1 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
3 * config/rs6000/altivec.md: Delete contraint arguments to
4 define_expand, define_split, and define_peephole2, and in
5 define_insn_and_split if always unused.
6 * config/rs6000/darwin.md: Ditto.
7 * config/rs6000/dfp.md: Ditto.
8 * config/rs6000/rs6000.md: Ditto.
9 * config/rs6000/sync.md: Ditto.
10 * config/rs6000/vector.md: Ditto.
11 * config/rs6000/vsx.md: Ditto.
13 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
15 * config/rs6000/altivec.md: Write output control strings as braced
16 blocks instead of double-quoted strings.
17 * config/rs6000/darwin.md: Ditto.
18 * config/rs6000/rs6000.md: Ditto.
19 * config/rs6000/vector.md: Ditto.
20 * config/rs6000/vsx.md: Ditto.
22 2018-02-21 Jason Merrill <jason@redhat.com>
24 PR c++/84314 - ICE with templates and fastcall attribute.
25 * attribs.c (build_type_attribute_qual_variant): Remove assert.
27 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
29 * ipa-cp.c (determine_versionability): Fix comment typos.
31 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
34 * ipa-cp.c (determine_versionability): Do not version functions caling
37 2018-02-21 Martin Liska <mliska@suse.cz>
40 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
41 Add "native" as a possible value.
42 * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT): Define
43 the macro when native cpu detection is available.
45 2018-02-21 Martin Liska <mliska@suse.cz>
48 * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
49 Add "native" as a possible value.
50 * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
51 when native cpu detection is available.
53 2018-02-21 Jakub Jelinek <jakub@redhat.com>
54 Martin Sebor <msebor@redhat.com>
56 PR tree-optimization/84478
57 * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
59 * gimple-fold.c (get_range_strlen): Make minlen const and assume it
60 can't be NULL. Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
61 support which is conservatively correct, for 2 only stay conservative
62 for maxlen. Formatting and comment capitalization fixes. Add STRICT
63 argument to the 2 argument get_range_strlen, adjust 6 arg
64 get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
66 (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
67 (gimple_fold_builtin_strlen): Pass true as last argument to
70 2018-02-20 Martin Sebor <msebor@redhat.com>
73 * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
74 (builtin_memref::set_base_and_offset): Same. Handle inner references.
75 (builtin_memref::builtin_memref): Factor out parts into
76 set_base_and_offset and call it.
78 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
81 * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
82 is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
83 greater precision. If to_mode is a MODE_PARTIAL_INT, stop the
84 search at the associated MODE_INT.
86 2018-02-20 Jeff Law <law@redhat.com>
89 PR tree-optimization/81592
91 * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
92 for range data rather than using global data.
93 * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
94 range data rather than using global data.
95 * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
96 pass it to children as needed.
97 (struct directive::fmtresult): Similarly.
98 (struct directive::set_width): Similarly.
99 (struct directive::set_precision): Similarly.
100 (format_integer, format_directive, parse_directive): Similarly.
101 (format_none): Accept unnamed vr_values parameter.
102 (format_percent, format_floating, format_character): Similarly.
103 (format_string, format_plain): Similarly.
104 * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
105 the EVRP range analyzer for range data rather than using global data.
106 * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
107 gimple-ssa-evrp-analyze.h
108 (class sprintf_dom_walker): Add after_dom_children member function.
109 Add evrp_range_analyzer member.
110 (sprintf_dom_walker::before_dom_children): Call into the EVRP
111 range analyzer as needed.
112 (sprintf_dom_walker::after_dom_children): New member function.
113 * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
115 (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
116 (evrp_range_analyzer::pop_to_marker): Likewise.
118 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
120 PR tree-optimization/84419
121 * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
122 with the required type if its current type is compatible but
125 2018-02-20 Jakub Jelinek <jakub@redhat.com>
128 * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
131 2018-02-20 Martin Liska <mliska@suse.cz>
134 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
135 possible values if we don't have a hint.
137 2018-02-20 Martin Liska <mliska@suse.cz>
141 * final.c (shorten_branches): Build align_tab array with one
143 * opts.c (finish_options): Add alignment option limit check.
144 (MAX_CODE_ALIGN): Likewise.
145 (MAX_CODE_ALIGN_VALUE): Likewise.
146 * doc/invoke.texi: Document maximum allowed option value for
147 all -falign-* options.
149 2018-02-19 Jakub Jelinek <jakub@redhat.com>
152 * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
153 * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
154 * var-tracking.c (emit_note_insn_var_location): Remove all references
155 to NOTE_INSN_CALL_ARG_LOCATION.
156 (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
157 the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
158 Use copy_rtx_if_shared.
159 * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
160 NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
161 (dwarf2out_var_location): Remove handling of
162 NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
164 * final.c (final_scan_insn): Remove all references to
165 NOTE_INSN_CALL_ARG_LOCATION.
166 (rest_of_clean_state): Likewise. Remove REG_CALL_ARG_LOCATION notes
167 before dumping final insns.
168 * except.c (emit_note_eh_region_end): Remove all references to
169 NOTE_INSN_CALL_ARG_LOCATION.
170 * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
171 * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
172 * config/arc/arc.c (hwloop_optimize): Likewise.
173 * config/arm/arm.c (create_fix_barrier): Likewise.
174 * config/s390/s390.c (s390_chunkify_start): Likewise.
175 * config/sh/sh.c (find_barrier): Likewise.
176 * config/i386/i386.c (rest_of_insert_endbranch,
177 ix86_seh_fixup_eh_fallthru): Likewise.
178 * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
179 * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
180 * config/frv/frv.c (frv_function_prologue): Likewise.
181 * emit-rtl.c (try_split): Likewise. Copy over REG_CALL_ARG_LOCATION
183 (note_outside_basic_block_p): Remove all references to
184 NOTE_INSN_CALL_ARG_LOCATION.
185 * gengtype.c (adjust_field_rtx_def): Likewise.
186 * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
188 * jump.c (cleanup_barriers, delete_related_insns): Likewise.
189 * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
192 * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
195 PR tree-optimization/84452
196 * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
197 expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
200 2018-02-19 Martin Liska <mliska@suse.cz>
203 * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
205 2018-02-19 Martin Liska <mliska@suse.cz>
206 Richard Sandiford <richard.sandiford@linaro.org>
208 PR tree-optimization/82491
209 * gimple-fold.c (get_base_constructor): Make earlier bail out
212 2018-02-19 Carl Love <cel@us.ibm.com>
214 * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
215 BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
217 * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
220 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
222 * config/rl78/rl78.md (movdf): New define expand.
224 2018-02-19 Martin Liska <mliska@suse.cz>
227 * doc/invoke.texi: Fix typo.
228 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
230 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
232 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
233 handle rs6000_single_float and rs6000_double_float specially for
236 2018-02-16 Jeff Law <law@redhat.com>
238 * config/rx/rx.c (add_pop_cfi_notes): New function.;
241 2018-02-16 Jakub Jelinek <jakub@redhat.com>
244 * ipa-inline.c (inline_small_functions): Fix a typo.
246 2018-02-16 Nathan Sidwell <nathan@acm.org>
248 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
250 2018-02-16 Carl Love <cel@us.ibm.com>
252 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
253 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
254 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
255 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
256 expansion to P8V_BUILTIN_VEC_FLOAT2.
258 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
260 PR rtl-optimization/70023
261 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
262 src_regno into account.
264 2018-02-16 Carl Love <cel@us.ibm.com>
266 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
267 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
268 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
269 * config/rs6000/rs6000.c: Remove case statements for
270 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
271 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
272 and P9V_BUILTIN_VEC_VINSERT4B.
273 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
274 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
275 * config/rs6000/vsx.md:
276 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
279 2018-02-16 Carl Love <cel@us.ibm.com>
281 * config/rs6000/altivec.h: Add builtin names vec_extract4b
283 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
285 * config/rs6000/rs6000-c.c: Add the definitions for
286 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
287 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
288 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
289 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
290 definition for insert4b and define insn *insert3b_internal.
291 * doc/extend.texi: Add documentation for vec_extract4b.
293 2018-02-16 Nathan Sidwell <nathan@acm.org>
295 * doc/extend.texi (Backwards Compatibility): Mention friend
296 injection. Note for-scope is deprecated.
297 * doc/invoke.texi (-ffriend-injection): Deprecate.
299 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
301 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
302 that moved to I2, also allow destinations that are a paradoxical
303 subreg (instead of a normal reg).
305 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
308 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
311 2018-02-16 Richard Biener <rguenther@suse.de>
313 PR tree-optimization/84037
314 PR tree-optimization/84016
316 * config/i386/i386.c (ix86_builtin_vectorization_cost):
317 Adjust vec_construct for the fact we need additional higher latency
318 128bit inserts for AVX256 and AVX512 vector builds.
319 (ix86_add_stmt_cost): Scale vector construction cost for
322 2018-02-16 Richard Biener <rguenther@suse.de>
324 PR tree-optimization/84417
325 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
326 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
327 (non_rewritable_lvalue_p): Likewise, use poly-ints.
329 2018-02-16 Martin Liska <mliska@suse.cz>
332 * internal-fn.def (ASAN_CHECK): Set proper flags.
333 (ASAN_MARK): Likewise.
335 2018-02-16 Julia Koval <julia.koval@intel.com>
337 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
340 2018-02-16 Jakub Jelinek <jakub@redhat.com>
343 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
344 Use ++iter rather than iter++ for std::list iterators.
345 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
346 defer deleting them until all nodes in the forest are processed. Do
347 free even leaf nodes. Change to_process into auto_vec.
350 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
351 * vec.h (vec_default_construct): Use memset instead of placement new
352 if BROKEN_VALUE_INITIALIZATION is defined.
353 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
354 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
357 PR rtl-optimization/83723
358 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
359 * lra.c (lra_substitute_pseudo): Likewise. If true, use
360 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
362 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
364 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
366 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
368 PR rtl-optimization/81443
369 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
370 from inner REGs to paradoxical SUBREGs.
372 2018-02-16 Richard Biener <rguenther@suse.de>
374 PR tree-optimization/84399
375 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
376 For operands we can analyze at their definition make sure we can
377 analyze them at each use as well.
379 2018-02-16 Richard Biener <rguenther@suse.de>
381 PR tree-optimization/84190
382 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
383 volatile accesses if the decl isn't volatile.
385 2018-02-15 Jason Merrill <jason@redhat.com>
387 PR c++/84314 - ICE with templates and fastcall attribute.
388 * attribs.c (build_type_attribute_qual_variant): Don't clobber
389 TYPE_CANONICAL on an existing type.
391 2018-02-15 Jakub Jelinek <jakub@redhat.com>
393 PR tree-optimization/84383
394 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
395 dstoff nor call operand_equal_p if dstbase is NULL.
397 PR tree-optimization/84334
398 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
399 also a CONSTANT_CLASS_P, punt.
401 2018-02-14 Jim Wilson <jimw@sifive.com>
403 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
404 first SMALL_OPERAND check. New local min_second_step. Move assert
405 to where locals are set. Add TARGET_RVC support.
406 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
408 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
410 * doc/invoke.texi: Correct -Wformat-overflow code sample.
412 2018-02-14 Martin Sebor <msebor@redhat.com>
414 PR tree-optimization/83698
415 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
416 arrays constrain the offset range to their bounds.
417 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
418 (builtin_access::overlap): Avoid setting the size of overlap if it's
420 (maybe_diag_overlap): Also consider arrays when deciding what values
421 of offsets to include in diagnostics.
423 2018-02-14 Martin Sebor <msebor@redhat.com>
426 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
427 that correspond to the kind of a declaration.
429 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
432 * config/pa/pa.md: Load address of PIC label using the linkage table
433 if the label is nonlocal.
435 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
437 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
438 warning message if user requests -maltivec=be.
439 * doc/invoke.texi: Document deprecation of -maltivec=be.
441 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
444 * config/rs6000/rs6000-c.c: Update definitions for
445 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
446 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
448 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
451 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
452 add _get_ssp intrinsics. Remove argument from
453 __builtin_ia32_rdssp[d|q].
454 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
455 * config/i386/i386-builtin.def: Remove argument from
456 __builtin_ia32_rdssp[d|q].
457 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
458 ix86_expand_special_args_builtin for _rdssp[d|q].
459 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
460 Clear register before usage.
461 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
462 Add documentation for new _get_ssp and _inc_ssp intrinsics.
464 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
466 PR tree-optimization/84357
467 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
468 operand 1 of an ARRAY_REF too.
470 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
473 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
474 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
476 (set_of_reg): New struct.
477 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
478 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
479 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
481 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
482 Split into bitclr, bitset, bitinvert patterns if appropriate.
483 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
484 use rx_fuse_in_memory_bitop.
485 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
486 to named insn, correct maximum insn length.
488 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
491 * machmode.def: Define a complex mode for PARTIAL_INT.
492 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
494 * doc/rtl.texi: Document CSPImode.
495 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
497 (msp430_hard_regno_nregs_with_padding): Likewise.
499 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
502 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
504 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
506 PR rtl-optimization/84169
507 * combine.c (try_combine): New variable split_i2i3. Set it to true if
508 we generated a parallel as new i3 and we split that to new i2 and i3
509 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
510 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
511 those to i2, not i1. Partially rewrite this scan code.
513 2018-02-13 Jakub Jelinek <jakub@redhat.com>
516 * stor-layout.c (place_field): For variable length fields, adjust
517 offset_align afterwards not just based on the field's alignment,
518 but also on the size.
521 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
522 of exps and logs in the use_exp2 case.
524 2018-02-13 Jeff Law <law@redhat.com>
526 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
529 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
532 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
536 * common.opt (gas-loc-support, gas-locview-support): New.
537 (ginline-points, ginternal-reset-location-views): New.
538 * doc/invoke.texi: Document them. Use @itemx where intended.
539 (gvariable-location-views): Adjust.
540 * target.def (reset_location_view): New.
541 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
542 (TARGET_RESET_LOCATION_VIEW): New.
543 * doc/tm.texi: Rebuilt.
544 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
545 (dwarf2out_default_as_locview_support): New.
546 (output_asm_line_debug_info): Use option variables.
547 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
548 (output_loc_list): Likewise.
549 (add_high_low_attributes): Check option variables.
550 Don't output entry view attribute in strict mode.
551 (gen_inlined_subroutine_die): Check option variables.
552 (dwarf2out_inline_entry): Likewise.
553 (init_sections_and_labels): Likewise.
554 (dwarf2out_early_finish): Likewise.
555 (maybe_reset_location_view): New, from...
556 (dwarf2out_var_location): ... here. Call it.
557 * debug.h (dwarf2out_default_as_loc_support): Declare.
558 (dwarf2out_default_as_locview_support): Declare.
559 * hooks.c (hook_int_rtx_insn_0): New.
560 * hooks.h (hook_int_rtx_insn_0): Declare.
561 * toplev.c (process_options): Take -gas-loc-support and
562 -gas-locview-support from dwarf2out. Enable
563 -gvariable-location-views by default only with locview
564 assembler support. Enable -ginternal-reset-location-views by
565 default only if the target defines the corresponding hook.
566 Enable -ginline-points by default if location views are
567 enabled; force it disabled if statement frontiers are
569 * tree-inline.c (expand_call_inline): Check option variables.
570 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
572 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
574 PR tree-optimization/84321
575 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
576 handling. Also check whether the anti-range contains any values
577 that satisfy the mask; switch to a VR_RANGE if not.
579 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
582 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
584 2018-02-13 Martin Jambor <mjambor@suse.cz>
587 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
588 of call statements, also set location of a load to a temporary.
590 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
592 * config/rl78/rl78.c (add_vector_labels): New function.
593 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
594 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
595 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
596 which checks that no arguments are passed.
597 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
598 * doc/extend.texi: Documentation for the new attribute.
600 2018-02-13 Andreas Schwab <schwab@suse.de>
602 * config/riscv/linux.h (CPP_SPEC): Define.
604 2018-02-13 Jakub Jelinek <jakub@redhat.com>
607 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
608 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
609 OPTION_MASK_ISA_AES as first argument to def_builtin_const
610 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
611 instead of OPTION_MASK_ISA_PCLMUL as first argument to
612 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
613 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
614 temporarily for AES and PCLMUL builtins.
616 PR tree-optimization/84339
617 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
618 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
622 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
623 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
624 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
626 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
628 * omp-simd-clone.h: New file.
629 * omp-simd-clone.c: Include omp-simd-clone.h.
630 (expand_simd_clones): No longer static.
631 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
632 cgraph.h and omp-simd-clone.h.
633 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
634 (vect_recog_widen_shift_pattern): Formatting fix.
635 (vect_pattern_recog_1): Don't check optab for calls.
638 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
639 operands[2] into a REG before using gen_lowpart on it.
641 2018-02-12 Jeff Law <law@redhat.com>
644 * config/sh/sh.c (find_barrier): Consider a sibling call
647 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
648 successfully back substituting a reg.
650 2018-02-12 Richard Biener <rguenther@suse.de>
652 PR tree-optimization/84037
653 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
654 parameter, move visited init to caller.
655 (vect_slp_analyze_operations): Separate cost from validity
656 check, initialize visited once for all instances.
657 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
659 * tree-vect-stmts.c (vect_model_simple_cost): Make early
661 (vect_model_promotion_demotion_cost): Likewise.
662 (vectorizable_bswap): Guard cost modeling with !slp_node
663 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
665 (vectorizable_call): Likewise.
666 (vectorizable_conversion): Likewise.
667 (vectorizable_assignment): Likewise.
668 (vectorizable_shift): Likewise.
669 (vectorizable_operation): Likewise.
670 (vectorizable_store): Likewise.
671 (vectorizable_load): Likewise.
672 (vectorizable_condition): Likewise.
673 (vectorizable_comparison): Likewise.
675 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
678 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
679 (ASAN_MARK): Fix fnspec to account for return value, change pointer
680 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
682 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
685 * params.def (inline-min-speedup): Increase from 8 to 15.
686 (max-inline-insns-auto): Decrease from 40 to 30.
687 * ipa-split.c (consider_split): Add some buffer for function to
688 be considered inlining candidate.
689 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
692 2018-02-12 Richard Biener <rguenther@suse.de>
694 PR tree-optimization/84037
695 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
696 matched stmts if we cannot swap the non-matched ones.
698 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
700 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
701 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
702 _mm_maskz_scalef_round_ss): New intrinsics.
703 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
704 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
705 __builtin_ia32_scalefss_round): Remove.
706 (__builtin_ia32_scalefsd_mask_round,
707 __builtin_ia32_scalefss_mask_round): New intrinsics.
708 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
709 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
710 ((match_operand:VF_128 2 "<round_nimm_predicate>"
711 "<round_constraint>")): Changed to ...
712 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
713 "<round_scalar_constraint>")): ... this.
714 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
715 %0, %1, %2<round_op3>}"): Changed to ...
716 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
717 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
718 %2<round_scalar_mask_op3>}"): ... this.
719 * config/i386/subst.md (round_scalar_nimm_predicate): New.
721 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
723 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
724 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
725 (_mm_maskz_sqrt_round_ss): New intrinsics.
726 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
727 (__builtin_ia32_sqrtsd_mask_round)
728 (__builtin_ia32_sqrtss_mask_round): New builtins.
729 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
730 (__builtin_ia32_sqrtss_round): Remove.
731 (__builtin_ia32_sqrtsd_mask_round)
732 (__builtin_ia32_sqrtss_mask_round): New builtins.
733 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
734 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
735 ((match_operand:VF_128 1 "vector_operand"
736 "xBm,<round_constraint>")): Changed to ...
737 ((match_operand:VF_128 1 "vector_operand"
738 "xBm,<round_scalar_constraint>")): ... this.
739 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
740 %0, %2, %<iptr>1<round_op3>}): Changed to ...
741 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
742 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
743 %<iptr>1<round_scalar_mask_op3>}): ... this.
744 ((set_attr "prefix" "<round_prefix>")): Changed to ...
745 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
747 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
750 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
751 Cast vec_cmpeq result to correct type.
752 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
753 Cast vec_cmpgt result to correct type.
755 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
757 * final.c (final_scan_insn_1): Renamed from...
758 (final_scan_insn): ... this. New wrapper, to recover
759 seen from the outermost call in recursive ones.
760 * config/sparc/sparc.c (output_return): Drop seen from call.
761 (output_sibcall): Likewise.
762 * config/visium/visium.c (output_branch): Likewise.
764 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
766 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
769 2018-02-10 Alan Modra <amodra@gmail.com>
772 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
773 Specify LR as an input.
775 2018-02-10 Jakub Jelinek <jakub@redhat.com>
778 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
779 remove_member_access_dummy_vars): New functions.
780 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
781 lower_omp_1, execute_lower_omp): Use them.
783 PR rtl-optimization/84308
784 * shrink-wrap.c (spread_components): Release todo vector.
786 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
788 PR rtl-optimization/57193
789 * ira-color.c (struct allocno_color_data): Add member
790 conflict_allocno_hard_prefs.
791 (update_conflict_allocno_hard_prefs): New.
792 (bucket_allocno_compare_func): Add a preference based on
793 conflict_allocno_hard_prefs.
794 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
795 (color_allocnos): Remove a dead code. Initiate
796 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
798 2018-02-09 Jakub Jelinek <jakub@redhat.com>
801 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
802 constraint from =wa to wa. Avoid a subreg on the output operand,
803 instead use a pseudo and subreg it in a move.
804 (p9_xxbrd_<mode>): Changed to ...
805 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
806 (p9_xxbrd_v2df): New expander.
807 (p9_xxbrw_<mode>): Changed to ...
808 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
809 (p9_xxbrw_v4sf): New expander.
811 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
813 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
815 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
818 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
819 multiply in 32-bit mode.
820 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
821 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
824 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
826 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
827 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
828 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
829 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
831 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
834 * dwarf2out.c (is_trivial_indirect_ref): New function.
835 (dwarf2out_late_global_decl): Do not generate a location
836 attribute for variables that have a non-trivial DECL_VALUE_EXPR
837 and that are not defined in the current unit.
839 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
841 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
842 instead of a libcall for UNORDERED.
844 2018-02-09 Tamar Christina <tamar.christina@arm.com>
847 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
848 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
850 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
853 * config/s390/s390.c (s390_set_current_function): Invoke
854 s390_indirect_branch_settings also if fndecl didn't change.
856 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
858 * config/rs6000/rs6000.md (blockage): Set length to zero.
860 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
862 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
864 2018-02-09 Jakub Jelinek <jakub@redhat.com>
867 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
868 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
872 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
873 PARALLEL incoming that failed vt_get_decl_and_offset check.
876 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
877 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
879 (get_variable_section): For decls in named .bss* sections pass true as
880 second argument to bss_initializer_p.
882 2018-02-09 Marek Polacek <polacek@redhat.com>
883 Jakub Jelinek <jakub@redhat.com>
886 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
887 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
888 Sync some changes from cxx_fold_indirect_ref.
890 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
892 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
894 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
895 (BLOCK_INLINE_ENTRY_LABEL): New.
896 (dwarf2out_var_location): Disregard inline entry markers.
897 (inline_entry_data): New struct.
898 (inline_entry_data_hasher): New hashtable type.
899 (inline_entry_data_hasher::hash): New.
900 (inline_entry_data_hasher::equal): New.
901 (inline_entry_data_table): New variable.
902 (add_high_low_attributes): Add DW_AT_entry_pc and
903 DW_AT_GNU_entry_view attributes if a pending entry is found
904 in inline_entry_data_table. Add old entry_pc attribute only
905 if debug nonbinding markers are disabled.
906 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
908 (block_within_block_p, dwarf2out_inline_entry): New.
909 (dwarf2out_finish): Check that no entries remained in
910 inline_entry_data_table.
911 * final.c (reemit_insn_block_notes): Handle inline entry notes.
912 (final_scan_insn, notice_source_line): Likewise.
913 (rest_of_clean_state): Skip inline entry markers.
914 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
916 * gimple.c (gimple_build_debug_inline_entry): New.
917 * gimple.h (enum gimple_debug_subcode): Add
918 GIMPLE_DEBUG_INLINE_ENTRY.
919 (gimple_build_debug_inline_entry): Declare.
920 (gimple_debug_inline_entry_p): New.
921 (gimple_debug_nonbind_marker_p): Adjust.
922 * insn-notes.def (INLINE_ENTRY): New.
923 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
924 inline entry marker notes.
925 (print_insn): Likewise.
926 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
927 (INSN_DEBUG_MARKER_KIND): Likewise.
928 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
929 * tree-inline.c (expand_call_inline): Build and insert
930 debug_inline_entry stmt.
931 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
932 inline entry blocks early, if nonbind markers are enabled.
933 (dump_scope_block): Dump fragment info.
934 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
935 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
936 (gimple_build_debug_inline_entry): New.
937 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
938 Enable/disable inline entry points too.
939 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
940 (DEBUG_INSN): Describe inline entry markers.
942 * common.opt (gvariable-location-views): New.
943 (gvariable-location-views=incompat5): New.
944 * config.in: Rebuilt.
945 * configure: Rebuilt.
946 * configure.ac: Test assembler for view support.
947 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
948 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
949 * dwarf2out.c (var_loc_view): New typedef.
950 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
951 (dwarf2out_locviews_in_attribute): New.
952 (dwarf2out_locviews_in_loclist): New.
953 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
954 (enum dw_line_info_opcode): Add LI_adv_address.
955 (struct dw_line_info_table): Add view.
956 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
957 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
958 (zero_view_p): New variable.
959 (ZERO_VIEW_P): New macro.
960 (output_asm_line_debug_info): New.
961 (struct var_loc_node): Add view.
962 (add_AT_view_list, AT_loc_list): New.
963 (add_var_loc_to_decl): Add view param. Test it against last.
964 (new_loc_list): Add view params. Record them.
965 (AT_loc_list_ptr): Handle loc and view lists.
966 (view_list_to_loc_list_val_node): New.
967 (print_dw_val): Handle dw_val_class_view_list.
968 (size_of_die): Likewise.
969 (value_format): Likewise.
970 (loc_list_has_views): New.
971 (gen_llsym): Set vl_symbol too.
972 (maybe_gen_llsym, skip_loc_list_entry): New.
973 (dwarf2out_maybe_output_loclist_view_pair): New.
974 (output_loc_list): Output view list or entries too.
975 (output_view_list_offset): New.
976 (output_die): Handle dw_val_class_view_list.
977 (output_dwarf_version): New.
978 (output_compilation_unit_header): Use it.
979 (output_skeleton_debug_sections): Likewise.
980 (output_rnglists, output_line_info): Likewise.
981 (output_pubnames, output_aranges): Update version comments.
982 (output_one_line_info_table): Output view numbers in asm comments.
983 (dw_loc_list): Determine current endview, pass it to new_loc_list.
