data-refs.c: Include internal-fn.h.
3390 (vect_gather_scatter_fn_p): New function.
3391 (vect_describe_gather_scatter_call): Likewise.
3392 (vect_check_gather_scatter): Try using internal functions for
3393 gather loads. Recognize existing calls to a gather load function.
3394 (vect_analyze_data_refs): Consider using gather loads if
3395 supports_vec_gather_load_p.
3396 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3397 (vect_get_gather_scatter_offset_type): Likewise.
3398 (vect_convert_mask_for_vectype): Likewise.
3399 (vect_add_conversion_to_patterm): Likewise.
3400 (vect_try_gather_scatter_pattern): Likewise.
3401 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3402 (vect_vect_recog_func_ptrs): Add it.
3403 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3404 internal_fn_mask_index and internal_gather_scatter_fn_p.
3405 (check_load_store_masking): Take the gather_scatter_info as an
3406 argument and handle gather loads.
3407 (vect_get_gather_scatter_ops): New function.
3408 (vectorizable_call): Check internal_load_fn_p.
3409 (vectorizable_load): Likewise. Handle gather load internal
3411 (vectorizable_store): Update call to check_load_store_masking.
3412 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3413 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3414 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3415 (aarch64_gather_scale_operand_d): New predicates.
3416 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3417 (mask_gather_load<mode>): New insns.
3419 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3420 Alan Hayward <alan.hayward@arm.com>
3421 David Sherwood <david.sherwood@arm.com>
3423 * optabs.def (fold_left_plus_optab): New optab.
3424 * doc/md.texi (fold_left_plus_@var{m}): Document.
3425 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3426 * internal-fn.c (fold_left_direct): Define.
3427 (expand_fold_left_optab_fn): Likewise.
3428 (direct_fold_left_optab_supported_p): Likewise.
3429 * fold-const-call.c (fold_const_fold_left): New function.
3430 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3431 * tree-parloops.c (valid_reduction_p): New function.
3432 (gather_scalar_reductions): Use it.
3433 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3434 (vect_finish_replace_stmt): Declare.
3435 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3436 (needs_fold_left_reduction_p): New function, split out from...
3437 (vect_is_simple_reduction): ...here. Accept reductions that
3438 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3439 (vect_force_simple_reduction): Also store the reduction type in
3440 the assignment's STMT_VINFO_REDUC_TYPE.
3441 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3442 (merge_with_identity): New function.
3443 (vect_expand_fold_left): Likewise.
3444 (vectorize_fold_left_reduction): Likewise.
3445 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3446 scalar phi in place for it. Check for target support and reject
3447 cases that would reassociate the operation. Defer the transform
3448 phase to vectorize_fold_left_reduction.
3449 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3450 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3451 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3453 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3455 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3456 call to ifc_temp_var.
3458 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3459 Alan Hayward <alan.hayward@arm.com>
3460 David Sherwood <david.sherwood@arm.com>
3462 * target.def (legitimize_address_displacement): Take the original
3463 offset as a poly_int.
3464 * targhooks.h (default_legitimize_address_displacement): Update
3466 * targhooks.c (default_legitimize_address_displacement): Likewise.
3467 * doc/tm.texi: Regenerate.
3468 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3469 as an argument, moving assert of ad->disp == ad->disp_term to...
3470 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3471 Try calling targetm.legitimize_address_displacement before expanding
3472 the address rather than afterwards, and adjust for the new interface.
3473 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3474 Match the new hook interface. Handle SVE addresses.
3475 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3478 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3480 * Makefile.in (OBJS): Add early-remat.o.
3481 * target.def (select_early_remat_modes): New hook.
3482 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3483 * doc/tm.texi: Regenerate.
3484 * targhooks.h (default_select_early_remat_modes): Declare.
3485 * targhooks.c (default_select_early_remat_modes): New function.
3486 * timevar.def (TV_EARLY_REMAT): New timevar.
3487 * passes.def (pass_early_remat): New pass.
3488 * tree-pass.h (make_pass_early_remat): Declare.
3489 * early-remat.c: New file.
3490 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3492 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3494 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3495 Alan Hayward <alan.hayward@arm.com>
3496 David Sherwood <david.sherwood@arm.com>
3498 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3499 vfm1 with a bound_epilog parameter.
3500 (vect_do_peeling): Update calls accordingly, and move the prologue
3501 call earlier in the function. Treat the base bound_epilog as 0 for
3502 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3503 this base when peeling for gaps.
3504 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3505 with fully-masked loops.
3506 (vect_estimate_min_profitable_iters): Handle the single peeled
3507 iteration in that case.
3509 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3510 Alan Hayward <alan.hayward@arm.com>
3511 David Sherwood <david.sherwood@arm.com>
3513 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3514 single-element interleaving even if the size is not a power of 2.
3515 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3516 accesses for single-element interleaving if the group size is
3519 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3520 Alan Hayward <alan.hayward@arm.com>
3521 David Sherwood <david.sherwood@arm.com>
3523 * doc/md.texi (fold_extract_last_@var{m}): Document.
3524 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3525 * optabs.def (fold_extract_last_optab): New optab.
3526 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3527 * internal-fn.c (fold_extract_direct): New macro.
3528 (expand_fold_extract_optab_fn): Likewise.
3529 (direct_fold_extract_optab_supported_p): Likewise.
3530 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3531 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3532 EXTRACT_LAST_REDUCTION.
3533 (get_initial_def_for_reduction): Do not create an initial vector
3534 for EXTRACT_LAST_REDUCTION reductions.
3535 (vectorizable_reduction): Leave the scalar phi in place for
3536 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3537 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3538 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3539 transform phase to vectorizable_condition.
3540 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3542 (vect_finish_stmt_generation): ...here.
3543 (vect_finish_replace_stmt): New function.
3544 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3545 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3547 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3549 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3550 Alan Hayward <alan.hayward@arm.com>
3551 David Sherwood <david.sherwood@arm.com>
3553 * doc/md.texi (extract_last_@var{m}): Document.
3554 * optabs.def (extract_last_optab): New optab.
3555 * internal-fn.def (EXTRACT_LAST): New internal function.
3556 * internal-fn.c (cond_unary_direct): New macro.
3557 (expand_cond_unary_optab_fn): Likewise.
3558 (direct_cond_unary_optab_supported_p): Likewise.
3559 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3560 loops using EXTRACT_LAST.
3561 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3562 (extract_last_<mode>): ...this optab.
3563 (vec_extract<mode><Vel>): Update accordingly.
3565 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3566 Alan Hayward <alan.hayward@arm.com>
3567 David Sherwood <david.sherwood@arm.com>
3569 * target.def (empty_mask_is_expensive): New hook.
3570 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3571 * doc/tm.texi: Regenerate.
3572 * targhooks.h (default_empty_mask_is_expensive): Declare.
3573 * targhooks.c (default_empty_mask_is_expensive): New function.
3574 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3575 if the target says that empty masks are expensive.
3576 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3578 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3580 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3581 Alan Hayward <alan.hayward@arm.com>
3582 David Sherwood <david.sherwood@arm.com>
3584 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3585 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3586 (vect_use_loop_mask_for_alignment_p): New function.
3587 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3588 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3589 niters_skip argument. Make sure that the first niters_skip elements
3590 of the first iteration are inactive.
3591 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3592 Update call to vect_set_loop_masks_directly.
3593 (get_misalign_in_elems): New function, split out from...
3594 (vect_gen_prolog_loop_niters): ...here.
3595 (vect_update_init_of_dr): Take a code argument that specifies whether
3596 the adjustment should be added or subtracted.
3597 (vect_update_init_of_drs): Likewise.
3598 (vect_prepare_for_masked_peels): New function.
3599 (vect_do_peeling): Skip prologue peeling if we're using a mask
3600 instead. Update call to vect_update_inits_of_drs.
3601 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3603 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3604 alignment. Do not include the number of peeled iterations in
3605 the minimum threshold in that case.
3606 (vectorizable_induction): Adjust the start value down by
3607 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3608 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3609 Take the number of skipped iterations into account when calculating
3611 * tree-vect-stmts.c (vect_gen_while_not): New function.
3613 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3614 Alan Hayward <alan.hayward@arm.com>
3615 David Sherwood <david.sherwood@arm.com>
3617 * doc/sourcebuild.texi (vect_fully_masked): Document.
3618 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3620 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3622 (vect_analyze_loop_2): ...here. Don't check the vectorization
3623 factor against the number of loop iterations if the loop is
3626 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3627 Alan Hayward <alan.hayward@arm.com>
3628 David Sherwood <david.sherwood@arm.com>
3630 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3631 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3632 (dump_groups): Update accordingly.
3633 (iv_use::mem_type): New member variable.
3634 (address_p): New function.
3635 (record_use): Add a mem_type argument and initialize the new
3637 (record_group_use): Add a mem_type argument. Use address_p.
3638 Remove obsolete null checks of base_object. Update call to record_use.
3639 (find_interesting_uses_op): Update call to record_group_use.
3640 (find_interesting_uses_cond): Likewise.
3641 (find_interesting_uses_address): Likewise.
3642 (get_mem_type_for_internal_fn): New function.
3643 (find_address_like_use): Likewise.
3644 (find_interesting_uses_stmt): Try find_address_like_use before
3645 calling find_interesting_uses_op.
3646 (addr_offset_valid_p): Use the iv mem_type field as the type
3647 of the addressed memory.
3648 (add_autoinc_candidates): Likewise.
3649 (get_address_cost): Likewise.
3650 (split_small_address_groups_p): Use address_p.
3651 (split_address_groups): Likewise.
3652 (add_iv_candidate_for_use): Likewise.
3653 (autoinc_possible_for_pair): Likewise.
3654 (rewrite_groups): Likewise.
3655 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3656 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3657 (get_alias_ptr_type_for_ptr_address): New function.
3658 (rewrite_use_address): Rewrite address uses in calls that were
3659 identified by find_address_like_use.
3661 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3662 Alan Hayward <alan.hayward@arm.com>
3663 David Sherwood <david.sherwood@arm.com>
3665 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3667 * gimple-expr.h (is_gimple_addressable: Likewise.
3668 * gimple-expr.c (is_gimple_address): Likewise.
3669 * internal-fn.c (expand_call_mem_ref): New function.
3670 (expand_mask_load_optab_fn): Use it.
3671 (expand_mask_store_optab_fn): Likewise.
3673 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3674 Alan Hayward <alan.hayward@arm.com>
3675 David Sherwood <david.sherwood@arm.com>
3677 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3678 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3679 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3680 (cond_umax@var{mode}): Document.
3681 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3682 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3683 (cond_umin_optab, cond_umax_optab): New optabs.
3684 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3685 (COND_IOR, COND_XOR): New internal functions.
3686 * internal-fn.h (get_conditional_internal_fn): Declare.
3687 * internal-fn.c (cond_binary_direct): New macro.
3688 (expand_cond_binary_optab_fn): Likewise.
3689 (direct_cond_binary_optab_supported_p): Likewise.
3690 (get_conditional_internal_fn): New function.
3691 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3692 Cope with reduction statements that are vectorized as calls rather
3694 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3695 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3696 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3697 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3698 (UNSPEC_COND_EOR): New unspecs.
3699 (optab): Add mappings for them.
3700 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3701 (sve_int_op, sve_fp_op): New int attributes.
3703 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3704 Alan Hayward <alan.hayward@arm.com>
3705 David Sherwood <david.sherwood@arm.com>
3707 * optabs.def (while_ult_optab): New optab.
3708 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3709 * internal-fn.def (WHILE_ULT): New internal function.
3710 * internal-fn.h (direct_internal_fn_supported_p): New override
3711 that takes two types as argument.
3712 * internal-fn.c (while_direct): New macro.
3713 (expand_while_optab_fn): New function.
3714 (convert_optab_supported_p): Likewise.
3715 (direct_while_optab_supported_p): New macro.
3716 * wide-int.h (wi::udiv_ceil): New function.
3717 * tree-vectorizer.h (rgroup_masks): New structure.
3718 (vec_loop_masks): New typedef.
3719 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3721 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3722 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3723 (vect_max_vf): New function.
3724 (slpeel_make_loop_iterate_ntimes): Delete.
3725 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3726 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3727 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3728 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3729 internal-fn.h, stor-layout.h and optabs-query.h.
3730 (vect_set_loop_mask): New function.
3731 (add_preheader_seq): Likewise.
3732 (add_header_seq): Likewise.
3733 (interleave_supported_p): Likewise.
3734 (vect_maybe_permute_loop_masks): Likewise.
3735 (vect_set_loop_masks_directly): Likewise.
3736 (vect_set_loop_condition_masked): Likewise.
3737 (vect_set_loop_condition_unmasked): New function, split out from
3738 slpeel_make_loop_iterate_ntimes.
3739 (slpeel_make_loop_iterate_ntimes): Rename to..
3740 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3741 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3742 (vect_do_peeling): Update call accordingly.
3743 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3745 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3746 mask_compare_type, can_fully_mask_p and fully_masked_p.
3747 (release_vec_loop_masks): New function.
3748 (_loop_vec_info): Use it to free the loop masks.
3749 (can_produce_all_loop_masks_p): New function.
3750 (vect_get_max_nscalars_per_iter): Likewise.
3751 (vect_verify_full_masking): Likewise.
3752 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3753 retries, and free the mask rgroups before retrying. Check loop-wide
3754 reasons for disallowing fully-masked loops. Make the final decision
3755 about whether use a fully-masked loop or not.
3756 (vect_estimate_min_profitable_iters): Do not assume that peeling
3757 for the number of iterations will be needed for fully-masked loops.
3758 (vectorizable_reduction): Disable fully-masked loops.
3759 (vectorizable_live_operation): Likewise.
3760 (vect_halve_mask_nunits): New function.
3761 (vect_double_mask_nunits): Likewise.
3762 (vect_record_loop_mask): Likewise.
3763 (vect_get_loop_mask): Likewise.
3764 (vect_transform_loop): Handle the case in which the final loop
3765 iteration might handle a partial vector. Call vect_set_loop_condition
3766 instead of slpeel_make_loop_iterate_ntimes.
3767 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3768 (check_load_store_masking): New function.
3769 (prepare_load_store_mask): Likewise.
3770 (vectorizable_store): Handle fully-masked loops.
3771 (vectorizable_load): Likewise.
3772 (supportable_widening_operation): Use vect_halve_mask_nunits for
3774 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3775 (vect_gen_while): New function.
3776 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3777 (aarch64_uqdec<mode>): New insn.
3779 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3780 Alan Hayward <alan.hayward@arm.com>
3781 David Sherwood <david.sherwood@arm.com>
3783 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3784 (reduc_xor_scal_optab): New optabs.
3785 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3786 (reduc_xor_scal_@var{m}): Document.
3787 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3788 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3790 * fold-const-call.c (fold_const_call): Handle them.
3791 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3792 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3793 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3794 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3795 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3796 (UNSPEC_XORV): New unspecs.
3797 (optab): Add entries for them.
3798 (BITWISEV): New int iterator.
3799 (bit_reduc_op): New int attributes.
3801 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3802 Alan Hayward <alan.hayward@arm.com>
3803 David Sherwood <david.sherwood@arm.com>
3805 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3806 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3807 * optabs.def (vec_shl_insert_optab): New optab.
3808 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3809 (duplicate_and_interleave): Likewise.