984 Call maybe_gen_llsym.
985 (loc_list_from_tree_1): Adjust.
986 (add_AT_location_description): Create view list attribute if
987 needed, check it's absent otherwise.
988 (convert_cfa_to_fb_loc_list): Adjust.
989 (maybe_emit_file): Call output_asm_line_debug_info for test.
990 (dwarf2out_var_location): Reset views as needed. Precompute
991 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
993 (new_line_info_table): Reset next view.
994 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
995 (dwarf2out_source_line): Likewise. Output view resets and labels to
996 the assembler, or select appropriate line info opcodes.
997 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
998 (optimize_string_length): Catch it. Adjust.
999 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
1000 dw_val_class_view_list, and remove it if no longer needed.
1001 (hash_loc_list): Hash view numbers.
1002 (loc_list_hasher::equal): Compare them.
1003 (optimize_location_lists): Check whether a view list symbol is
1004 needed, and whether the locview attribute is present, and
1005 whether they match. Remove the locview attribute if no longer
1007 (index_location_lists): Call skip_loc_list_entry for test.
1008 (dwarf2out_finish): Call output_asm_line_debug_info for test.
1009 Use output_dwarf_version.
1010 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
1011 (struct dw_val_node): Add val_view_list.
1012 * final.c (SEEN_NEXT_VIEW): New.
1013 (set_next_view_needed): New.
1014 (clear_next_view_needed): New.
1015 (maybe_output_next_view): New.
1016 (final_start_function): Rename to...
1017 (final_start_function_1): ... this. Take pointer to FIRST,
1018 add SEEN parameter. Emit param bindings in the initial view.
1019 (final_start_function): Reintroduce SEEN-less interface.
1020 (final): Rename to...
1021 (final_1): ... this. Take SEEN parameter. Output final pending
1022 next view at the end.
1023 (final): Reintroduce seen-less interface.
1024 (final_scan_insn): Output pending next view before switching
1025 sections or ending a block. Mark the next view as needed when
1026 outputting variable locations. Notify debug backend of section
1027 changes, and of location view changes.
1028 (rest_of_handle_final): Adjust.
1029 * toplev.c (process_options): Autodetect value for debug variable
1030 location views option. Warn on incompat5 without -gdwarf-5.
1031 * doc/invoke.texi (gvariable-location-views): New.
1032 (gvariable-location-views=incompat5): New.
1033 (gno-variable-location-views): New.
1035 2018-02-08 David Malcolm <dmalcolm@redhat.com>
1037 PR tree-optimization/84136
1038 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
1039 that the result of find_edge is non-NULL.
1041 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1044 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
1045 storing integer register in SImode. Fix cost of 256 and 512
1046 byte aligned SSE register store.
1048 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1050 * config/i386/i386.c (ix86_multiplication_cost): Fix
1051 multiplication cost for TARGET_AVX512DQ.
1053 2018-02-08 Marek Polacek <polacek@redhat.com>
1055 PR tree-optimization/84238
1056 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
1059 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1061 PR tree-optimization/84265
1062 * tree-vect-stmts.c (vectorizable_store): Don't treat
1063 VMAT_CONTIGUOUS accesses as grouped.
1064 (vectorizable_load): Likewise.
1066 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1068 PR tree-optimization/81635
1069 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
1070 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
1071 (test_round_for_mask): New functions.
1072 (wide_int_cc_tests): Call test_round_for_mask.
1073 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
1074 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
1075 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
1076 range returned by get_range_info.
1078 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1081 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
1082 * symtab.c: Include builtins.h
1083 (symtab_node::output_to_lto_symbol_table_p): Move here
1084 from lto-streamer-out.c:output_symbol_p.
1085 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
1086 (output_symbol_p): Move all logic to symtab.c
1087 (produce_symtab): Update.
1089 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1091 * config/s390/s390-opts.h (enum indirect_branch): Define.
1092 * config/s390/s390-protos.h (s390_return_addr_from_memory)
1093 (s390_indirect_branch_via_thunk)
1094 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
1095 (enum s390_indirect_branch_type): Define.
1096 * config/s390/s390.c (struct s390_frame_layout, struct
1097 machine_function): Remove.
1098 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
1099 (indirect_branch_table_label_no, indirect_branch_table_name):
1101 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
1102 (enum s390_indirect_branch_option): Define.
1103 (s390_return_addr_from_memory): New function.
1104 (s390_handle_string_attribute): New function.
1105 (s390_attribute_table): Add new attribute handler.
1106 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
1107 (s390_indirect_branch_via_thunk): New function.
1108 (s390_indirect_branch_via_inline_thunk): New function.
1109 (s390_function_ok_for_sibcall): When jumping via thunk disallow
1110 sibling call optimization for non z10 compiles.
1111 (s390_emit_call): Force indirect branch target to be a single
1112 register. Add r1 clobber for non-z10 compiles.
1113 (s390_emit_epilogue): Emit return jump via return_use expander.
1114 (s390_reorg): Handle JUMP_INSNs as execute targets.
1115 (s390_option_override_internal): Perform validity checks for the
1116 new command line options.
1117 (s390_indirect_branch_attrvalue): New function.
1118 (s390_indirect_branch_settings): New function.
1119 (s390_set_current_function): Invoke s390_indirect_branch_settings.
1120 (s390_output_indirect_thunk_function): New function.
1121 (s390_code_end): Implement target hook.
1122 (s390_case_values_threshold): Implement target hook.
1123 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
1125 * config/s390/s390.h (struct s390_frame_layout)
1126 (struct machine_function): Move here from s390.c.
1127 (TARGET_INDIRECT_BRANCH_NOBP_RET)
1128 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
1129 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
1130 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
1131 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
1132 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
1133 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
1134 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
1135 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
1136 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
1137 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
1138 (mnemonic attribute): Add values which aren't recognized
1140 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
1141 pattern for branch conversion. Fix mnemonic attribute.
1142 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
1143 indirect branch via thunk if requested.
1144 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
1145 ("*indirect_jump"): Disable for branch conversion using out of
1147 ("indirect_jump_via_thunk<mode>_z10")
1148 ("indirect_jump_via_thunk<mode>")
1149 ("indirect_jump_via_inlinethunk<mode>_z10")
1150 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
1151 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
1152 ("casesi_jump_via_inlinethunk<mode>_z10")
1153 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
1154 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
1155 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
1156 ("*indirect2_jump"): Disable for branch conversion.
1157 ("casesi_jump"): Turn into expander and expand patterns for branch
1159 ("return_use"): New expander.
1160 ("*return"): Emit return via thunk and rename it to ...
1161 ("*return<mode>"): ... this one.
1162 * config/s390/s390.opt: Add new options and and enum for the
1165 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1167 * lra-constraints.c (match_reload): Unconditionally use
1168 gen_lowpart_SUBREG, rather than selecting between that
1169 and equivalent gen_rtx_SUBREG code.
1171 2018-02-08 Richard Biener <rguenther@suse.de>
1173 PR tree-optimization/84233
1174 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
1175 changed flag instead of boguously re-using phi_inserted.
1177 2018-02-08 Martin Jambor <mjambor@suse.cz>
1179 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
1180 static local variables.
1182 2018-02-08 Richard Biener <rguenther@suse.de>
1184 PR tree-optimization/84278
1185 * tree-vect-stmts.c (vectorizable_store): When looking for
1186 smaller vector types to perform grouped strided loads/stores
1187 make sure the mode is supported by the target.
1188 (vectorizable_load): Likewise.
1190 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1192 * config/aarch64/aarch64.c (aarch64_components_for_bb):
1193 Increase LDP/STP opportunities by adding adjacent callee-saves.
1195 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1197 PR rtl-optimization/84068
1198 PR rtl-optimization/83459
1199 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
1201 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
1203 PR tree-optimization/84224
1204 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
1205 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
1208 2018-02-07 Iain Sandoe <iain@codesourcery.com>
1211 * config/rs6000/altivec.md (*restore_world): Remove LR use.
1212 * config/rs6000/predicates.md (restore_world_operation): Adjust op
1213 count, remove one USE.
1215 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
1217 * doc/install.texi (Configuration): Document the
1218 --with-long-double-format={ibm,ieee} PowerPC configuration
1222 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
1223 Convert from define_expand to be define_insn_and_split. Rework
1224 float/double/_Float128 conversions to QI/HI/SImode to work with
1225 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
1226 conversions to QI/HImode types did a store and then a load to
1227 truncate the value. For conversions to VSX registers, don't split
1228 the insn, instead emit the code directly. Use the code iterator
1229 any_fix to combine signed and unsigned conversions.
1230 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
1231 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
1232 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1233 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
1234 (fix_<mode>di2_hw): Likewise.
1235 (fixuns_<mode>di2_hw): Likewise.
1236 (fix_<mode>si2_hw): Likewise.
1237 (fixuns_<mode>si2_hw): Likewise.
1238 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
1239 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1240 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1241 fix<uns>_trunc<SFDF:mode>si2_p8.
1242 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1244 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1245 (fix<uns>_<mode>_mem): Likewise.
1246 (fctiw<u>z_<mode>_mem): Likewise.
1247 (fix<uns>_<mode>_mem): Likewise.
1248 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1249 the register allocator from doing a direct move to the GPRs to do
1250 a store, and instead use the ISA 3.0 store byte/half-word from
1251 vector register instruction. For IEEE 128-bit floating point,
1252 also optimize stores of 32-bit ints.
1253 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1255 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1257 * genextract.c (push_pathstr_operand): New function to support
1259 (walk_rtx): Call push_pathstr_operand.
1260 (print_path): Support [a-zA-Z].
1262 2018-02-07 Richard Biener <rguenther@suse.de>
1264 PR tree-optimization/84037
1265 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1266 (cse_and_gimplify_to_preheader): Declare.
1267 (vect_get_place_in_interleaving_chain): Likewise.
1268 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1270 (_loop_vec_info::~_loop_vec_info): Delete it.
1271 (cse_and_gimplify_to_preheader): New function.
1272 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1273 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1274 (vectorizable_load): Likewise. For grouped stores always base
1275 the IV on the first element.
1276 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1277 condition before gimplifying.
1279 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1281 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1282 *DIV_EXPR and *MOD_EXPR.
1284 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1287 * config/i386/i386.c (ix86_option_override_internal): Mask out
1288 the CF_SET bit when checking -fcf-protection.
1290 2018-02-07 Tom de Vries <tom@codesourcery.com>
1293 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1296 2018-02-07 Richard Biener <rguenther@suse.de>
1298 PR tree-optimization/84204
1299 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1302 PR tree-optimization/84205
1303 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1304 special-case isl_ast_op_zdiv_r.
1306 PR tree-optimization/84223
1307 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1308 Only add conditions from within the region.
1309 (gather_bbs::after_dom_children): Adjust.
1311 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1314 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1315 * config/avr/avr.md: Only post-reload split REG-REG moves if
1316 either register is REGERAL_REG_P.
1318 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1320 PR tree-optimization/84235
1321 * tree-ssa-scopedtables.c
1322 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1323 if the subtraction is performed in floating point type where NaNs are
1324 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1325 build 1. Formatting fix.
1327 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1330 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1331 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1332 and skip it regardless of bb boundaries. Use CALL_P macro,
1333 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1334 unnecessarily, formatting fix.
1336 2018-02-06 Michael Collison <michael.collison@arm.com>
1338 * config/arm/thumb2.md:
1339 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1340 (*thumb_mov_notscc): Ditto.
1342 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1345 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1346 unsigned_fix, not "s".
1348 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1350 * configure.ac (gcc_fn_eh_frame_ro): New function.
1351 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1352 correct .eh_frame permissions.
1353 * configure: Regenerate.
1355 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1357 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1360 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1362 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1363 Display warning message for -mno-speculate-indirect-jumps.
1365 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1367 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1369 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1371 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1373 PR tree-optimization/84225
1374 * tree-eh.c (find_trapping_overflow): Only call
1375 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1377 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1380 * config/i386/i386.c: Reimplement the check of possible options
1381 -mibt/-mshstk conbination. Change error messages.
1382 * doc/invoke.texi: Fix a typo: remove extra '='.
1384 2018-02-06 Marek Polacek <polacek@redhat.com>
1386 PR tree-optimization/84228
1387 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1389 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1392 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1393 emitted arch directives.
1394 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1395 __ARM_FEATURE_COPROC before changing architectures.
1397 2018-02-06 Richard Biener <rguenther@suse.de>
1399 * config/i386/i386.c (print_reg): Fix typo.
1400 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1402 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1404 * configure: Regenerate.
1406 2018-02-05 Martin Sebor <msebor@redhat.com>
1408 PR tree-optimization/83369
1409 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1412 2018-02-05 Martin Liska <mliska@suse.cz>
1414 * doc/invoke.texi: Cherry-pick upstream r323995.
1416 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1418 * ira.c (ira_init_register_move_cost): Adjust comment.
1420 2018-02-05 Martin Liska <mliska@suse.cz>
1422 PR gcov-profile/84137
1423 * doc/gcov.texi: Fix typo in documentation.
1425 2018-02-05 Martin Liska <mliska@suse.cz>
1427 PR gcov-profile/83879
1428 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1429 using dlopen functionality.
1431 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1433 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1434 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1435 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1436 _mm_maskz_range_round_ss): New intrinsics.
1437 (__builtin_ia32_rangesd128_round)
1438 (__builtin_ia32_rangess128_round): Remove.
1439 (__builtin_ia32_rangesd128_mask_round,
1440 __builtin_ia32_rangess128_mask_round): New builtins.
1441 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1442 __builtin_ia32_rangess128_round): Remove.
1443 (__builtin_ia32_rangesd128_mask_round,
1444 __builtin_ia32_rangess128_mask_round): New builtins.
1445 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1446 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1447 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1448 "<round_saeonly_constraint>")): Changed to ...
1449 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1450 "<round_saeonly_scalar_constraint>")): ... this.
1451 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1452 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1453 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1454 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1455 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1457 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1459 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1461 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1462 Remove all values except native, 8540 and 8548.
1464 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1466 * config/i386/i386.c (ix86_output_function_return): Pass
1467 INVALID_REGNUM, instead of -1, as invalid register number to
1468 indirect_thunk_name and output_indirect_thunk.
1470 2018-02-02 Julia Koval <julia.koval@intel.com>
1472 * config.gcc: Add -march=icelake.
1473 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1474 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1475 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1476 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1477 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1478 (processor_target_table): Add icelake.
1479 (ix86_option_override_internal): Handle new PTAs.
1480 (get_builtin_code_for_version): Handle icelake.
1481 (M_INTEL_COREI7_ICELAKE): New.
1482 (fold_builtin_cpu): Handle icelake.
1483 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1484 * doc/invoke.texi: Add -march=icelake.
1486 2018-02-02 Julia Koval <julia.koval@intel.com>
1488 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1489 to wide_int_bitmask.
1490 * wide-int-bitmask.h: New.
1492 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1495 * config/i386/i386.md: Replace Pmode with word_mode in
1496 builtin_setjmp_setup and builtin_longjmp to support x32.
1498 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1502 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1504 (rs6000_supported_cpu_names): New static variable.
1505 (linux_cpu_translation_table): Likewise.
1506 (elf_platform) <cpu>: Define new static variable and use it.
1507 Translate kernel AT_PLATFORM name to canonical name if needed.
1508 Error if platform name is unknown.
1510 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1513 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1515 2018-02-01 Jeff Law <law@redhat.com>
1518 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1519 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1520 the scratch if RELEASE_VIA_POP is false.
1521 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1522 If we have to save a temporary register, decrement SIZE appropriately.
1523 Pass new arguments to release_scratch_register_on_entry.
1524 (ix86_adjust_stack_and_probe): Likewise.
1525 (ix86_emit_probe_stack_range): Pass new arguments to
1526 release_scratch_register_on_entry.
1528 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1530 PR rtl-optimization/84157
1531 * combine.c (change_zero_ext): Use REG_P predicate in
1532 front of HARD_REGISTER_P predicate.
1534 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1536 * config/avr/avr.c (avr_option_override): Move disabling of
1537 -fdelete-null-pointer-checks to...
1538 * common/config/avr/avr-common.c (avr_option_optimization_table):
1541 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1543 PR tree-optimization/81635
1544 * tree-data-ref.c (split_constant_offset_1): For types that
1545 wrap on overflow, try to use range info to prove that wrapping
1548 2018-02-01 Renlin Li <renlin.li@arm.com>
1551 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1553 (aarch64_register_move_cost): Likewise.
1554 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1556 (REG_CLASS_NAMES): Likewise.
1557 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1558 TAILCALL_ADDR_REGS. Remove IP registers.
1559 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1561 2018-02-01 Richard Biener <rguenther@suse.de>
1563 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1564 for specifying RPO order and allow NULL for that.
1565 * domwalk.c (dom_walker::dom_walker): Likewise.
1566 (dom_walker::walk): Handle NULL RPO order.
1567 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1569 (rewrite_update_dom_walker): Likewise.
1570 (mark_def_dom_walker): Likewise.
1572 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1574 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1575 (aarch64_maybe_expand_sve_subreg_move): Declare.
1576 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1577 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1579 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1580 that are semantically a reverse operation.
1581 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1582 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1583 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1585 (aarch64_can_change_mode_class): For big-endian, forbid changes
1586 between two SVE modes if they have different element sizes.
1588 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1590 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1591 the TImode handling for big-endian targets.
1593 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1595 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1596 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1598 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1599 Remove BSWAP handing for big-endian targets and use the form of
1600 LD1RQ appropariate for the mode.
1602 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1604 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1605 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1608 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1611 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1612 check for operands that need to go through aarch64_sve_reload_be.
1614 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1616 PR tree-optimization/81661
1617 PR tree-optimization/84117
1618 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1619 * tree-eh.c: Include gimplify.h.
1620 (find_trapping_overflow, replace_trapping_overflow,
1621 rewrite_to_non_trapping_overflow): New functions.
1622 * tree-vect-loop.c: Include tree-eh.h.
1623 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1624 * tree-data-ref.c: Include tree-eh.h.
1625 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1627 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1629 PR rtl-optimization/84123
1630 * combine.c (change_zero_ext): Check if hard register satisfies
1631 can_change_dest_mode before calling gen_lowpart_SUBREG.
1633 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1636 * ira.c (ira_init_register_move_cost): Remove assert.
1638 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1640 PR rtl-optimization/84071
1641 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1642 * doc/tm.texi: Regenerate.
1644 2018-01-31 Richard Biener <rguenther@suse.de>
1646 PR tree-optimization/84132
1647 * tree-data-ref.c (analyze_miv_subscript): Properly
1648 check whether evolution_function_is_affine_multivariate_p
1649 before calling gcd_of_steps_may_divide_p.
1651 2018-01-31 Julia Koval <julia.koval@intel.com>
1654 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1655 * config/i386/i386.md (rdpid_rex64) New.
1656 (rdpid): Make 32bit only.
1658 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1661 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1662 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1664 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1667 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1669 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1671 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1673 PR rtl-optimization/84071
1674 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1675 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1677 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1679 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1680 (arc_attribute_table): Add 'aux' attribute.
1681 (arc_in_small_data_p): Consider aux like variables.
1682 (arc_is_aux_reg_p): New function.
1683 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1684 (arc_get_aux_arg): New function.
1685 (prepare_move_operands): Handle aux-register access.
1686 (arc_handle_aux_attribute): New function.
1687 * doc/extend.texi (ARC Variable attributes): Add subsection.
1689 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1691 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1692 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1693 (arc_attribute_table): Add 'uncached' attribute.
1694 (arc_print_operand): Print '.di' flag for uncached memory
1696 (arc_in_small_data_p): Do not consider for small data the uncached
1698 (arc_is_uncached_mem_p): New function.
1699 * config/arc/predicates.md (compact_store_memory_operand): Check
1700 for uncached memory accesses.
1701 (nonvol_nonimm_operand): Likewise.
1702 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1704 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1707 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1708 falign-loops=): Add Optimization flag.
1710 2018-01-30 Jeff Law <law@redhat.com>
1713 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1714 INT_REGISTERS_SAVED. Check it prior to calling
1715 get_scratch_register_on_entry.
1716 (ix86_adjust_stack_and_probe): Similarly.
1717 (ix86_emit_probe_stack_range): Similarly.
1718 (ix86_expand_prologue): Corresponding changes.
1720 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1723 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1724 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1726 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1729 * lra-constraints.c (curr_insn_transform): Process AND in the
1732 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1734 PR rtl-optimization/83986
1735 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1736 dependence against last_pending_memory_flush in addition to
1739 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1741 PR tree-optimization/81611
1742 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1745 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1748 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1751 2018-01-30 Richard Biener <rguenther@suse.de>
1752 Jakub Jelinek <jakub@redhat.com>
1754 PR tree-optimization/84111
1755 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1756 inner loops added during recursion, as they don't have up-to-date
1759 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1762 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1763 (can_inline_edge_by_limits_p): ... here.
1764 (can_early_inline_edge_p, check_callers,
1765 update_caller_keys, update_callee_keys, recursive_inlining,
1766 add_new_edges_to_heap, speculation_useful_p,
1767 inline_small_functions,
1768 inline_small_functions, flatten_function,
1769 inline_to_all_callers_1): Update.
1771 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1773 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1776 2018-01-30 Richard Biener <rguenther@suse.de>
1778 PR tree-optimization/83008
1779 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1780 invariant and constant vector uses in stmts when they need
1783 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1786 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1787 * configure: Regenerate.
1789 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1791 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1793 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1794 Use gen_rtx_REG rather than gen_lowpart.
1796 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1798 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1799 rather than 0 when creating partial subregs.
1801 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1803 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1806 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1809 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1810 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1811 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1812 flags. This restores the settings used before the 2017-07-24.
1813 Turning off pre increment/decrement/modify allows IVOPTS to
1814 optimize DF/SF loops where the index is an int.
1816 2018-01-29 Richard Biener <rguenther@suse.de>
1817 Kelvin Nilsen <kelvin@gcc.gnu.org>
1820 * tree-vect-stmts.c (vectorizable_call): Don't call
1821 targetm.vectorize_builtin_md_vectorized_function if callee is
1824 2018-01-22 Carl Love <cel@us.ibm.com>
1826 * doc/extend.tex: Fix typo in second arg in
1827 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1829 2018-01-29 Richard Biener <rguenther@suse.de>
1831 PR tree-optimization/84086
1832 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1833 (flush_ssaname_freelist): When SSA names were released reset
1834 the SCEV hash table.
1836 2018-01-29 Richard Biener <rguenther@suse.de>
1838 PR tree-optimization/84057
1839 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1840 removed paths when removing edges.
1842 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1844 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1845 -mfunction-return=@var{choice}.
1847 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1850 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1852 (layout::print_source_line): Likewise.
1853 (test_get_line_width_without_trailing_whitespace): Add test cases.
1855 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1858 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1861 2018-01-26 Jim Wilson <jimw@sifive.com>
1863 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1865 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1868 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1870 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1871 and CMP + SUB-immediate -> SUBS.
1873 2018-01-26 Martin Sebor <msebor@redhat.com>
1875 PR tree-optimization/83896
1876 * tree-ssa-strlen.c (get_string_len): Rename...
1877 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1878 Avoid assuming length is constant.
1879 (handle_char_store): Use HOST_WIDE_INT for string length.
1881 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1884 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1885 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1887 2018-01-26 Richard Biener <rguenther@suse.de>
1889 PR rtl-optimization/84003
1890 * dse.c (record_store): Only record redundant stores when
1891 the earlier store aliases at least all accesses the later one does.
1893 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1895 PR rtl-optimization/83985
1896 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1897 REG_CFA_RESTORE insns.
1898 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1899 REG_CFA_RESTORE insns here.
1902 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1903 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1905 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1907 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1908 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1909 (arc_init): Likewise.
1910 (arc_override_options): Likewise.
1911 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1913 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1915 * config/arc/arc.h (TARGET_DBNZ): Define.
1916 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1917 properly set the tune attribute.
1918 (dbnz): Use TARGET_DBNZ guard.
1919 * config/arc/arc.opt (mtune): Add core3 option.
1921 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1923 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1924 recognize new pic like addresses.
1925 (arc_delegitimize_address): Clean up.
1927 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1929 * config/arc/arc-arches.def: Option mrf16 valid for all
1931 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1932 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1933 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1934 * config/arc/arc-tables.opt: Regenerate.
1935 * config/arc/arc.c (arc_conditional_register_usage): Handle
1936 reduced register file case.
1937 (arc_file_start): Set must have build attributes.
1938 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1940 * config/arc/arc.opt (mrf16): Add new option.
1941 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1942 * config/arc/genmultilib.awk: Handle new mrf16 option.
1943 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1944 * config/arc/t-multilib: Regenerate.
1945 * doc/invoke.texi (ARC Options): Document mrf16 option.
1947 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1949 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1950 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1951 (arc_attribute_table): Add 'secure_call' attribute.
1952 (arc_print_operand): Print secure call operand.
1953 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1955 (arc_is_secure_call_p): New function. * config/arc/arc.md
1956 (call_i): Add support for sjli instruction.
1957 (call_value_i): Likewise.
1958 * config/arc/constraints.md (Csc): New constraint.
1960 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1961 John Eric Martin <John.Martin@emmicro-us.com>
1963 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1964 * config/arc/arc.c (_arc_jli_section): New struct.
1965 (arc_jli_section): New type.
1966 (rc_jli_sections): New static variable.
1967 (arc_handle_jli_attribute): New function.
1968 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1969 (arc_file_end): New function.
1970 (TARGET_ASM_FILE_END): Define.
1971 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1972 (arc_add_jli_section): New function.
1973 (jli_call_scan): Likewise.
1974 (arc_reorg): Call jli_call_scan.
1975 (arc_output_addsi): Remove 'S' from printing asm operand.
1976 (arc_is_jli_call_p): New function.
1977 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1979 (movhi_insn): Likewise.
1980 (movsi_insn): Likewise.
1981 (movsi_set_cc_insn): Likewise.
1982 (loadqi_update): Likewise.
1983 (load_zeroextendqisi_update): Likewise.
1984 (load_signextendqisi_update): Likewise.
1985 (loadhi_update): Likewise.
1986 (load_zeroextendhisi_update): Likewise.
1987 (load_signextendhisi_update): Likewise.
1988 (loadsi_update): Likewise.
1989 (loadsf_update): Likewise.
1990 (movsicc_insn): Likewise.
1991 (bset_insn): Likewise.
1992 (bxor_insn): Likewise.
1993 (bclr_insn): Likewise.
1994 (bmsk_insn): Likewise.
1995 (bicsi3_insn): Likewise.
1996 (cmpsi_cc_c_insn): Likewise.
1997 (movsi_ne): Likewise.
1998 (movsi_cond_exec): Likewise.
1999 (clrsbsi2): Likewise.
2010 (call_i): Remove 'S' asm letter, add jli instruction.
2011 (call_value_i): Likewise.
2012 * config/arc/arc.op (mjli-always): New option.
2013 * config/arc/constraints.md (Cji): New constraint.
2014 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
2016 (subsf3_fpx): Likewise.
2017 (mulsf3_fpx): Likewise.
2018 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
2020 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
2022 * doc/invoke.texi (ARC): Document mjli-always option.
2024 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
2026 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
2027 avoid addition with 0 and use incw and decw where possible.
2029 2018-01-26 Richard Biener <rguenther@suse.de>
2031 PR tree-optimization/81082
2032 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
2033 association if it requires casting to unsigned.
2034 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
2035 from fold_plusminus_mult_expr to catch important cases late when
2036 range info is available.
2038 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2040 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
2041 * configure.ac (hidden_linkonce): New test.
2042 * configure: Regenerate.
2043 * config.in: Regenerate.
2045 2018-01-26 Julia Koval <julia.koval@intel.com>
2047 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
2048 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
2049 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
2050 _mm_mask_bitshuffle_epi64_mask): Fix type.
2051 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
2052 USI_FTYPE_V4DI_V4DI_USI): Remove.
2053 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
2054 __builtin_ia32_vpshufbitqmb256_mask,
2055 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
2056 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
2057 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
2059 2018-01-26 Alan Modra <amodra@gmail.com>
2062 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
2063 UNSPEC_VBPERMQ. Sort other unspecs.
2065 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
2067 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
2069 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
2072 * predict.c (drop_profile): Do not push/pop cfun; update also
2074 (handle_missing_profiles): Fix logic looking for zero profiles.
2076 2018-01-25 Jakub Jelinek <jakub@redhat.com>
2079 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
2080 on functions with #pragma omp declare simd or functions with simd
2082 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
2083 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
2084 Remove trailing \n from warning_at calls.
2086 2018-01-25 Tom de Vries <tom@codesourcery.com>
2089 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2090 for neutered workers.
2092 2018-01-24 Joseph Myers <joseph@codesourcery.com>
2095 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
2096 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
2098 2018-01-24 Jeff Law <law@redhat.com>
2101 * i386.c (get_probe_interval): Move to earlier point.
2102 (ix86_compute_frame_layout): If -fstack-clash-protection and
2103 the frame is larger than the probe interval, then use pushes
2104 to save registers rather than reg->mem moves.
2105 (ix86_expand_prologue): Remove conditional for int_registers_saved
2108 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
2111 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
2112 min/max for never referenced object.
2114 2018-01-24 Jakub Jelinek <jakub@redhat.com>
2117 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
2119 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
2120 attributes from DECL_ATTRIBUTES (decl) without affecting
2121 DECL_ATTRIBUTES (current_function_decl).
2122 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
2123 functions with non-NULL DECL_ABSTRACT_ORIGIN.
2125 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
2127 PR tree-optimization/83979
2128 * fold-const.c (fold_comparison): Use constant_boolean_node
2129 instead of boolean_{true,false}_node.
2131 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
2133 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
2136 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2138 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2139 Simplify the clause that sets the length attribute.
2140 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2141 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
2142 clause that sets the length attribute.
2143 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2145 2018-01-24 Tom de Vries <tom@codesourcery.com>
2148 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
2149 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
2150 Add strict parameter.
2151 (prevent_branch_around_nothing): Insert dummy insn between branch to
2152 label and label with no ptx insn inbetween.
2153 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
2155 2018-01-24 Tom de Vries <tom@codesourcery.com>
2158 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2159 for neutered threads in warp.
2160 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
2162 2018-01-24 Richard Biener <rguenther@suse.de>
2164 PR tree-optimization/83176
2165 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
2168 2018-01-24 Richard Biener <rguenther@suse.de>
2170 PR tree-optimization/82819
2171 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
2172 code generating pluses that are no-ops in the target precision.
2174 2018-01-24 Richard Biener <rguenther@suse.de>
2177 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
2179 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2181 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
2182 to merge probabilities.
2183 * predict.c (probably_never_executed): Also mark as cold functions
2184 with global 0 profile and guessed local profile.
2185 * profile-count.c (profile_probability::combine_with_count): New
2187 * profile-count.h (profile_probability::operator*,
2188 profile_probability::operator*=, profile_probability::operator/,
2189 profile_probability::operator/=): Reduce precision to adjusted
2190 and set value to guessed on contradictory divisions.
2191 (profile_probability::combine_with_freq): Remove.
2192 (profile_probability::combine_wiht_count): Declare.
2193 (profile_count::force_nonzero):: Set to adjusted.
2194 (profile_count::probability_in):: Set quality to adjusted.
2195 * tree-ssa-tail-merge.c (replace_block_by): Use
2198 2018-01-23 Andrew Waterman <andrew@sifive.com>
2199 Jim Wilson <jimw@sifive.com>
2201 * config/riscv/riscv.c (riscv_stack_boundary): New.
2202 (riscv_option_override): Set riscv_stack_boundary. Handle
2203 riscv_preferred_stack_boundary_arg.
2204 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
2205 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
2206 (STACK_BOUNDARY): Set to riscv_stack_boundary.
2207 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
2208 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
2209 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
2211 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
2214 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
2215 of struct ix86_frame.
2216 (ix86_expand_epilogue): Likewise. Add a local variable for
2217 the reg_save_offset field in struct ix86_frame.
2219 2018-01-23 Bin Cheng <bin.cheng@arm.com>
2221 PR tree-optimization/82604
2222 * tree-loop-distribution.c (enum partition_kind): New enum item
2223 PKIND_PARTIAL_MEMSET.
2224 (partition_builtin_p): Support above new enum item.
2225 (generate_code_for_partition): Ditto.
2226 (compute_access_range): Differentiate cases that equality can be
2227 proven at all loops, the innermost loops or no loops.
2228 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
2229 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
2230 (finalize_partitions, distribute_loop): Don't fuse partition of
2231 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
2232 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
2235 2018-01-23 Martin Liska <mliska@suse.cz>
2237 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
2238 order to ignore the predictor.
2239 (PRED_POLYMORPHIC_CALL): Likewise.
2240 (PRED_RECURSIVE_CALL): Likewise.
2242 2018-01-23 Martin Liska <mliska@suse.cz>
2244 * tree-profile.c (tree_profiling): Print function header to
2245 aware reader which function we are working on.
2246 * value-prof.c (gimple_find_values_to_profile): Do not print
2247 not interesting value histograms.
2249 2018-01-23 Martin Liska <mliska@suse.cz>
2251 * profile-count.h (enum profile_quality): Add
2252 profile_uninitialized as the first value. Do not number values
2253 as they are zero based.
2254 (profile_count::verify): Update sanity check.
2255 (profile_probability::verify): Likewise.
2257 2018-01-23 Nathan Sidwell <nathan@acm.org>
2259 * doc/invoke.texi (ffor-scope): Deprecate.
2261 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2263 PR tree-optimization/83510
2264 * domwalk.c (set_all_edges_as_executable): New function.
2265 (dom_walker::dom_walker): Convert bool param
2266 "skip_unreachable_blocks" to enum reachability. Move setup of
2267 edge flags to set_all_edges_as_executable and only do it when
2268 reachability is REACHABLE_BLOCKS.
2269 * domwalk.h (enum dom_walker::reachability): New enum.
2270 (dom_walker::dom_walker): Convert bool param
2271 "skip_unreachable_blocks" to enum reachability.
2272 (set_all_edges_as_executable): New decl.
2273 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2274 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2276 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2277 but converting true to REACHABLE_BLOCKS.
2278 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2280 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2281 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2282 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2284 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2285 if check_all_array_refs will be called.
2287 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2289 * tree.c (selftest::test_location_wrappers): Add more test
2292 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2294 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2295 (selftest::test_bit_in_range): Likewise.
2297 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2300 * doc/sourcebuild.texi (vect_float): Say that the selector
2301 only describes the situation when -funsafe-math-optimizations is on.
2302 (vect_float_strict): Document.
2304 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2306 PR tree-optimization/83965
2307 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2308 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2309 instead of checking only for a reduction.
2310 (vect_recog_widen_sum_pattern): Likewise.
2312 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2314 * predict.c (probably_never_executed): Only use precise profile info.
2315 (compute_function_frequency): Skip after inlining hack since we now
2316 have quality checking.
2318 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2320 * profile-count.h (profile_probability::very_unlikely,
2321 profile_probability::unlikely, profile_probability::even): Set
2322 precision to guessed.
2324 2018-01-23 Richard Biener <rguenther@suse.de>
2326 PR tree-optimization/83963
2327 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2328 Properly terminate dominator walk when crossing the exit edge not
2329 when visiting its source block.
2331 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2334 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2335 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2337 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2339 PR tree-optimization/83957
2340 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2341 semicolon after for body surrounded by braces.
2343 PR tree-optimization/83081
2344 * profile-count.h (profile_probability::split): New method.
2345 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2346 Use profile_probability::split.
2347 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2348 when splitting a single conditional jump into 2.
2350 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2352 PR tree-optimization/69452
2353 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2356 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2358 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2359 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2360 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2362 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2364 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2365 * config/rl78/rl78.md (movdi): New define_expand.
2366 * config/rl78/rl78.c (rl78_split_movdi): New function.
2368 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2371 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2373 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2374 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2375 128-bit to produce an UNSPEC move to get the double word with the
2376 signbit and then a shift directly to do signbit.
2377 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2378 implementation with a new version that just does either a direct
2379 move or a regular move. Move memory interface to separate insns.
2380 Move insns so they are next to the expander.
2381 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2382 with signbit move. Split big and little endian case.
2383 (signbit<mode>2_dm_mem_le): Likewise.
2384 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2385 (signbit<mode>2_dm2): Likewise.
2387 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2389 * config/rl78/rl78.md (anddi3): New define_expand.
2391 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2393 * config/rl78/rl78.md (umindi3): New define_expand.
2395 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2397 * config/rl78/rl78.md (smindi3): New define_expand.
2399 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2401 * config/rl78/rl78.md (smaxdi3): New define_expand.
2403 2018-01-22 Carl Love <cel@us.ibm.com>
2405 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2406 LVX_V1TI): Add macro expansion.
2407 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2408 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2409 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2410 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2411 Change check to determine if the instruction is a byte reversing
2412 entry. Fix typo in comment.
2413 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2414 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2415 Add def_builtin calls for new builtins.
2416 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2417 Add define_insn expansion.
2419 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2421 * config/rl78/rl78.md (umaxdi3): New define_expand.
2423 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2425 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2426 for non-QImode registers.
2428 2018-01-22 Richard Biener <rguenther@suse.de>
2430 PR tree-optimization/83963
2431 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2432 including the loop exit block.
2433 (scop_detection::merge_sese): Likewise.
2434 (scop_detection::add_scop): Do it here instead.
2436 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2438 * doc/sourcebuild.texi (arm_softfloat): Document.
2440 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2443 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2444 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2445 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2447 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2448 David Edelsohn <dje.gcc@gmail.com>
2451 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2452 Change "crset eq" to "crset 2".
2453 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2454 (*call_indirect_aix<mode>_nospec): Likewise.
2455 (*call_value_indirect_aix<mode>_nospec): Likewise.
2456 (*call_indirect_elfv2<mode>_nospec): Likewise.
2457 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2458 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2459 change assembly output from . to $.
2460 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2461 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2462 (*tablejump<mode>_internal1_nospec): Likewise.
2464 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2467 * config/sh/sh_optimize_sett_clrt.cc:
2468 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2470 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2472 PR tree-optimization/83940
2473 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2474 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2475 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2476 use it to pass back the definition type.
2477 (vect_check_store_rhs): Likewise rhs_dt_out.
2478 (vect_build_gather_load_calls): Add a mask_dt argument and use
2479 it instead of a call to vect_is_simple_use.
2480 (vectorizable_store): Update calls to vect_check_load_store_mask
2481 and vect_check_store_rhs. Use the dt returned by the latter instead
2482 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2483 instead of calls to vect_is_simple_use. Pass the scalar rather
2484 than the vector operand to vect_is_simple_use when handling
2485 second and subsequent copies of an rhs value.
2486 (vectorizable_load): Update calls to vect_check_load_store_mask
2487 and vect_build_gather_load_calls. Use the cached mask_dt and
2488 gs_info.offset_dt instead of calls to vect_is_simple_use.
2490 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2493 * tree-emutls.c: Include gimplify.h.
2494 (lower_emutls_2): New function.
2495 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2496 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2497 it before further processing.
2500 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2501 UINTVAL (trueop1) instead of INTVAL (op1).
2503 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2507 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2508 INCOMING_FRAME_SP_OFFSET if not defined.
2509 (scan_trace): Add ENTRY argument. If true and
2510 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2511 emit a note to adjust the CFA offset.
2512 (create_cfi_notes): Adjust scan_trace callers.
2513 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2514 INCOMING_FRAME_SP_OFFSET in the CIE.
2515 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2516 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2518 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2519 * doc/tm.texi: Regenerated.
2521 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2523 PR rtl-optimization/83147
2524 * lra-constraints.c (remove_inheritance_pseudos): Use
2525 lra_substitute_pseudo_within_insn.
2527 2018-01-19 Tom de Vries <tom@codesourcery.com>
2528 Cesar Philippidis <cesar@codesourcery.com>
2531 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2533 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2536 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2537 spaces for function labels.
2539 2018-01-19 Martin Liska <mliska@suse.cz>
2541 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2542 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2543 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2544 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2545 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2546 (PRED_CONST_RETURN): Change from 69 to 65.
2547 (PRED_NULL_RETURN): Change from 91 to 71.
2548 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2549 (PRED_LOOP_GUARD): Change from 66 to 73.
2551 2018-01-19 Martin Liska <mliska@suse.cz>
2553 * predict.c (predict_insn_def): Add new assert.
2554 (struct branch_predictor): Change type to signed integer.
2555 (test_prediction_value_range): Amend test to cover
2557 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2558 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2559 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2560 (PRED_LOOP_IV_COMPARE): Likewise.
2561 * predict.h (PROB_UNINITIALIZED): Define new constant.
2563 2018-01-19 Martin Liska <mliska@suse.cz>
2565 * predict.c (dump_prediction): Add new format for
2566 analyze_brprob.py script which is enabled with -details
2568 * profile-count.h (precise_p): New function.
2570 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2572 PR tree-optimization/83922
2573 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2574 there are no statements that need masking.
2575 (vect_active_double_reduction_p): New function.
2576 (vect_analyze_loop_operations): Use it when handling phis that
2577 are not in the loop header.
2579 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2581 PR tree-optimization/83914
2582 * tree-vect-loop.c (vectorizable_induction): Don't convert
2583 init_expr or apply the peeling adjustment for inductions
2584 that are nested within the vectorized loop.
2586 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2588 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2591 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2595 * function.h (gimplify_parameters): Add gimple_seq * argument.
2596 * function.c: Include gimple.h and options.h.
2597 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2598 for the added local temporaries if needed.
2599 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2600 if there are any parameter cleanups, wrap whole body into a
2601 try/finally with the cleanups.
2603 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2606 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2607 Use GET_MODE_CLASS for scalar floating point.
2609 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2613 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2614 Fix call of call_cgraph_insertion_hooks.
2616 2018-01-18 Martin Sebor <msebor@redhat.com>
2618 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2620 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2623 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2626 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2629 * common.opt: (-ffile-prefix-map): New option.
2630 * opts.c (common_handle_option): Defer it.
2631 * opts-global.c (handle_common_deferred_options): Handle it.
2632 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2633 * file-prefix-map.h: New file.
2634 (remap_debug_filename, add_debug_prefix_map): ...here.
2635 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2636 * final.c (debug_prefix_map, add_debug_prefix_map
2637 remap_debug_filename): Move to...
2638 * file-prefix-map.c: New file.
2639 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2640 generalize, get rid of alloca(), use strrchr() instead of strchr().
2641 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2642 Implement in terms of add_prefix_map().
2643 (remap_macro_filename, remap_debug_filename): Implement in term of
2645 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2646 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2647 * dbxout.c: Include file-prefix-map.h.
2648 * varasm.c: Likewise.
2649 * vmsdbgout.c: Likewise.
2650 * xcoffout.c: Likewise.
2651 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2652 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2653 * doc/invoke.texi (-ffile-prefix-map): Document.
2654 (-fdebug-prefix-map): Update description.
2656 2018-01-18 Martin Liska <mliska@suse.cz>
2658 * config/i386/i386.c (indirect_thunk_name): Document that also
2660 (output_indirect_thunk): Document why both instructions
2661 (pause and lfence) are generated.
2663 2018-01-18 Richard Biener <rguenther@suse.de>
2665 PR tree-optimization/83887
2666 * graphite-scop-detection.c
2667 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2668 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2669 (scop_detection::merge_sese): Re-implement with a flood-fill
2670 algorithm that properly finds a SESE region if it exists.
2672 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2675 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2676 pointer_diff optimizations use view_convert instead of convert.
2678 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2680 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2681 Generate different code for -mno-speculate-indirect-jumps.
2682 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2683 (*call_indirect_aix<mode>): Disable for
2684 -mno-speculate-indirect-jumps.
2685 (*call_indirect_aix<mode>_nospec): New define_insn.
2686 (*call_value_indirect_aix<mode>): Disable for
2687 -mno-speculate-indirect-jumps.
2688 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2689 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2690 -mno-speculate-indirect-jumps.
2691 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2693 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2695 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2696 long double type, set the flags for noting the default long double
2697 type, even if we don't pass or return a long double type.
2699 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2702 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2705 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2707 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2708 support for merge[hl].
2709 (fold_mergehl_helper): New helper function.
2710 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2711 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2712 (altivec_vmrglw_direct): Add xxmrglw insn.
2714 2018-01-17 Andrew Waterman <andrew@sifive.com>
2716 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2717 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2719 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2722 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2723 call the lto_location_cache before reading the
2724 DECL_SOURCE_LOCATION of the types.
2726 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2727 Richard Sandiford <richard.sandiford@linaro.org>
2729 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2730 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2731 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2732 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2733 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2735 * config/aarch64/constraints.md (aarch64_movti_operand):
2737 * config/aarch64/predicates.md (Uti): Add new constraint.
2739 2018-01-17 Carl Love <cel@us.ibm.com>
2741 * config/rs6000/vsx.md (define_expand xl_len_r,
2742 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2743 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2745 (define_expand, define_insn): Move the shift left from the
2746 define_insn to the define_expand for lxvl and stxvl instructions.
2747 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2748 and XL_LEN_R definitions to PURE.
2750 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2752 * config/i386/i386.c (indirect_thunk_name): Declare regno
2753 as unsigned int. Compare regno with INVALID_REGNUM.
2754 (output_indirect_thunk): Ditto.
2755 (output_indirect_thunk_function): Ditto.
2756 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2757 in the call to output_indirect_thunk_function.
2759 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2762 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2763 rather than the size of inner_type to determine the stack slot size
2764 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2766 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2769 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2772 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2774 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2775 endian Linux systems to optionally enable multilibs for selecting
2776 the long double type if the user configured an explicit type.
2777 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2778 have no long double multilibs if not defined.
2779 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2780 warn if the user used -mabi={ieee,ibm}longdouble and we built
2781 multilibs for long double.
2782 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2783 appropriate multilib option.
2784 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2786 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2787 for building long double multilibs.
2788 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2790 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2792 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2795 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2797 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2800 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2803 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2806 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2808 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2809 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2811 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2813 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2814 different rtl trees depending on TARGET_64BIT.
2815 (rs6000_gen_lvx): Likewise.
2817 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2819 * config/visium/visium.md (nop): Tweak comment.
2820 (hazard_nop): Likewise.
2822 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2824 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2825 -mspeculate-indirect-jumps.
2826 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2827 for -mno-speculate-indirect-jumps.
2828 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2829 (*call_value_indirect_elfv2<mode>): Disable for
2830 -mno-speculate-indirect-jumps.
2831 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2832 (indirect_jump): Emit different RTL for
2833 -mno-speculate-indirect-jumps.
2834 (*indirect_jump<mode>): Disable for
2835 -mno-speculate-indirect-jumps.
2836 (*indirect_jump<mode>_nospec): New define_insn.
2837 (tablejump): Emit different RTL for
2838 -mno-speculate-indirect-jumps.
2839 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2840 (tablejumpsi_nospec): New define_expand.
2841 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2842 (tablejumpdi_nospec): New define_expand.
2843 (*tablejump<mode>_internal1): Disable for
2844 -mno-speculate-indirect-jumps.
2845 (*tablejump<mode>_internal1_nospec): New define_insn.
2846 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2849 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2851 * caller-save.c (insert_save): Drop unnecessary parameter. All
2854 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2855 Richard Biener <rguenth@suse.de>
2858 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2859 return early, inline manually is_gimple_sizepos. Make sure if we
2860 call gimplify_expr we don't end up with a gimple constant.
2861 * tree.c (variably_modified_type_p): Don't return true for
2862 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2863 * gimplify.h (is_gimple_sizepos): Remove.
2865 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2867 PR tree-optimization/83857
2868 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2869 vectorizable_live_operation for pure SLP statements.
2870 (vectorizable_live_operation): Handle PHIs.
2872 2018-01-16 Richard Biener <rguenther@suse.de>
2874 PR tree-optimization/83867
2875 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2876 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2878 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2881 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2882 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2883 If off is not INTEGER_CST, issue a may not be aligned warning
2884 rather than isn't aligned. Use isn%'t rather than isn't.
2885 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2887 <case MULT_EXPR>: Improve the case when bottom and one of the
2888 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2889 operand, in that case check if the other operand is multiple of
2890 bottom divided by the INTEGER_CST operand.
2892 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2895 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2896 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2897 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2898 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2899 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2900 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2901 (pa_function_arg_size): New function.
2903 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2905 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2906 in a separate statement.
2908 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2910 PR tree-optimization/83847
2911 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2912 group gathers and scatters.
2914 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2916 PR rtl-optimization/86620
2917 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2919 PR rtl-optimization/83213
2920 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2921 to last if both are JUMP_INSNs.
2923 PR tree-optimization/83843
2924 * gimple-ssa-store-merging.c
2925 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2926 store_immediate_info for bswap/nop orig_stores.
2928 2018-01-15 Andrew Waterman <andrew@sifive.com>
2930 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2932 <UDIV>: Increase cost if !TARGET_DIV.
2934 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2936 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2937 (define_attr "cr_logical_3op"): New.
2938 (cceq_ior_compare): Adjust.
2939 (cceq_ior_compare_complement): Adjust.
2940 (*cceq_rev_compare): Adjust.
2941 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2942 (is_cracked_insn): Adjust.
2943 (insn_must_be_first_in_group): Adjust.
2944 * config/rs6000/40x.md: Adjust.
2945 * config/rs6000/440.md: Adjust.
2946 * config/rs6000/476.md: Adjust.
2947 * config/rs6000/601.md: Adjust.
2948 * config/rs6000/603.md: Adjust.
2949 * config/rs6000/6xx.md: Adjust.
2950 * config/rs6000/7450.md: Adjust.
2951 * config/rs6000/7xx.md: Adjust.
2952 * config/rs6000/8540.md: Adjust.
2953 * config/rs6000/cell.md: Adjust.
2954 * config/rs6000/e300c2c3.md: Adjust.
2955 * config/rs6000/e500mc.md: Adjust.
2956 * config/rs6000/e500mc64.md: Adjust.
2957 * config/rs6000/e5500.md: Adjust.
2958 * config/rs6000/e6500.md: Adjust.
2959 * config/rs6000/mpc.md: Adjust.
2960 * config/rs6000/power4.md: Adjust.
2961 * config/rs6000/power5.md: Adjust.
2962 * config/rs6000/power6.md: Adjust.
2963 * config/rs6000/power7.md: Adjust.
2964 * config/rs6000/power8.md: Adjust.
2965 * config/rs6000/power9.md: Adjust.
2966 * config/rs6000/rs64.md: Adjust.
2967 * config/rs6000/titan.md: Adjust.
2969 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2971 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2972 ix86_indirect_branch_register logic.
2974 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2976 * config/i386/constraints.md (Bs): Update
2977 ix86_indirect_branch_register check. Don't check
2978 ix86_indirect_branch_register with GOT_memory_operand.
2980 * config/i386/predicates.md (GOT_memory_operand): Don't check
2981 ix86_indirect_branch_register here.
2982 (GOT32_symbol_operand): Likewise.
2984 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2986 * config/i386/predicates.md (constant_call_address_operand):
2987 Rewrite ix86_indirect_branch_register logic.
2988 (sibcall_insn_operand): Likewise.
2990 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2992 * config/i386/constraints.md (Bs): Replace
2993 ix86_indirect_branch_thunk_register with
2994 ix86_indirect_branch_register.
2996 * config/i386/i386.md (indirect_jump): Likewise.
2997 (tablejump): Likewise.
2998 (*sibcall_memory): Likewise.
2999 (*sibcall_value_memory): Likewise.
3000 Peepholes of indirect call and jump via memory: Likewise.
3001 * config/i386/i386.opt: Likewise.
3002 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3003 (GOT_memory_operand): Likewise.
3004 (call_insn_operand): Likewise.
3005 (sibcall_insn_operand): Likewise.
3006 (GOT32_symbol_operand): Likewise.
3008 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3011 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
3012 type rather than type addr's type points to.
3013 (expand_omp_atomic_mutex): Likewise.
3014 (expand_omp_atomic): Likewise.
3016 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3019 * config/i386/i386.c (output_indirect_thunk_function): Use
3020 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
3021 for __x86_return_thunk.
3023 2018-01-15 Richard Biener <rguenther@suse.de>
3026 * expmed.c (extract_bit_field_1): Fix typo.
3028 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3031 * config/arm/iterators.md (VF): New mode iterator.
3032 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
3033 Remove integer-related logic from pattern.
3034 (neon_vabd<mode>_3): Likewise.
3036 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3039 * common.opt (fstrict-overflow): No longer an alias.
3040 (fwrapv-pointer): New option.
3041 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
3042 also for pointer types based on flag_wrapv_pointer.
3043 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
3044 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
3045 opts->x_flag_wrapv got set.
3046 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
3047 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
3048 POINTER_TYPE_OVERFLOW_UNDEFINED.
3049 * match.pd: Likewise in address comparison pattern.
3050 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
3052 2018-01-15 Richard Biener <rguenther@suse.de>
3055 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
3056 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
3057 Reset type names to their identifier if their TYPE_DECL doesn't
3058 have linkage (and thus is used for ODR and devirt).
3059 (save_debug_info_for_decl): Remove.
3060 (save_debug_info_for_type): Likewise.
3061 (add_tree_to_fld_list): Adjust.
3062 * tree-pretty-print.c (dump_generic_node): Make dumping of
3063 type names more robust.
3065 2018-01-15 Richard Biener <rguenther@suse.de>
3067 * BASE-VER: Bump to 8.0.1.