3810 * tree-vect-loop.c: Include internal-fn.h.
3811 (neutral_op_for_slp_reduction): New function, split out from
3812 get_initial_defs_for_reduction.
3813 (get_initial_def_for_reduction): Handle option 2 for variable-length
3814 vectors by loading the neutral value into a vector and then shifting
3815 the initial value into element 0.
3816 (get_initial_defs_for_reduction): Replace the code argument with
3817 the neutral value calculated by neutral_op_for_slp_reduction.
3818 Use gimple_build_vector for constant-length vectors.
3819 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3820 but the first group_size elements have a neutral value.
3821 Use duplicate_and_interleave otherwise.
3822 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3823 Update call to get_initial_defs_for_reduction. Handle SLP
3824 reductions for variable-length vectors by creating one vector
3825 result for each scalar result, with the elements associated
3826 with other scalar results stubbed out with the neutral value.
3827 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3828 Require IFN_VEC_SHL_INSERT for double reductions on
3829 variable-length vectors, or SLP reductions that have
3830 a neutral value. Require can_duplicate_and_interleave_p
3831 support for variable-length unchained SLP reductions if there
3832 is no neutral value, such as for MIN/MAX reductions. Also require
3833 the number of vector elements to be a multiple of the number of
3834 SLP statements when doing variable-length unchained SLP reductions.
3835 Update call to vect_create_epilog_for_reduction.
3836 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3837 and remove initial values.
3838 (duplicate_and_interleave): Make public.
3839 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3840 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3842 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3843 Alan Hayward <alan.hayward@arm.com>
3844 David Sherwood <david.sherwood@arm.com>
3846 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3847 (can_duplicate_and_interleave_p): New function.
3848 (vect_get_and_check_slp_defs): Take the vector of statements
3849 rather than just the current one. Remove excess parentheses.
3850 Restriction rejectinon of vect_constant_def and vect_external_def
3851 for variable-length vectors to boolean types, or types for which
3852 can_duplicate_and_interleave_p is false.
3853 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3854 (duplicate_and_interleave): New function.
3855 (vect_get_constant_vectors): Use gimple_build_vector for
3856 constant-length vectors and suitable variable-length constant
3857 vectors. Use duplicate_and_interleave for other variable-length
3858 vectors. Don't defer the update when inserting new statements.
3860 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3861 Alan Hayward <alan.hayward@arm.com>
3862 David Sherwood <david.sherwood@arm.com>
3864 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3865 min_profitable_iters doesn't go negative.
3867 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3868 Alan Hayward <alan.hayward@arm.com>
3869 David Sherwood <david.sherwood@arm.com>
3871 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3872 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3873 * optabs.def (vec_mask_load_lanes_optab): New optab.
3874 (vec_mask_store_lanes_optab): Likewise.
3875 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3876 (MASK_STORE_LANES): Likewise.
3877 * internal-fn.c (mask_load_lanes_direct): New macro.
3878 (mask_store_lanes_direct): Likewise.
3879 (expand_mask_load_optab_fn): Handle masked operations.
3880 (expand_mask_load_lanes_optab_fn): New macro.
3881 (expand_mask_store_optab_fn): Handle masked operations.
3882 (expand_mask_store_lanes_optab_fn): New macro.
3883 (direct_mask_load_lanes_optab_supported_p): Likewise.
3884 (direct_mask_store_lanes_optab_supported_p): Likewise.
3885 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3887 (vect_load_lanes_supported): Likewise.
3888 * tree-vect-data-refs.c (strip_conversion): New function.
3889 (can_group_stmts_p): Likewise.
3890 (vect_analyze_data_ref_accesses): Use it instead of checking
3891 for a pair of assignments.
3892 (vect_store_lanes_supported): Take a masked_p parameter.
3893 (vect_load_lanes_supported): Likewise.
3894 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3895 vect_store_lanes_supported and vect_load_lanes_supported.
3896 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3897 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3898 parameter. Don't allow gaps for masked accesses.
3899 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3900 and vect_load_lanes_supported.
3901 (get_load_store_type): Take a masked_p parameter and update
3902 call to get_group_load_store_type.
3903 (vectorizable_store): Update call to get_load_store_type.
3904 Handle IFN_MASK_STORE_LANES.
3905 (vectorizable_load): Update call to get_load_store_type.
3906 Handle IFN_MASK_LOAD_LANES.
3908 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3909 Alan Hayward <alan.hayward@arm.com>
3910 David Sherwood <david.sherwood@arm.com>
3912 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3914 * config/aarch64/aarch64-protos.h
3915 (aarch64_sve_struct_memory_operand_p): Declare.
3916 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3917 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3918 (VPRED, vpred): Handle SVE structure modes.
3919 * config/aarch64/constraints.md (Utx): New constraint.
3920 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3921 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3922 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3923 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3924 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3925 structure modes. Split into pieces after RA.
3926 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3927 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3929 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3930 SVE structure modes.
3931 (aarch64_classify_address): Likewise.
3932 (sizetochar): Move earlier in file.
3933 (aarch64_print_operand): Handle SVE register lists.
3934 (aarch64_array_mode): New function.
3935 (aarch64_sve_struct_memory_operand_p): Likewise.
3936 (TARGET_ARRAY_MODE): Redefine.
3938 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3939 Alan Hayward <alan.hayward@arm.com>
3940 David Sherwood <david.sherwood@arm.com>
3942 * target.def (array_mode): New target hook.
3943 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3944 * doc/tm.texi: Regenerate.
3945 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3946 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3947 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3949 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3952 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3953 Alan Hayward <alan.hayward@arm.com>
3954 David Sherwood <david.sherwood@arm.com>
3956 * fold-const.c (fold_binary_loc): Check the argument types
3957 rather than the result type when testing for a vector operation.
3959 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3961 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
3962 * doc/tm.texi: Regenerate.
3964 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3965 Alan Hayward <alan.hayward@arm.com>
3966 David Sherwood <david.sherwood@arm.com>
3968 * doc/invoke.texi (-msve-vector-bits=): Document new option.
3969 (sve): Document new AArch64 extension.
3970 * doc/md.texi (w): Extend the description of the AArch64
3971 constraint to include SVE vectors.
3972 (Upl, Upa): Document new AArch64 predicate constraints.
3973 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
3975 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
3976 (msve-vector-bits=): New option.
3977 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
3978 SVE when these are disabled.
3979 (sve): New extension.
3980 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
3981 modes. Adjust their number of units based on aarch64_sve_vg.
3982 (MAX_BITSIZE_MODE_ANY_MODE): Define.
3983 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
3984 aarch64_addr_query_type.
3985 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
3986 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
3987 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
3988 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
3989 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
3990 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
3991 (aarch64_simd_imm_zero_p): Delete.
3992 (aarch64_check_zero_based_sve_index_immediate): Declare.
3993 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
3994 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
3995 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
3996 (aarch64_sve_float_mul_immediate_p): Likewise.
3997 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
3999 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
4000 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
4001 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
4002 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
4003 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
4004 (aarch64_regmode_natural_size): Likewise.
4005 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
4006 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
4008 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
4009 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
4010 for VG and the SVE predicate registers.
4011 (V_ALIASES): Add a "z"-prefixed alias.
4012 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
4013 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
4014 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
4015 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
4016 (REG_CLASS_NAMES): Add entries for them.
4017 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
4018 and the predicate registers.
4019 (aarch64_sve_vg): Declare.
4020 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
4021 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
4022 (REGMODE_NATURAL_SIZE): Define.
4023 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
4025 * config/aarch64/aarch64.c: Include cfgrtl.h.
4026 (simd_immediate_info): Add a constructor for series vectors,
4027 and an associated step field.
4028 (aarch64_sve_vg): New variable.
4029 (aarch64_dbx_register_number): Handle VG and the predicate registers.
4030 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
4031 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
4032 (VEC_ANY_DATA, VEC_STRUCT): New constants.
4033 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
4034 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
4035 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
4036 (aarch64_get_mask_mode): New functions.
4037 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
4038 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
4039 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
4040 predicate modes and predicate registers. Explicitly restrict
4041 GPRs to modes of 16 bytes or smaller. Only allow FP registers
4042 to store a vector mode if it is recognized by
4043 aarch64_classify_vector_mode.
4044 (aarch64_regmode_natural_size): New function.
4045 (aarch64_hard_regno_caller_save_mode): Return the original mode
4047 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
4048 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
4049 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
4050 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
4052 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
4053 does not overlap dest if the function is frame-related. Handle
4055 (aarch64_split_add_offset): New function.
4056 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
4057 them aarch64_add_offset.
4058 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
4059 and update call to aarch64_sub_sp.
4060 (aarch64_add_cfa_expression): New function.
4061 (aarch64_expand_prologue): Pass extra temporary registers to the
4062 functions above. Handle the case in which we need to emit new
4063 DW_CFA_expressions for registers that were originally saved
4064 relative to the stack pointer, but now have to be expressed
4065 relative to the frame pointer.
4066 (aarch64_output_mi_thunk): Pass extra temporary registers to the
4068 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
4069 IP0 and IP1 values for SVE frames.
4070 (aarch64_expand_vec_series): New function.
4071 (aarch64_expand_sve_widened_duplicate): Likewise.
4072 (aarch64_expand_sve_const_vector): Likewise.
4073 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
4074 Handle SVE constants. Use emit_move_insn to move a force_const_mem
4075 into the register, rather than emitting a SET directly.
4076 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
4077 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
4078 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
4079 (offset_9bit_signed_scaled_p): New functions.
4080 (aarch64_replicate_bitmask_imm): New function.
4081 (aarch64_bitmask_imm): Use it.
4082 (aarch64_cannot_force_const_mem): Reject expressions involving
4083 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
4084 (aarch64_classify_index): Handle SVE indices, by requiring
4085 a plain register index with a scale that matches the element size.
4086 (aarch64_classify_address): Handle SVE addresses. Assert that
4087 the mode of the address is VOIDmode or an integer mode.
4088 Update call to aarch64_classify_symbol.
4089 (aarch64_classify_symbolic_expression): Update call to
4090 aarch64_classify_symbol.
4091 (aarch64_const_vec_all_in_range_p): New function.
4092 (aarch64_print_vector_float_operand): Likewise.
4093 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
4094 "vN" for FP registers with SVE modes. Handle (const ...) vectors
4095 and the FP immediates 1.0 and 0.5.
4096 (aarch64_print_address_internal): Handle SVE addresses.
4097 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
4098 (aarch64_regno_regclass): Handle predicate registers.
4099 (aarch64_secondary_reload): Handle big-endian reloads of SVE
4101 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
4102 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
4103 (aarch64_convert_sve_vector_bits): New function.
4104 (aarch64_override_options): Use it to handle -msve-vector-bits=.
4105 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4107 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
4108 Handle SVE vector and predicate modes. Accept VL-based constants
4109 that need only one temporary register, and VL offsets that require
4110 no temporary registers.
4111 (aarch64_conditional_register_usage): Mark the predicate registers
4112 as fixed if SVE isn't available.
4113 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
4114 Return true for SVE vector and predicate modes.
4115 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
4116 rather than an unsigned int. Handle SVE modes.
4117 (aarch64_preferred_simd_mode): Update call accordingly. Handle
4119 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
4121 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4122 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4123 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4124 (aarch64_sve_float_mul_immediate_p): New functions.
4125 (aarch64_sve_valid_immediate): New function.
4126 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
4127 Explicitly reject structure modes. Check for INDEX constants.
4128 Handle PTRUE and PFALSE constants.
4129 (aarch64_check_zero_based_sve_index_immediate): New function.
4130 (aarch64_simd_imm_zero_p): Delete.
4131 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
4132 vector modes. Accept constants in the range of CNT[BHWD].
4133 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
4134 ask for an Advanced SIMD mode.
4135 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
4136 (aarch64_simd_vector_alignment): Handle SVE predicates.
4137 (aarch64_vectorize_preferred_vector_alignment): New function.
4138 (aarch64_simd_vector_alignment_reachable): Use it instead of
4140 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
4141 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
4143 (MAX_VECT_LEN): Delete.
4144 (expand_vec_perm_d): Add a vec_flags field.
4145 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
4146 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
4147 (aarch64_evpc_ext): Don't apply a big-endian lane correction
4149 (aarch64_evpc_rev): Rename to...
4150 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
4151 (aarch64_evpc_rev_global): New function.
4152 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
4153 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
4155 (aarch64_evpc_sve_tbl): New function.
4156 (aarch64_expand_vec_perm_const_1): Update after rename of
4157 aarch64_evpc_rev. Handle SVE permutes too, trying
4158 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
4159 than aarch64_evpc_tbl.
4160 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
4161 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
4162 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
4163 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
4164 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
4165 (aarch64_expand_sve_vcond): New functions.
4166 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
4167 of aarch64_vector_mode_p.
4168 (aarch64_dwarf_poly_indeterminate_value): New function.
4169 (aarch64_compute_pressure_classes): Likewise.
4170 (aarch64_can_change_mode_class): Likewise.
4171 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
4172 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
4173 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
4174 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
4175 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
4176 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
4177 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
4178 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
4180 (Dn, Dl, Dr): Accept const as well as const_vector.
4181 (Dz): Likewise. Compare against CONST0_RTX.
4182 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
4183 of "vector" where appropriate.
4184 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
4185 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
4186 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
4187 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
4188 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
4189 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
4190 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
4191 (v_int_equiv): Extend to SVE modes.
4192 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
4194 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
4195 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4196 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4197 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4198 (SVE_COND_FP_CMP): New int iterators.
4199 (perm_hilo): Handle the new unpack unspecs.
4200 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4202 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4203 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4204 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4205 (aarch64_equality_operator, aarch64_constant_vector_operand)
4206 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4207 (aarch64_sve_nonimmediate_operand): Likewise.
4208 (aarch64_sve_general_operand): Likewise.
4209 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4210 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4211 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4212 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4213 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4214 (aarch64_sve_float_arith_immediate): Likewise.
4215 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4216 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4217 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4218 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4219 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4220 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4221 (aarch64_sve_float_arith_operand): Likewise.
4222 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4223 (aarch64_sve_float_mul_operand): Likewise.
4224 (aarch64_sve_vec_perm_operand): Likewise.
4225 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4226 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4227 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4228 as well as const_vector.
4229 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4230 in file. Use CONST0_RTX and CONSTM1_RTX.
4231 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4232 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4233 Use aarch64_simd_imm_zero.
4234 * config/aarch64/aarch64-sve.md: New file.
4235 * config/aarch64/aarch64.md: Include it.
4236 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4237 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4238 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4239 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4240 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4241 (sve): New attribute.
4242 (enabled): Disable instructions with the sve attribute unless
4244 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4245 aarch64_expand_mov_immediate.
4246 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4247 CNT[BHSD] immediates.
4248 (movti): Split CONST_POLY_INT moves into two halves.
4249 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4250 Split additions that need a temporary here if the destination
4251 is the stack pointer.
4252 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4253 (*add<mode>3_poly_1): New instruction.
4254 (set_clobber_cc): New expander.
4256 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4258 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4259 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4260 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4261 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4262 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4263 Change innermode from fixed_mode_size to machine_mode.
4264 (simplify_subreg): Update call accordingly. Handle a constant-sized
4265 subreg of a variable-length CONST_VECTOR.