3069 2018-01-14 Martin Sebor <msebor@redhat.com>
3072 * builtins.c (check_access): Avoid warning when the no-warning bit
3075 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
3077 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
3078 * ira-color (allocno_hard_regs_compare): Likewise.
3080 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
3083 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
3084 Use .pushsection/.popsection.
3086 2018-01-14 Martin Sebor <msebor@redhat.com>
3089 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
3091 2018-01-14 Jakub Jelinek <jakub@redhat.com>
3093 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
3094 entry from extra_headers.
3095 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
3096 extra_headers, make the list bitwise identical to the i?86-*-* one.
3098 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3100 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
3101 -mcmodel=large with -mindirect-branch=thunk,
3102 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
3103 -mfunction-return=thunk-extern.
3104 * doc/invoke.texi: Document -mcmodel=large is incompatible with
3105 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
3106 -mfunction-return=thunk and -mfunction-return=thunk-extern.
3108 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3110 * config/i386/i386.c (print_reg): Print the name of the full
3111 integer register without '%'.
3112 (ix86_print_operand): Handle 'V'.
3113 * doc/extend.texi: Document 'V' modifier.
3115 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3117 * config/i386/constraints.md (Bs): Disallow memory operand for
3118 -mindirect-branch-register.
3120 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3121 (GOT_memory_operand): Likewise.
3122 (call_insn_operand): Likewise.
3123 (sibcall_insn_operand): Likewise.
3124 (GOT32_symbol_operand): Likewise.
3125 * config/i386/i386.md (indirect_jump): Call convert_memory_address
3126 for -mindirect-branch-register.
3127 (tablejump): Likewise.
3128 (*sibcall_memory): Likewise.
3129 (*sibcall_value_memory): Likewise.
3130 Disallow peepholes of indirect call and jump via memory for
3131 -mindirect-branch-register.
3132 (*call_pop): Replace m with Bw.
3133 (*call_value_pop): Likewise.
3134 (*sibcall_pop_memory): Replace m with Bs.
3135 * config/i386/i386.opt (mindirect-branch-register): New option.
3136 * doc/invoke.texi: Document -mindirect-branch-register option.
3138 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3140 * config/i386/i386-protos.h (ix86_output_function_return): New.
3141 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
3142 set function_return_type.
3143 (indirect_thunk_name): Add ret_p to indicate thunk for function
3145 (output_indirect_thunk_function): Pass false to
3146 indirect_thunk_name.
3147 (ix86_output_indirect_branch_via_reg): Likewise.
3148 (ix86_output_indirect_branch_via_push): Likewise.
3149 (output_indirect_thunk_function): Create alias for function
3150 return thunk if regno < 0.
3151 (ix86_output_function_return): New function.
3152 (ix86_handle_fndecl_attribute): Handle function_return.
3153 (ix86_attribute_table): Add function_return.
3154 * config/i386/i386.h (machine_function): Add
3155 function_return_type.
3156 * config/i386/i386.md (simple_return_internal): Use
3157 ix86_output_function_return.
3158 (simple_return_internal_long): Likewise.
3159 * config/i386/i386.opt (mfunction-return=): New option.
3160 (indirect_branch): Mention -mfunction-return=.
3161 * doc/extend.texi: Document function_return function attribute.
3162 * doc/invoke.texi: Document -mfunction-return= option.
3164 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3166 * config/i386/i386-opts.h (indirect_branch): New.
3167 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
3168 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
3169 with local indirect jump when converting indirect call and jump.
3170 (ix86_set_indirect_branch_type): New.
3171 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
3172 (indirectlabelno): New.
3173 (indirect_thunk_needed): Likewise.
3174 (indirect_thunk_bnd_needed): Likewise.
3175 (indirect_thunks_used): Likewise.
3176 (indirect_thunks_bnd_used): Likewise.
3177 (INDIRECT_LABEL): Likewise.
3178 (indirect_thunk_name): Likewise.
3179 (output_indirect_thunk): Likewise.
3180 (output_indirect_thunk_function): Likewise.
3181 (ix86_output_indirect_branch_via_reg): Likewise.
3182 (ix86_output_indirect_branch_via_push): Likewise.
3183 (ix86_output_indirect_branch): Likewise.
3184 (ix86_output_indirect_jmp): Likewise.
3185 (ix86_code_end): Call output_indirect_thunk_function if needed.
3186 (ix86_output_call_insn): Call ix86_output_indirect_branch if
3188 (ix86_handle_fndecl_attribute): Handle indirect_branch.
3189 (ix86_attribute_table): Add indirect_branch.
3190 * config/i386/i386.h (machine_function): Add indirect_branch_type
3191 and has_local_indirect_jump.
3192 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
3194 (tablejump): Likewise.
3195 (*indirect_jump): Use ix86_output_indirect_jmp.
3196 (*tablejump_1): Likewise.
3197 (simple_return_indirect_internal): Likewise.
3198 * config/i386/i386.opt (mindirect-branch=): New option.
3199 (indirect_branch): New.
3202 (thunk-inline): Likewise.
3203 (thunk-extern): Likewise.
3204 * doc/extend.texi: Document indirect_branch function attribute.
3205 * doc/invoke.texi: Document -mindirect-branch= option.
3207 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
3210 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
3212 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
3214 * ipa-inline.c (want_inline_small_function_p): Return false if
3215 inlining has already failed with CIF_FINAL_ERROR.
3216 (update_caller_keys): Call want_inline_small_function_p before
3218 (update_callee_keys): Likewise.
3220 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
3222 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
3224 (rs6000_quadword_masked_address_p): Likewise.
3225 (quad_aligned_load_p): Likewise.
3226 (quad_aligned_store_p): Likewise.
3227 (const_load_sequence_p): Add comment to describe the outer-most loop.
3228 (mimic_memory_attributes_and_flags): New function.
3229 (rs6000_gen_stvx): Likewise.
3230 (replace_swapped_aligned_store): Likewise.
3231 (rs6000_gen_lvx): Likewise.
3232 (replace_swapped_aligned_load): Likewise.
3233 (replace_swapped_load_constant): Capitalize argument name in
3234 comment describing this function.
3235 (rs6000_analyze_swaps): Add a third pass to search for vector loads
3236 and stores that access quad-word aligned addresses and replace
3237 with stvx or lvx instructions when appropriate.
3238 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
3239 New function prototype.
3240 (rs6000_quadword_masked_address_p): Likewise.
3241 (rs6000_gen_lvx): Likewise.
3242 (rs6000_gen_stvx): Likewise.
3243 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3244 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3245 when memory address is aligned.
3246 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3247 this split to select lvx instruction when memory address is aligned.
3248 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3249 instruction when memory address is aligned.
3250 (*vsx_le_perm_load_v16qi): Likewise.
3251 (four unnamed splitters): Modify to select the stvx instruction
3252 when memory is aligned.
3254 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3256 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3257 which appears in the queue multiple times.
3259 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3260 Alan Hayward <alan.hayward@arm.com>
3261 David Sherwood <david.sherwood@arm.com>
3263 * tree-vectorizer.h (vec_lower_bound): New structure.
3264 (_loop_vec_info): Add check_nonzero and lower_bounds.
3265 (LOOP_VINFO_CHECK_NONZERO): New macro.
3266 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3267 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3268 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3269 fields. Make seg_len the distance travelled, not including the
3271 (dr_direction_indicator): Declare.
3272 (dr_zero_step_indicator): Likewise.
3273 (dr_known_forward_stride_p): Likewise.
3274 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3276 (runtime_alias_check_p): Allow runtime alias checks with
3278 (operator ==): Compare access_size and align.
3279 (prune_runtime_alias_test_list): Rework for new distinction between
3280 the access_size and seg_len.
3281 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3283 (get_segment_min_max): New function.
3284 (create_intersect_range_checks): Use it.
3285 (dr_step_indicator): New function.
3286 (dr_direction_indicator): Likewise.
3287 (dr_zero_step_indicator): Likewise.
3288 (dr_known_forward_stride_p): Likewise.
3289 * tree-loop-distribution.c (data_ref_segment_size): Return
3290 DR_STEP * (niters - 1).
3291 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3293 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3294 (vect_preserves_scalar_order_p): New function, split out from...
3295 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3296 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3297 (vect_vfa_access_size): New function.
3298 (vect_vfa_align): Likewise.
3299 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3300 (dump_lower_bound): New function.
3301 (vect_check_lower_bound): Likewise.
3302 (vect_small_gap_p): Likewise.
3303 (vectorizable_with_step_bound_p): Likewise.
3304 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3305 depencies if the vectorization factor is 1. Convert the checks
3306 for nonzero steps into checks on the bounds of DR_STEP. Try using
3307 a bunds check for variable steps if the minimum required step is
3308 relatively small. Update calls to the dr_with_seg_len
3309 constructor and to vect_compile_time_alias.
3310 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3312 (vect_loop_versioning): Call it.
3313 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3315 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3317 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3318 Alan Hayward <alan.hayward@arm.com>
3319 David Sherwood <david.sherwood@arm.com>
3321 * doc/sourcebuild.texi (vect_scatter_store): Document.
3322 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3324 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3326 * genopinit.c (main): Add supports_vec_scatter_store and
3327 supports_vec_scatter_store_cached to target_optabs.
3328 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3329 IFN_MASK_SCATTER_STORE.
3330 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3332 * internal-fn.h (internal_store_fn_p): Declare.
3333 (internal_fn_stored_value_index): Likewise.
3334 * internal-fn.c (scatter_store_direct): New macro.
3335 (expand_scatter_store_optab_fn): New function.
3336 (direct_scatter_store_optab_supported_p): New macro.
3337 (internal_store_fn_p): New function.
3338 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3339 IFN_MASK_SCATTER_STORE.
3340 (internal_fn_mask_index): Likewise.
3341 (internal_fn_stored_value_index): New function.
3342 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3344 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3345 * optabs-query.c (supports_vec_scatter_store_p): New function.
3346 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3347 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3348 true for scatter stores.
3349 (vect_gather_scatter_fn_p): Handle scatter stores too.
3350 (vect_check_gather_scatter): Consider using scatter stores if
3351 supports_vec_scatter_store_p.
3352 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3354 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3355 internal_fn_stored_value_index.
3356 (check_load_store_masking): Handle scatter stores too.
3357 (vect_get_store_rhs): Make public.
3358 (vectorizable_call): Use internal_store_fn_p.
3359 (vectorizable_store): Handle scatter store internal functions.
3360 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3361 when deciding whether the end of the group has been reached.
3362 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3363 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3364 (mask_scatter_store<mode>): New insns.
3366 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3367 Alan Hayward <alan.hayward@arm.com>
3368 David Sherwood <david.sherwood@arm.com>
3370 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3371 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3372 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3374 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3375 Use vect_truncate_gather_scatter_offset if we can't treat the
3376 operation as a normal gather load or scatter store.
3377 (get_group_load_store_type): Take the gather_scatter_info
3378 as argument. Try using a gather load or scatter store for
3379 single-element groups.
3380 (get_load_store_type): Update calls to get_group_load_store_type
3381 and vect_use_strided_gather_scatters_p.
3383 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3384 Alan Hayward <alan.hayward@arm.com>
3385 David Sherwood <david.sherwood@arm.com>
3387 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3388 optional tree argument.
3389 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3391 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3392 but continue to use the current value as a fallback.
3393 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3394 to compare the updates.
3395 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3396 (get_load_store_type): Use it when handling a strided access.
3397 (vect_get_strided_load_store_ops): New function.
3398 (vect_get_data_ptr_increment): Likewise.
3399 (vectorizable_load): Handle strided gather loads. Always pass
3400 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3402 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3403 Alan Hayward <alan.hayward@arm.com>
3404 David Sherwood <david.sherwood@arm.com>
3406 * doc/md.texi (gather_load@var{m}): Document.
3407 (mask_gather_load@var{m}): Likewise.
3408 * genopinit.c (main): Add supports_vec_gather_load and
3409 supports_vec_gather_load_cached to target_optabs.
3410 * optabs-tree.c (init_tree_optimization_optabs): Use
3411 ggc_cleared_alloc to allocate target_optabs.
3412 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3413 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3415 * internal-fn.h (internal_load_fn_p): Declare.
3416 (internal_gather_scatter_fn_p): Likewise.
3417 (internal_fn_mask_index): Likewise.
3418 (internal_gather_scatter_fn_supported_p): Likewise.
3419 * internal-fn.c (gather_load_direct): New macro.
3420 (expand_gather_load_optab_fn): New function.
3421 (direct_gather_load_optab_supported_p): New macro.
3422 (direct_internal_fn_optab): New function.
3423 (internal_load_fn_p): Likewise.
3424 (internal_gather_scatter_fn_p): Likewise.
3425 (internal_fn_mask_index): Likewise.
3426 (internal_gather_scatter_fn_supported_p): Likewise.
3427 * optabs-query.c (supports_at_least_one_mode_p): New function.
3428 (supports_vec_gather_load_p): Likewise.
3429 * optabs-query.h (supports_vec_gather_load_p): Declare.
3430 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3431 and memory_type field.
3432 (NUM_PATTERNS): Bump to 15.
3433 * tree-vect-data-refs.c: Include internal-fn.h.
3434 (vect_gather_scatter_fn_p): New function.
3435 (vect_describe_gather_scatter_call): Likewise.
3436 (vect_check_gather_scatter): Try using internal functions for
3437 gather loads. Recognize existing calls to a gather load function.
3438 (vect_analyze_data_refs): Consider using gather loads if
3439 supports_vec_gather_load_p.
3440 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3441 (vect_get_gather_scatter_offset_type): Likewise.
3442 (vect_convert_mask_for_vectype): Likewise.
3443 (vect_add_conversion_to_patterm): Likewise.
3444 (vect_try_gather_scatter_pattern): Likewise.
3445 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3446 (vect_vect_recog_func_ptrs): Add it.
3447 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3448 internal_fn_mask_index and internal_gather_scatter_fn_p.
3449 (check_load_store_masking): Take the gather_scatter_info as an
3450 argument and handle gather loads.
3451 (vect_get_gather_scatter_ops): New function.
3452 (vectorizable_call): Check internal_load_fn_p.
3453 (vectorizable_load): Likewise. Handle gather load internal
3455 (vectorizable_store): Update call to check_load_store_masking.
3456 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3457 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3458 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3459 (aarch64_gather_scale_operand_d): New predicates.
3460 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3461 (mask_gather_load<mode>): New insns.
3463 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3464 Alan Hayward <alan.hayward@arm.com>
3465 David Sherwood <david.sherwood@arm.com>
3467 * optabs.def (fold_left_plus_optab): New optab.
3468 * doc/md.texi (fold_left_plus_@var{m}): Document.
3469 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3470 * internal-fn.c (fold_left_direct): Define.
3471 (expand_fold_left_optab_fn): Likewise.
3472 (direct_fold_left_optab_supported_p): Likewise.
3473 * fold-const-call.c (fold_const_fold_left): New function.
3474 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3475 * tree-parloops.c (valid_reduction_p): New function.
3476 (gather_scalar_reductions): Use it.
3477 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3478 (vect_finish_replace_stmt): Declare.
3479 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3480 (needs_fold_left_reduction_p): New function, split out from...
3481 (vect_is_simple_reduction): ...here. Accept reductions that
3482 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3483 (vect_force_simple_reduction): Also store the reduction type in
3484 the assignment's STMT_VINFO_REDUC_TYPE.
3485 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3486 (merge_with_identity): New function.
3487 (vect_expand_fold_left): Likewise.
3488 (vectorize_fold_left_reduction): Likewise.
3489 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3490 scalar phi in place for it. Check for target support and reject
3491 cases that would reassociate the operation. Defer the transform
3492 phase to vectorize_fold_left_reduction.
3493 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3494 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3495 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3497 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3499 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3500 call to ifc_temp_var.
3502 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3503 Alan Hayward <alan.hayward@arm.com>
3504 David Sherwood <david.sherwood@arm.com>
3506 * target.def (legitimize_address_displacement): Take the original
3507 offset as a poly_int.
3508 * targhooks.h (default_legitimize_address_displacement): Update
3510 * targhooks.c (default_legitimize_address_displacement): Likewise.
3511 * doc/tm.texi: Regenerate.
3512 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3513 as an argument, moving assert of ad->disp == ad->disp_term to...
3514 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3515 Try calling targetm.legitimize_address_displacement before expanding
3516 the address rather than afterwards, and adjust for the new interface.
3517 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3518 Match the new hook interface. Handle SVE addresses.
3519 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3522 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3524 * Makefile.in (OBJS): Add early-remat.o.
3525 * target.def (select_early_remat_modes): New hook.
3526 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3527 * doc/tm.texi: Regenerate.
3528 * targhooks.h (default_select_early_remat_modes): Declare.
3529 * targhooks.c (default_select_early_remat_modes): New function.
3530 * timevar.def (TV_EARLY_REMAT): New timevar.
3531 * passes.def (pass_early_remat): New pass.
3532 * tree-pass.h (make_pass_early_remat): Declare.
3533 * early-remat.c: New file.
3534 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3536 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3538 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3539 Alan Hayward <alan.hayward@arm.com>
3540 David Sherwood <david.sherwood@arm.com>
3542 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3543 vfm1 with a bound_epilog parameter.
3544 (vect_do_peeling): Update calls accordingly, and move the prologue
3545 call earlier in the function. Treat the base bound_epilog as 0 for
3546 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3547 this base when peeling for gaps.
3548 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3549 with fully-masked loops.
3550 (vect_estimate_min_profitable_iters): Handle the single peeled
3551 iteration in that case.
3553 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3554 Alan Hayward <alan.hayward@arm.com>
3555 David Sherwood <david.sherwood@arm.com>
3557 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3558 single-element interleaving even if the size is not a power of 2.
3559 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3560 accesses for single-element interleaving if the group size is
3563 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3564 Alan Hayward <alan.hayward@arm.com>
3565 David Sherwood <david.sherwood@arm.com>
3567 * doc/md.texi (fold_extract_last_@var{m}): Document.
3568 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3569 * optabs.def (fold_extract_last_optab): New optab.
3570 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3571 * internal-fn.c (fold_extract_direct): New macro.
3572 (expand_fold_extract_optab_fn): Likewise.
3573 (direct_fold_extract_optab_supported_p): Likewise.
3574 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3575 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3576 EXTRACT_LAST_REDUCTION.
3577 (get_initial_def_for_reduction): Do not create an initial vector
3578 for EXTRACT_LAST_REDUCTION reductions.
3579 (vectorizable_reduction): Leave the scalar phi in place for
3580 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3581 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3582 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3583 transform phase to vectorizable_condition.
3584 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3586 (vect_finish_stmt_generation): ...here.
3587 (vect_finish_replace_stmt): New function.
3588 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3589 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3591 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3593 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3594 Alan Hayward <alan.hayward@arm.com>
3595 David Sherwood <david.sherwood@arm.com>
3597 * doc/md.texi (extract_last_@var{m}): Document.
3598 * optabs.def (extract_last_optab): New optab.
3599 * internal-fn.def (EXTRACT_LAST): New internal function.
3600 * internal-fn.c (cond_unary_direct): New macro.
3601 (expand_cond_unary_optab_fn): Likewise.
3602 (direct_cond_unary_optab_supported_p): Likewise.
3603 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3604 loops using EXTRACT_LAST.
3605 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3606 (extract_last_<mode>): ...this optab.
3607 (vec_extract<mode><Vel>): Update accordingly.
3609 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3610 Alan Hayward <alan.hayward@arm.com>
3611 David Sherwood <david.sherwood@arm.com>
3613 * target.def (empty_mask_is_expensive): New hook.
3614 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3615 * doc/tm.texi: Regenerate.
3616 * targhooks.h (default_empty_mask_is_expensive): Declare.
3617 * targhooks.c (default_empty_mask_is_expensive): New function.
3618 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3619 if the target says that empty masks are expensive.
3620 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3622 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3624 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3625 Alan Hayward <alan.hayward@arm.com>
3626 David Sherwood <david.sherwood@arm.com>
3628 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3629 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3630 (vect_use_loop_mask_for_alignment_p): New function.
3631 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3632 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3633 niters_skip argument. Make sure that the first niters_skip elements
3634 of the first iteration are inactive.
3635 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3636 Update call to vect_set_loop_masks_directly.
3637 (get_misalign_in_elems): New function, split out from...
3638 (vect_gen_prolog_loop_niters): ...here.
3639 (vect_update_init_of_dr): Take a code argument that specifies whether
3640 the adjustment should be added or subtracted.
3641 (vect_update_init_of_drs): Likewise.
3642 (vect_prepare_for_masked_peels): New function.
3643 (vect_do_peeling): Skip prologue peeling if we're using a mask
3644 instead. Update call to vect_update_inits_of_drs.
3645 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3647 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3648 alignment. Do not include the number of peeled iterations in
3649 the minimum threshold in that case.
3650 (vectorizable_induction): Adjust the start value down by
3651 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3652 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3653 Take the number of skipped iterations into account when calculating
3655 * tree-vect-stmts.c (vect_gen_while_not): New function.
3657 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3658 Alan Hayward <alan.hayward@arm.com>
3659 David Sherwood <david.sherwood@arm.com>
3661 * doc/sourcebuild.texi (vect_fully_masked): Document.
3662 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3664 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3666 (vect_analyze_loop_2): ...here. Don't check the vectorization
3667 factor against the number of loop iterations if the loop is
3670 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3671 Alan Hayward <alan.hayward@arm.com>
3672 David Sherwood <david.sherwood@arm.com>
3674 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3675 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3676 (dump_groups): Update accordingly.
3677 (iv_use::mem_type): New member variable.
3678 (address_p): New function.
3679 (record_use): Add a mem_type argument and initialize the new
3681 (record_group_use): Add a mem_type argument. Use address_p.
3682 Remove obsolete null checks of base_object. Update call to record_use.
3683 (find_interesting_uses_op): Update call to record_group_use.
3684 (find_interesting_uses_cond): Likewise.
3685 (find_interesting_uses_address): Likewise.
3686 (get_mem_type_for_internal_fn): New function.
3687 (find_address_like_use): Likewise.
3688 (find_interesting_uses_stmt): Try find_address_like_use before
3689 calling find_interesting_uses_op.
3690 (addr_offset_valid_p): Use the iv mem_type field as the type
3691 of the addressed memory.
3692 (add_autoinc_candidates): Likewise.
3693 (get_address_cost): Likewise.
3694 (split_small_address_groups_p): Use address_p.
3695 (split_address_groups): Likewise.
3696 (add_iv_candidate_for_use): Likewise.
3697 (autoinc_possible_for_pair): Likewise.
3698 (rewrite_groups): Likewise.
3699 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3700 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3701 (get_alias_ptr_type_for_ptr_address): New function.
3702 (rewrite_use_address): Rewrite address uses in calls that were
3703 identified by find_address_like_use.
3705 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3706 Alan Hayward <alan.hayward@arm.com>
3707 David Sherwood <david.sherwood@arm.com>
3709 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3711 * gimple-expr.h (is_gimple_addressable: Likewise.
3712 * gimple-expr.c (is_gimple_address): Likewise.
3713 * internal-fn.c (expand_call_mem_ref): New function.
3714 (expand_mask_load_optab_fn): Use it.
3715 (expand_mask_store_optab_fn): Likewise.
3717 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3718 Alan Hayward <alan.hayward@arm.com>
3719 David Sherwood <david.sherwood@arm.com>
3721 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3722 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3723 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3724 (cond_umax@var{mode}): Document.
3725 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3726 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3727 (cond_umin_optab, cond_umax_optab): New optabs.
3728 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3729 (COND_IOR, COND_XOR): New internal functions.
3730 * internal-fn.h (get_conditional_internal_fn): Declare.
3731 * internal-fn.c (cond_binary_direct): New macro.
3732 (expand_cond_binary_optab_fn): Likewise.
3733 (direct_cond_binary_optab_supported_p): Likewise.
3734 (get_conditional_internal_fn): New function.
3735 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3736 Cope with reduction statements that are vectorized as calls rather
3738 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3739 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3740 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3741 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3742 (UNSPEC_COND_EOR): New unspecs.
3743 (optab): Add mappings for them.
3744 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3745 (sve_int_op, sve_fp_op): New int attributes.
3747 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3748 Alan Hayward <alan.hayward@arm.com>
3749 David Sherwood <david.sherwood@arm.com>
3751 * optabs.def (while_ult_optab): New optab.
3752 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3753 * internal-fn.def (WHILE_ULT): New internal function.
3754 * internal-fn.h (direct_internal_fn_supported_p): New override
3755 that takes two types as argument.
3756 * internal-fn.c (while_direct): New macro.
3757 (expand_while_optab_fn): New function.
3758 (convert_optab_supported_p): Likewise.
3759 (direct_while_optab_supported_p): New macro.
3760 * wide-int.h (wi::udiv_ceil): New function.
3761 * tree-vectorizer.h (rgroup_masks): New structure.
3762 (vec_loop_masks): New typedef.
3763 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3765 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3766 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3767 (vect_max_vf): New function.
3768 (slpeel_make_loop_iterate_ntimes): Delete.
3769 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3770 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3771 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3772 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3773 internal-fn.h, stor-layout.h and optabs-query.h.
3774 (vect_set_loop_mask): New function.
3775 (add_preheader_seq): Likewise.
3776 (add_header_seq): Likewise.
3777 (interleave_supported_p): Likewise.
3778 (vect_maybe_permute_loop_masks): Likewise.
3779 (vect_set_loop_masks_directly): Likewise.
3780 (vect_set_loop_condition_masked): Likewise.
3781 (vect_set_loop_condition_unmasked): New function, split out from
3782 slpeel_make_loop_iterate_ntimes.
3783 (slpeel_make_loop_iterate_ntimes): Rename to..
3784 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3785 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3786 (vect_do_peeling): Update call accordingly.
3787 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3789 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3790 mask_compare_type, can_fully_mask_p and fully_masked_p.
3791 (release_vec_loop_masks): New function.
3792 (_loop_vec_info): Use it to free the loop masks.
3793 (can_produce_all_loop_masks_p): New function.
3794 (vect_get_max_nscalars_per_iter): Likewise.
3795 (vect_verify_full_masking): Likewise.
3796 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3797 retries, and free the mask rgroups before retrying. Check loop-wide
3798 reasons for disallowing fully-masked loops. Make the final decision
3799 about whether use a fully-masked loop or not.
3800 (vect_estimate_min_profitable_iters): Do not assume that peeling
3801 for the number of iterations will be needed for fully-masked loops.
3802 (vectorizable_reduction): Disable fully-masked loops.
3803 (vectorizable_live_operation): Likewise.
3804 (vect_halve_mask_nunits): New function.
3805 (vect_double_mask_nunits): Likewise.
3806 (vect_record_loop_mask): Likewise.