4267 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4268 Alan Hayward <alan.hayward@arm.com>
4269 David Sherwood <david.sherwood@arm.com>
4271 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4272 (add_offset_to_base): New function, split out from...
4273 (create_mem_ref): ...here. When handling a scale other than 1,
4274 check first whether the address is valid without the offset.
4275 Add it into the base if so, leaving the index and scale as-is.
4277 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4280 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4281 fold_for_warn before checking if arg2 is INTEGER_CST.
4283 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4285 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4286 (store_multiple_operation): Delete.
4287 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4288 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4289 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4290 guarded by TARGET_STRING.
4291 (rs6000_output_load_multiple): Delete.
4292 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4293 OPTION_MASK_STRING / TARGET_STRING handling.
4294 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4295 (const rs6000_opt_masks) <"string">: Change mask to 0.
4296 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4297 (MASK_STRING): Delete.
4298 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4300 (load_multiple): Delete.
4307 (store_multiple): Delete.
4314 (movmemsi_8reg): Delete.
4315 (corresponding unnamed define_insn): Delete.
4316 (movmemsi_6reg): Delete.
4317 (corresponding unnamed define_insn): Delete.
4318 (movmemsi_4reg): Delete.
4319 (corresponding unnamed define_insn): Delete.
4320 (movmemsi_2reg): Delete.
4321 (corresponding unnamed define_insn): Delete.
4322 (movmemsi_1reg): Delete.
4323 (corresponding unnamed define_insn): Delete.
4324 * config/rs6000/rs6000.opt (mno-string): New.
4325 (mstring): Replace by deprecation warning stub.
4326 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4328 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4330 * regrename.c (regrename_do_replace): If replacing the same
4331 reg multiple times, try to reuse last created gen_raw_REG.
4334 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4335 main to workaround a bug in GDB.
4337 2018-01-12 Tom de Vries <tom@codesourcery.com>
4340 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4342 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4344 PR rtl-optimization/80481
4345 * ira-color.c (get_cap_member): New function.
4346 (allocnos_conflict_by_live_ranges_p): Use it.
4347 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4348 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4350 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4353 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4354 (*saddl_se_1): Ditto.
4356 (*ssubl_se_1): Ditto.
4358 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4360 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4361 rather than wi::to_widest for DR_INITs.
4362 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4363 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4364 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4366 (vect_analyze_group_access_1): Note that here.
4368 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4370 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4371 polynomial type sizes.
4373 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4375 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4376 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4377 (gimple_add_tmp_var): Likewise.
4379 2018-01-12 Martin Liska <mliska@suse.cz>
4381 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4382 (gimple_alloc_sizes): Likewise.
4383 (dump_gimple_statistics): Use PRIu64 in printf format.
4384 * gimple.h: Change uint64_t to int.
4386 2018-01-12 Martin Liska <mliska@suse.cz>
4388 * tree-core.h: Use uint64_t instead of int.
4389 * tree.c (tree_node_counts): Likewise.
4390 (tree_node_sizes): Likewise.
4391 (dump_tree_statistics): Use PRIu64 in printf format.
4393 2018-01-12 Martin Liska <mliska@suse.cz>
4395 * Makefile.in: As qsort_chk is implemented in vec.c, add
4396 vec.o to linkage of gencfn-macros.
4397 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4398 passing the info to record_node_allocation_statistics.
4399 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4401 * ggc-common.c (struct ggc_usage): Add operator== and use
4402 it in operator< and compare function.
4403 * mem-stats.h (struct mem_usage): Likewise.
4404 * vec.c (struct vec_usage): Remove operator< and compare
4405 function. Can be simply inherited.
4407 2018-01-12 Martin Jambor <mjambor@suse.cz>
4410 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4411 * tree-ssa-math-opts.c: Include domwalk.h.
4412 (convert_mult_to_fma_1): New function.
4413 (fma_transformation_info): New type.
4414 (fma_deferring_state): Likewise.
4415 (cancel_fma_deferring): New function.
4416 (result_of_phi): Likewise.
4417 (last_fma_candidate_feeds_initial_phi): Likewise.
4418 (convert_mult_to_fma): Added deferring logic, split actual
4419 transformation to convert_mult_to_fma_1.
4420 (math_opts_dom_walker): New type.
4421 (math_opts_dom_walker::after_dom_children): New method, body moved
4422 here from pass_optimize_widening_mul::execute, added deferring logic
4424 (pass_optimize_widening_mul::execute): Moved most of code to
4425 math_opts_dom_walker::after_dom_children.
4426 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4427 * config/i386/i386.c (ix86_option_override_internal): Added
4428 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4430 2018-01-12 Richard Biener <rguenther@suse.de>
4433 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4434 inline instance vars.
4436 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4439 * config/rx/rx.c (rx_is_restricted_memory_address):
4442 2018-01-12 Richard Biener <rguenther@suse.de>
4444 PR tree-optimization/80846
4445 * target.def (split_reduction): New target hook.
4446 * targhooks.c (default_split_reduction): New function.
4447 * targhooks.h (default_split_reduction): Declare.
4448 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4449 target requests first reduce vectors by combining low and high
4451 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4452 (get_vectype_for_scalar_type_and_size): Export.
4453 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4454 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4455 * doc/tm.texi: Regenerate.
4456 * config/i386/i386.c (ix86_split_reduction): Implement
4457 TARGET_VECTORIZE_SPLIT_REDUCTION.
4459 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4462 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4463 in PIC mode except for TARGET_VXWORKS_RTP.
4464 * config/sparc/sparc.c: Include cfgrtl.h.
4465 (TARGET_INIT_PIC_REG): Define.
4466 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4467 (sparc_pic_register_p): New predicate.
4468 (sparc_legitimate_address_p): Use it.
4469 (sparc_legitimize_pic_address): Likewise.
4470 (sparc_delegitimize_address): Likewise.
4471 (sparc_mode_dependent_address_p): Likewise.
4472 (gen_load_pcrel_sym): Remove 4th parameter.
4473 (load_got_register): Adjust call to above. Remove obsolete stuff.
4474 (sparc_expand_prologue): Do not call load_got_register here.
4475 (sparc_flat_expand_prologue): Likewise.
4476 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4477 (sparc_use_pseudo_pic_reg): New function.
4478 (sparc_init_pic_reg): Likewise.
4479 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4480 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4482 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4484 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4485 Add item for branch_cost.
4487 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4489 PR rtl-optimization/83565
4490 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4491 not extend the result to a larger mode for rotate operations.
4492 (num_sign_bit_copies1): Likewise.
4494 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4497 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4499 Use values-Xc.o for -pedantic.
4500 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4502 2018-01-12 Martin Liska <mliska@suse.cz>
4505 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4507 (possible_polymorphic_call_targets): Use it.
4508 (ipa_devirt): Likewise.
4510 2018-01-12 Martin Liska <mliska@suse.cz>
4512 * profile-count.h (enum profile_quality): Use 0 as invalid
4513 enum value of profile_quality.
4515 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4517 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4518 -mext-string options.
4520 2018-01-12 Richard Biener <rguenther@suse.de>
4522 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4523 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4524 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4526 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4528 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4530 * configure.ac (--with-long-double-format): Add support for the
4531 configuration option to change the default long double format on
4533 * config.gcc (powerpc*-linux*-*): Likewise.
4534 * configure: Regenerate.
4535 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4536 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4537 used without modification.
4539 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4541 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4542 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4543 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4544 MISC_BUILTIN_SPEC_BARRIER.
4545 (rs6000_init_builtins): Likewise.
4546 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4548 (speculation_barrier): New define_insn.
4549 * doc/extend.texi: Document __builtin_speculation_barrier.
4551 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4554 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4555 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4556 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4558 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4559 integral modes instead of "ss" and "sd".
4560 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4561 vectors with 32-bit and 64-bit elements.
4562 (vecdupssescalarmodesuffix): New mode attribute.
4563 (vec_dup<mode>): Use it.
4565 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4568 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4569 frame if argument is passed on stack.
4571 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4574 * ree.c (combine_reaching_defs): Optimize also
4575 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4576 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4578 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4581 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4583 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4586 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4587 after they are computed.
4589 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4591 PR tree-optimization/83695
4592 * gimple-loop-linterchange.cc
4593 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4594 reset cached scev information after interchange.
4595 (pass_linterchange::execute): Remove call to scev_reset_htab.
4597 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4599 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4600 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4601 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4602 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4603 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4604 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4605 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4606 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4607 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4608 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4609 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4610 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4611 (V_lane_reg): Likewise.
4612 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4614 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4615 (vfmal_lane_low<mode>_intrinsic,
4616 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4617 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4618 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4619 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4620 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4621 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4623 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4625 * config/arm/arm-cpus.in (fp16fml): New feature.
4626 (ALL_SIMD): Add fp16fml.
4627 (armv8.2-a): Add fp16fml as an option.
4628 (armv8.3-a): Likewise.
4629 (armv8.4-a): Add fp16fml as part of fp16.
4630 * config/arm/arm.h (TARGET_FP16FML): Define.
4631 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4633 * config/arm/arm-modes.def (V2HF): Define.
4634 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4635 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4636 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4637 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4638 vfmsl_low, vfmsl_high): New set of builtins.
4639 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4640 (vfml_op): New code attribute.
4641 (VFMLHALVES): New int iterator.
4642 (VFML, VFMLSEL): New mode attributes.
4643 (V_reg): Define mapping for V2HF.
4644 (V_hi, V_lo): New mode attributes.
4645 (VF_constraint): Likewise.
4646 (vfml_half, vfml_half_selector): New int attributes.
4647 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4649 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4650 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4652 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4653 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4654 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4655 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4657 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4658 Document new effective target and option set.
4660 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4662 * config/arm/arm-cpus.in (armv8_4): New feature.
4663 (ARMv8_4a): New fgroup.
4664 (armv8.4-a): New arch.
4665 * config/arm/arm-tables.opt: Regenerate.
4666 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4667 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4668 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4669 Add matching rules for -march=armv8.4-a and extensions.
4670 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4672 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4675 * config/rx/rx.md (BW): New mode attribute.
4676 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4678 2018-01-11 Richard Biener <rguenther@suse.de>
4680 PR tree-optimization/83435
4681 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4682 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4683 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4685 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4686 Alan Hayward <alan.hayward@arm.com>
4687 David Sherwood <david.sherwood@arm.com>
4689 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4691 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4692 (aarch64_print_address_internal): Use it to check for a zero offset.
4694 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4695 Alan Hayward <alan.hayward@arm.com>
4696 David Sherwood <david.sherwood@arm.com>
4698 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4699 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4700 Return a poly_int64 rather than a HOST_WIDE_INT.
4701 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4702 rather than a HOST_WIDE_INT.
4703 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4704 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4705 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4706 final_offset from HOST_WIDE_INT to poly_int64.
4707 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4708 to_constant when getting the number of units in an Advanced SIMD
4710 (aarch64_builtin_vectorized_function): Check for a constant number
4712 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4714 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4715 attribute instead of GET_MODE_NUNITS.
4716 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4717 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4718 GET_MODE_SIZE for fixed-size registers.
4719 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4720 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4721 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4722 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4723 (aarch64_print_operand, aarch64_print_address_internal)
4724 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4725 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4726 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4727 Handle polynomial GET_MODE_SIZE.
4728 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4729 wider than SImode without modification.
4730 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4731 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4732 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4733 passing and returning SVE modes.
4734 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4735 rather than GEN_INT.
4736 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4737 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4738 (aarch64_allocate_and_probe_stack_space): Likewise.
4739 (aarch64_layout_frame): Cope with polynomial offsets.
4740 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4741 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4743 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4744 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4745 poly_int64 rather than a HOST_WIDE_INT.
4746 (aarch64_get_separate_components, aarch64_process_components)
4747 (aarch64_expand_prologue, aarch64_expand_epilogue)
4748 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4749 (aarch64_anchor_offset): New function, split out from...
4750 (aarch64_legitimize_address): ...here.
4751 (aarch64_builtin_vectorization_cost): Handle polynomial
4752 TYPE_VECTOR_SUBPARTS.
4753 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4755 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4756 number of elements from the PARALLEL rather than the mode.
4757 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4758 rather than GET_MODE_BITSIZE.
4759 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4760 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4761 (aarch64_expand_vec_perm_const_1): Handle polynomial
4762 d->perm.length () and d->perm elements.
4763 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4764 Apply to_constant to d->perm elements.
4765 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4766 polynomial CONST_VECTOR_NUNITS.
4767 (aarch64_move_pointer): Take amount as a poly_int64 rather
4769 (aarch64_progress_pointer): Avoid temporary variable.
4770 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4771 the mode attribute instead of GET_MODE.
4773 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4774 Alan Hayward <alan.hayward@arm.com>
4775 David Sherwood <david.sherwood@arm.com>
4777 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4778 x exists before using it.
4779 (aarch64_add_constant_internal): Rename to...
4780 (aarch64_add_offset_1): ...this. Replace regnum with separate
4781 src and dest rtxes. Handle the case in which they're different,
4782 including when the offset is zero. Replace scratchreg with an rtx.
4783 Use 2 additions if there is no spare register into which we can
4784 move a 16-bit constant.
4785 (aarch64_add_constant): Delete.
4786 (aarch64_add_offset): Replace reg with separate src and dest
4787 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4788 Use aarch64_add_offset_1.
4789 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4790 an rtx rather than an int. Take the delta as a poly_int64
4791 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4792 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4793 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4794 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4795 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4797 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4798 aarch64_add_constant.
4800 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4802 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4803 Use scalar_float_mode.
4805 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4807 * config/aarch64/aarch64-simd.md
4808 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4809 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4810 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4811 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4812 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4813 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4814 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4815 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4816 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4817 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4819 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4822 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4823 targ_options->x_arm_arch_string is non NULL.
4825 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4827 * config/aarch64/aarch64.h
4828 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4830 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4833 * expmed.c (emit_store_flag_force): Swap if const op0
4834 and change VOIDmode to mode of op0.
4836 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4838 PR rtl-optimization/83761
4839 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4840 than bytes to mode_for_size.
4842 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4845 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4846 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4849 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4852 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4853 when in layout mode.
4854 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4855 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4858 2018-01-10 Michael Collison <michael.collison@arm.com>
4860 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4861 * config/aarch64/aarch64-option-extension.def: Add
4862 AARCH64_OPT_EXTENSION of 'fp16fml'.
4863 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4864 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4865 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4866 * config/aarch64/constraints.md (Ui7): New constraint.
4867 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4868 (VFMLA_SEL_W): Ditto.
4871 (VFMLA16_LOW): New int iterator.
4872 (VFMLA16_HIGH): Ditto.
4873 (UNSPEC_FMLAL): New unspec.
4874 (UNSPEC_FMLSL): Ditto.
4875 (UNSPEC_FMLAL2): Ditto.
4876 (UNSPEC_FMLSL2): Ditto.
4877 (f16mac): New code attribute.
4878 * config/aarch64/aarch64-simd-builtins.def
4879 (aarch64_fmlal_lowv2sf): Ditto.
4880 (aarch64_fmlsl_lowv2sf): Ditto.
4881 (aarch64_fmlalq_lowv4sf): Ditto.
4882 (aarch64_fmlslq_lowv4sf): Ditto.
4883 (aarch64_fmlal_highv2sf): Ditto.