3807 (vect_get_loop_mask): Likewise.
3808 (vect_transform_loop): Handle the case in which the final loop
3809 iteration might handle a partial vector. Call vect_set_loop_condition
3810 instead of slpeel_make_loop_iterate_ntimes.
3811 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3812 (check_load_store_masking): New function.
3813 (prepare_load_store_mask): Likewise.
3814 (vectorizable_store): Handle fully-masked loops.
3815 (vectorizable_load): Likewise.
3816 (supportable_widening_operation): Use vect_halve_mask_nunits for
3818 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3819 (vect_gen_while): New function.
3820 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3821 (aarch64_uqdec<mode>): New insn.
3823 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3824 Alan Hayward <alan.hayward@arm.com>
3825 David Sherwood <david.sherwood@arm.com>
3827 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3828 (reduc_xor_scal_optab): New optabs.
3829 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3830 (reduc_xor_scal_@var{m}): Document.
3831 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3832 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3834 * fold-const-call.c (fold_const_call): Handle them.
3835 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3836 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3837 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3838 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3839 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3840 (UNSPEC_XORV): New unspecs.
3841 (optab): Add entries for them.
3842 (BITWISEV): New int iterator.
3843 (bit_reduc_op): New int attributes.
3845 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3846 Alan Hayward <alan.hayward@arm.com>
3847 David Sherwood <david.sherwood@arm.com>
3849 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3850 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3851 * optabs.def (vec_shl_insert_optab): New optab.
3852 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3853 (duplicate_and_interleave): Likewise.
3854 * tree-vect-loop.c: Include internal-fn.h.
3855 (neutral_op_for_slp_reduction): New function, split out from
3856 get_initial_defs_for_reduction.
3857 (get_initial_def_for_reduction): Handle option 2 for variable-length
3858 vectors by loading the neutral value into a vector and then shifting
3859 the initial value into element 0.
3860 (get_initial_defs_for_reduction): Replace the code argument with
3861 the neutral value calculated by neutral_op_for_slp_reduction.
3862 Use gimple_build_vector for constant-length vectors.
3863 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3864 but the first group_size elements have a neutral value.
3865 Use duplicate_and_interleave otherwise.
3866 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3867 Update call to get_initial_defs_for_reduction. Handle SLP
3868 reductions for variable-length vectors by creating one vector
3869 result for each scalar result, with the elements associated
3870 with other scalar results stubbed out with the neutral value.
3871 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3872 Require IFN_VEC_SHL_INSERT for double reductions on
3873 variable-length vectors, or SLP reductions that have
3874 a neutral value. Require can_duplicate_and_interleave_p
3875 support for variable-length unchained SLP reductions if there
3876 is no neutral value, such as for MIN/MAX reductions. Also require
3877 the number of vector elements to be a multiple of the number of
3878 SLP statements when doing variable-length unchained SLP reductions.
3879 Update call to vect_create_epilog_for_reduction.
3880 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3881 and remove initial values.
3882 (duplicate_and_interleave): Make public.
3883 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3884 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3886 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3887 Alan Hayward <alan.hayward@arm.com>
3888 David Sherwood <david.sherwood@arm.com>
3890 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3891 (can_duplicate_and_interleave_p): New function.
3892 (vect_get_and_check_slp_defs): Take the vector of statements
3893 rather than just the current one. Remove excess parentheses.
3894 Restriction rejectinon of vect_constant_def and vect_external_def
3895 for variable-length vectors to boolean types, or types for which
3896 can_duplicate_and_interleave_p is false.
3897 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3898 (duplicate_and_interleave): New function.
3899 (vect_get_constant_vectors): Use gimple_build_vector for
3900 constant-length vectors and suitable variable-length constant
3901 vectors. Use duplicate_and_interleave for other variable-length
3902 vectors. Don't defer the update when inserting new statements.
3904 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3905 Alan Hayward <alan.hayward@arm.com>
3906 David Sherwood <david.sherwood@arm.com>
3908 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3909 min_profitable_iters doesn't go negative.
3911 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3912 Alan Hayward <alan.hayward@arm.com>
3913 David Sherwood <david.sherwood@arm.com>
3915 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3916 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3917 * optabs.def (vec_mask_load_lanes_optab): New optab.
3918 (vec_mask_store_lanes_optab): Likewise.
3919 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3920 (MASK_STORE_LANES): Likewise.
3921 * internal-fn.c (mask_load_lanes_direct): New macro.
3922 (mask_store_lanes_direct): Likewise.
3923 (expand_mask_load_optab_fn): Handle masked operations.
3924 (expand_mask_load_lanes_optab_fn): New macro.
3925 (expand_mask_store_optab_fn): Handle masked operations.
3926 (expand_mask_store_lanes_optab_fn): New macro.
3927 (direct_mask_load_lanes_optab_supported_p): Likewise.
3928 (direct_mask_store_lanes_optab_supported_p): Likewise.
3929 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3931 (vect_load_lanes_supported): Likewise.
3932 * tree-vect-data-refs.c (strip_conversion): New function.
3933 (can_group_stmts_p): Likewise.
3934 (vect_analyze_data_ref_accesses): Use it instead of checking
3935 for a pair of assignments.
3936 (vect_store_lanes_supported): Take a masked_p parameter.
3937 (vect_load_lanes_supported): Likewise.
3938 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3939 vect_store_lanes_supported and vect_load_lanes_supported.
3940 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3941 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3942 parameter. Don't allow gaps for masked accesses.
3943 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3944 and vect_load_lanes_supported.
3945 (get_load_store_type): Take a masked_p parameter and update
3946 call to get_group_load_store_type.
3947 (vectorizable_store): Update call to get_load_store_type.
3948 Handle IFN_MASK_STORE_LANES.
3949 (vectorizable_load): Update call to get_load_store_type.
3950 Handle IFN_MASK_LOAD_LANES.
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 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3958 * config/aarch64/aarch64-protos.h
3959 (aarch64_sve_struct_memory_operand_p): Declare.
3960 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3961 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3962 (VPRED, vpred): Handle SVE structure modes.
3963 * config/aarch64/constraints.md (Utx): New constraint.
3964 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3965 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3966 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3967 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3968 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3969 structure modes. Split into pieces after RA.
3970 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3971 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3973 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3974 SVE structure modes.
3975 (aarch64_classify_address): Likewise.
3976 (sizetochar): Move earlier in file.
3977 (aarch64_print_operand): Handle SVE register lists.
3978 (aarch64_array_mode): New function.
3979 (aarch64_sve_struct_memory_operand_p): Likewise.
3980 (TARGET_ARRAY_MODE): Redefine.
3982 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3983 Alan Hayward <alan.hayward@arm.com>
3984 David Sherwood <david.sherwood@arm.com>
3986 * target.def (array_mode): New target hook.
3987 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
3988 * doc/tm.texi: Regenerate.
3989 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
3990 * hooks.c (hook_optmode_mode_uhwi_none): New function.
3991 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
3993 * stor-layout.c (mode_for_array): Likewise. Support polynomial
3996 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3997 Alan Hayward <alan.hayward@arm.com>
3998 David Sherwood <david.sherwood@arm.com>
4000 * fold-const.c (fold_binary_loc): Check the argument types
4001 rather than the result type when testing for a vector operation.
4003 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4005 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
4006 * doc/tm.texi: Regenerate.
4008 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4009 Alan Hayward <alan.hayward@arm.com>
4010 David Sherwood <david.sherwood@arm.com>
4012 * doc/invoke.texi (-msve-vector-bits=): Document new option.
4013 (sve): Document new AArch64 extension.
4014 * doc/md.texi (w): Extend the description of the AArch64
4015 constraint to include SVE vectors.
4016 (Upl, Upa): Document new AArch64 predicate constraints.
4017 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
4019 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
4020 (msve-vector-bits=): New option.
4021 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
4022 SVE when these are disabled.
4023 (sve): New extension.
4024 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
4025 modes. Adjust their number of units based on aarch64_sve_vg.
4026 (MAX_BITSIZE_MODE_ANY_MODE): Define.
4027 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
4028 aarch64_addr_query_type.
4029 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
4030 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
4031 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
4032 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
4033 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
4034 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
4035 (aarch64_simd_imm_zero_p): Delete.
4036 (aarch64_check_zero_based_sve_index_immediate): Declare.
4037 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4038 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4039 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4040 (aarch64_sve_float_mul_immediate_p): Likewise.
4041 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4043 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
4044 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
4045 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
4046 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
4047 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
4048 (aarch64_regmode_natural_size): Likewise.
4049 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
4050 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
4052 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
4053 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
4054 for VG and the SVE predicate registers.
4055 (V_ALIASES): Add a "z"-prefixed alias.
4056 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
4057 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
4058 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
4059 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
4060 (REG_CLASS_NAMES): Add entries for them.
4061 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
4062 and the predicate registers.
4063 (aarch64_sve_vg): Declare.
4064 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
4065 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
4066 (REGMODE_NATURAL_SIZE): Define.
4067 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
4069 * config/aarch64/aarch64.c: Include cfgrtl.h.
4070 (simd_immediate_info): Add a constructor for series vectors,
4071 and an associated step field.
4072 (aarch64_sve_vg): New variable.
4073 (aarch64_dbx_register_number): Handle VG and the predicate registers.
4074 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
4075 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
4076 (VEC_ANY_DATA, VEC_STRUCT): New constants.
4077 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
4078 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
4079 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
4080 (aarch64_get_mask_mode): New functions.
4081 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
4082 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
4083 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
4084 predicate modes and predicate registers. Explicitly restrict
4085 GPRs to modes of 16 bytes or smaller. Only allow FP registers
4086 to store a vector mode if it is recognized by
4087 aarch64_classify_vector_mode.
4088 (aarch64_regmode_natural_size): New function.
4089 (aarch64_hard_regno_caller_save_mode): Return the original mode
4091 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
4092 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
4093 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
4094 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
4096 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
4097 does not overlap dest if the function is frame-related. Handle
4099 (aarch64_split_add_offset): New function.
4100 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
4101 them aarch64_add_offset.
4102 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
4103 and update call to aarch64_sub_sp.
4104 (aarch64_add_cfa_expression): New function.
4105 (aarch64_expand_prologue): Pass extra temporary registers to the
4106 functions above. Handle the case in which we need to emit new
4107 DW_CFA_expressions for registers that were originally saved
4108 relative to the stack pointer, but now have to be expressed
4109 relative to the frame pointer.
4110 (aarch64_output_mi_thunk): Pass extra temporary registers to the
4112 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
4113 IP0 and IP1 values for SVE frames.
4114 (aarch64_expand_vec_series): New function.
4115 (aarch64_expand_sve_widened_duplicate): Likewise.
4116 (aarch64_expand_sve_const_vector): Likewise.
4117 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
4118 Handle SVE constants. Use emit_move_insn to move a force_const_mem
4119 into the register, rather than emitting a SET directly.
4120 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
4121 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
4122 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
4123 (offset_9bit_signed_scaled_p): New functions.
4124 (aarch64_replicate_bitmask_imm): New function.
4125 (aarch64_bitmask_imm): Use it.
4126 (aarch64_cannot_force_const_mem): Reject expressions involving
4127 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
4128 (aarch64_classify_index): Handle SVE indices, by requiring
4129 a plain register index with a scale that matches the element size.
4130 (aarch64_classify_address): Handle SVE addresses. Assert that
4131 the mode of the address is VOIDmode or an integer mode.
4132 Update call to aarch64_classify_symbol.
4133 (aarch64_classify_symbolic_expression): Update call to
4134 aarch64_classify_symbol.
4135 (aarch64_const_vec_all_in_range_p): New function.
4136 (aarch64_print_vector_float_operand): Likewise.
4137 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
4138 "vN" for FP registers with SVE modes. Handle (const ...) vectors
4139 and the FP immediates 1.0 and 0.5.
4140 (aarch64_print_address_internal): Handle SVE addresses.
4141 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
4142 (aarch64_regno_regclass): Handle predicate registers.
4143 (aarch64_secondary_reload): Handle big-endian reloads of SVE
4145 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
4146 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
4147 (aarch64_convert_sve_vector_bits): New function.
4148 (aarch64_override_options): Use it to handle -msve-vector-bits=.
4149 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4151 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
4152 Handle SVE vector and predicate modes. Accept VL-based constants
4153 that need only one temporary register, and VL offsets that require
4154 no temporary registers.
4155 (aarch64_conditional_register_usage): Mark the predicate registers
4156 as fixed if SVE isn't available.
4157 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
4158 Return true for SVE vector and predicate modes.
4159 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
4160 rather than an unsigned int. Handle SVE modes.
4161 (aarch64_preferred_simd_mode): Update call accordingly. Handle
4163 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
4165 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4166 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4167 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4168 (aarch64_sve_float_mul_immediate_p): New functions.
4169 (aarch64_sve_valid_immediate): New function.
4170 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
4171 Explicitly reject structure modes. Check for INDEX constants.
4172 Handle PTRUE and PFALSE constants.
4173 (aarch64_check_zero_based_sve_index_immediate): New function.
4174 (aarch64_simd_imm_zero_p): Delete.
4175 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
4176 vector modes. Accept constants in the range of CNT[BHWD].
4177 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
4178 ask for an Advanced SIMD mode.
4179 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
4180 (aarch64_simd_vector_alignment): Handle SVE predicates.
4181 (aarch64_vectorize_preferred_vector_alignment): New function.
4182 (aarch64_simd_vector_alignment_reachable): Use it instead of
4184 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
4185 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
4187 (MAX_VECT_LEN): Delete.
4188 (expand_vec_perm_d): Add a vec_flags field.
4189 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
4190 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
4191 (aarch64_evpc_ext): Don't apply a big-endian lane correction
4193 (aarch64_evpc_rev): Rename to...
4194 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
4195 (aarch64_evpc_rev_global): New function.
4196 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
4197 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
4199 (aarch64_evpc_sve_tbl): New function.
4200 (aarch64_expand_vec_perm_const_1): Update after rename of
4201 aarch64_evpc_rev. Handle SVE permutes too, trying
4202 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
4203 than aarch64_evpc_tbl.
4204 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
4205 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
4206 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
4207 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
4208 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
4209 (aarch64_expand_sve_vcond): New functions.
4210 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
4211 of aarch64_vector_mode_p.
4212 (aarch64_dwarf_poly_indeterminate_value): New function.
4213 (aarch64_compute_pressure_classes): Likewise.
4214 (aarch64_can_change_mode_class): Likewise.
4215 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
4216 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
4217 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
4218 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
4219 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
4220 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
4221 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
4222 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
4224 (Dn, Dl, Dr): Accept const as well as const_vector.
4225 (Dz): Likewise. Compare against CONST0_RTX.
4226 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
4227 of "vector" where appropriate.
4228 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
4229 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
4230 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
4231 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
4232 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
4233 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
4234 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
4235 (v_int_equiv): Extend to SVE modes.
4236 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
4238 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
4239 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4240 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4241 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4242 (SVE_COND_FP_CMP): New int iterators.
4243 (perm_hilo): Handle the new unpack unspecs.
4244 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4246 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4247 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4248 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4249 (aarch64_equality_operator, aarch64_constant_vector_operand)
4250 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4251 (aarch64_sve_nonimmediate_operand): Likewise.
4252 (aarch64_sve_general_operand): Likewise.
4253 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4254 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4255 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4256 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4257 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4258 (aarch64_sve_float_arith_immediate): Likewise.
4259 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4260 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4261 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4262 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4263 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4264 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4265 (aarch64_sve_float_arith_operand): Likewise.
4266 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4267 (aarch64_sve_float_mul_operand): Likewise.
4268 (aarch64_sve_vec_perm_operand): Likewise.
4269 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4270 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4271 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4272 as well as const_vector.
4273 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4274 in file. Use CONST0_RTX and CONSTM1_RTX.
4275 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4276 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4277 Use aarch64_simd_imm_zero.
4278 * config/aarch64/aarch64-sve.md: New file.
4279 * config/aarch64/aarch64.md: Include it.
4280 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4281 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4282 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4283 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4284 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4285 (sve): New attribute.
4286 (enabled): Disable instructions with the sve attribute unless
4288 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4289 aarch64_expand_mov_immediate.
4290 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4291 CNT[BHSD] immediates.
4292 (movti): Split CONST_POLY_INT moves into two halves.
4293 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4294 Split additions that need a temporary here if the destination
4295 is the stack pointer.
4296 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4297 (*add<mode>3_poly_1): New instruction.
4298 (set_clobber_cc): New expander.
4300 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4302 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4303 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4304 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4305 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4306 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4307 Change innermode from fixed_mode_size to machine_mode.
4308 (simplify_subreg): Update call accordingly. Handle a constant-sized
4309 subreg of a variable-length CONST_VECTOR.
4311 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4312 Alan Hayward <alan.hayward@arm.com>
4313 David Sherwood <david.sherwood@arm.com>
4315 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4316 (add_offset_to_base): New function, split out from...
4317 (create_mem_ref): ...here. When handling a scale other than 1,
4318 check first whether the address is valid without the offset.
4319 Add it into the base if so, leaving the index and scale as-is.
4321 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4324 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4325 fold_for_warn before checking if arg2 is INTEGER_CST.
4327 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4329 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4330 (store_multiple_operation): Delete.
4331 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4332 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4333 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4334 guarded by TARGET_STRING.
4335 (rs6000_output_load_multiple): Delete.
4336 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4337 OPTION_MASK_STRING / TARGET_STRING handling.
4338 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4339 (const rs6000_opt_masks) <"string">: Change mask to 0.
4340 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4341 (MASK_STRING): Delete.
4342 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4344 (load_multiple): Delete.
4351 (store_multiple): Delete.
4358 (movmemsi_8reg): Delete.
4359 (corresponding unnamed define_insn): Delete.
4360 (movmemsi_6reg): Delete.
4361 (corresponding unnamed define_insn): Delete.
4362 (movmemsi_4reg): Delete.
4363 (corresponding unnamed define_insn): Delete.
4364 (movmemsi_2reg): Delete.
4365 (corresponding unnamed define_insn): Delete.
4366 (movmemsi_1reg): Delete.
4367 (corresponding unnamed define_insn): Delete.
4368 * config/rs6000/rs6000.opt (mno-string): New.
4369 (mstring): Replace by deprecation warning stub.
4370 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4372 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4374 * regrename.c (regrename_do_replace): If replacing the same
4375 reg multiple times, try to reuse last created gen_raw_REG.
4378 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4379 main to workaround a bug in GDB.
4381 2018-01-12 Tom de Vries <tom@codesourcery.com>
4384 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4386 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4388 PR rtl-optimization/80481
4389 * ira-color.c (get_cap_member): New function.
4390 (allocnos_conflict_by_live_ranges_p): Use it.
4391 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4392 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4394 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4397 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4398 (*saddl_se_1): Ditto.
4400 (*ssubl_se_1): Ditto.
4402 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4404 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4405 rather than wi::to_widest for DR_INITs.
4406 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4407 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4408 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4410 (vect_analyze_group_access_1): Note that here.
4412 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4414 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4415 polynomial type sizes.
4417 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4419 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4420 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4421 (gimple_add_tmp_var): Likewise.
4423 2018-01-12 Martin Liska <mliska@suse.cz>
4425 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4426 (gimple_alloc_sizes): Likewise.
4427 (dump_gimple_statistics): Use PRIu64 in printf format.
4428 * gimple.h: Change uint64_t to int.
4430 2018-01-12 Martin Liska <mliska@suse.cz>
4432 * tree-core.h: Use uint64_t instead of int.
4433 * tree.c (tree_node_counts): Likewise.
4434 (tree_node_sizes): Likewise.
4435 (dump_tree_statistics): Use PRIu64 in printf format.
4437 2018-01-12 Martin Liska <mliska@suse.cz>
4439 * Makefile.in: As qsort_chk is implemented in vec.c, add
4440 vec.o to linkage of gencfn-macros.
4441 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4442 passing the info to record_node_allocation_statistics.
4443 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4445 * ggc-common.c (struct ggc_usage): Add operator== and use
4446 it in operator< and compare function.
4447 * mem-stats.h (struct mem_usage): Likewise.
4448 * vec.c (struct vec_usage): Remove operator< and compare
4449 function. Can be simply inherited.
4451 2018-01-12 Martin Jambor <mjambor@suse.cz>
4454 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4455 * tree-ssa-math-opts.c: Include domwalk.h.
4456 (convert_mult_to_fma_1): New function.
4457 (fma_transformation_info): New type.
4458 (fma_deferring_state): Likewise.
4459 (cancel_fma_deferring): New function.
4460 (result_of_phi): Likewise.
4461 (last_fma_candidate_feeds_initial_phi): Likewise.
4462 (convert_mult_to_fma): Added deferring logic, split actual
4463 transformation to convert_mult_to_fma_1.
4464 (math_opts_dom_walker): New type.
4465 (math_opts_dom_walker::after_dom_children): New method, body moved
4466 here from pass_optimize_widening_mul::execute, added deferring logic
4468 (pass_optimize_widening_mul::execute): Moved most of code to
4469 math_opts_dom_walker::after_dom_children.
4470 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4471 * config/i386/i386.c (ix86_option_override_internal): Added
4472 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4474 2018-01-12 Richard Biener <rguenther@suse.de>
4477 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4478 inline instance vars.
4480 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4483 * config/rx/rx.c (rx_is_restricted_memory_address):
4486 2018-01-12 Richard Biener <rguenther@suse.de>
4488 PR tree-optimization/80846
4489 * target.def (split_reduction): New target hook.
4490 * targhooks.c (default_split_reduction): New function.
4491 * targhooks.h (default_split_reduction): Declare.
4492 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4493 target requests first reduce vectors by combining low and high
4495 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4496 (get_vectype_for_scalar_type_and_size): Export.
4497 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4498 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4499 * doc/tm.texi: Regenerate.
4500 * config/i386/i386.c (ix86_split_reduction): Implement
4501 TARGET_VECTORIZE_SPLIT_REDUCTION.
4503 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4506 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4507 in PIC mode except for TARGET_VXWORKS_RTP.
4508 * config/sparc/sparc.c: Include cfgrtl.h.
4509 (TARGET_INIT_PIC_REG): Define.
4510 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4511 (sparc_pic_register_p): New predicate.
4512 (sparc_legitimate_address_p): Use it.
4513 (sparc_legitimize_pic_address): Likewise.
4514 (sparc_delegitimize_address): Likewise.
4515 (sparc_mode_dependent_address_p): Likewise.
4516 (gen_load_pcrel_sym): Remove 4th parameter.
4517 (load_got_register): Adjust call to above. Remove obsolete stuff.
4518 (sparc_expand_prologue): Do not call load_got_register here.
4519 (sparc_flat_expand_prologue): Likewise.
4520 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4521 (sparc_use_pseudo_pic_reg): New function.
4522 (sparc_init_pic_reg): Likewise.
4523 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4524 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4526 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4528 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4529 Add item for branch_cost.
4531 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4533 PR rtl-optimization/83565
4534 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4535 not extend the result to a larger mode for rotate operations.
4536 (num_sign_bit_copies1): Likewise.
4538 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4541 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4543 Use values-Xc.o for -pedantic.
4544 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4546 2018-01-12 Martin Liska <mliska@suse.cz>
4549 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4551 (possible_polymorphic_call_targets): Use it.
4552 (ipa_devirt): Likewise.
4554 2018-01-12 Martin Liska <mliska@suse.cz>
4556 * profile-count.h (enum profile_quality): Use 0 as invalid
4557 enum value of profile_quality.
4559 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4561 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4562 -mext-string options.
4564 2018-01-12 Richard Biener <rguenther@suse.de>
4566 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4567 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4568 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4570 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4572 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4574 * configure.ac (--with-long-double-format): Add support for the
4575 configuration option to change the default long double format on
4577 * config.gcc (powerpc*-linux*-*): Likewise.
4578 * configure: Regenerate.
4579 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4580 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4581 used without modification.
4583 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4585 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4586 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4587 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4588 MISC_BUILTIN_SPEC_BARRIER.
4589 (rs6000_init_builtins): Likewise.
4590 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4592 (speculation_barrier): New define_insn.
4593 * doc/extend.texi: Document __builtin_speculation_barrier.
4595 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4598 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4599 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4600 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4602 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4603 integral modes instead of "ss" and "sd".
4604 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4605 vectors with 32-bit and 64-bit elements.
4606 (vecdupssescalarmodesuffix): New mode attribute.
4607 (vec_dup<mode>): Use it.
4609 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4612 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4613 frame if argument is passed on stack.
4615 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4618 * ree.c (combine_reaching_defs): Optimize also
4619 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4620 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4622 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4625 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4627 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4630 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4631 after they are computed.
4633 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4635 PR tree-optimization/83695
4636 * gimple-loop-linterchange.cc
4637 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4638 reset cached scev information after interchange.
4639 (pass_linterchange::execute): Remove call to scev_reset_htab.
4641 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4643 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4644 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4645 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4646 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4647 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4648 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4649 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4650 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4651 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4652 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4653 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4654 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4655 (V_lane_reg): Likewise.
4656 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4658 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4659 (vfmal_lane_low<mode>_intrinsic,
4660 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4661 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4662 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4663 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4664 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4665 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4667 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4669 * config/arm/arm-cpus.in (fp16fml): New feature.
4670 (ALL_SIMD): Add fp16fml.
4671 (armv8.2-a): Add fp16fml as an option.
4672 (armv8.3-a): Likewise.
4673 (armv8.4-a): Add fp16fml as part of fp16.
4674 * config/arm/arm.h (TARGET_FP16FML): Define.
4675 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4677 * config/arm/arm-modes.def (V2HF): Define.
4678 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4679 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4680 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4681 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4682 vfmsl_low, vfmsl_high): New set of builtins.
4683 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4684 (vfml_op): New code attribute.
4685 (VFMLHALVES): New int iterator.
4686 (VFML, VFMLSEL): New mode attributes.
4687 (V_reg): Define mapping for V2HF.
4688 (V_hi, V_lo): New mode attributes.
4689 (VF_constraint): Likewise.
4690 (vfml_half, vfml_half_selector): New int attributes.
4691 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4693 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4694 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4696 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4697 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4698 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4699 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4701 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4702 Document new effective target and option set.
4704 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4706 * config/arm/arm-cpus.in (armv8_4): New feature.
4707 (ARMv8_4a): New fgroup.
4708 (armv8.4-a): New arch.
4709 * config/arm/arm-tables.opt: Regenerate.
4710 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4711 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4712 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4713 Add matching rules for -march=armv8.4-a and extensions.
4714 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4716 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4719 * config/rx/rx.md (BW): New mode attribute.
4720 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4722 2018-01-11 Richard Biener <rguenther@suse.de>
4724 PR tree-optimization/83435
4725 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4726 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4727 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4729 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4730 Alan Hayward <alan.hayward@arm.com>
4731 David Sherwood <david.sherwood@arm.com>
4733 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4735 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4736 (aarch64_print_address_internal): Use it to check for a zero offset.
4738 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4739 Alan Hayward <alan.hayward@arm.com>
4740 David Sherwood <david.sherwood@arm.com>
4742 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4743 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4744 Return a poly_int64 rather than a HOST_WIDE_INT.
4745 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4746 rather than a HOST_WIDE_INT.
4747 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4748 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4749 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4750 final_offset from HOST_WIDE_INT to poly_int64.
4751 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4752 to_constant when getting the number of units in an Advanced SIMD
4754 (aarch64_builtin_vectorized_function): Check for a constant number
4756 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4758 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4759 attribute instead of GET_MODE_NUNITS.
4760 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4761 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4762 GET_MODE_SIZE for fixed-size registers.
4763 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4764 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4765 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4766 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4767 (aarch64_print_operand, aarch64_print_address_internal)
4768 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4769 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4770 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4771 Handle polynomial GET_MODE_SIZE.
4772 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4773 wider than SImode without modification.
4774 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4775 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4776 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4777 passing and returning SVE modes.
4778 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4779 rather than GEN_INT.
4780 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4781 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4782 (aarch64_allocate_and_probe_stack_space): Likewise.
4783 (aarch64_layout_frame): Cope with polynomial offsets.
4784 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4785 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4787 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4788 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4789 poly_int64 rather than a HOST_WIDE_INT.
4790 (aarch64_get_separate_components, aarch64_process_components)
4791 (aarch64_expand_prologue, aarch64_expand_epilogue)
4792 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4793 (aarch64_anchor_offset): New function, split out from...
4794 (aarch64_legitimize_address): ...here.
4795 (aarch64_builtin_vectorization_cost): Handle polynomial
4796 TYPE_VECTOR_SUBPARTS.
4797 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4799 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4800 number of elements from the PARALLEL rather than the mode.
4801 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4802 rather than GET_MODE_BITSIZE.
4803 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4804 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4805 (aarch64_expand_vec_perm_const_1): Handle polynomial
4806 d->perm.length () and d->perm elements.
4807 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4808 Apply to_constant to d->perm elements.
4809 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4810 polynomial CONST_VECTOR_NUNITS.
4811 (aarch64_move_pointer): Take amount as a poly_int64 rather
4813 (aarch64_progress_pointer): Avoid temporary variable.
4814 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4815 the mode attribute instead of GET_MODE.
4817 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4818 Alan Hayward <alan.hayward@arm.com>
4819 David Sherwood <david.sherwood@arm.com>
4821 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4822 x exists before using it.
4823 (aarch64_add_constant_internal): Rename to...
4824 (aarch64_add_offset_1): ...this. Replace regnum with separate
4825 src and dest rtxes. Handle the case in which they're different,
4826 including when the offset is zero. Replace scratchreg with an rtx.
4827 Use 2 additions if there is no spare register into which we can
4828 move a 16-bit constant.
4829 (aarch64_add_constant): Delete.
4830 (aarch64_add_offset): Replace reg with separate src and dest
4831 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4832 Use aarch64_add_offset_1.
4833 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4834 an rtx rather than an int. Take the delta as a poly_int64
4835 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4836 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4837 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4838 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4839 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4841 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4842 aarch64_add_constant.
4844 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4846 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4847 Use scalar_float_mode.
4849 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4851 * config/aarch64/aarch64-simd.md
4852 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4853 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4854 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4855 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4856 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4857 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4858 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4859 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4860 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4861 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4863 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4866 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4867 targ_options->x_arm_arch_string is non NULL.
4869 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4871 * config/aarch64/aarch64.h
4872 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4874 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4877 * expmed.c (emit_store_flag_force): Swap if const op0
4878 and change VOIDmode to mode of op0.
4880 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4882 PR rtl-optimization/83761
4883 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4884 than bytes to mode_for_size.
4886 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4889 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4890 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4893 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4896 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4897 when in layout mode.
4898 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4899 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4902 2018-01-10 Michael Collison <michael.collison@arm.com>
4904 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4905 * config/aarch64/aarch64-option-extension.def: Add
4906 AARCH64_OPT_EXTENSION of 'fp16fml'.
4907 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4908 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4909 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4910 * config/aarch64/constraints.md (Ui7): New constraint.
4911 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4912 (VFMLA_SEL_W): Ditto.
4915 (VFMLA16_LOW): New int iterator.
4916 (VFMLA16_HIGH): Ditto.
4917 (UNSPEC_FMLAL): New unspec.
4918 (UNSPEC_FMLSL): Ditto.
4919 (UNSPEC_FMLAL2): Ditto.
4920 (UNSPEC_FMLSL2): Ditto.
4921 (f16mac): New code attribute.
4922 * config/aarch64/aarch64-simd-builtins.def
4923 (aarch64_fmlal_lowv2sf): Ditto.
4924 (aarch64_fmlsl_lowv2sf): Ditto.
4925 (aarch64_fmlalq_lowv4sf): Ditto.
4926 (aarch64_fmlslq_lowv4sf): Ditto.
4927 (aarch64_fmlal_highv2sf): Ditto.
4928 (aarch64_fmlsl_highv2sf): Ditto.
4929 (aarch64_fmlalq_highv4sf): Ditto.
4930 (aarch64_fmlslq_highv4sf): Ditto.
4931 (aarch64_fmlal_lane_lowv2sf): Ditto.
4932 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4933 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4934 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4935 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4936 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4937 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4938 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4939 (aarch64_fmlal_lane_highv2sf): Ditto.
4940 (aarch64_fmlsl_lane_highv2sf): Ditto.
4941 (aarch64_fmlal_laneq_highv2sf): Ditto.
4942 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4943 (aarch64_fmlalq_lane_highv4sf): Ditto.
4944 (aarch64_fmlsl_lane_highv4sf): Ditto.
4945 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4946 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4947 * config/aarch64/aarch64-simd.md:
4948 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4949 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4950 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4951 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4952 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4953 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4954 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4955 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4956 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4957 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4958 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4959 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4960 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4961 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4962 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4963 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4964 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4965 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4966 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4967 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4968 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4969 (vfmlsl_low_u32): Ditto.
4970 (vfmlalq_low_u32): Ditto.
4971 (vfmlslq_low_u32): Ditto.
4972 (vfmlal_high_u32): Ditto.
4973 (vfmlsl_high_u32): Ditto.
4974 (vfmlalq_high_u32): Ditto.
4975 (vfmlslq_high_u32): Ditto.
4976 (vfmlal_lane_low_u32): Ditto.
4977 (vfmlsl_lane_low_u32): Ditto.
4978 (vfmlal_laneq_low_u32): Ditto.
4979 (vfmlsl_laneq_low_u32): Ditto.
4980 (vfmlalq_lane_low_u32): Ditto.
4981 (vfmlslq_lane_low_u32): Ditto.
4982 (vfmlalq_laneq_low_u32): Ditto.
4983 (vfmlslq_laneq_low_u32): Ditto.
4984 (vfmlal_lane_high_u32): Ditto.
4985 (vfmlsl_lane_high_u32): Ditto.
4986 (vfmlal_laneq_high_u32): Ditto.
4987 (vfmlsl_laneq_high_u32): Ditto.
4988 (vfmlalq_lane_high_u32): Ditto.
4989 (vfmlslq_lane_high_u32): Ditto.
4990 (vfmlalq_laneq_high_u32): Ditto.
4991 (vfmlslq_laneq_high_u32): Ditto.
4992 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
4993 (AARCH64_FL_FOR_ARCH8_4): New.
4994 (AARCH64_ISA_F16FML): New ISA flag.
4995 (TARGET_F16FML): New feature flag for fp16fml.
4996 (doc/invoke.texi): Document new fp16fml option.
4998 2018-01-10 Michael Collison <michael.collison@arm.com>
5000 * config/aarch64/aarch64-builtins.c:
5001 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
5002 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5003 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
5004 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
5005 (AARCH64_ISA_SHA3): New ISA flag.
5006 (TARGET_SHA3): New feature flag for sha3.
5007 * config/aarch64/iterators.md (sha512_op): New int attribute.
5008 (CRYPTO_SHA512): New int iterator.
5009 (UNSPEC_SHA512H): New unspec.
5010 (UNSPEC_SHA512H2): Ditto.
5011 (UNSPEC_SHA512SU0): Ditto.
5012 (UNSPEC_SHA512SU1): Ditto.
5013 * config/aarch64/aarch64-simd-builtins.def
5014 (aarch64_crypto_sha512hqv2di): New builtin.
5015 (aarch64_crypto_sha512h2qv2di): Ditto.
5016 (aarch64_crypto_sha512su0qv2di): Ditto.
5017 (aarch64_crypto_sha512su1qv2di): Ditto.
5018 (aarch64_eor3qv8hi): Ditto.
5019 (aarch64_rax1qv2di): Ditto.
5020 (aarch64_xarqv2di): Ditto.
5021 (aarch64_bcaxqv8hi): Ditto.
5022 * config/aarch64/aarch64-simd.md:
5023 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
5024 (aarch64_crypto_sha512su0qv2di): Ditto.
5025 (aarch64_crypto_sha512su1qv2di): Ditto.
5026 (aarch64_eor3qv8hi): Ditto.
5027 (aarch64_rax1qv2di): Ditto.
5028 (aarch64_xarqv2di): Ditto.
5029 (aarch64_bcaxqv8hi): Ditto.
5030 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
5031 (vsha512h2q_u64): Ditto.
5032 (vsha512su0q_u64): Ditto.
5033 (vsha512su1q_u64): Ditto.
5034 (veor3q_u16): Ditto.
5035 (vrax1q_u64): Ditto.
5037 (vbcaxq_u16): Ditto.
5038 * config/arm/types.md (crypto_sha512): New type attribute.
5039 (crypto_sha3): Ditto.
5040 (doc/invoke.texi): Document new sha3 option.
5042 2018-01-10 Michael Collison <michael.collison@arm.com>
5044 * config/aarch64/aarch64-builtins.c:
5045 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
5046 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5047 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
5048 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
5049 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
5050 (AARCH64_ISA_SM4): New ISA flag.
5051 (TARGET_SM4): New feature flag for sm4.
5052 * config/aarch64/aarch64-simd-builtins.def
5053 (aarch64_sm3ss1qv4si): Ditto.
5054 (aarch64_sm3tt1aq4si): Ditto.
5055 (aarch64_sm3tt1bq4si): Ditto.
5056 (aarch64_sm3tt2aq4si): Ditto.
5057 (aarch64_sm3tt2bq4si): Ditto.
5058 (aarch64_sm3partw1qv4si): Ditto.
5059 (aarch64_sm3partw2qv4si): Ditto.
5060 (aarch64_sm4eqv4si): Ditto.
5061 (aarch64_sm4ekeyqv4si): Ditto.
5062 * config/aarch64/aarch64-simd.md:
5063 (aarch64_sm3ss1qv4si): Ditto.
5064 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
5065 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
5066 (aarch64_sm4eqv4si): Ditto.
5067 (aarch64_sm4ekeyqv4si): Ditto.
5068 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
5069 (sm3part_op): Ditto.
5070 (CRYPTO_SM3TT): Ditto.
5071 (CRYPTO_SM3PART): Ditto.
5072 (UNSPEC_SM3SS1): New unspec.
5073 (UNSPEC_SM3TT1A): Ditto.
5074 (UNSPEC_SM3TT1B): Ditto.
5075 (UNSPEC_SM3TT2A): Ditto.
5076 (UNSPEC_SM3TT2B): Ditto.
5077 (UNSPEC_SM3PARTW1): Ditto.
5078 (UNSPEC_SM3PARTW2): Ditto.
5079 (UNSPEC_SM4E): Ditto.
5080 (UNSPEC_SM4EKEY): Ditto.
5081 * config/aarch64/constraints.md (Ui2): New constraint.
5082 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
5083 * config/arm/types.md (crypto_sm3): New type attribute.
5084 (crypto_sm4): Ditto.
5085 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
5086 (vsm3tt1aq_u32): Ditto.
5087 (vsm3tt1bq_u32): Ditto.
5088 (vsm3tt2aq_u32): Ditto.
5089 (vsm3tt2bq_u32): Ditto.
5090 (vsm3partw1q_u32): Ditto.
5091 (vsm3partw2q_u32): Ditto.
5092 (vsm4eq_u32): Ditto.
5093 (vsm4ekeyq_u32): Ditto.
5094 (doc/invoke.texi): Document new sm4 option.
5096 2018-01-10 Michael Collison <michael.collison@arm.com>
5098 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
5099 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
5100 (AARCH64_FL_FOR_ARCH8_4): New.
5101 (AARCH64_FL_V8_4): New flag.
5102 (doc/invoke.texi): Document new armv8.4-a option.
5104 2018-01-10 Michael Collison <michael.collison@arm.com>
5106 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5107 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
5108 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
5109 * config/aarch64/aarch64-option-extension.def: Add
5110 AARCH64_OPT_EXTENSION of 'sha2'.
5111 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
5112 (crypto): Disable sha2 and aes if crypto disabled.
5113 (crypto): Enable aes and sha2 if enabled.
5114 (simd): Disable sha2 and aes if simd disabled.
5115 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
5117 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
5118 (TARGET_SHA2): New feature flag for sha2.
5119 (TARGET_AES): New feature flag for aes.
5120 * config/aarch64/aarch64-simd.md:
5121 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
5122 conditional on TARGET_AES.
5123 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
5124 (aarch64_crypto_sha1hsi): Make pattern conditional
5126 (aarch64_crypto_sha1hv4si): Ditto.
5127 (aarch64_be_crypto_sha1hv4si): Ditto.
5128 (aarch64_crypto_sha1su1v4si): Ditto.
5129 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
5130 (aarch64_crypto_sha1su0v4si): Ditto.
5131 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
5132 (aarch64_crypto_sha256su0v4si): Ditto.
5133 (aarch64_crypto_sha256su1v4si): Ditto.
5134 (doc/invoke.texi): Document new aes and sha2 options.
5136 2018-01-10 Martin Sebor <msebor@redhat.com>
5138 PR tree-optimization/83781
5139 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
5142 2018-01-11 Martin Sebor <msebor@gmail.com>
5143 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5145 PR tree-optimization/83501
5146 PR tree-optimization/81703
5148 * tree-ssa-strlen.c (get_string_cst): Rename...
5149 (get_string_len): ...to this. Handle global constants.
5150 (handle_char_store): Adjust.
5152 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
5153 Jim Wilson <jimw@sifive.com>
5155 * config/riscv/riscv-protos.h (riscv_output_return): New.
5156 * config/riscv/riscv.c (struct machine_function): New naked_p field.
5157 (riscv_attribute_table, riscv_output_return),
5158 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
5159 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
5160 (riscv_compute_frame_info): Only compute frame->mask if not a naked
5162 (riscv_expand_prologue): Add early return for naked function.
5163 (riscv_expand_epilogue): Likewise.
5164 (riscv_function_ok_for_sibcall): Return false for naked function.
5165 (riscv_set_current_function): New.
5166 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
5167 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
5168 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
5169 * doc/extend.texi (RISC-V Function Attributes): New.
5171 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
5173 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
5174 check for 128-bit long double before checking TCmode.
5175 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
5176 128-bit long doubles before checking TFmode or TCmode.
5177 (FLOAT128_IBM_P): Likewise.
5179 2018-01-10 Martin Sebor <msebor@redhat.com>
5181 PR tree-optimization/83671
5182 * builtins.c (c_strlen): Unconditionally return zero for the empty
5184 Use -Warray-bounds for warnings.
5185 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
5186 for non-constant array indices with COMPONENT_REF, arrays of
5187 arrays, and pointers to arrays.
5188 (gimple_fold_builtin_strlen): Determine and set length range for
5189 non-constant character arrays.
5191 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
5194 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
5197 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
5199 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
5201 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5204 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
5205 VECTOR_MEM_ALTIVEC_OR_VSX_P.
5206 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
5207 indexed_or_indirect_operand predicate.
5208 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
5209 (*vsx_le_perm_load_v8hi): Likewise.
5210 (*vsx_le_perm_load_v16qi): Likewise.
5211 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
5212 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
5213 (*vsx_le_perm_store_v8hi): Likewise.
5214 (*vsx_le_perm_store_v16qi): Likewise.
5215 (eight unnamed splitters): Likewise.
5217 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5219 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
5220 * config/rs6000/emmintrin.h: Likewise.
5221 * config/rs6000/mmintrin.h: Likewise.
5222 * config/rs6000/xmmintrin.h: Likewise.
5224 2018-01-10 David Malcolm <dmalcolm@redhat.com>
5227 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
5229 * tree.c (tree_nop_conversion): Return true for location wrapper
5231 (maybe_wrap_with_location): New function.
5232 (selftest::check_strip_nops): New function.
5233 (selftest::test_location_wrappers): New function.
5234 (selftest::tree_c_tests): Call it.
5235 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
5236 (maybe_wrap_with_location): New decl.
5237 (EXPR_LOCATION_WRAPPER_P): New macro.
5238 (location_wrapper_p): New inline function.
5239 (tree_strip_any_location_wrapper): New inline function.
5241 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5244 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5245 stack_realign_offset for the largest alignment of stack slot
5247 (ix86_find_max_used_stack_alignment): New function.
5248 (ix86_finalize_stack_frame_flags): Use it. Set
5249 max_used_stack_alignment if we don't realign stack.
5250 * config/i386/i386.h (machine_function): Add
5251 max_used_stack_alignment.
5253 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5255 * config/arm/arm.opt (-mbranch-cost): New option.
5256 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5259 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5262 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5263 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5265 2018-01-10 Richard Biener <rguenther@suse.de>
5268 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5269 early out so it also covers the case where we have a non-NULL
5272 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5274 PR tree-optimization/83753
5275 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5276 for non-strided grouped accesses if the number of elements is 1.
5278 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5281 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5282 * i386.h (TARGET_USE_GATHER): Define.
5283 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5285 2018-01-10 Martin Liska <mliska@suse.cz>
5288 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5289 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5291 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5292 CLEANUP_NO_PARTITIONING is not set.
5294 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5296 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5297 for vectors, as a partial revert of r254296.
5298 * rtl.h (const_vec_p): Delete.
5299 (const_vec_duplicate_p): Don't test for vector CONSTs.
5300 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5301 * expmed.c (make_tree): Likewise.
5304 * common.md (E, F): Use CONSTANT_P instead of checking for
5306 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5307 checking for CONST_VECTOR.
5309 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5312 * predict.c (force_edge_cold): Handle in more sane way edges
5315 2018-01-09 Carl Love <cel@us.ibm.com>
5317 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5319 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5320 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5321 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5322 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5323 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5324 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5325 * config/rs6000/rs6000-protos.h: Add extern defition for
5326 rs6000_generate_float2_double_code.
5327 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5329 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5330 (float2_v2df): Add define_expand.
5332 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5335 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5336 op_mode in the force_to_mode call.
5338 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5340 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5341 instead of checking each element individually.
5342 (aarch64_evpc_uzp): Likewise.
5343 (aarch64_evpc_zip): Likewise.
5344 (aarch64_evpc_ext): Likewise.
5345 (aarch64_evpc_rev): Likewise.
5346 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5347 instead of checking each element individually. Return true without
5349 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5350 whether all selected elements come from the same input, instead of
5351 checking each element individually. Remove calls to gen_rtx_REG,
5352 start_sequence and end_sequence and instead assert that no rtl is
5355 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5357 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5358 order of HIGH and CONST checks.
5360 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5362 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5363 if the destination isn't an SSA_NAME.
5365 2018-01-09 Richard Biener <rguenther@suse.de>
5367 PR tree-optimization/83668
5368 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5370 (canonicalize_loop_form): ... here, renamed from ...
5371 (canonicalize_loop_closed_ssa_form): ... this and amended to
5372 swap successor edges for loop exit blocks to make us use
5373 the RPO order we need for initial schedule generation.
5375 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5377 PR tree-optimization/64811
5378 * match.pd: When optimizing comparisons with Inf, avoid
5379 introducing or losing exceptions from comparisons with NaN.
5381 2018-01-09 Martin Liska <mliska@suse.cz>
5384 * asan.c (shadow_mem_size): Add gcc_assert.
5386 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5388 Don't save registers in main().
5391 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5392 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5393 * config/avr/avr.c (avr_set_current_function): Don't error if
5394 naked, OS_task or OS_main are specified at the same time.
5395 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5397 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5399 * common/config/avr/avr-common.c (avr_option_optimization_table):
5400 Switch on -mmain-is-OS_task for optimizing compilations.
5402 2018-01-09 Richard Biener <rguenther@suse.de>
5404 PR tree-optimization/83572
5405 * graphite.c: Include cfganal.h.
5406 (graphite_transform_loops): Connect infinite loops to exit
5407 and remove fake edges at the end.
5409 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5411 * ipa-inline.c (edge_badness): Revert accidental checkin.
5413 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5416 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5417 symbols; not inline clones.
5419 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5422 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5423 hard registers. Formatting fixes.
5425 PR preprocessor/83722
5426 * gcc.c (try_generate_repro): Pass
5427 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5428 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5431 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5432 Kito Cheng <kito.cheng@gmail.com>
5434 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5435 (riscv_leaf_function_p): Delete.
5436 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5438 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5440 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5442 (do_ifelse): New function.
5443 (do_isel): New function.
5444 (do_sub3): New function.
5445 (do_add3): New function.
5446 (do_load_mask_compare): New function.
5447 (do_overlap_load_compare): New function.
5448 (expand_compare_loop): New function.
5449 (expand_block_compare): Call expand_compare_loop() when appropriate.
5450 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5452 (-mblock-compare-inline-loop-limit): New option.
5454 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5457 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5458 Reverse order of second and third operands in first alternative.
5459 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5460 of first and second elements in UNSPEC_VPERMR vector.
5461 (altivec_expand_vec_perm_le): Likewise.
5463 2018-01-08 Jeff Law <law@redhat.com>
5465 PR rtl-optimizatin/81308
5466 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5467 (process_switch): If group_case_labels makes a change, then set
5469 (pass_convert_switch::execute): If a switch is converted, then
5470 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5472 PR rtl-optimization/81308
5473 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5476 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5478 PR target/83663 - Revert r255946
5479 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5480 generation for cases where splatting a value is not useful.
5481 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5482 across a vec_duplicate and a paradoxical subreg forming a vector
5483 mode to a vec_concat.
5485 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5487 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5488 -march=armv8.3-a variants.
5489 * config/arm/t-multilib: Likewise.
5490 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5492 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5494 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5496 (cceq_ior_compare_complement): Give it a name so I can use it, and
5497 change boolean_or_operator predicate to boolean_operator so it can
5498 be used to generate a crand.
5499 (eqne): New code iterator.
5500 (bd/bd_neg): New code_attrs.
5501 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5502 a single define_insn.
5503 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5504 decrement (bdnzt/bdnzf/bdzt/bdzf).
5505 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5506 with the new names of the branch decrement patterns, and added the
5507 names of the branch decrement conditional patterns.
5509 2018-01-08 Richard Biener <rguenther@suse.de>
5511 PR tree-optimization/83563
5512 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5515 2018-01-08 Richard Biener <rguenther@suse.de>
5518 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5520 2018-01-08 Richard Biener <rguenther@suse.de>
5522 PR tree-optimization/83685
5523 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5524 references to abnormals.
5526 2018-01-08 Richard Biener <rguenther@suse.de>
5529 * dwarf2out.c (output_indirect_strings): Handle empty
5530 skeleton_debug_str_hash.
5531 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5533 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5535 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5536 (emit_store_direct): Likewise.
5537 (arc_trampoline_adjust_address): Likewise.
5538 (arc_asm_trampoline_template): New function.
5539 (arc_initialize_trampoline): Use asm_trampoline_template.
5540 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5541 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5542 * config/arc/arc.md (flush_icache): Delete pattern.
5544 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5546 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5547 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5550 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5553 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5554 by not USED_FOR_TARGET.
5555 (make_pass_resolve_sw_modes): Likewise.
5557 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5559 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5562 2018-01-08 Richard Biener <rguenther@suse.de>
5565 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5567 2018-01-08 Richard Biener <rguenther@suse.de>
5570 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5572 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5575 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5576 basic blocks with a small number of successors.
5577 (convert_control_dep_chain_into_preds): Improve handling of
5579 (dump_predicates): Split apart into...
5580 (dump_pred_chain): ...here...
5581 (dump_pred_info): ...and here.
5582 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5583 (can_chain_union_be_invalidated_p): Improve check for invalidation
5585 (uninit_uses_cannot_happen): Avoid unnecessary if
5586 convert_control_dep_chain_into_preds yielded nothing.
5588 2018-01-06 Martin Sebor <msebor@redhat.com>
5590 PR tree-optimization/83640
5591 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5592 subtracting negative offset from size.
5593 (builtin_access::overlap): Adjust offset bounds of the access to fall
5594 within the size of the object if possible.
5596 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5598 PR rtl-optimization/83699
5599 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5600 extract_bit_field_as_subreg to cases in which the extracted
5601 value is also a vector.
5603 * lra-constraints.c (process_alt_operands): Test for the equivalence
5604 substitutions when detecting a possible reload cycle.
5606 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5609 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5610 by default if flag_selective_schedling{,2}. Formatting fixes.
5612 PR rtl-optimization/83682
5613 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5614 if it has non-VECTOR_MODE element mode.
5615 (vec_duplicate_p): Likewise.
5618 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5619 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5621 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5624 * config/i386/i386-builtin.def
5625 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5626 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5627 Require also OPTION_MASK_ISA_AVX512F in addition to
5628 OPTION_MASK_ISA_GFNI.
5629 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5630 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5631 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5632 to OPTION_MASK_ISA_GFNI.
5633 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5634 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5635 OPTION_MASK_ISA_AVX512BW.
5636 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5637 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5638 addition to OPTION_MASK_ISA_GFNI.
5639 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5640 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5641 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5642 to OPTION_MASK_ISA_GFNI.
5643 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5644 a requirement for all ISAs rather than any of them with a few
5646 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5648 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5649 bitmasks to be enabled with 3 exceptions, instead of requiring any
5650 enabled ISA with lots of exceptions.
5651 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5652 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5653 Change avx512bw in isa attribute to avx512f.
5654 * config/i386/sgxintrin.h: Add license boilerplate.
5655 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5656 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5657 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5658 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5660 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5661 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5662 temporarily sse2 rather than sse if not enabled already.
5665 * config/i386/sse.md (VI248_VLBW): Rename to ...
5666 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5667 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5668 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5669 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5670 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5671 mode iterator instead of VI248_VLBW.
5673 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5675 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5676 (record_modified): Skip clobbers; add debug output.
5677 (param_change_prob): Use sreal frequencies.