4884 (aarch64_fmlsl_highv2sf): Ditto.
4885 (aarch64_fmlalq_highv4sf): Ditto.
4886 (aarch64_fmlslq_highv4sf): Ditto.
4887 (aarch64_fmlal_lane_lowv2sf): Ditto.
4888 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4889 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4890 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4891 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4892 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4893 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4894 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4895 (aarch64_fmlal_lane_highv2sf): Ditto.
4896 (aarch64_fmlsl_lane_highv2sf): Ditto.
4897 (aarch64_fmlal_laneq_highv2sf): Ditto.
4898 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4899 (aarch64_fmlalq_lane_highv4sf): Ditto.
4900 (aarch64_fmlsl_lane_highv4sf): Ditto.
4901 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4902 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4903 * config/aarch64/aarch64-simd.md:
4904 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4905 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4906 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4907 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4908 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4909 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4910 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4911 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4912 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4913 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4914 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4915 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4916 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4917 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4918 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4919 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4920 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4921 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4922 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4923 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4924 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4925 (vfmlsl_low_u32): Ditto.
4926 (vfmlalq_low_u32): Ditto.
4927 (vfmlslq_low_u32): Ditto.
4928 (vfmlal_high_u32): Ditto.
4929 (vfmlsl_high_u32): Ditto.
4930 (vfmlalq_high_u32): Ditto.
4931 (vfmlslq_high_u32): Ditto.
4932 (vfmlal_lane_low_u32): Ditto.
4933 (vfmlsl_lane_low_u32): Ditto.
4934 (vfmlal_laneq_low_u32): Ditto.
4935 (vfmlsl_laneq_low_u32): Ditto.
4936 (vfmlalq_lane_low_u32): Ditto.
4937 (vfmlslq_lane_low_u32): Ditto.
4938 (vfmlalq_laneq_low_u32): Ditto.
4939 (vfmlslq_laneq_low_u32): Ditto.
4940 (vfmlal_lane_high_u32): Ditto.
4941 (vfmlsl_lane_high_u32): Ditto.
4942 (vfmlal_laneq_high_u32): Ditto.
4943 (vfmlsl_laneq_high_u32): Ditto.
4944 (vfmlalq_lane_high_u32): Ditto.
4945 (vfmlslq_lane_high_u32): Ditto.
4946 (vfmlalq_laneq_high_u32): Ditto.
4947 (vfmlslq_laneq_high_u32): Ditto.
4948 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4949 (AARCH64_FL_FOR_ARCH8_4): New.
4950 (AARCH64_ISA_F16FML): New ISA flag.
4951 (TARGET_F16FML): New feature flag for fp16fml.
4952 (doc/invoke.texi): Document new fp16fml option.
4954 2018-01-10 Michael Collison <michael.collison@arm.com>
4956 * config/aarch64/aarch64-builtins.c:
4957 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
4958 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4959 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
4960 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
4961 (AARCH64_ISA_SHA3): New ISA flag.
4962 (TARGET_SHA3): New feature flag for sha3.
4963 * config/aarch64/iterators.md (sha512_op): New int attribute.
4964 (CRYPTO_SHA512): New int iterator.
4965 (UNSPEC_SHA512H): New unspec.
4966 (UNSPEC_SHA512H2): Ditto.
4967 (UNSPEC_SHA512SU0): Ditto.
4968 (UNSPEC_SHA512SU1): Ditto.
4969 * config/aarch64/aarch64-simd-builtins.def
4970 (aarch64_crypto_sha512hqv2di): New builtin.
4971 (aarch64_crypto_sha512h2qv2di): Ditto.
4972 (aarch64_crypto_sha512su0qv2di): Ditto.
4973 (aarch64_crypto_sha512su1qv2di): Ditto.
4974 (aarch64_eor3qv8hi): Ditto.
4975 (aarch64_rax1qv2di): Ditto.
4976 (aarch64_xarqv2di): Ditto.
4977 (aarch64_bcaxqv8hi): Ditto.
4978 * config/aarch64/aarch64-simd.md:
4979 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
4980 (aarch64_crypto_sha512su0qv2di): Ditto.
4981 (aarch64_crypto_sha512su1qv2di): Ditto.
4982 (aarch64_eor3qv8hi): Ditto.
4983 (aarch64_rax1qv2di): Ditto.
4984 (aarch64_xarqv2di): Ditto.
4985 (aarch64_bcaxqv8hi): Ditto.
4986 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
4987 (vsha512h2q_u64): Ditto.
4988 (vsha512su0q_u64): Ditto.
4989 (vsha512su1q_u64): Ditto.
4990 (veor3q_u16): Ditto.
4991 (vrax1q_u64): Ditto.
4993 (vbcaxq_u16): Ditto.
4994 * config/arm/types.md (crypto_sha512): New type attribute.
4995 (crypto_sha3): Ditto.
4996 (doc/invoke.texi): Document new sha3 option.
4998 2018-01-10 Michael Collison <michael.collison@arm.com>
5000 * config/aarch64/aarch64-builtins.c:
5001 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
5002 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5003 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
5004 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
5005 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
5006 (AARCH64_ISA_SM4): New ISA flag.
5007 (TARGET_SM4): New feature flag for sm4.
5008 * config/aarch64/aarch64-simd-builtins.def
5009 (aarch64_sm3ss1qv4si): Ditto.
5010 (aarch64_sm3tt1aq4si): Ditto.
5011 (aarch64_sm3tt1bq4si): Ditto.
5012 (aarch64_sm3tt2aq4si): Ditto.
5013 (aarch64_sm3tt2bq4si): Ditto.
5014 (aarch64_sm3partw1qv4si): Ditto.
5015 (aarch64_sm3partw2qv4si): Ditto.
5016 (aarch64_sm4eqv4si): Ditto.
5017 (aarch64_sm4ekeyqv4si): Ditto.
5018 * config/aarch64/aarch64-simd.md:
5019 (aarch64_sm3ss1qv4si): Ditto.
5020 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
5021 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
5022 (aarch64_sm4eqv4si): Ditto.
5023 (aarch64_sm4ekeyqv4si): Ditto.
5024 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
5025 (sm3part_op): Ditto.
5026 (CRYPTO_SM3TT): Ditto.
5027 (CRYPTO_SM3PART): Ditto.
5028 (UNSPEC_SM3SS1): New unspec.
5029 (UNSPEC_SM3TT1A): Ditto.
5030 (UNSPEC_SM3TT1B): Ditto.
5031 (UNSPEC_SM3TT2A): Ditto.
5032 (UNSPEC_SM3TT2B): Ditto.
5033 (UNSPEC_SM3PARTW1): Ditto.
5034 (UNSPEC_SM3PARTW2): Ditto.
5035 (UNSPEC_SM4E): Ditto.
5036 (UNSPEC_SM4EKEY): Ditto.
5037 * config/aarch64/constraints.md (Ui2): New constraint.
5038 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
5039 * config/arm/types.md (crypto_sm3): New type attribute.
5040 (crypto_sm4): Ditto.
5041 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
5042 (vsm3tt1aq_u32): Ditto.
5043 (vsm3tt1bq_u32): Ditto.
5044 (vsm3tt2aq_u32): Ditto.
5045 (vsm3tt2bq_u32): Ditto.
5046 (vsm3partw1q_u32): Ditto.
5047 (vsm3partw2q_u32): Ditto.
5048 (vsm4eq_u32): Ditto.
5049 (vsm4ekeyq_u32): Ditto.
5050 (doc/invoke.texi): Document new sm4 option.
5052 2018-01-10 Michael Collison <michael.collison@arm.com>
5054 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
5055 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
5056 (AARCH64_FL_FOR_ARCH8_4): New.
5057 (AARCH64_FL_V8_4): New flag.
5058 (doc/invoke.texi): Document new armv8.4-a option.
5060 2018-01-10 Michael Collison <michael.collison@arm.com>
5062 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5063 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
5064 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
5065 * config/aarch64/aarch64-option-extension.def: Add
5066 AARCH64_OPT_EXTENSION of 'sha2'.
5067 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
5068 (crypto): Disable sha2 and aes if crypto disabled.
5069 (crypto): Enable aes and sha2 if enabled.
5070 (simd): Disable sha2 and aes if simd disabled.
5071 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
5073 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
5074 (TARGET_SHA2): New feature flag for sha2.
5075 (TARGET_AES): New feature flag for aes.
5076 * config/aarch64/aarch64-simd.md:
5077 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
5078 conditional on TARGET_AES.
5079 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
5080 (aarch64_crypto_sha1hsi): Make pattern conditional
5082 (aarch64_crypto_sha1hv4si): Ditto.
5083 (aarch64_be_crypto_sha1hv4si): Ditto.
5084 (aarch64_crypto_sha1su1v4si): Ditto.
5085 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
5086 (aarch64_crypto_sha1su0v4si): Ditto.
5087 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
5088 (aarch64_crypto_sha256su0v4si): Ditto.
5089 (aarch64_crypto_sha256su1v4si): Ditto.
5090 (doc/invoke.texi): Document new aes and sha2 options.
5092 2018-01-10 Martin Sebor <msebor@redhat.com>
5094 PR tree-optimization/83781
5095 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
5098 2018-01-11 Martin Sebor <msebor@gmail.com>
5099 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5101 PR tree-optimization/83501
5102 PR tree-optimization/81703
5104 * tree-ssa-strlen.c (get_string_cst): Rename...
5105 (get_string_len): ...to this. Handle global constants.
5106 (handle_char_store): Adjust.
5108 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
5109 Jim Wilson <jimw@sifive.com>
5111 * config/riscv/riscv-protos.h (riscv_output_return): New.
5112 * config/riscv/riscv.c (struct machine_function): New naked_p field.
5113 (riscv_attribute_table, riscv_output_return),
5114 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
5115 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
5116 (riscv_compute_frame_info): Only compute frame->mask if not a naked
5118 (riscv_expand_prologue): Add early return for naked function.
5119 (riscv_expand_epilogue): Likewise.
5120 (riscv_function_ok_for_sibcall): Return false for naked function.
5121 (riscv_set_current_function): New.
5122 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
5123 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
5124 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
5125 * doc/extend.texi (RISC-V Function Attributes): New.
5127 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
5129 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
5130 check for 128-bit long double before checking TCmode.
5131 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
5132 128-bit long doubles before checking TFmode or TCmode.
5133 (FLOAT128_IBM_P): Likewise.
5135 2018-01-10 Martin Sebor <msebor@redhat.com>
5137 PR tree-optimization/83671
5138 * builtins.c (c_strlen): Unconditionally return zero for the empty
5140 Use -Warray-bounds for warnings.
5141 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
5142 for non-constant array indices with COMPONENT_REF, arrays of
5143 arrays, and pointers to arrays.
5144 (gimple_fold_builtin_strlen): Determine and set length range for
5145 non-constant character arrays.
5147 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
5150 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
5153 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
5155 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
5157 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5160 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
5161 VECTOR_MEM_ALTIVEC_OR_VSX_P.
5162 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
5163 indexed_or_indirect_operand predicate.
5164 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
5165 (*vsx_le_perm_load_v8hi): Likewise.
5166 (*vsx_le_perm_load_v16qi): Likewise.
5167 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
5168 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
5169 (*vsx_le_perm_store_v8hi): Likewise.
5170 (*vsx_le_perm_store_v16qi): Likewise.
5171 (eight unnamed splitters): Likewise.
5173 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5175 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
5176 * config/rs6000/emmintrin.h: Likewise.
5177 * config/rs6000/mmintrin.h: Likewise.
5178 * config/rs6000/xmmintrin.h: Likewise.
5180 2018-01-10 David Malcolm <dmalcolm@redhat.com>
5183 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
5185 * tree.c (tree_nop_conversion): Return true for location wrapper
5187 (maybe_wrap_with_location): New function.
5188 (selftest::check_strip_nops): New function.
5189 (selftest::test_location_wrappers): New function.
5190 (selftest::tree_c_tests): Call it.
5191 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
5192 (maybe_wrap_with_location): New decl.
5193 (EXPR_LOCATION_WRAPPER_P): New macro.
5194 (location_wrapper_p): New inline function.
5195 (tree_strip_any_location_wrapper): New inline function.
5197 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5200 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5201 stack_realign_offset for the largest alignment of stack slot
5203 (ix86_find_max_used_stack_alignment): New function.
5204 (ix86_finalize_stack_frame_flags): Use it. Set
5205 max_used_stack_alignment if we don't realign stack.
5206 * config/i386/i386.h (machine_function): Add
5207 max_used_stack_alignment.
5209 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5211 * config/arm/arm.opt (-mbranch-cost): New option.
5212 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5215 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5218 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5219 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5221 2018-01-10 Richard Biener <rguenther@suse.de>
5224 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5225 early out so it also covers the case where we have a non-NULL
5228 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5230 PR tree-optimization/83753
5231 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5232 for non-strided grouped accesses if the number of elements is 1.
5234 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5237 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5238 * i386.h (TARGET_USE_GATHER): Define.
5239 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5241 2018-01-10 Martin Liska <mliska@suse.cz>
5244 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5245 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5247 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5248 CLEANUP_NO_PARTITIONING is not set.
5250 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5252 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5253 for vectors, as a partial revert of r254296.
5254 * rtl.h (const_vec_p): Delete.
5255 (const_vec_duplicate_p): Don't test for vector CONSTs.
5256 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5257 * expmed.c (make_tree): Likewise.
5260 * common.md (E, F): Use CONSTANT_P instead of checking for
5262 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5263 checking for CONST_VECTOR.
5265 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5268 * predict.c (force_edge_cold): Handle in more sane way edges
5271 2018-01-09 Carl Love <cel@us.ibm.com>
5273 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5275 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5276 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5277 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5278 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5279 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5280 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5281 * config/rs6000/rs6000-protos.h: Add extern defition for
5282 rs6000_generate_float2_double_code.
5283 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5285 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5286 (float2_v2df): Add define_expand.
5288 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5291 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5292 op_mode in the force_to_mode call.
5294 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5296 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5297 instead of checking each element individually.
5298 (aarch64_evpc_uzp): Likewise.
5299 (aarch64_evpc_zip): Likewise.
5300 (aarch64_evpc_ext): Likewise.
5301 (aarch64_evpc_rev): Likewise.
5302 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5303 instead of checking each element individually. Return true without
5305 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5306 whether all selected elements come from the same input, instead of
5307 checking each element individually. Remove calls to gen_rtx_REG,
5308 start_sequence and end_sequence and instead assert that no rtl is
5311 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5313 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5314 order of HIGH and CONST checks.
5316 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5318 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5319 if the destination isn't an SSA_NAME.
5321 2018-01-09 Richard Biener <rguenther@suse.de>
5323 PR tree-optimization/83668
5324 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5326 (canonicalize_loop_form): ... here, renamed from ...
5327 (canonicalize_loop_closed_ssa_form): ... this and amended to
5328 swap successor edges for loop exit blocks to make us use
5329 the RPO order we need for initial schedule generation.
5331 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5333 PR tree-optimization/64811
5334 * match.pd: When optimizing comparisons with Inf, avoid
5335 introducing or losing exceptions from comparisons with NaN.
5337 2018-01-09 Martin Liska <mliska@suse.cz>
5340 * asan.c (shadow_mem_size): Add gcc_assert.
5342 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5344 Don't save registers in main().
5347 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5348 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5349 * config/avr/avr.c (avr_set_current_function): Don't error if
5350 naked, OS_task or OS_main are specified at the same time.
5351 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5353 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5355 * common/config/avr/avr-common.c (avr_option_optimization_table):
5356 Switch on -mmain-is-OS_task for optimizing compilations.
5358 2018-01-09 Richard Biener <rguenther@suse.de>
5360 PR tree-optimization/83572
5361 * graphite.c: Include cfganal.h.
5362 (graphite_transform_loops): Connect infinite loops to exit
5363 and remove fake edges at the end.
5365 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5367 * ipa-inline.c (edge_badness): Revert accidental checkin.
5369 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5372 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5373 symbols; not inline clones.
5375 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5378 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5379 hard registers. Formatting fixes.
5381 PR preprocessor/83722
5382 * gcc.c (try_generate_repro): Pass
5383 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5384 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5387 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5388 Kito Cheng <kito.cheng@gmail.com>
5390 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5391 (riscv_leaf_function_p): Delete.
5392 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5394 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5396 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5398 (do_ifelse): New function.
5399 (do_isel): New function.
5400 (do_sub3): New function.
5401 (do_add3): New function.
5402 (do_load_mask_compare): New function.
5403 (do_overlap_load_compare): New function.
5404 (expand_compare_loop): New function.
5405 (expand_block_compare): Call expand_compare_loop() when appropriate.
5406 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5408 (-mblock-compare-inline-loop-limit): New option.
5410 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5413 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5414 Reverse order of second and third operands in first alternative.
5415 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5416 of first and second elements in UNSPEC_VPERMR vector.
5417 (altivec_expand_vec_perm_le): Likewise.
5419 2018-01-08 Jeff Law <law@redhat.com>
5421 PR rtl-optimizatin/81308
5422 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5423 (process_switch): If group_case_labels makes a change, then set
5425 (pass_convert_switch::execute): If a switch is converted, then
5426 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5428 PR rtl-optimization/81308
5429 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5432 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5434 PR target/83663 - Revert r255946
5435 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5436 generation for cases where splatting a value is not useful.
5437 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5438 across a vec_duplicate and a paradoxical subreg forming a vector
5439 mode to a vec_concat.
5441 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5443 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5444 -march=armv8.3-a variants.
5445 * config/arm/t-multilib: Likewise.
5446 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5448 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5450 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5452 (cceq_ior_compare_complement): Give it a name so I can use it, and
5453 change boolean_or_operator predicate to boolean_operator so it can
5454 be used to generate a crand.
5455 (eqne): New code iterator.
5456 (bd/bd_neg): New code_attrs.
5457 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5458 a single define_insn.
5459 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5460 decrement (bdnzt/bdnzf/bdzt/bdzf).
5461 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5462 with the new names of the branch decrement patterns, and added the
5463 names of the branch decrement conditional patterns.
5465 2018-01-08 Richard Biener <rguenther@suse.de>
5467 PR tree-optimization/83563
5468 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5471 2018-01-08 Richard Biener <rguenther@suse.de>
5474 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5476 2018-01-08 Richard Biener <rguenther@suse.de>
5478 PR tree-optimization/83685
5479 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5480 references to abnormals.
5482 2018-01-08 Richard Biener <rguenther@suse.de>
5485 * dwarf2out.c (output_indirect_strings): Handle empty
5486 skeleton_debug_str_hash.
5487 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5489 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5491 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5492 (emit_store_direct): Likewise.
5493 (arc_trampoline_adjust_address): Likewise.
5494 (arc_asm_trampoline_template): New function.
5495 (arc_initialize_trampoline): Use asm_trampoline_template.
5496 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5497 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5498 * config/arc/arc.md (flush_icache): Delete pattern.
5500 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5502 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5503 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5506 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5509 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5510 by not USED_FOR_TARGET.
5511 (make_pass_resolve_sw_modes): Likewise.
5513 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5515 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5518 2018-01-08 Richard Biener <rguenther@suse.de>
5521 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5523 2018-01-08 Richard Biener <rguenther@suse.de>
5526 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5528 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5531 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5532 basic blocks with a small number of successors.
5533 (convert_control_dep_chain_into_preds): Improve handling of
5535 (dump_predicates): Split apart into...
5536 (dump_pred_chain): ...here...
5537 (dump_pred_info): ...and here.
5538 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5539 (can_chain_union_be_invalidated_p): Improve check for invalidation
5541 (uninit_uses_cannot_happen): Avoid unnecessary if
5542 convert_control_dep_chain_into_preds yielded nothing.
5544 2018-01-06 Martin Sebor <msebor@redhat.com>
5546 PR tree-optimization/83640
5547 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5548 subtracting negative offset from size.
5549 (builtin_access::overlap): Adjust offset bounds of the access to fall
5550 within the size of the object if possible.
5552 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5554 PR rtl-optimization/83699
5555 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5556 extract_bit_field_as_subreg to cases in which the extracted
5557 value is also a vector.
5559 * lra-constraints.c (process_alt_operands): Test for the equivalence
5560 substitutions when detecting a possible reload cycle.
5562 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5565 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5566 by default if flag_selective_schedling{,2}. Formatting fixes.
5568 PR rtl-optimization/83682
5569 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5570 if it has non-VECTOR_MODE element mode.
5571 (vec_duplicate_p): Likewise.
5574 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5575 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5577 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5580 * config/i386/i386-builtin.def
5581 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5582 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5583 Require also OPTION_MASK_ISA_AVX512F in addition to
5584 OPTION_MASK_ISA_GFNI.
5585 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5586 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5587 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5588 to OPTION_MASK_ISA_GFNI.
5589 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5590 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5591 OPTION_MASK_ISA_AVX512BW.
5592 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5593 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5594 addition to OPTION_MASK_ISA_GFNI.
5595 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5596 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5597 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5598 to OPTION_MASK_ISA_GFNI.
5599 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5600 a requirement for all ISAs rather than any of them with a few
5602 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5604 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5605 bitmasks to be enabled with 3 exceptions, instead of requiring any
5606 enabled ISA with lots of exceptions.
5607 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5608 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5609 Change avx512bw in isa attribute to avx512f.
5610 * config/i386/sgxintrin.h: Add license boilerplate.
5611 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5612 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5613 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5614 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5616 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5617 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5618 temporarily sse2 rather than sse if not enabled already.
5621 * config/i386/sse.md (VI248_VLBW): Rename to ...
5622 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5623 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5624 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5625 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5626 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5627 mode iterator instead of VI248_VLBW.
5629 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5631 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5632 (record_modified): Skip clobbers; add debug output.
5633 (param_change_prob): Use sreal frequencies.
5635 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5637 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5638 punt for user-aligned variables.
5640 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5642 * tree-chrec.c (chrec_contains_symbols): Return true for
5645 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5648 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5649 of (x|y) == x for BICS pattern.
5651 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5653 PR tree-optimization/83605
5654 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5655 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5658 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5660 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5661 * config/epiphany/rtems.h: New file.
5663 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5664 Uros Bizjak <ubizjak@gmail.com>
5667 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5668 QIreg_operand instead of register_operand predicate.
5669 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5670 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5671 comments instead of -fmitigate[-_]rop.
5673 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5676 * cgraphunit.c (symbol_table::compile): Switch to text_section
5677 before calling assembly_start debug hook.
5678 * run-rtl-passes.c (run_rtl_passes): Likewise.
5681 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5683 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5684 range_int_cst_p rather than !symbolic_range_p before calling
5685 extract_range_from_multiplicative_op_1.
5687 2018-01-04 Jeff Law <law@redhat.com>
5689 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5690 redundant test in assertion.
5692 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5694 * doc/rtl.texi: Document machine_mode wrapper classes.
5696 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5698 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5701 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5703 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5704 the VEC_PERM_EXPR fold to fail.
5706 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5709 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5710 to switched_sections.
5712 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5715 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5718 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5721 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5722 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5724 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5727 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5728 is BLKmode and bitpos not zero or mode change is needed.
5730 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5733 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5736 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5739 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5740 instead of MULT rtx. Update all corresponding splitters.
5742 (*ssub<modesuffix>): Ditto.
5744 (*cmp_sadd_di): Update split patterns.
5745 (*cmp_sadd_si): Ditto.
5746 (*cmp_sadd_sidi): Ditto.
5747 (*cmp_ssub_di): Ditto.
5748 (*cmp_ssub_si): Ditto.
5749 (*cmp_ssub_sidi): Ditto.
5750 * config/alpha/predicates.md (const23_operand): New predicate.
5751 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5752 Look for ASHIFT, not MULT inner operand.
5753 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5755 2018-01-04 Martin Liska <mliska@suse.cz>
5757 PR gcov-profile/83669
5758 * gcov.c (output_intermediate_file): Add version to intermediate
5760 * doc/gcov.texi: Document new field 'version' in intermediate
5761 file format. Fix location of '-k' option of gcov command.
5763 2018-01-04 Martin Liska <mliska@suse.cz>
5766 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5768 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5770 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5772 2018-01-03 Martin Sebor <msebor@redhat.com>
5774 PR tree-optimization/83655
5775 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5776 checking calls with invalid arguments.
5778 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5780 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5781 (vectorizable_mask_load_store): Delete.
5782 (vectorizable_call): Return false for masked loads and stores.
5783 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5784 instead of gimple_assign_rhs1.
5785 (vectorizable_load): Handle IFN_MASK_LOAD.
5786 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5788 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5790 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5792 (vectorizable_mask_load_store): ...here.
5793 (vectorizable_load): ...and here.
5795 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5797 * tree-vect-stmts.c (vect_build_all_ones_mask)
5798 (vect_build_zero_merge_argument): New functions, split out from...
5799 (vectorizable_load): ...here.
5801 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5803 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5805 (vectorizable_mask_load_store): ...here.
5806 (vectorizable_store): ...and here.
5808 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5810 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5812 (vectorizable_mask_load_store): ...here.
5814 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5816 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5817 (vect_model_store_cost): Take a vec_load_store_type instead of a
5819 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5820 (vect_model_store_cost): Take a vec_load_store_type instead of a
5822 (vectorizable_mask_load_store): Update accordingly.
5823 (vectorizable_store): Likewise.
5824 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5826 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5828 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5829 IFN_MASK_LOAD calls here rather than...
5830 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5832 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5833 Alan Hayward <alan.hayward@arm.com>
5834 David Sherwood <david.sherwood@arm.com>
5836 * expmed.c (extract_bit_field_1): For vector extracts,
5837 fall back to extract_bit_field_as_subreg if vec_extract
5840 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5841 Alan Hayward <alan.hayward@arm.com>
5842 David Sherwood <david.sherwood@arm.com>
5844 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5845 they are variable or constant sized.
5846 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5847 slots for constant-sized data.
5849 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5850 Alan Hayward <alan.hayward@arm.com>
5851 David Sherwood <david.sherwood@arm.com>
5853 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5854 handling COND_EXPRs with boolean comparisons, try to find a better
5855 basis for the mask type than the boolean itself.
5857 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5859 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5860 is calculated and how it can be overridden.
5861 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5862 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5864 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5867 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5868 Alan Hayward <alan.hayward@arm.com>
5869 David Sherwood <david.sherwood@arm.com>
5871 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5872 Remove the mode argument.
5873 (aarch64_simd_valid_immediate): Remove the mode and inverse
5875 * config/aarch64/iterators.md (bitsize): New iterator.
5876 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5877 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5878 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5879 aarch64_simd_valid_immediate.
5880 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5881 (aarch64_reg_or_bic_imm): Likewise.
5882 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5883 with an insn_type enum and msl with a modifier_type enum.
5884 Replace element_width with a scalar_mode. Change the shift
5885 to unsigned int. Add constructors for scalar_float_mode and
5886 scalar_int_mode elements.
5887 (aarch64_vect_float_const_representable_p): Delete.
5888 (aarch64_can_const_movi_rtx_p)
5889 (aarch64_simd_scalar_immediate_valid_for_move)
5890 (aarch64_simd_make_constant): Update call to
5891 aarch64_simd_valid_immediate.
5892 (aarch64_advsimd_valid_immediate_hs): New function.
5893 (aarch64_advsimd_valid_immediate): Likewise.
5894 (aarch64_simd_valid_immediate): Remove mode and inverse
5895 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5896 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5897 and aarch64_float_const_representable_p on the result.
5898 (aarch64_output_simd_mov_immediate): Remove mode argument.
5899 Update call to aarch64_simd_valid_immediate and use of
5900 simd_immediate_info.
5901 (aarch64_output_scalar_simd_mov_immediate): Update call
5904 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5905 Alan Hayward <alan.hayward@arm.com>
5906 David Sherwood <david.sherwood@arm.com>
5908 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5909 (mode_nunits): Likewise CONST_MODE_NUNITS.
5910 * machmode.def (ADJUST_NUNITS): Document.
5911 * genmodes.c (mode_data::need_nunits_adj): New field.
5912 (blank_mode): Update accordingly.
5913 (adj_nunits): New variable.
5914 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5916 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5917 listed in adj_nunits.
5918 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5919 listed in adj_nunits. Don't emit case statements for such modes.
5920 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5921 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5922 nothing if adj_nunits is nonnull.
5923 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5924 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5925 (emit_mode_fbit): Update use of print_maybe_const_decl.
5926 (emit_move_size): Likewise. Treat the array as non-const
5928 (emit_mode_adjustments): Handle adj_nunits.
5930 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5932 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5933 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5934 (VECTOR_MODES): Use it.
5935 (make_vector_modes): Take the prefix as an argument.
5937 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5938 Alan Hayward <alan.hayward@arm.com>
5939 David Sherwood <david.sherwood@arm.com>
5941 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5942 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5943 for MODE_VECTOR_BOOL.
5944 * machmode.def (VECTOR_BOOL_MODE): Document.
5945 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5946 (make_vector_bool_mode): New function.
5947 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5949 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5950 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5952 * stor-layout.c (int_mode_for_mode): Likewise.
5953 * tree.c (build_vector_type_for_mode): Likewise.
5954 * varasm.c (output_constant_pool_2): Likewise.
5955 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
5956 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
5957 for MODE_VECTOR_BOOL.
5958 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
5959 of mode class checks.
5960 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
5961 instead of a list of mode class checks.
5962 (expand_vector_scalar_condition): Likewise.
5963 (type_for_widest_vector_mode): Handle BImode as an inner mode.
5965 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5966 Alan Hayward <alan.hayward@arm.com>
5967 David Sherwood <david.sherwood@arm.com>
5969 * machmode.h (mode_size): Change from unsigned short to
5971 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
5972 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
5973 or if measurement_type is not polynomial.
5974 (fixed_size_mode::includes_p): Check for constant-sized modes.
5975 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
5976 return a poly_uint16 rather than an unsigned short.
5977 (emit_mode_size): Change the type of mode_size from unsigned short
5978 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
5979 (emit_mode_adjustments): Cope with polynomial vector sizes.
5980 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5982 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5984 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
5985 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
5986 * caller-save.c (setup_save_areas): Likewise.
5987 (replace_reg_with_saved_mem): Likewise.
5988 * calls.c (emit_library_call_value_1): Likewise.
5989 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
5990 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
5991 (gen_lowpart_for_combine): Likewise.
5992 * convert.c (convert_to_integer_1): Likewise.