5679 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5681 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5682 punt for user-aligned variables.
5684 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5686 * tree-chrec.c (chrec_contains_symbols): Return true for
5689 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5692 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5693 of (x|y) == x for BICS pattern.
5695 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5697 PR tree-optimization/83605
5698 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5699 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5702 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5704 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5705 * config/epiphany/rtems.h: New file.
5707 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5708 Uros Bizjak <ubizjak@gmail.com>
5711 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5712 QIreg_operand instead of register_operand predicate.
5713 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5714 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5715 comments instead of -fmitigate[-_]rop.
5717 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5720 * cgraphunit.c (symbol_table::compile): Switch to text_section
5721 before calling assembly_start debug hook.
5722 * run-rtl-passes.c (run_rtl_passes): Likewise.
5725 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5727 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5728 range_int_cst_p rather than !symbolic_range_p before calling
5729 extract_range_from_multiplicative_op_1.
5731 2018-01-04 Jeff Law <law@redhat.com>
5733 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5734 redundant test in assertion.
5736 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5738 * doc/rtl.texi: Document machine_mode wrapper classes.
5740 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5742 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5745 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5747 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5748 the VEC_PERM_EXPR fold to fail.
5750 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5753 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5754 to switched_sections.
5756 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5759 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5762 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5765 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5766 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5768 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5771 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5772 is BLKmode and bitpos not zero or mode change is needed.
5774 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5777 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5780 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5783 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5784 instead of MULT rtx. Update all corresponding splitters.
5786 (*ssub<modesuffix>): Ditto.
5788 (*cmp_sadd_di): Update split patterns.
5789 (*cmp_sadd_si): Ditto.
5790 (*cmp_sadd_sidi): Ditto.
5791 (*cmp_ssub_di): Ditto.
5792 (*cmp_ssub_si): Ditto.
5793 (*cmp_ssub_sidi): Ditto.
5794 * config/alpha/predicates.md (const23_operand): New predicate.
5795 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5796 Look for ASHIFT, not MULT inner operand.
5797 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5799 2018-01-04 Martin Liska <mliska@suse.cz>
5801 PR gcov-profile/83669
5802 * gcov.c (output_intermediate_file): Add version to intermediate
5804 * doc/gcov.texi: Document new field 'version' in intermediate
5805 file format. Fix location of '-k' option of gcov command.
5807 2018-01-04 Martin Liska <mliska@suse.cz>
5810 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5812 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5814 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5816 2018-01-03 Martin Sebor <msebor@redhat.com>
5818 PR tree-optimization/83655
5819 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5820 checking calls with invalid arguments.
5822 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5824 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5825 (vectorizable_mask_load_store): Delete.
5826 (vectorizable_call): Return false for masked loads and stores.
5827 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5828 instead of gimple_assign_rhs1.
5829 (vectorizable_load): Handle IFN_MASK_LOAD.
5830 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5832 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5834 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5836 (vectorizable_mask_load_store): ...here.
5837 (vectorizable_load): ...and here.
5839 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5841 * tree-vect-stmts.c (vect_build_all_ones_mask)
5842 (vect_build_zero_merge_argument): New functions, split out from...
5843 (vectorizable_load): ...here.
5845 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5847 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5849 (vectorizable_mask_load_store): ...here.
5850 (vectorizable_store): ...and here.
5852 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5854 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5856 (vectorizable_mask_load_store): ...here.
5858 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5860 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5861 (vect_model_store_cost): Take a vec_load_store_type instead of a
5863 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5864 (vect_model_store_cost): Take a vec_load_store_type instead of a
5866 (vectorizable_mask_load_store): Update accordingly.
5867 (vectorizable_store): Likewise.
5868 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5870 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5872 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5873 IFN_MASK_LOAD calls here rather than...
5874 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5876 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5877 Alan Hayward <alan.hayward@arm.com>
5878 David Sherwood <david.sherwood@arm.com>
5880 * expmed.c (extract_bit_field_1): For vector extracts,
5881 fall back to extract_bit_field_as_subreg if vec_extract
5884 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5885 Alan Hayward <alan.hayward@arm.com>
5886 David Sherwood <david.sherwood@arm.com>
5888 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5889 they are variable or constant sized.
5890 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5891 slots for constant-sized data.
5893 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5894 Alan Hayward <alan.hayward@arm.com>
5895 David Sherwood <david.sherwood@arm.com>
5897 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5898 handling COND_EXPRs with boolean comparisons, try to find a better
5899 basis for the mask type than the boolean itself.
5901 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5903 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5904 is calculated and how it can be overridden.
5905 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5906 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5908 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5911 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5912 Alan Hayward <alan.hayward@arm.com>
5913 David Sherwood <david.sherwood@arm.com>
5915 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5916 Remove the mode argument.
5917 (aarch64_simd_valid_immediate): Remove the mode and inverse
5919 * config/aarch64/iterators.md (bitsize): New iterator.
5920 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5921 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5922 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5923 aarch64_simd_valid_immediate.
5924 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5925 (aarch64_reg_or_bic_imm): Likewise.
5926 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5927 with an insn_type enum and msl with a modifier_type enum.
5928 Replace element_width with a scalar_mode. Change the shift
5929 to unsigned int. Add constructors for scalar_float_mode and
5930 scalar_int_mode elements.
5931 (aarch64_vect_float_const_representable_p): Delete.
5932 (aarch64_can_const_movi_rtx_p)
5933 (aarch64_simd_scalar_immediate_valid_for_move)
5934 (aarch64_simd_make_constant): Update call to
5935 aarch64_simd_valid_immediate.
5936 (aarch64_advsimd_valid_immediate_hs): New function.
5937 (aarch64_advsimd_valid_immediate): Likewise.
5938 (aarch64_simd_valid_immediate): Remove mode and inverse
5939 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5940 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5941 and aarch64_float_const_representable_p on the result.
5942 (aarch64_output_simd_mov_immediate): Remove mode argument.
5943 Update call to aarch64_simd_valid_immediate and use of
5944 simd_immediate_info.
5945 (aarch64_output_scalar_simd_mov_immediate): Update call
5948 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5949 Alan Hayward <alan.hayward@arm.com>
5950 David Sherwood <david.sherwood@arm.com>
5952 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5953 (mode_nunits): Likewise CONST_MODE_NUNITS.
5954 * machmode.def (ADJUST_NUNITS): Document.
5955 * genmodes.c (mode_data::need_nunits_adj): New field.
5956 (blank_mode): Update accordingly.
5957 (adj_nunits): New variable.
5958 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5960 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5961 listed in adj_nunits.
5962 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5963 listed in adj_nunits. Don't emit case statements for such modes.
5964 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5965 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5966 nothing if adj_nunits is nonnull.
5967 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5968 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5969 (emit_mode_fbit): Update use of print_maybe_const_decl.
5970 (emit_move_size): Likewise. Treat the array as non-const
5972 (emit_mode_adjustments): Handle adj_nunits.
5974 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5976 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5977 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5978 (VECTOR_MODES): Use it.
5979 (make_vector_modes): Take the prefix as an argument.
5981 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5982 Alan Hayward <alan.hayward@arm.com>
5983 David Sherwood <david.sherwood@arm.com>
5985 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5986 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
5987 for MODE_VECTOR_BOOL.
5988 * machmode.def (VECTOR_BOOL_MODE): Document.
5989 * genmodes.c (VECTOR_BOOL_MODE): New macro.
5990 (make_vector_bool_mode): New function.
5991 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
5993 * lto-streamer-in.c (lto_input_mode_table): Likewise.
5994 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
5996 * stor-layout.c (int_mode_for_mode): Likewise.
5997 * tree.c (build_vector_type_for_mode): Likewise.
5998 * varasm.c (output_constant_pool_2): Likewise.
5999 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
6000 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
6001 for MODE_VECTOR_BOOL.
6002 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
6003 of mode class checks.
6004 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
6005 instead of a list of mode class checks.
6006 (expand_vector_scalar_condition): Likewise.
6007 (type_for_widest_vector_mode): Handle BImode as an inner mode.
6009 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6010 Alan Hayward <alan.hayward@arm.com>
6011 David Sherwood <david.sherwood@arm.com>
6013 * machmode.h (mode_size): Change from unsigned short to
6015 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
6016 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6017 or if measurement_type is not polynomial.
6018 (fixed_size_mode::includes_p): Check for constant-sized modes.
6019 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
6020 return a poly_uint16 rather than an unsigned short.
6021 (emit_mode_size): Change the type of mode_size from unsigned short
6022 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
6023 (emit_mode_adjustments): Cope with polynomial vector sizes.
6024 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6026 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6028 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
6029 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
6030 * caller-save.c (setup_save_areas): Likewise.
6031 (replace_reg_with_saved_mem): Likewise.
6032 * calls.c (emit_library_call_value_1): Likewise.
6033 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
6034 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
6035 (gen_lowpart_for_combine): Likewise.
6036 * convert.c (convert_to_integer_1): Likewise.
6037 * cse.c (equiv_constant, cse_insn): Likewise.
6038 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
6039 (cselib_subst_to_values): Likewise.
6040 * dce.c (word_dce_process_block): Likewise.
6041 * df-problems.c (df_word_lr_mark_ref): Likewise.
6042 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
6043 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
6044 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
6045 (rtl_for_decl_location): Likewise.
6046 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
6047 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
6048 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
6049 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
6050 (expand_expr_real_1): Likewise.
6051 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
6052 (pad_below): Likewise.
6053 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6054 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
6055 * ira.c (get_subreg_tracking_sizes): Likewise.
6056 * ira-build.c (ira_create_allocno_objects): Likewise.
6057 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
6058 (ira_sort_regnos_for_alter_reg): Likewise.
6059 * ira-costs.c (record_operand_costs): Likewise.
6060 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
6061 (resolve_simple_move): Likewise.
6062 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
6063 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
6064 (lra_constraints): Likewise.
6065 (CONST_POOL_OK_P): Reject variable-sized modes.
6066 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
6067 (add_pseudo_to_slot, lra_spill): Likewise.
6068 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6069 * optabs-query.c (get_best_extraction_insn): Likewise.
6070 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6071 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
6072 (expand_mult_highpart, valid_multiword_target_p): Likewise.
6073 * recog.c (offsettable_address_addr_space_p): Likewise.
6074 * regcprop.c (maybe_mode_change): Likewise.
6075 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
6076 * regrename.c (build_def_use): Likewise.
6077 * regstat.c (dump_reg_info): Likewise.
6078 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
6079 (find_reloads, find_reloads_subreg_address): Likewise.
6080 * reload1.c (eliminate_regs_1): Likewise.
6081 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
6082 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
6083 (simplify_binary_operation_1, simplify_subreg): Likewise.
6084 * targhooks.c (default_function_arg_padding): Likewise.
6085 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
6086 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
6087 (verify_gimple_assign_ternary): Likewise.
6088 * tree-inline.c (estimate_move_cost): Likewise.
6089 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6090 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
6091 (get_address_cost_ainc): Likewise.
6092 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
6093 (vect_supportable_dr_alignment): Likewise.
6094 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6095 (vectorizable_reduction): Likewise.
6096 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
6097 (vectorizable_operation, vectorizable_load): Likewise.
6098 * tree.c (build_same_sized_truth_vector_type): Likewise.
6099 * valtrack.c (cleanup_auto_inc_dec): Likewise.
6100 * var-tracking.c (emit_note_insn_var_location): Likewise.
6101 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
6102 (ADDR_VEC_ALIGN): Likewise.
6104 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6105 Alan Hayward <alan.hayward@arm.com>
6106 David Sherwood <david.sherwood@arm.com>
6108 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
6110 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6111 or if measurement_type is polynomial.
6112 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
6113 * combine.c (make_extraction): Likewise.
6114 * dse.c (find_shift_sequence): Likewise.
6115 * dwarf2out.c (mem_loc_descriptor): Likewise.
6116 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
6117 (extract_bit_field, extract_low_bits): Likewise.
6118 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
6119 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
6120 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
6121 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
6122 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6123 * reload.c (find_reloads): Likewise.
6124 * reload1.c (alter_reg): Likewise.
6125 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
6126 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
6127 * tree-if-conv.c (predicate_mem_writes): Likewise.
6128 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
6129 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
6130 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
6131 * valtrack.c (dead_debug_insert_temp): Likewise.
6132 * varasm.c (mergeable_constant_section): Likewise.
6133 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
6135 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6136 Alan Hayward <alan.hayward@arm.com>
6137 David Sherwood <david.sherwood@arm.com>
6139 * expr.c (expand_assignment): Cope with polynomial mode sizes
6140 when assigning to a CONCAT.
6142 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6143 Alan Hayward <alan.hayward@arm.com>
6144 David Sherwood <david.sherwood@arm.com>
6146 * machmode.h (mode_precision): Change from unsigned short to
6148 (mode_to_precision): Return a poly_uint16 rather than an unsigned
6150 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
6151 or if measurement_type is not polynomial.
6152 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
6153 in which the mode is already known to be a scalar_int_mode.
6154 * genmodes.c (emit_mode_precision): Change the type of mode_precision
6155 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
6157 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6158 for GET_MODE_PRECISION.
6159 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6160 for GET_MODE_PRECISION.
6161 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
6163 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
6164 (expand_field_assignment, make_extraction): Likewise.
6165 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
6166 (get_last_value): Likewise.
6167 * convert.c (convert_to_integer_1): Likewise.
6168 * cse.c (cse_insn): Likewise.
6169 * expr.c (expand_expr_real_1): Likewise.
6170 * lra-constraints.c (simplify_operand_subreg): Likewise.
6171 * optabs-query.c (can_atomic_load_p): Likewise.
6172 * optabs.c (expand_atomic_load): Likewise.
6173 (expand_atomic_store): Likewise.
6174 * ree.c (combine_reaching_defs): Likewise.
6175 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
6176 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
6177 * tree.h (type_has_mode_precision_p): Likewise.
6178 * ubsan.c (instrument_si_overflow): Likewise.
6180 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6181 Alan Hayward <alan.hayward@arm.com>
6182 David Sherwood <david.sherwood@arm.com>
6184 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
6185 polynomial numbers of units.
6186 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
6187 (valid_vector_subparts_p): New function.
6188 (build_vector_type): Remove temporary shim and take the number
6189 of units as a poly_uint64 rather than an int.
6190 (build_opaque_vector_type): Take the number of units as a
6191 poly_uint64 rather than an int.
6192 * tree.c (build_vector_from_ctor): Handle polynomial
6193 TYPE_VECTOR_SUBPARTS.
6194 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
6195 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
6196 (build_vector_from_val): If the number of units is variable,
6197 use build_vec_duplicate_cst for constant operands and
6198 VEC_DUPLICATE_EXPR otherwise.
6199 (make_vector_type): Remove temporary is_constant ().
6200 (build_vector_type, build_opaque_vector_type): Take the number of
6201 units as a poly_uint64 rather than an int.
6202 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
6204 * cfgexpand.c (expand_debug_expr): Likewise.
6205 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
6206 (store_constructor, expand_expr_real_1): Likewise.
6207 (const_scalar_mask_from_tree): Likewise.
6208 * fold-const-call.c (fold_const_reduction): Likewise.
6209 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
6210 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
6211 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
6212 (fold_relational_const): Likewise.
6213 (native_interpret_vector): Likewise. Change the size from an
6214 int to an unsigned int.
6215 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
6216 TYPE_VECTOR_SUBPARTS.
6217 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
6218 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
6219 duplicating a non-constant operand into a variable-length vector.
6220 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
6221 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
6222 * ipa-icf.c (sem_variable::equals): Likewise.
6223 * match.pd: Likewise.
6224 * omp-simd-clone.c (simd_clone_subparts): Likewise.
6225 * print-tree.c (print_node): Likewise.
6226 * stor-layout.c (layout_type): Likewise.
6227 * targhooks.c (default_builtin_vectorization_cost): Likewise.
6228 * tree-cfg.c (verify_gimple_comparison): Likewise.
6229 (verify_gimple_assign_binary): Likewise.
6230 (verify_gimple_assign_ternary): Likewise.
6231 (verify_gimple_assign_single): Likewise.
6232 * tree-pretty-print.c (dump_generic_node): Likewise.
6233 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6234 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
6235 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
6236 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
6237 (vect_shift_permute_load_chain): Likewise.
6238 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
6239 (expand_vector_condition, optimize_vector_constructor): Likewise.
6240 (lower_vec_perm, get_compute_type): Likewise.
6241 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6242 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6243 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6244 (vect_recog_mask_conversion_pattern): Likewise.
6245 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6246 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6247 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6248 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6249 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6250 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6251 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6252 (supportable_widening_operation): Likewise.
6253 (supportable_narrowing_operation): Likewise.
6254 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6256 * varasm.c (output_constant): Likewise.
6258 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6259 Alan Hayward <alan.hayward@arm.com>
6260 David Sherwood <david.sherwood@arm.com>
6262 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6263 so that both the length == 3 and length != 3 cases set up their
6264 own permute vectors. Add comments explaining why we know the
6265 number of elements is constant.
6266 (vect_permute_load_chain): Likewise.
6268 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6269 Alan Hayward <alan.hayward@arm.com>
6270 David Sherwood <david.sherwood@arm.com>
6272 * machmode.h (mode_nunits): Change from unsigned char to
6274 (ONLY_FIXED_SIZE_MODES): New macro.
6275 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6276 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6277 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6279 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6280 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6281 or if measurement_type is not polynomial.
6282 * genmodes.c (ZERO_COEFFS): New macro.
6283 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6285 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6286 Use ZERO_COEFFS when emitting initializers.
6287 * data-streamer.h (bp_pack_poly_value): New function.
6288 (bp_unpack_poly_value): Likewise.
6289 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6290 for GET_MODE_NUNITS.
6291 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6292 for GET_MODE_NUNITS.
6293 * tree.c (make_vector_type): Remove temporary shim and make
6294 the real function take the number of units as a poly_uint64
6296 (build_vector_type_for_mode): Handle polynomial nunits.
6297 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6298 * emit-rtl.c (const_vec_series_p_1): Likewise.
6299 (gen_rtx_CONST_VECTOR): Likewise.
6300 * fold-const.c (test_vec_duplicate_folding): Likewise.
6301 * genrecog.c (validate_pattern): Likewise.
6302 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6303 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6304 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6305 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6306 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6307 * rtlanal.c (subreg_get_info): Likewise.
6308 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6309 (vect_grouped_load_supported): Likewise.
6310 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6311 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6312 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6313 (simplify_const_unary_operation, simplify_binary_operation_1)
6314 (simplify_const_binary_operation, simplify_ternary_operation)
6315 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6316 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6317 instead of CONST_VECTOR_NUNITS.
6318 * varasm.c (output_constant_pool_2): Likewise.
6319 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6320 explicit-encoded elements in the XVEC for variable-length vectors.
6322 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6324 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6326 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6327 Alan Hayward <alan.hayward@arm.com>
6328 David Sherwood <david.sherwood@arm.com>
6330 * coretypes.h (fixed_size_mode): Declare.
6331 (fixed_size_mode_pod): New typedef.
6332 * builtins.h (target_builtins::x_apply_args_mode)
6333 (target_builtins::x_apply_result_mode): Change type to
6334 fixed_size_mode_pod.
6335 * builtins.c (apply_args_size, apply_result_size, result_vector)
6336 (expand_builtin_apply_args_1, expand_builtin_apply)
6337 (expand_builtin_return): Update accordingly.
6339 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6341 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6342 * cselib.c (cselib_hash_rtx): Likewise.
6343 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6344 CONST_VECTOR encoding.
6346 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6347 Jeff Law <law@redhat.com>
6350 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6351 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6352 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6353 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6356 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6357 explicitly probe *sp in a noreturn function if there were any callee
6358 register saves or frame pointer is needed.
6360 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6363 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6364 BLKmode for ternary, binary or unary expressions.
6367 * var-tracking.c (delete_vta_debug_insn): New inline function.
6368 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6369 insns from get_insns () to NULL instead of each bb separately.
6370 Use delete_vta_debug_insn. No longer static.
6371 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6372 delete_vta_debug_insns callers.
6373 * rtl.h (delete_vta_debug_insns): Declare.
6374 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6375 instead of variable_tracking_main.
6377 2018-01-03 Martin Sebor <msebor@redhat.com>
6379 PR tree-optimization/83603
6380 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6381 arguments past the endof the argument list in functions declared
6382 without a prototype.
6383 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6384 Avoid checking when arguments are null.
6386 2018-01-03 Martin Sebor <msebor@redhat.com>
6389 * doc/extend.texi (attribute const): Fix a typo.
6390 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6391 issuing -Wsuggest-attribute for void functions.
6393 2018-01-03 Martin Sebor <msebor@redhat.com>
6395 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6396 offset_int::from instead of wide_int::to_shwi.
6397 (maybe_diag_overlap): Remove assertion.
6398 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6399 * gimple-ssa-sprintf.c (format_directive): Same.
6400 (parse_directive): Same.
6401 (sprintf_dom_walker::compute_format_length): Same.
6402 (try_substitute_return_value): Same.
6404 2018-01-03 Jeff Law <law@redhat.com>
6407 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6408 non-constant residual for zero at runtime and avoid probing in
6409 that case. Reorganize code for trailing problem to mirror handling
6412 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6414 PR tree-optimization/83501
6415 * tree-ssa-strlen.c (get_string_cst): New.
6416 (handle_char_store): Call get_string_cst.
6418 2018-01-03 Martin Liska <mliska@suse.cz>
6420 PR tree-optimization/83593
6421 * tree-ssa-strlen.c: Include tree-cfg.h.
6422 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6423 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6424 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6426 (strlen_dom_walker::before_dom_children): Call
6427 gimple_purge_dead_eh_edges. Dump tranformation with details
6429 (strlen_dom_walker::before_dom_children): Update call by adding
6430 new argument cleanup_eh.
6431 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6433 2018-01-03 Martin Liska <mliska@suse.cz>
6436 * cif-code.def (VARIADIC_THUNK): New enum value.
6437 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6440 2018-01-03 Jan Beulich <jbeulich@suse.com>
6442 * sse.md (mov<mode>_internal): Tighten condition for when to use
6443 vmovdqu<ssescalarsize> for TI and OI modes.
6445 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6447 Update copyright years.
6449 2018-01-03 Martin Liska <mliska@suse.cz>
6452 * ipa-visibility.c (function_and_variable_visibility): Skip
6453 functions with noipa attribure.
6455 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6457 * gcc.c (process_command): Update copyright notice dates.
6458 * gcov-dump.c (print_version): Ditto.
6459 * gcov.c (print_version): Ditto.
6460 * gcov-tool.c (print_version): Ditto.
6461 * gengtype.c (create_file): Ditto.
6462 * doc/cpp.texi: Bump @copying's copyright year.
6463 * doc/cppinternals.texi: Ditto.
6464 * doc/gcc.texi: Ditto.
6465 * doc/gccint.texi: Ditto.
6466 * doc/gcov.texi: Ditto.
6467 * doc/install.texi: Ditto.
6468 * doc/invoke.texi: Ditto.
6470 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6472 * vector-builder.h (vector_builder::m_full_nelts): Change from
6473 unsigned int to poly_uint64.
6474 (vector_builder::full_nelts): Update prototype accordingly.
6475 (vector_builder::new_vector): Likewise.
6476 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6477 (vector_builder::operator ==): Likewise.
6478 (vector_builder::finalize): Likewise.
6479 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6480 Take the number of elements as a poly_uint64 rather than an
6482 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6483 from unsigned int to poly_uint64.
6484 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6485 (vec_perm_indices::new_vector): Likewise.
6486 (vec_perm_indices::length): Likewise.
6487 (vec_perm_indices::nelts_per_input): Likewise.
6488 (vec_perm_indices::input_nelts): Likewise.
6489 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6490 number of elements per input as a poly_uint64 rather than an
6491 unsigned int. Use the original encoding for variable-length
6492 vectors, rather than clamping each individual element.
6493 For the second and subsequent elements in each pattern,
6494 clamp the step and base before clamping their sum.
6495 (vec_perm_indices::series_p): Handle polynomial element counts.
6496 (vec_perm_indices::all_in_range_p): Likewise.
6497 (vec_perm_indices_to_tree): Likewise.
6498 (vec_perm_indices_to_rtx): Likewise.
6499 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6500 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6501 (tree_vector_builder::new_binary_operation): Handle polynomial
6502 element counts. Return false if we need to know the number
6503 of elements at compile time.
6504 * fold-const.c (fold_vec_perm): Punt if the number of elements
6505 isn't known at compile time.
6507 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6509 * vec-perm-indices.h (vec_perm_builder): Change element type
6510 from HOST_WIDE_INT to poly_int64.
6511 (vec_perm_indices::element_type): Update accordingly.
6512 (vec_perm_indices::clamp): Handle polynomial element_types.
6513 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6514 (vec_perm_indices::all_in_range_p): Likewise.
6515 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6517 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6518 polynomial vec_perm_indices element types.
6519 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6520 * fold-const.c (fold_vec_perm): Likewise.
6521 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6522 * tree-vect-generic.c (lower_vec_perm): Likewise.
6523 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6524 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6525 element type to HOST_WIDE_INT.
6527 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6528 Alan Hayward <alan.hayward@arm.com>
6529 David Sherwood <david.sherwood@arm.com>
6531 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6532 rather than an int. Use plus_constant.
6533 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6534 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6536 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6537 Alan Hayward <alan.hayward@arm.com>
6538 David Sherwood <david.sherwood@arm.com>
6540 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6541 a HOST_WIDE_INT to a poly_int64.
6543 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6544 Alan Hayward <alan.hayward@arm.com>
6545 David Sherwood <david.sherwood@arm.com>
6547 * calls.c (load_register_parameters): Cope with polynomial
6548 mode sizes. Require a constant size for BLKmode parameters
6549 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6550 forces a parameter to be padded at the lsb end in order to
6551 fill a complete number of words, require the parameter size
6552 to be ordered wrt UNITS_PER_WORD.
6554 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6555 Alan Hayward <alan.hayward@arm.com>
6556 David Sherwood <david.sherwood@arm.com>
6558 * reload1.c (spill_stack_slot_width): Change element type
6559 from unsigned int to poly_uint64_pod.
6560 (alter_reg): Treat mode sizes as polynomial.
6562 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6563 Alan Hayward <alan.hayward@arm.com>
6564 David Sherwood <david.sherwood@arm.com>
6566 * reload.c (complex_word_subreg_p): New function.
6567 (reload_inner_reg_of_subreg, push_reload): Use it.
6569 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6570 Alan Hayward <alan.hayward@arm.com>
6571 David Sherwood <david.sherwood@arm.com>
6573 * lra-constraints.c (process_alt_operands): Reject matched
6574 operands whose sizes aren't ordered.
6575 (match_reload): Refer to this check here.