5993 * cse.c (equiv_constant, cse_insn): Likewise.
5994 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
5995 (cselib_subst_to_values): Likewise.
5996 * dce.c (word_dce_process_block): Likewise.
5997 * df-problems.c (df_word_lr_mark_ref): Likewise.
5998 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
5999 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
6000 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
6001 (rtl_for_decl_location): Likewise.
6002 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
6003 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
6004 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
6005 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
6006 (expand_expr_real_1): Likewise.
6007 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
6008 (pad_below): Likewise.
6009 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6010 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
6011 * ira.c (get_subreg_tracking_sizes): Likewise.
6012 * ira-build.c (ira_create_allocno_objects): Likewise.
6013 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
6014 (ira_sort_regnos_for_alter_reg): Likewise.
6015 * ira-costs.c (record_operand_costs): Likewise.
6016 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
6017 (resolve_simple_move): Likewise.
6018 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
6019 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
6020 (lra_constraints): Likewise.
6021 (CONST_POOL_OK_P): Reject variable-sized modes.
6022 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
6023 (add_pseudo_to_slot, lra_spill): Likewise.
6024 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6025 * optabs-query.c (get_best_extraction_insn): Likewise.
6026 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6027 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
6028 (expand_mult_highpart, valid_multiword_target_p): Likewise.
6029 * recog.c (offsettable_address_addr_space_p): Likewise.
6030 * regcprop.c (maybe_mode_change): Likewise.
6031 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
6032 * regrename.c (build_def_use): Likewise.
6033 * regstat.c (dump_reg_info): Likewise.
6034 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
6035 (find_reloads, find_reloads_subreg_address): Likewise.
6036 * reload1.c (eliminate_regs_1): Likewise.
6037 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
6038 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
6039 (simplify_binary_operation_1, simplify_subreg): Likewise.
6040 * targhooks.c (default_function_arg_padding): Likewise.
6041 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
6042 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
6043 (verify_gimple_assign_ternary): Likewise.
6044 * tree-inline.c (estimate_move_cost): Likewise.
6045 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6046 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
6047 (get_address_cost_ainc): Likewise.
6048 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
6049 (vect_supportable_dr_alignment): Likewise.
6050 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6051 (vectorizable_reduction): Likewise.
6052 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
6053 (vectorizable_operation, vectorizable_load): Likewise.
6054 * tree.c (build_same_sized_truth_vector_type): Likewise.
6055 * valtrack.c (cleanup_auto_inc_dec): Likewise.
6056 * var-tracking.c (emit_note_insn_var_location): Likewise.
6057 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
6058 (ADDR_VEC_ALIGN): Likewise.
6060 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6061 Alan Hayward <alan.hayward@arm.com>
6062 David Sherwood <david.sherwood@arm.com>
6064 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
6066 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6067 or if measurement_type is polynomial.
6068 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
6069 * combine.c (make_extraction): Likewise.
6070 * dse.c (find_shift_sequence): Likewise.
6071 * dwarf2out.c (mem_loc_descriptor): Likewise.
6072 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
6073 (extract_bit_field, extract_low_bits): Likewise.
6074 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
6075 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
6076 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
6077 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
6078 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6079 * reload.c (find_reloads): Likewise.
6080 * reload1.c (alter_reg): Likewise.
6081 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
6082 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
6083 * tree-if-conv.c (predicate_mem_writes): Likewise.
6084 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
6085 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
6086 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
6087 * valtrack.c (dead_debug_insert_temp): Likewise.
6088 * varasm.c (mergeable_constant_section): Likewise.
6089 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
6091 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6092 Alan Hayward <alan.hayward@arm.com>
6093 David Sherwood <david.sherwood@arm.com>
6095 * expr.c (expand_assignment): Cope with polynomial mode sizes
6096 when assigning to a CONCAT.
6098 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6099 Alan Hayward <alan.hayward@arm.com>
6100 David Sherwood <david.sherwood@arm.com>
6102 * machmode.h (mode_precision): Change from unsigned short to
6104 (mode_to_precision): Return a poly_uint16 rather than an unsigned
6106 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
6107 or if measurement_type is not polynomial.
6108 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
6109 in which the mode is already known to be a scalar_int_mode.
6110 * genmodes.c (emit_mode_precision): Change the type of mode_precision
6111 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
6113 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6114 for GET_MODE_PRECISION.
6115 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6116 for GET_MODE_PRECISION.
6117 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
6119 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
6120 (expand_field_assignment, make_extraction): Likewise.
6121 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
6122 (get_last_value): Likewise.
6123 * convert.c (convert_to_integer_1): Likewise.
6124 * cse.c (cse_insn): Likewise.
6125 * expr.c (expand_expr_real_1): Likewise.
6126 * lra-constraints.c (simplify_operand_subreg): Likewise.
6127 * optabs-query.c (can_atomic_load_p): Likewise.
6128 * optabs.c (expand_atomic_load): Likewise.
6129 (expand_atomic_store): Likewise.
6130 * ree.c (combine_reaching_defs): Likewise.
6131 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
6132 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
6133 * tree.h (type_has_mode_precision_p): Likewise.
6134 * ubsan.c (instrument_si_overflow): Likewise.
6136 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6137 Alan Hayward <alan.hayward@arm.com>
6138 David Sherwood <david.sherwood@arm.com>
6140 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
6141 polynomial numbers of units.
6142 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
6143 (valid_vector_subparts_p): New function.
6144 (build_vector_type): Remove temporary shim and take the number
6145 of units as a poly_uint64 rather than an int.
6146 (build_opaque_vector_type): Take the number of units as a
6147 poly_uint64 rather than an int.
6148 * tree.c (build_vector_from_ctor): Handle polynomial
6149 TYPE_VECTOR_SUBPARTS.
6150 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
6151 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
6152 (build_vector_from_val): If the number of units is variable,
6153 use build_vec_duplicate_cst for constant operands and
6154 VEC_DUPLICATE_EXPR otherwise.
6155 (make_vector_type): Remove temporary is_constant ().
6156 (build_vector_type, build_opaque_vector_type): Take the number of
6157 units as a poly_uint64 rather than an int.
6158 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
6160 * cfgexpand.c (expand_debug_expr): Likewise.
6161 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
6162 (store_constructor, expand_expr_real_1): Likewise.
6163 (const_scalar_mask_from_tree): Likewise.
6164 * fold-const-call.c (fold_const_reduction): Likewise.
6165 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
6166 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
6167 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
6168 (fold_relational_const): Likewise.
6169 (native_interpret_vector): Likewise. Change the size from an
6170 int to an unsigned int.
6171 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
6172 TYPE_VECTOR_SUBPARTS.
6173 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
6174 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
6175 duplicating a non-constant operand into a variable-length vector.
6176 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
6177 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
6178 * ipa-icf.c (sem_variable::equals): Likewise.
6179 * match.pd: Likewise.
6180 * omp-simd-clone.c (simd_clone_subparts): Likewise.
6181 * print-tree.c (print_node): Likewise.
6182 * stor-layout.c (layout_type): Likewise.
6183 * targhooks.c (default_builtin_vectorization_cost): Likewise.
6184 * tree-cfg.c (verify_gimple_comparison): Likewise.
6185 (verify_gimple_assign_binary): Likewise.
6186 (verify_gimple_assign_ternary): Likewise.
6187 (verify_gimple_assign_single): Likewise.
6188 * tree-pretty-print.c (dump_generic_node): Likewise.
6189 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6190 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
6191 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
6192 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
6193 (vect_shift_permute_load_chain): Likewise.
6194 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
6195 (expand_vector_condition, optimize_vector_constructor): Likewise.
6196 (lower_vec_perm, get_compute_type): Likewise.
6197 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6198 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6199 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6200 (vect_recog_mask_conversion_pattern): Likewise.
6201 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6202 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6203 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6204 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6205 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6206 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6207 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6208 (supportable_widening_operation): Likewise.
6209 (supportable_narrowing_operation): Likewise.
6210 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6212 * varasm.c (output_constant): Likewise.
6214 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6215 Alan Hayward <alan.hayward@arm.com>
6216 David Sherwood <david.sherwood@arm.com>
6218 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6219 so that both the length == 3 and length != 3 cases set up their
6220 own permute vectors. Add comments explaining why we know the
6221 number of elements is constant.
6222 (vect_permute_load_chain): Likewise.
6224 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6225 Alan Hayward <alan.hayward@arm.com>
6226 David Sherwood <david.sherwood@arm.com>
6228 * machmode.h (mode_nunits): Change from unsigned char to
6230 (ONLY_FIXED_SIZE_MODES): New macro.
6231 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6232 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6233 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6235 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6236 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6237 or if measurement_type is not polynomial.
6238 * genmodes.c (ZERO_COEFFS): New macro.
6239 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6241 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6242 Use ZERO_COEFFS when emitting initializers.
6243 * data-streamer.h (bp_pack_poly_value): New function.
6244 (bp_unpack_poly_value): Likewise.
6245 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6246 for GET_MODE_NUNITS.
6247 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6248 for GET_MODE_NUNITS.
6249 * tree.c (make_vector_type): Remove temporary shim and make
6250 the real function take the number of units as a poly_uint64
6252 (build_vector_type_for_mode): Handle polynomial nunits.
6253 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6254 * emit-rtl.c (const_vec_series_p_1): Likewise.
6255 (gen_rtx_CONST_VECTOR): Likewise.
6256 * fold-const.c (test_vec_duplicate_folding): Likewise.
6257 * genrecog.c (validate_pattern): Likewise.
6258 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6259 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6260 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6261 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6262 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6263 * rtlanal.c (subreg_get_info): Likewise.
6264 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6265 (vect_grouped_load_supported): Likewise.
6266 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6267 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6268 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6269 (simplify_const_unary_operation, simplify_binary_operation_1)
6270 (simplify_const_binary_operation, simplify_ternary_operation)
6271 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6272 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6273 instead of CONST_VECTOR_NUNITS.
6274 * varasm.c (output_constant_pool_2): Likewise.
6275 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6276 explicit-encoded elements in the XVEC for variable-length vectors.
6278 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6280 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6282 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6283 Alan Hayward <alan.hayward@arm.com>
6284 David Sherwood <david.sherwood@arm.com>
6286 * coretypes.h (fixed_size_mode): Declare.
6287 (fixed_size_mode_pod): New typedef.
6288 * builtins.h (target_builtins::x_apply_args_mode)
6289 (target_builtins::x_apply_result_mode): Change type to
6290 fixed_size_mode_pod.
6291 * builtins.c (apply_args_size, apply_result_size, result_vector)
6292 (expand_builtin_apply_args_1, expand_builtin_apply)
6293 (expand_builtin_return): Update accordingly.
6295 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6297 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6298 * cselib.c (cselib_hash_rtx): Likewise.
6299 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6300 CONST_VECTOR encoding.
6302 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6303 Jeff Law <law@redhat.com>
6306 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6307 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6308 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6309 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6312 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6313 explicitly probe *sp in a noreturn function if there were any callee
6314 register saves or frame pointer is needed.
6316 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6319 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6320 BLKmode for ternary, binary or unary expressions.
6323 * var-tracking.c (delete_vta_debug_insn): New inline function.
6324 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6325 insns from get_insns () to NULL instead of each bb separately.
6326 Use delete_vta_debug_insn. No longer static.
6327 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6328 delete_vta_debug_insns callers.
6329 * rtl.h (delete_vta_debug_insns): Declare.
6330 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6331 instead of variable_tracking_main.
6333 2018-01-03 Martin Sebor <msebor@redhat.com>
6335 PR tree-optimization/83603
6336 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6337 arguments past the endof the argument list in functions declared
6338 without a prototype.
6339 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6340 Avoid checking when arguments are null.
6342 2018-01-03 Martin Sebor <msebor@redhat.com>
6345 * doc/extend.texi (attribute const): Fix a typo.
6346 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6347 issuing -Wsuggest-attribute for void functions.
6349 2018-01-03 Martin Sebor <msebor@redhat.com>
6351 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6352 offset_int::from instead of wide_int::to_shwi.
6353 (maybe_diag_overlap): Remove assertion.
6354 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6355 * gimple-ssa-sprintf.c (format_directive): Same.
6356 (parse_directive): Same.
6357 (sprintf_dom_walker::compute_format_length): Same.
6358 (try_substitute_return_value): Same.
6360 2018-01-03 Jeff Law <law@redhat.com>
6363 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6364 non-constant residual for zero at runtime and avoid probing in
6365 that case. Reorganize code for trailing problem to mirror handling
6368 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6370 PR tree-optimization/83501
6371 * tree-ssa-strlen.c (get_string_cst): New.
6372 (handle_char_store): Call get_string_cst.
6374 2018-01-03 Martin Liska <mliska@suse.cz>
6376 PR tree-optimization/83593
6377 * tree-ssa-strlen.c: Include tree-cfg.h.
6378 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6379 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6380 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6382 (strlen_dom_walker::before_dom_children): Call
6383 gimple_purge_dead_eh_edges. Dump tranformation with details
6385 (strlen_dom_walker::before_dom_children): Update call by adding
6386 new argument cleanup_eh.
6387 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6389 2018-01-03 Martin Liska <mliska@suse.cz>
6392 * cif-code.def (VARIADIC_THUNK): New enum value.
6393 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6396 2018-01-03 Jan Beulich <jbeulich@suse.com>
6398 * sse.md (mov<mode>_internal): Tighten condition for when to use
6399 vmovdqu<ssescalarsize> for TI and OI modes.
6401 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6403 Update copyright years.
6405 2018-01-03 Martin Liska <mliska@suse.cz>
6408 * ipa-visibility.c (function_and_variable_visibility): Skip
6409 functions with noipa attribure.
6411 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6413 * gcc.c (process_command): Update copyright notice dates.
6414 * gcov-dump.c (print_version): Ditto.
6415 * gcov.c (print_version): Ditto.
6416 * gcov-tool.c (print_version): Ditto.
6417 * gengtype.c (create_file): Ditto.
6418 * doc/cpp.texi: Bump @copying's copyright year.
6419 * doc/cppinternals.texi: Ditto.
6420 * doc/gcc.texi: Ditto.
6421 * doc/gccint.texi: Ditto.
6422 * doc/gcov.texi: Ditto.
6423 * doc/install.texi: Ditto.
6424 * doc/invoke.texi: Ditto.
6426 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6428 * vector-builder.h (vector_builder::m_full_nelts): Change from
6429 unsigned int to poly_uint64.
6430 (vector_builder::full_nelts): Update prototype accordingly.
6431 (vector_builder::new_vector): Likewise.
6432 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6433 (vector_builder::operator ==): Likewise.
6434 (vector_builder::finalize): Likewise.
6435 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6436 Take the number of elements as a poly_uint64 rather than an
6438 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6439 from unsigned int to poly_uint64.
6440 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6441 (vec_perm_indices::new_vector): Likewise.
6442 (vec_perm_indices::length): Likewise.
6443 (vec_perm_indices::nelts_per_input): Likewise.
6444 (vec_perm_indices::input_nelts): Likewise.
6445 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6446 number of elements per input as a poly_uint64 rather than an
6447 unsigned int. Use the original encoding for variable-length
6448 vectors, rather than clamping each individual element.
6449 For the second and subsequent elements in each pattern,
6450 clamp the step and base before clamping their sum.
6451 (vec_perm_indices::series_p): Handle polynomial element counts.
6452 (vec_perm_indices::all_in_range_p): Likewise.
6453 (vec_perm_indices_to_tree): Likewise.
6454 (vec_perm_indices_to_rtx): Likewise.
6455 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6456 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6457 (tree_vector_builder::new_binary_operation): Handle polynomial
6458 element counts. Return false if we need to know the number
6459 of elements at compile time.
6460 * fold-const.c (fold_vec_perm): Punt if the number of elements
6461 isn't known at compile time.
6463 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6465 * vec-perm-indices.h (vec_perm_builder): Change element type
6466 from HOST_WIDE_INT to poly_int64.
6467 (vec_perm_indices::element_type): Update accordingly.
6468 (vec_perm_indices::clamp): Handle polynomial element_types.
6469 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6470 (vec_perm_indices::all_in_range_p): Likewise.
6471 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6473 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6474 polynomial vec_perm_indices element types.
6475 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6476 * fold-const.c (fold_vec_perm): Likewise.
6477 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6478 * tree-vect-generic.c (lower_vec_perm): Likewise.
6479 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6480 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6481 element type to HOST_WIDE_INT.
6483 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6484 Alan Hayward <alan.hayward@arm.com>
6485 David Sherwood <david.sherwood@arm.com>
6487 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6488 rather than an int. Use plus_constant.
6489 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6490 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6492 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6493 Alan Hayward <alan.hayward@arm.com>
6494 David Sherwood <david.sherwood@arm.com>
6496 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6497 a HOST_WIDE_INT to a poly_int64.
6499 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6500 Alan Hayward <alan.hayward@arm.com>
6501 David Sherwood <david.sherwood@arm.com>
6503 * calls.c (load_register_parameters): Cope with polynomial
6504 mode sizes. Require a constant size for BLKmode parameters
6505 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6506 forces a parameter to be padded at the lsb end in order to
6507 fill a complete number of words, require the parameter size
6508 to be ordered wrt UNITS_PER_WORD.
6510 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6511 Alan Hayward <alan.hayward@arm.com>
6512 David Sherwood <david.sherwood@arm.com>
6514 * reload1.c (spill_stack_slot_width): Change element type
6515 from unsigned int to poly_uint64_pod.
6516 (alter_reg): Treat mode sizes as polynomial.
6518 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6519 Alan Hayward <alan.hayward@arm.com>
6520 David Sherwood <david.sherwood@arm.com>
6522 * reload.c (complex_word_subreg_p): New function.
6523 (reload_inner_reg_of_subreg, push_reload): Use it.
6525 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6526 Alan Hayward <alan.hayward@arm.com>
6527 David Sherwood <david.sherwood@arm.com>
6529 * lra-constraints.c (process_alt_operands): Reject matched
6530 operands whose sizes aren't ordered.
6531 (match_reload): Refer to this check here.
6533 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6534 Alan Hayward <alan.hayward@arm.com>
6535 David Sherwood <david.sherwood@arm.com>
6537 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6538 that the mode size is in the set {1, 2, 4, 8, 16}.
6540 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6541 Alan Hayward <alan.hayward@arm.com>
6542 David Sherwood <david.sherwood@arm.com>
6544 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6545 Use plus_constant instead of gen_rtx_PLUS.
6547 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6548 Alan Hayward <alan.hayward@arm.com>
6549 David Sherwood <david.sherwood@arm.com>
6551 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6552 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6553 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6554 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6555 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6556 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6557 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6558 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6559 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6560 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6562 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6563 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6564 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6565 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6566 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6567 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6568 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6569 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6570 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6571 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6573 * expr.c (emit_move_resolve_push): Treat the input and result
6574 of PUSH_ROUNDING as a poly_int64.
6575 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6576 (emit_push_insn): Likewise.
6577 * lra-eliminations.c (mark_not_eliminable): Likewise.
6578 * recog.c (push_operand): Likewise.
6579 * reload1.c (elimination_effects): Likewise.
6580 * rtlanal.c (nonzero_bits1): Likewise.
6581 * calls.c (store_one_arg): Likewise. Require the padding to be
6582 known at compile time.
6584 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6585 Alan Hayward <alan.hayward@arm.com>
6586 David Sherwood <david.sherwood@arm.com>
6588 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6589 Use plus_constant instead of gen_rtx_PLUS.
6591 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6592 Alan Hayward <alan.hayward@arm.com>
6593 David Sherwood <david.sherwood@arm.com>
6595 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6598 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6599 Alan Hayward <alan.hayward@arm.com>
6600 David Sherwood <david.sherwood@arm.com>
6602 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6603 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6604 via stack temporaries. Treat the mode size as polynomial too.
6606 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6607 Alan Hayward <alan.hayward@arm.com>
6608 David Sherwood <david.sherwood@arm.com>
6610 * expr.c (expand_expr_real_2): When handling conversions involving
6611 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6612 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6613 as a poly_uint64 too.
6615 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6616 Alan Hayward <alan.hayward@arm.com>
6617 David Sherwood <david.sherwood@arm.com>
6619 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6621 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6622 Alan Hayward <alan.hayward@arm.com>
6623 David Sherwood <david.sherwood@arm.com>
6625 * combine.c (can_change_dest_mode): Handle polynomial
6626 REGMODE_NATURAL_SIZE.
6627 * expmed.c (store_bit_field_1): Likewise.
6628 * expr.c (store_constructor): Likewise.
6629 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6630 and polynomial REGMODE_NATURAL_SIZE.
6631 (gen_lowpart_common): Likewise.
6632 * reginfo.c (record_subregs_of_mode): Likewise.
6633 * rtlanal.c (read_modify_subreg_p): Likewise.
6635 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6636 Alan Hayward <alan.hayward@arm.com>
6637 David Sherwood <david.sherwood@arm.com>
6639 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6640 numbers of elements.
6642 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6643 Alan Hayward <alan.hayward@arm.com>
6644 David Sherwood <david.sherwood@arm.com>
6646 * match.pd: Cope with polynomial numbers of vector elements.
6648 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6649 Alan Hayward <alan.hayward@arm.com>
6650 David Sherwood <david.sherwood@arm.com>
6652 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6653 in a POINTER_PLUS_EXPR.
6655 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6656 Alan Hayward <alan.hayward@arm.com>
6657 David Sherwood <david.sherwood@arm.com>
6659 * omp-simd-clone.c (simd_clone_subparts): New function.
6660 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6661 (ipa_simd_modify_function_body): Likewise.
6663 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6664 Alan Hayward <alan.hayward@arm.com>
6665 David Sherwood <david.sherwood@arm.com>
6667 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6668 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6669 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6670 (expand_vector_condition, vector_element): Likewise.
6671 (subparts_gt): New function.
6672 (get_compute_type): Use subparts_gt.
6673 (count_type_subparts): Delete.
6674 (expand_vector_operations_1): Use subparts_gt instead of
6675 count_type_subparts.
6677 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6678 Alan Hayward <alan.hayward@arm.com>
6679 David Sherwood <david.sherwood@arm.com>
6681 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6682 (vect_compile_time_alias): ...this new function. Do the calculation
6683 on poly_ints rather than trees.
6684 (vect_prune_runtime_alias_test_list): Update call accordingly.
6686 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6687 Alan Hayward <alan.hayward@arm.com>
6688 David Sherwood <david.sherwood@arm.com>
6690 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6692 (vect_schedule_slp_instance): Likewise.
6694 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6695 Alan Hayward <alan.hayward@arm.com>
6696 David Sherwood <david.sherwood@arm.com>
6698 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6699 constant and extern definitions for variable-length vectors.
6700 (vect_get_constant_vectors): Note that the number of units
6701 is known to be constant.
6703 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6704 Alan Hayward <alan.hayward@arm.com>
6705 David Sherwood <david.sherwood@arm.com>
6707 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6708 of units as polynomial. Choose between WIDE and NARROW based
6711 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6712 Alan Hayward <alan.hayward@arm.com>
6713 David Sherwood <david.sherwood@arm.com>
6715 * tree-vect-stmts.c (simd_clone_subparts): New function.
6716 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6718 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6719 Alan Hayward <alan.hayward@arm.com>
6720 David Sherwood <david.sherwood@arm.com>
6722 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6723 vectors as polynomial. Use build_index_vector for
6726 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6727 Alan Hayward <alan.hayward@arm.com>
6728 David Sherwood <david.sherwood@arm.com>
6730 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6731 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6732 for variable-length vectors.
6733 (vectorizable_mask_load_store): Treat the number of units as
6734 polynomial, asserting that it is constant if the condition has
6735 already been enforced.
6736 (vectorizable_store, vectorizable_load): Likewise.
6738 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6739 Alan Hayward <alan.hayward@arm.com>
6740 David Sherwood <david.sherwood@arm.com>
6742 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6743 of units as polynomial. Punt if we can't tell at compile time
6744 which vector contains the final result.
6746 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6747 Alan Hayward <alan.hayward@arm.com>
6748 David Sherwood <david.sherwood@arm.com>
6750 * tree-vect-loop.c (vectorizable_induction): Treat the number
6751 of units as polynomial. Punt on SLP inductions. Use an integer
6752 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6753 cast of such a series for variable-length floating-point
6756 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6757 Alan Hayward <alan.hayward@arm.com>
6758 David Sherwood <david.sherwood@arm.com>
6760 * tree.h (build_index_vector): Declare.
6761 * tree.c (build_index_vector): New function.
6762 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6763 of units as polynomial, forcibly converting it to a constant if
6764 vectorizable_reduction has already enforced the condition.
6765 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6766 to create a {1,2,3,...} vector.
6767 (vectorizable_reduction): Treat the number of units as polynomial.
6768 Choose vectype_in based on the largest scalar element size rather
6769 than the smallest number of units. Enforce the restrictions
6772 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6773 Alan Hayward <alan.hayward@arm.com>
6774 David Sherwood <david.sherwood@arm.com>
6776 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6777 number of units as polynomial.
6779 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6780 Alan Hayward <alan.hayward@arm.com>
6781 David Sherwood <david.sherwood@arm.com>
6783 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6784 * target.def (autovectorize_vector_sizes): Return the vector sizes
6785 by pointer, using vector_sizes rather than a bitmask.
6786 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6787 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6788 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6790 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6791 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6792 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6793 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6794 * omp-general.c (omp_max_vf): Likewise.
6795 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6796 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6797 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6798 * tree-vect-slp.c (vect_slp_bb): Likewise.
6799 * doc/tm.texi: Regenerate.
6800 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6802 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6803 the vector size as a poly_uint64 rather than an unsigned int.
6804 (current_vector_size): Change from an unsigned int to a poly_uint64.
6805 (get_vectype_for_scalar_type): Update accordingly.
6806 * tree.h (build_truth_vector_type): Take the size and number of
6807 units as a poly_uint64 rather than an unsigned int.
6808 (build_vector_type): Add a temporary overload that takes
6809 the number of units as a poly_uint64 rather than an unsigned int.
6810 * tree.c (make_vector_type): Likewise.
6811 (build_truth_vector_type): Take the number of units as a poly_uint64
6812 rather than an unsigned int.
6814 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6815 Alan Hayward <alan.hayward@arm.com>
6816 David Sherwood <david.sherwood@arm.com>
6818 * target.def (get_mask_mode): Take the number of units and length
6819 as poly_uint64s rather than unsigned ints.
6820 * targhooks.h (default_get_mask_mode): Update accordingly.
6821 * targhooks.c (default_get_mask_mode): Likewise.
6822 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6823 * doc/tm.texi: Regenerate.
6825 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6826 Alan Hayward <alan.hayward@arm.com>
6827 David Sherwood <david.sherwood@arm.com>
6829 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6830 * omp-general.c (omp_max_vf): Likewise.
6831 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6832 (expand_omp_simd): Handle polynomial safelen.
6833 * omp-low.c (omplow_simd_context): Add a default constructor.
6834 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6835 (lower_rec_simd_input_clauses): Update accordingly.
6836 (lower_rec_input_clauses): Likewise.
6838 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6839 Alan Hayward <alan.hayward@arm.com>
6840 David Sherwood <david.sherwood@arm.com>
6842 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6843 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6844 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6845 (vect_analyze_slp_cost): Likewise.
6846 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6847 (vect_model_load_cost): Likewise.
6849 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6850 Alan Hayward <alan.hayward@arm.com>
6851 David Sherwood <david.sherwood@arm.com>
6853 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6854 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6855 from an unsigned int * to a poly_uint64_pod *.
6856 (calculate_unrolling_factor): New function.
6857 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6859 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6860 Alan Hayward <alan.hayward@arm.com>
6861 David Sherwood <david.sherwood@arm.com>
6863 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6864 from an unsigned int to a poly_uint64.
6865 (_loop_vec_info::slp_unrolling_factor): Likewise.
6866 (_loop_vec_info::vectorization_factor): Change from an int
6868 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6869 (vect_get_num_vectors): New function.
6870 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6871 (vect_get_num_copies): Use vect_get_num_vectors.
6872 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6873 to an unsigned int *.
6874 (vect_analyze_data_refs): Change min_vf from an int * to a
6876 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6877 than an unsigned HOST_WIDE_INT.
6878 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6879 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6880 to an unsigned int *.
6881 (vect_analyze_data_ref_dependences): Likewise.
6882 (vect_compute_data_ref_alignment): Handle polynomial vf.
6883 (vect_enhance_data_refs_alignment): Likewise.
6884 (vect_prune_runtime_alias_test_list): Likewise.
6885 (vect_shift_permute_load_chain): Likewise.
6886 (vect_supportable_dr_alignment): Likewise.
6887 (dependence_distance_ge_vf): Take the vectorization factor as a
6888 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6889 (vect_analyze_data_refs): Change min_vf from an int * to a
6891 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6892 vfm1 as a poly_uint64 rather than an int. Make the same change
6893 for the returned bound_scalar.
6894 (vect_gen_vector_loop_niters): Handle polynomial vf.
6895 (vect_do_peeling): Likewise. Update call to
6896 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6897 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6899 * tree-vect-loop.c (vect_determine_vectorization_factor)
6900 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6901 (vect_get_known_peeling_cost): Likewise.
6902 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6903 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6904 (vect_transform_loop): Likewise. Use the lowest possible VF when
6905 updating the upper bounds of the loop.
6906 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6908 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6909 polynomial unroll factors.
6910 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6911 (vect_make_slp_decision): Likewise.
6912 (vect_supported_load_permutation_p): Likewise, and polynomial
6914 (vect_analyze_slp_cost): Handle polynomial vf.
6915 (vect_slp_analyze_node_operations): Likewise.
6916 (vect_slp_analyze_bb_1): Likewise.
6917 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6918 than an unsigned HOST_WIDE_INT.
6919 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6920 (vectorizable_load): Handle polynomial vf.
6921 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6923 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6925 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6926 Alan Hayward <alan.hayward@arm.com>
6927 David Sherwood <david.sherwood@arm.com>
6929 * match.pd: Handle bit operations involving three constants
6930 and try to fold one pair.
6932 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6934 * tree-vect-loop-manip.c: Include gimple-fold.h.
6935 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6936 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6937 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6938 Add a path that uses a step of VF instead of 1, but disable it
6940 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6941 and niters_no_overflow parameters. Update calls to
6942 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6943 Create a new SSA name if the latter choses to use a ste other
6944 than zero, and return it via niters_vector_mult_vf_var.