6577 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6578 Alan Hayward <alan.hayward@arm.com>
6579 David Sherwood <david.sherwood@arm.com>
6581 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6582 that the mode size is in the set {1, 2, 4, 8, 16}.
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 * var-tracking.c (adjust_mems): 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 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6596 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6597 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6598 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6599 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6600 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6601 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6602 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6603 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6604 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6606 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6607 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6608 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6609 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6610 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6611 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6612 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6613 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6614 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6615 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6617 * expr.c (emit_move_resolve_push): Treat the input and result
6618 of PUSH_ROUNDING as a poly_int64.
6619 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6620 (emit_push_insn): Likewise.
6621 * lra-eliminations.c (mark_not_eliminable): Likewise.
6622 * recog.c (push_operand): Likewise.
6623 * reload1.c (elimination_effects): Likewise.
6624 * rtlanal.c (nonzero_bits1): Likewise.
6625 * calls.c (store_one_arg): Likewise. Require the padding to be
6626 known at compile time.
6628 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6629 Alan Hayward <alan.hayward@arm.com>
6630 David Sherwood <david.sherwood@arm.com>
6632 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6633 Use plus_constant instead of gen_rtx_PLUS.
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 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
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 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6647 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6648 via stack temporaries. Treat the mode size as polynomial too.
6650 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6651 Alan Hayward <alan.hayward@arm.com>
6652 David Sherwood <david.sherwood@arm.com>
6654 * expr.c (expand_expr_real_2): When handling conversions involving
6655 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6656 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6657 as a poly_uint64 too.
6659 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6660 Alan Hayward <alan.hayward@arm.com>
6661 David Sherwood <david.sherwood@arm.com>
6663 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6665 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6666 Alan Hayward <alan.hayward@arm.com>
6667 David Sherwood <david.sherwood@arm.com>
6669 * combine.c (can_change_dest_mode): Handle polynomial
6670 REGMODE_NATURAL_SIZE.
6671 * expmed.c (store_bit_field_1): Likewise.
6672 * expr.c (store_constructor): Likewise.
6673 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6674 and polynomial REGMODE_NATURAL_SIZE.
6675 (gen_lowpart_common): Likewise.
6676 * reginfo.c (record_subregs_of_mode): Likewise.
6677 * rtlanal.c (read_modify_subreg_p): Likewise.
6679 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6680 Alan Hayward <alan.hayward@arm.com>
6681 David Sherwood <david.sherwood@arm.com>
6683 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6684 numbers of elements.
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 * match.pd: Cope with polynomial numbers of vector elements.
6692 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6693 Alan Hayward <alan.hayward@arm.com>
6694 David Sherwood <david.sherwood@arm.com>
6696 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6697 in a POINTER_PLUS_EXPR.
6699 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6700 Alan Hayward <alan.hayward@arm.com>
6701 David Sherwood <david.sherwood@arm.com>
6703 * omp-simd-clone.c (simd_clone_subparts): New function.
6704 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6705 (ipa_simd_modify_function_body): Likewise.
6707 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6708 Alan Hayward <alan.hayward@arm.com>
6709 David Sherwood <david.sherwood@arm.com>
6711 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6712 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6713 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6714 (expand_vector_condition, vector_element): Likewise.
6715 (subparts_gt): New function.
6716 (get_compute_type): Use subparts_gt.
6717 (count_type_subparts): Delete.
6718 (expand_vector_operations_1): Use subparts_gt instead of
6719 count_type_subparts.
6721 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6722 Alan Hayward <alan.hayward@arm.com>
6723 David Sherwood <david.sherwood@arm.com>
6725 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6726 (vect_compile_time_alias): ...this new function. Do the calculation
6727 on poly_ints rather than trees.
6728 (vect_prune_runtime_alias_test_list): Update call accordingly.
6730 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6731 Alan Hayward <alan.hayward@arm.com>
6732 David Sherwood <david.sherwood@arm.com>
6734 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6736 (vect_schedule_slp_instance): 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-slp.c (vect_get_and_check_slp_defs): Reject
6743 constant and extern definitions for variable-length vectors.
6744 (vect_get_constant_vectors): Note that the number of units
6745 is known to be constant.
6747 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6748 Alan Hayward <alan.hayward@arm.com>
6749 David Sherwood <david.sherwood@arm.com>
6751 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6752 of units as polynomial. Choose between WIDE and NARROW based
6755 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6756 Alan Hayward <alan.hayward@arm.com>
6757 David Sherwood <david.sherwood@arm.com>
6759 * tree-vect-stmts.c (simd_clone_subparts): New function.
6760 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6762 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6763 Alan Hayward <alan.hayward@arm.com>
6764 David Sherwood <david.sherwood@arm.com>
6766 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6767 vectors as polynomial. Use build_index_vector for
6770 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6771 Alan Hayward <alan.hayward@arm.com>
6772 David Sherwood <david.sherwood@arm.com>
6774 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6775 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6776 for variable-length vectors.
6777 (vectorizable_mask_load_store): Treat the number of units as
6778 polynomial, asserting that it is constant if the condition has
6779 already been enforced.
6780 (vectorizable_store, vectorizable_load): Likewise.
6782 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6783 Alan Hayward <alan.hayward@arm.com>
6784 David Sherwood <david.sherwood@arm.com>
6786 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6787 of units as polynomial. Punt if we can't tell at compile time
6788 which vector contains the final result.
6790 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6791 Alan Hayward <alan.hayward@arm.com>
6792 David Sherwood <david.sherwood@arm.com>
6794 * tree-vect-loop.c (vectorizable_induction): Treat the number
6795 of units as polynomial. Punt on SLP inductions. Use an integer
6796 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6797 cast of such a series for variable-length floating-point
6800 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6801 Alan Hayward <alan.hayward@arm.com>
6802 David Sherwood <david.sherwood@arm.com>
6804 * tree.h (build_index_vector): Declare.
6805 * tree.c (build_index_vector): New function.
6806 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6807 of units as polynomial, forcibly converting it to a constant if
6808 vectorizable_reduction has already enforced the condition.
6809 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6810 to create a {1,2,3,...} vector.
6811 (vectorizable_reduction): Treat the number of units as polynomial.
6812 Choose vectype_in based on the largest scalar element size rather
6813 than the smallest number of units. Enforce the restrictions
6816 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6817 Alan Hayward <alan.hayward@arm.com>
6818 David Sherwood <david.sherwood@arm.com>
6820 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6821 number of units as polynomial.
6823 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6824 Alan Hayward <alan.hayward@arm.com>
6825 David Sherwood <david.sherwood@arm.com>
6827 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6828 * target.def (autovectorize_vector_sizes): Return the vector sizes
6829 by pointer, using vector_sizes rather than a bitmask.
6830 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6831 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6832 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6834 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6835 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6836 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6837 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6838 * omp-general.c (omp_max_vf): Likewise.
6839 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6840 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6841 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6842 * tree-vect-slp.c (vect_slp_bb): Likewise.
6843 * doc/tm.texi: Regenerate.
6844 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6846 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6847 the vector size as a poly_uint64 rather than an unsigned int.
6848 (current_vector_size): Change from an unsigned int to a poly_uint64.
6849 (get_vectype_for_scalar_type): Update accordingly.
6850 * tree.h (build_truth_vector_type): Take the size and number of
6851 units as a poly_uint64 rather than an unsigned int.
6852 (build_vector_type): Add a temporary overload that takes
6853 the number of units as a poly_uint64 rather than an unsigned int.
6854 * tree.c (make_vector_type): Likewise.
6855 (build_truth_vector_type): Take the number of units as a poly_uint64
6856 rather than an unsigned int.
6858 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6859 Alan Hayward <alan.hayward@arm.com>
6860 David Sherwood <david.sherwood@arm.com>
6862 * target.def (get_mask_mode): Take the number of units and length
6863 as poly_uint64s rather than unsigned ints.
6864 * targhooks.h (default_get_mask_mode): Update accordingly.
6865 * targhooks.c (default_get_mask_mode): Likewise.
6866 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6867 * doc/tm.texi: Regenerate.
6869 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6870 Alan Hayward <alan.hayward@arm.com>
6871 David Sherwood <david.sherwood@arm.com>
6873 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6874 * omp-general.c (omp_max_vf): Likewise.
6875 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6876 (expand_omp_simd): Handle polynomial safelen.
6877 * omp-low.c (omplow_simd_context): Add a default constructor.
6878 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6879 (lower_rec_simd_input_clauses): Update accordingly.
6880 (lower_rec_input_clauses): Likewise.
6882 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6883 Alan Hayward <alan.hayward@arm.com>
6884 David Sherwood <david.sherwood@arm.com>
6886 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6887 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6888 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6889 (vect_analyze_slp_cost): Likewise.
6890 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6891 (vect_model_load_cost): Likewise.
6893 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6894 Alan Hayward <alan.hayward@arm.com>
6895 David Sherwood <david.sherwood@arm.com>
6897 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6898 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6899 from an unsigned int * to a poly_uint64_pod *.
6900 (calculate_unrolling_factor): New function.
6901 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6903 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6904 Alan Hayward <alan.hayward@arm.com>
6905 David Sherwood <david.sherwood@arm.com>
6907 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6908 from an unsigned int to a poly_uint64.
6909 (_loop_vec_info::slp_unrolling_factor): Likewise.
6910 (_loop_vec_info::vectorization_factor): Change from an int
6912 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6913 (vect_get_num_vectors): New function.
6914 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6915 (vect_get_num_copies): Use vect_get_num_vectors.
6916 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6917 to an unsigned int *.
6918 (vect_analyze_data_refs): Change min_vf from an int * to a
6920 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6921 than an unsigned HOST_WIDE_INT.
6922 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6923 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6924 to an unsigned int *.
6925 (vect_analyze_data_ref_dependences): Likewise.
6926 (vect_compute_data_ref_alignment): Handle polynomial vf.
6927 (vect_enhance_data_refs_alignment): Likewise.
6928 (vect_prune_runtime_alias_test_list): Likewise.
6929 (vect_shift_permute_load_chain): Likewise.
6930 (vect_supportable_dr_alignment): Likewise.
6931 (dependence_distance_ge_vf): Take the vectorization factor as a
6932 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6933 (vect_analyze_data_refs): Change min_vf from an int * to a
6935 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6936 vfm1 as a poly_uint64 rather than an int. Make the same change
6937 for the returned bound_scalar.
6938 (vect_gen_vector_loop_niters): Handle polynomial vf.
6939 (vect_do_peeling): Likewise. Update call to
6940 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6941 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6943 * tree-vect-loop.c (vect_determine_vectorization_factor)
6944 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6945 (vect_get_known_peeling_cost): Likewise.
6946 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6947 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6948 (vect_transform_loop): Likewise. Use the lowest possible VF when
6949 updating the upper bounds of the loop.
6950 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6952 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6953 polynomial unroll factors.
6954 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6955 (vect_make_slp_decision): Likewise.
6956 (vect_supported_load_permutation_p): Likewise, and polynomial
6958 (vect_analyze_slp_cost): Handle polynomial vf.
6959 (vect_slp_analyze_node_operations): Likewise.
6960 (vect_slp_analyze_bb_1): Likewise.
6961 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6962 than an unsigned HOST_WIDE_INT.
6963 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6964 (vectorizable_load): Handle polynomial vf.
6965 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6967 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6969 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6970 Alan Hayward <alan.hayward@arm.com>
6971 David Sherwood <david.sherwood@arm.com>
6973 * match.pd: Handle bit operations involving three constants
6974 and try to fold one pair.
6976 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6978 * tree-vect-loop-manip.c: Include gimple-fold.h.
6979 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6980 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6981 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6982 Add a path that uses a step of VF instead of 1, but disable it
6984 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6985 and niters_no_overflow parameters. Update calls to
6986 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
6987 Create a new SSA name if the latter choses to use a ste other
6988 than zero, and return it via niters_vector_mult_vf_var.
6989 * tree-vect-loop.c (vect_transform_loop): Update calls to
6990 vect_do_peeling, vect_gen_vector_loop_niters and
6991 slpeel_make_loop_iterate_ntimes.
6992 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
6993 (vect_gen_vector_loop_niters): Update declarations after above changes.
6995 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
6997 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
6998 128-bit round to integer instructions.
6999 (ceil<mode>2): Likewise.
7000 (btrunc<mode>2): Likewise.
7001 (round<mode>2): Likewise.
7003 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7005 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
7006 unaligned VSX load/store on P8/P9.
7007 (expand_block_clear): Allow the use of unaligned VSX
7008 load/store on P8/P9.
7010 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
7012 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
7014 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
7015 swap associated with both a load and a store.
7017 2018-01-02 Andrew Waterman <andrew@sifive.com>
7019 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
7020 * config/riscv/riscv.md (clear_cache): Use it.
7022 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
7024 * web.c: Remove out-of-date comment.
7026 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7028 * expr.c (fixup_args_size_notes): Check that any existing
7029 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
7030 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
7031 (emit_single_push_insn): ...here.
7033 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7035 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
7036 (const_vector_encoded_nelts): New function.
7037 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
7038 (const_vector_int_elt, const_vector_elt): Declare.
7039 * emit-rtl.c (const_vector_int_elt_1): New function.
7040 (const_vector_elt): Likewise.
7041 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
7042 of CONST_VECTOR_ELT.
7044 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7046 * expr.c: Include rtx-vector-builder.h.
7047 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
7048 directly on the tree encoding.
7049 (const_vector_from_tree): Likewise.
7050 * optabs.c: Include rtx-vector-builder.h.
7051 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
7052 sequence of "u" values.
7053 * vec-perm-indices.c: Include rtx-vector-builder.h.
7054 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
7055 directly on the vec_perm_indices encoding.
7057 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7059 * doc/rtl.texi (const_vector): Describe new encoding scheme.
7060 * Makefile.in (OBJS): Add rtx-vector-builder.o.
7061 * rtx-vector-builder.h: New file.
7062 * rtx-vector-builder.c: Likewise.
7063 * rtl.h (rtx_def::u2): Add a const_vector field.
7064 (CONST_VECTOR_NPATTERNS): New macro.
7065 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
7066 (CONST_VECTOR_DUPLICATE_P): Likewise.
7067 (CONST_VECTOR_STEPPED_P): Likewise.
7068 (CONST_VECTOR_ENCODED_ELT): Likewise.
7069 (const_vec_duplicate_p): Check for a duplicated vector encoding.
7070 (unwrap_const_vec_duplicate): Likewise.
7071 (const_vec_series_p): Check for a non-duplicated vector encoding.
7072 Say that the function only returns true for integer vectors.
7073 * emit-rtl.c: Include rtx-vector-builder.h.
7074 (gen_const_vec_duplicate_1): Delete.
7075 (gen_const_vector): Call gen_const_vec_duplicate instead of
7076 gen_const_vec_duplicate_1.
7077 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
7078 (gen_const_vec_duplicate): Use rtx_vector_builder.
7079 (gen_const_vec_series): Likewise.
7080 (gen_rtx_CONST_VECTOR): Likewise.
7081 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
7082 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7083 Build a new vector rather than modifying a CONST_VECTOR in-place.
7084 (handle_special_swappables): Update call accordingly.
7085 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
7086 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7087 Build a new vector rather than modifying a CONST_VECTOR in-place.
7088 (handle_special_swappables): Update call accordingly.
7090 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7092 * simplify-rtx.c (simplify_const_binary_operation): Use
7093 CONST_VECTOR_ELT instead of XVECEXP.
7095 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7097 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
7098 the selector elements to be different from the data elements
7099 if the selector is a VECTOR_CST.
7100 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
7101 ssizetype for the selector.
7103 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7105 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
7106 before testing each element individually.
7107 * tree-vect-generic.c (lower_vec_perm): Likewise.
7109 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7111 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
7112 * selftest-run-tests.c (selftest::run_tests): Call it.
7113 * vector-builder.h (vector_builder::operator ==): New function.
7114 (vector_builder::operator !=): Likewise.
7115 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
7116 (vec_perm_indices::all_from_input_p): New function.
7117 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7118 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
7119 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
7120 instead of reading the VECTOR_CST directly. Detect whether both
7121 vector inputs are the same before constructing the vec_perm_indices,
7122 and update the number of inputs argument accordingly. Use the
7123 utility functions added above. Only construct sel2 if we need to.
7125 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7127 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
7128 the broadcast of the low byte.
7129 (expand_mult_highpart): Use an explicit encoding for the permutes.
7130 * optabs-query.c (can_mult_highpart_p): Likewise.
7131 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
7132 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7133 (vectorizable_bswap): Likewise.
7134 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
7135 explicit encoding for the power-of-2 permutes.
7136 (vect_permute_store_chain): Likewise.
7137 (vect_grouped_load_supported): Likewise.
7138 (vect_permute_load_chain): Likewise.
7140 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7142 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
7143 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
7144 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
7145 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7146 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
7147 (vect_gen_perm_mask_any): Likewise.
7149 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7151 * int-vector-builder.h: New file.
7152 * vec-perm-indices.h: Include int-vector-builder.h.
7153 (vec_perm_indices): Redefine as an int_vector_builder.
7154 (auto_vec_perm_indices): Delete.
7155 (vec_perm_builder): Redefine as a stand-alone class.
7156 (vec_perm_indices::vec_perm_indices): New function.
7157 (vec_perm_indices::clamp): Likewise.
7158 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
7159 (vec_perm_indices::new_vector): New function.
7160 (vec_perm_indices::new_expanded_vector): Update for new
7161 vec_perm_indices class.
7162 (vec_perm_indices::rotate_inputs): New function.
7163 (vec_perm_indices::all_in_range_p): Operate directly on the
7164 encoded form, without computing elided elements.
7165 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
7166 encoding. Update for new vec_perm_indices class.
7167 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
7168 the given vec_perm_builder.
7169 (expand_vec_perm_var): Update vec_perm_builder constructor.
7170 (expand_mult_highpart): Use vec_perm_builder instead of
7171 auto_vec_perm_indices.
7172 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
7173 vec_perm_indices instead of auto_vec_perm_indices. Use a single
7174 or double series encoding as appropriate.
7175 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
7176 vec_perm_indices instead of auto_vec_perm_indices.
7177 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7178 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7179 (vect_permute_store_chain): Likewise.
7180 (vect_grouped_load_supported): Likewise.
7181 (vect_permute_load_chain): Likewise.
7182 (vect_shift_permute_load_chain): Likewise.
7183 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7184 (vect_transform_slp_perm_load): Likewise.
7185 (vect_schedule_slp_instance): Likewise.
7186 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7187 (vectorizable_mask_load_store): Likewise.
7188 (vectorizable_bswap): Likewise.
7189 (vectorizable_store): Likewise.
7190 (vectorizable_load): Likewise.
7191 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
7192 vec_perm_indices instead of auto_vec_perm_indices. Use
7193 tree_to_vec_perm_builder to read the vector from a tree.
7194 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
7195 vec_perm_builder instead of a vec_perm_indices.
7196 (have_whole_vector_shift): Use vec_perm_builder and
7197 vec_perm_indices instead of auto_vec_perm_indices. Leave the
7198 truncation to calc_vec_perm_mask_for_shift.
7199 (vect_create_epilog_for_reduction): Likewise.
7200 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
7201 from auto_vec_perm_indices to vec_perm_indices.
7202 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7203 instead of changing individual elements.
7204 (aarch64_vectorize_vec_perm_const): Use new_vector to install
7205 the vector in d.perm.
7206 * config/arm/arm.c (expand_vec_perm_d::perm): Change
7207 from auto_vec_perm_indices to vec_perm_indices.
7208 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7209 instead of changing individual elements.
7210 (arm_vectorize_vec_perm_const): Use new_vector to install
7211 the vector in d.perm.
7212 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
7213 Update vec_perm_builder constructor.
7214 (rs6000_expand_interleave): Likewise.
7215 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
7216 (rs6000_expand_interleave): Likewise.
7218 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7220 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
7221 to qimode could truncate the indices.
7222 * optabs.c (expand_vec_perm_var): Likewise.
7224 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7226 * Makefile.in (OBJS): Add vec-perm-indices.o.
7227 * vec-perm-indices.h: New file.
7228 * vec-perm-indices.c: Likewise.
7229 * target.h (vec_perm_indices): Replace with a forward class
7231 (auto_vec_perm_indices): Move to vec-perm-indices.h.
7232 * optabs.h: Include vec-perm-indices.h.
7233 (expand_vec_perm): Delete.
7234 (selector_fits_mode_p, expand_vec_perm_var): Declare.
7235 (expand_vec_perm_const): Declare.
7236 * target.def (vec_perm_const_ok): Replace with...
7237 (vec_perm_const): ...this new hook.
7238 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
7239 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7240 * doc/tm.texi: Regenerate.
7241 * optabs.def (vec_perm_const): Delete.
7242 * doc/md.texi (vec_perm_const): Likewise.
7243 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7244 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7245 expand_vec_perm for constant permutation vectors. Assert that
7246 the mode of variable permutation vectors is the integer equivalent
7247 of the mode that is being permuted.
7248 * optabs-query.h (selector_fits_mode_p): Declare.
7249 * optabs-query.c: Include vec-perm-indices.h.
7250 (selector_fits_mode_p): New function.
7251 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7252 is defined, instead of checking whether the vec_perm_const_optab
7253 exists. Use targetm.vectorize.vec_perm_const instead of
7254 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7255 fit in the vector mode before using a variable permute.
7256 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7257 vec_perm_indices instead of an rtx.
7258 (expand_vec_perm): Replace with...
7259 (expand_vec_perm_const): ...this new function. Take the selector
7260 as a vec_perm_indices rather than an rtx. Also take the mode of
7261 the selector. Update call to shift_amt_for_vec_perm_mask.
7262 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7263 Use vec_perm_indices::new_expanded_vector to expand the original
7264 selector into bytes. Check whether the indices fit in the vector
7265 mode before using a variable permute.
7266 (expand_vec_perm_var): Make global.
7267 (expand_mult_highpart): Use expand_vec_perm_const.
7268 * fold-const.c: Includes vec-perm-indices.h.
7269 * tree-ssa-forwprop.c: Likewise.
7270 * tree-vect-data-refs.c: Likewise.
7271 * tree-vect-generic.c: Likewise.
7272 * tree-vect-loop.c: Likewise.
7273 * tree-vect-slp.c: Likewise.
7274 * tree-vect-stmts.c: Likewise.
7275 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7277 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7278 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7279 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7280 (aarch64_vectorize_vec_perm_const): ...this new function.
7281 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7282 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7283 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7284 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7285 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7286 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7287 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7289 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7290 check for NEON modes.
7291 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7292 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7293 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7294 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7296 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7297 the old VEC_PERM_CONST conditions.
7298 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7299 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7300 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7301 (ia64_vectorize_vec_perm_const_ok): Merge into...
7302 (ia64_vectorize_vec_perm_const): ...this new function.
7303 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7304 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7305 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7306 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7307 * config/mips/mips.c (mips_expand_vec_perm_const)
7308 (mips_vectorize_vec_perm_const_ok): Merge into...
7309 (mips_vectorize_vec_perm_const): ...this new function.
7310 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7311 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7312 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7313 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7314 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7315 (rs6000_expand_vec_perm_const): Delete.
7316 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7318 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7319 (altivec_expand_vec_perm_const_le): Take each operand individually.
7320 Operate on constant selectors rather than rtxes.
7321 (altivec_expand_vec_perm_const): Likewise. Update call to
7322 altivec_expand_vec_perm_const_le.
7323 (rs6000_expand_vec_perm_const): Delete.
7324 (rs6000_vectorize_vec_perm_const_ok): Delete.
7325 (rs6000_vectorize_vec_perm_const): New function.
7326 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7327 an element count and rtx array.
7328 (rs6000_expand_extract_even): Update call accordingly.
7329 (rs6000_expand_interleave): Likewise.
7330 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7331 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7332 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7333 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7334 (rs6000_expand_vec_perm_const): Delete.
7335 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7336 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7337 (altivec_expand_vec_perm_const_le): Take each operand individually.
7338 Operate on constant selectors rather than rtxes.
7339 (altivec_expand_vec_perm_const): Likewise. Update call to
7340 altivec_expand_vec_perm_const_le.
7341 (rs6000_expand_vec_perm_const): Delete.
7342 (rs6000_vectorize_vec_perm_const_ok): Delete.
7343 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7344 reference to the SPE evmerge intructions.
7345 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7346 an element count and rtx array.
7347 (rs6000_expand_extract_even): Update call accordingly.
7348 (rs6000_expand_interleave): Likewise.
7349 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7350 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7352 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7354 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7356 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7357 vector mode and that that mode matches the mode of the data
7359 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7360 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7361 directly using expand_vec_perm_1 when forcing selectors into
7363 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7365 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7367 * optabs-query.h (can_vec_perm_p): Delete.
7368 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7369 * optabs-query.c (can_vec_perm_p): Split into...
7370 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7371 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7372 particular selector is valid.
7373 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7374 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7375 (vect_grouped_load_supported): Likewise.
7376 (vect_shift_permute_load_chain): Likewise.
7377 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7378 (vect_transform_slp_perm_load): Likewise.
7379 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7380 (vectorizable_bswap): Likewise.
7381 (vect_gen_perm_mask_checked): Likewise.
7382 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7383 implementations of variable permutation vectors into account
7384 when deciding which selector to use.
7385 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7386 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7387 with a false third argument.
7388 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7389 to test whether the constant selector is valid and can_vec_perm_var_p
7390 to test whether a variable selector is valid.
7392 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7394 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7395 * optabs-query.c (can_vec_perm_p): Likewise.
7396 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7397 instead of vec_perm_indices.
7398 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7399 (vect_gen_perm_mask_checked): Likewise,
7400 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7401 (vect_gen_perm_mask_checked): Likewise,
7403 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7405 * optabs-query.h (qimode_for_vec_perm): Declare.
7406 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7407 (qimode_for_vec_perm): ...this new function.
7408 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7410 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7412 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7413 does not have a conditional at the top.
7415 2018-01-02 Richard Biener <rguenther@suse.de>
7417 * ipa-inline.c (big_speedup_p): Fix expression.
7419 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7422 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7425 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7429 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7430 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7431 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7432 cond_taken_branch_cost 3->4.
7434 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7436 PR tree-optimization/83581
7437 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7438 TODO_cleanup_cfg if any changes have been made.
7441 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7442 convert_modes if target mode has the right side, but different mode
7446 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7447 last argument when extracting from CONCAT. If either from_real or
7448 from_imag is NULL, use expansion through memory. If result is not
7449 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7450 the parts directly to inner mode, if even that fails, use expansion
7454 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7455 check for bswap in mode rather than HImode and use that in expand_unop
7458 Copyright (C) 2018 Free Software Foundation, Inc.
7460 Copying and distribution of this file, with or without modification,
7461 are permitted in any medium without royalty provided the copyright
7462 notice and this notice are preserved.