6945 * tree-vect-loop.c (vect_transform_loop): Update calls to
6946 vect_do_peeling, vect_gen_vector_loop_niters and
6947 slpeel_make_loop_iterate_ntimes.
6948 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6949 (vect_gen_vector_loop_niters): Update declarations after above changes.
6951 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6953 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6954 128-bit round to integer instructions.
6955 (ceil<mode>2): Likewise.
6956 (btrunc<mode>2): Likewise.
6957 (round<mode>2): Likewise.
6959 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6961 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
6962 unaligned VSX load/store on P8/P9.
6963 (expand_block_clear): Allow the use of unaligned VSX
6964 load/store on P8/P9.
6966 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
6968 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
6970 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
6971 swap associated with both a load and a store.
6973 2018-01-02 Andrew Waterman <andrew@sifive.com>
6975 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
6976 * config/riscv/riscv.md (clear_cache): Use it.
6978 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
6980 * web.c: Remove out-of-date comment.
6982 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6984 * expr.c (fixup_args_size_notes): Check that any existing
6985 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
6986 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
6987 (emit_single_push_insn): ...here.
6989 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6991 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
6992 (const_vector_encoded_nelts): New function.
6993 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
6994 (const_vector_int_elt, const_vector_elt): Declare.
6995 * emit-rtl.c (const_vector_int_elt_1): New function.
6996 (const_vector_elt): Likewise.
6997 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
6998 of CONST_VECTOR_ELT.
7000 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7002 * expr.c: Include rtx-vector-builder.h.
7003 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
7004 directly on the tree encoding.
7005 (const_vector_from_tree): Likewise.
7006 * optabs.c: Include rtx-vector-builder.h.
7007 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
7008 sequence of "u" values.
7009 * vec-perm-indices.c: Include rtx-vector-builder.h.
7010 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
7011 directly on the vec_perm_indices encoding.
7013 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7015 * doc/rtl.texi (const_vector): Describe new encoding scheme.
7016 * Makefile.in (OBJS): Add rtx-vector-builder.o.
7017 * rtx-vector-builder.h: New file.
7018 * rtx-vector-builder.c: Likewise.
7019 * rtl.h (rtx_def::u2): Add a const_vector field.
7020 (CONST_VECTOR_NPATTERNS): New macro.
7021 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
7022 (CONST_VECTOR_DUPLICATE_P): Likewise.
7023 (CONST_VECTOR_STEPPED_P): Likewise.
7024 (CONST_VECTOR_ENCODED_ELT): Likewise.
7025 (const_vec_duplicate_p): Check for a duplicated vector encoding.
7026 (unwrap_const_vec_duplicate): Likewise.
7027 (const_vec_series_p): Check for a non-duplicated vector encoding.
7028 Say that the function only returns true for integer vectors.
7029 * emit-rtl.c: Include rtx-vector-builder.h.
7030 (gen_const_vec_duplicate_1): Delete.
7031 (gen_const_vector): Call gen_const_vec_duplicate instead of
7032 gen_const_vec_duplicate_1.
7033 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
7034 (gen_const_vec_duplicate): Use rtx_vector_builder.
7035 (gen_const_vec_series): Likewise.
7036 (gen_rtx_CONST_VECTOR): Likewise.
7037 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
7038 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7039 Build a new vector rather than modifying a CONST_VECTOR in-place.
7040 (handle_special_swappables): Update call accordingly.
7041 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
7042 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7043 Build a new vector rather than modifying a CONST_VECTOR in-place.
7044 (handle_special_swappables): Update call accordingly.
7046 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7048 * simplify-rtx.c (simplify_const_binary_operation): Use
7049 CONST_VECTOR_ELT instead of XVECEXP.
7051 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7053 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
7054 the selector elements to be different from the data elements
7055 if the selector is a VECTOR_CST.
7056 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
7057 ssizetype for the selector.
7059 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7061 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
7062 before testing each element individually.
7063 * tree-vect-generic.c (lower_vec_perm): Likewise.
7065 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7067 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
7068 * selftest-run-tests.c (selftest::run_tests): Call it.
7069 * vector-builder.h (vector_builder::operator ==): New function.
7070 (vector_builder::operator !=): Likewise.
7071 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
7072 (vec_perm_indices::all_from_input_p): New function.
7073 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7074 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
7075 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
7076 instead of reading the VECTOR_CST directly. Detect whether both
7077 vector inputs are the same before constructing the vec_perm_indices,
7078 and update the number of inputs argument accordingly. Use the
7079 utility functions added above. Only construct sel2 if we need to.
7081 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7083 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
7084 the broadcast of the low byte.
7085 (expand_mult_highpart): Use an explicit encoding for the permutes.
7086 * optabs-query.c (can_mult_highpart_p): Likewise.
7087 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
7088 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7089 (vectorizable_bswap): Likewise.
7090 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
7091 explicit encoding for the power-of-2 permutes.
7092 (vect_permute_store_chain): Likewise.
7093 (vect_grouped_load_supported): Likewise.
7094 (vect_permute_load_chain): Likewise.
7096 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7098 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
7099 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
7100 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
7101 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7102 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
7103 (vect_gen_perm_mask_any): Likewise.
7105 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7107 * int-vector-builder.h: New file.
7108 * vec-perm-indices.h: Include int-vector-builder.h.
7109 (vec_perm_indices): Redefine as an int_vector_builder.
7110 (auto_vec_perm_indices): Delete.
7111 (vec_perm_builder): Redefine as a stand-alone class.
7112 (vec_perm_indices::vec_perm_indices): New function.
7113 (vec_perm_indices::clamp): Likewise.
7114 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
7115 (vec_perm_indices::new_vector): New function.
7116 (vec_perm_indices::new_expanded_vector): Update for new
7117 vec_perm_indices class.
7118 (vec_perm_indices::rotate_inputs): New function.
7119 (vec_perm_indices::all_in_range_p): Operate directly on the
7120 encoded form, without computing elided elements.
7121 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
7122 encoding. Update for new vec_perm_indices class.
7123 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
7124 the given vec_perm_builder.
7125 (expand_vec_perm_var): Update vec_perm_builder constructor.
7126 (expand_mult_highpart): Use vec_perm_builder instead of
7127 auto_vec_perm_indices.
7128 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
7129 vec_perm_indices instead of auto_vec_perm_indices. Use a single
7130 or double series encoding as appropriate.
7131 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
7132 vec_perm_indices instead of auto_vec_perm_indices.
7133 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7134 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7135 (vect_permute_store_chain): Likewise.
7136 (vect_grouped_load_supported): Likewise.
7137 (vect_permute_load_chain): Likewise.
7138 (vect_shift_permute_load_chain): Likewise.
7139 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7140 (vect_transform_slp_perm_load): Likewise.
7141 (vect_schedule_slp_instance): Likewise.
7142 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7143 (vectorizable_mask_load_store): Likewise.
7144 (vectorizable_bswap): Likewise.
7145 (vectorizable_store): Likewise.
7146 (vectorizable_load): Likewise.
7147 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
7148 vec_perm_indices instead of auto_vec_perm_indices. Use
7149 tree_to_vec_perm_builder to read the vector from a tree.
7150 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
7151 vec_perm_builder instead of a vec_perm_indices.
7152 (have_whole_vector_shift): Use vec_perm_builder and
7153 vec_perm_indices instead of auto_vec_perm_indices. Leave the
7154 truncation to calc_vec_perm_mask_for_shift.
7155 (vect_create_epilog_for_reduction): Likewise.
7156 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
7157 from auto_vec_perm_indices to vec_perm_indices.
7158 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7159 instead of changing individual elements.
7160 (aarch64_vectorize_vec_perm_const): Use new_vector to install
7161 the vector in d.perm.
7162 * config/arm/arm.c (expand_vec_perm_d::perm): Change
7163 from auto_vec_perm_indices to vec_perm_indices.
7164 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7165 instead of changing individual elements.
7166 (arm_vectorize_vec_perm_const): Use new_vector to install
7167 the vector in d.perm.
7168 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
7169 Update vec_perm_builder constructor.
7170 (rs6000_expand_interleave): Likewise.
7171 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
7172 (rs6000_expand_interleave): Likewise.
7174 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7176 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
7177 to qimode could truncate the indices.
7178 * optabs.c (expand_vec_perm_var): Likewise.
7180 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7182 * Makefile.in (OBJS): Add vec-perm-indices.o.
7183 * vec-perm-indices.h: New file.
7184 * vec-perm-indices.c: Likewise.
7185 * target.h (vec_perm_indices): Replace with a forward class
7187 (auto_vec_perm_indices): Move to vec-perm-indices.h.
7188 * optabs.h: Include vec-perm-indices.h.
7189 (expand_vec_perm): Delete.
7190 (selector_fits_mode_p, expand_vec_perm_var): Declare.
7191 (expand_vec_perm_const): Declare.
7192 * target.def (vec_perm_const_ok): Replace with...
7193 (vec_perm_const): ...this new hook.
7194 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
7195 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7196 * doc/tm.texi: Regenerate.
7197 * optabs.def (vec_perm_const): Delete.
7198 * doc/md.texi (vec_perm_const): Likewise.
7199 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7200 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7201 expand_vec_perm for constant permutation vectors. Assert that
7202 the mode of variable permutation vectors is the integer equivalent
7203 of the mode that is being permuted.
7204 * optabs-query.h (selector_fits_mode_p): Declare.
7205 * optabs-query.c: Include vec-perm-indices.h.
7206 (selector_fits_mode_p): New function.
7207 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7208 is defined, instead of checking whether the vec_perm_const_optab
7209 exists. Use targetm.vectorize.vec_perm_const instead of
7210 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7211 fit in the vector mode before using a variable permute.
7212 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7213 vec_perm_indices instead of an rtx.
7214 (expand_vec_perm): Replace with...
7215 (expand_vec_perm_const): ...this new function. Take the selector
7216 as a vec_perm_indices rather than an rtx. Also take the mode of
7217 the selector. Update call to shift_amt_for_vec_perm_mask.
7218 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7219 Use vec_perm_indices::new_expanded_vector to expand the original
7220 selector into bytes. Check whether the indices fit in the vector
7221 mode before using a variable permute.
7222 (expand_vec_perm_var): Make global.
7223 (expand_mult_highpart): Use expand_vec_perm_const.
7224 * fold-const.c: Includes vec-perm-indices.h.
7225 * tree-ssa-forwprop.c: Likewise.
7226 * tree-vect-data-refs.c: Likewise.
7227 * tree-vect-generic.c: Likewise.
7228 * tree-vect-loop.c: Likewise.
7229 * tree-vect-slp.c: Likewise.
7230 * tree-vect-stmts.c: Likewise.
7231 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7233 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7234 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7235 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7236 (aarch64_vectorize_vec_perm_const): ...this new function.
7237 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7238 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7239 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7240 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7241 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7242 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7243 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7245 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7246 check for NEON modes.
7247 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7248 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7249 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7250 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7252 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7253 the old VEC_PERM_CONST conditions.
7254 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7255 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7256 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7257 (ia64_vectorize_vec_perm_const_ok): Merge into...
7258 (ia64_vectorize_vec_perm_const): ...this new function.
7259 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7260 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7261 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7262 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7263 * config/mips/mips.c (mips_expand_vec_perm_const)
7264 (mips_vectorize_vec_perm_const_ok): Merge into...
7265 (mips_vectorize_vec_perm_const): ...this new function.
7266 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7267 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7268 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7269 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7270 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7271 (rs6000_expand_vec_perm_const): Delete.
7272 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7274 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7275 (altivec_expand_vec_perm_const_le): Take each operand individually.
7276 Operate on constant selectors rather than rtxes.
7277 (altivec_expand_vec_perm_const): Likewise. Update call to
7278 altivec_expand_vec_perm_const_le.
7279 (rs6000_expand_vec_perm_const): Delete.
7280 (rs6000_vectorize_vec_perm_const_ok): Delete.
7281 (rs6000_vectorize_vec_perm_const): New function.
7282 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7283 an element count and rtx array.
7284 (rs6000_expand_extract_even): Update call accordingly.
7285 (rs6000_expand_interleave): Likewise.
7286 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7287 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7288 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7289 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7290 (rs6000_expand_vec_perm_const): Delete.
7291 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7292 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7293 (altivec_expand_vec_perm_const_le): Take each operand individually.
7294 Operate on constant selectors rather than rtxes.
7295 (altivec_expand_vec_perm_const): Likewise. Update call to
7296 altivec_expand_vec_perm_const_le.
7297 (rs6000_expand_vec_perm_const): Delete.
7298 (rs6000_vectorize_vec_perm_const_ok): Delete.
7299 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7300 reference to the SPE evmerge intructions.
7301 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7302 an element count and rtx array.
7303 (rs6000_expand_extract_even): Update call accordingly.
7304 (rs6000_expand_interleave): Likewise.
7305 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7306 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7308 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7310 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7312 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7313 vector mode and that that mode matches the mode of the data
7315 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7316 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7317 directly using expand_vec_perm_1 when forcing selectors into
7319 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7321 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7323 * optabs-query.h (can_vec_perm_p): Delete.
7324 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7325 * optabs-query.c (can_vec_perm_p): Split into...
7326 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7327 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7328 particular selector is valid.
7329 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7330 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7331 (vect_grouped_load_supported): Likewise.
7332 (vect_shift_permute_load_chain): Likewise.
7333 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7334 (vect_transform_slp_perm_load): Likewise.
7335 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7336 (vectorizable_bswap): Likewise.
7337 (vect_gen_perm_mask_checked): Likewise.
7338 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7339 implementations of variable permutation vectors into account
7340 when deciding which selector to use.
7341 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7342 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7343 with a false third argument.
7344 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7345 to test whether the constant selector is valid and can_vec_perm_var_p
7346 to test whether a variable selector is valid.
7348 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7350 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7351 * optabs-query.c (can_vec_perm_p): Likewise.
7352 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7353 instead of vec_perm_indices.
7354 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7355 (vect_gen_perm_mask_checked): Likewise,
7356 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7357 (vect_gen_perm_mask_checked): Likewise,
7359 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7361 * optabs-query.h (qimode_for_vec_perm): Declare.
7362 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7363 (qimode_for_vec_perm): ...this new function.
7364 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7366 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7368 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7369 does not have a conditional at the top.
7371 2018-01-02 Richard Biener <rguenther@suse.de>
7373 * ipa-inline.c (big_speedup_p): Fix expression.
7375 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7378 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7381 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7385 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7386 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7387 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7388 cond_taken_branch_cost 3->4.
7390 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7392 PR tree-optimization/83581
7393 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7394 TODO_cleanup_cfg if any changes have been made.
7397 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7398 convert_modes if target mode has the right side, but different mode
7402 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7403 last argument when extracting from CONCAT. If either from_real or
7404 from_imag is NULL, use expansion through memory. If result is not
7405 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7406 the parts directly to inner mode, if even that fails, use expansion
7410 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7411 check for bswap in mode rather than HImode and use that in expand_unop
7414 Copyright (C) 2018 Free Software Foundation, Inc.
7416 Copying and distribution of this file, with or without modification,
7417 are permitted in any medium without royalty provided the copyright
7418 notice and this notice are preserved.