1 2018-02-22 Jakub Jelinek <jakub@redhat.com>
3 PR tree-optimization/84503
4 * gimple-ssa-store-merging.c (merged_store_group::merge_into): Compute
5 width as info->bitpos + info->bitsize - start.
6 (merged_store_group::merge_overlapping): Simplify width computation.
7 (check_no_overlap): New function.
8 (imm_store_chain_info::try_coalesce_bswap): Compute expected
9 start + width and last_order of the group, fail if check_no_overlap
11 (imm_store_chain_info::coalesce_immediate_stores): Don't merge info
12 to group if check_no_overlap fails.
14 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
16 * config/rs6000/altivec.md: Delete contraint arguments to
17 define_expand, define_split, and define_peephole2, and in
18 define_insn_and_split if always unused.
19 * config/rs6000/darwin.md: Ditto.
20 * config/rs6000/dfp.md: Ditto.
21 * config/rs6000/rs6000.md: Ditto.
22 * config/rs6000/sync.md: Ditto.
23 * config/rs6000/vector.md: Ditto.
24 * config/rs6000/vsx.md: Ditto.
26 2018-02-21 Segher Boessenkool <segher@kernel.crashing.org>
28 * config/rs6000/altivec.md: Write output control strings as braced
29 blocks instead of double-quoted strings.
30 * config/rs6000/darwin.md: Ditto.
31 * config/rs6000/rs6000.md: Ditto.
32 * config/rs6000/vector.md: Ditto.
33 * config/rs6000/vsx.md: Ditto.
35 2018-02-21 Jason Merrill <jason@redhat.com>
37 PR c++/84314 - ICE with templates and fastcall attribute.
38 * attribs.c (build_type_attribute_qual_variant): Remove assert.
40 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
42 * ipa-cp.c (determine_versionability): Fix comment typos.
44 2018-02-21 Jan Hubicka <hubicka@ucw.cz>
47 * ipa-cp.c (determine_versionability): Do not version functions caling
50 2018-02-21 Martin Liska <mliska@suse.cz>
53 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch):
54 Add "native" as a possible value.
55 * config/aarch64/aarch64.h (HAVE_LOCAL_CPU_DETECT): Define
56 the macro when native cpu detection is available.
58 2018-02-21 Martin Liska <mliska@suse.cz>
61 * common/config/arm/arm-common.c (arm_print_hint_for_arch_option):
62 Add "native" as a possible value.
63 * config/arm/arm.h (HAVE_LOCAL_CPU_DETECT): Define the macro
64 when native cpu detection is available.
66 2018-02-21 Jakub Jelinek <jakub@redhat.com>
67 Martin Sebor <msebor@redhat.com>
69 PR tree-optimization/84478
70 * gimple-fold.h (get_range_strlen): Add a bool argument defaulted to
72 * gimple-fold.c (get_range_strlen): Make minlen const and assume it
73 can't be NULL. Change FUZZY from bool to int, for 1 add PHI/COND_EXPR
74 support which is conservatively correct, for 2 only stay conservative
75 for maxlen. Formatting and comment capitalization fixes. Add STRICT
76 argument to the 2 argument get_range_strlen, adjust 6 arg
77 get_range_strlen caller and clear minmaxlen[0] and [1] if it returned
79 (get_maxval_strlen): Adjust 6 arg get_range_strlen caller.
80 (gimple_fold_builtin_strlen): Pass true as last argument to
83 2018-02-20 Martin Sebor <msebor@redhat.com>
86 * gimple-ssa-warn-restrict.c (builtin_memref::extend_offset_range): New.
87 (builtin_memref::set_base_and_offset): Same. Handle inner references.
88 (builtin_memref::builtin_memref): Factor out parts into
89 set_base_and_offset and call it.
91 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
94 * optabs-query.c (find_widening_optab_handler_and_mode): If from_mode
95 is a scalar_int_mode, assert that to_mode is a scalar_int_mode with
96 greater precision. If to_mode is a MODE_PARTIAL_INT, stop the
97 search at the associated MODE_INT.
99 2018-02-20 Jeff Law <law@redhat.com>
102 PR tree-optimization/81592
104 * gimple-ssa-sprintf.c (format_integer): Query EVRP range analyzer
105 for range data rather than using global data.
106 * gimple-ssa-sprintf.c (get_int_range): Query EVRP range analyzer for
107 range data rather than using global data.
108 * gimple-ssa-sprintf.c (get_int_range): Accept vr_values parameter
109 pass it to children as needed.
110 (struct directive::fmtresult): Similarly.
111 (struct directive::set_width): Similarly.
112 (struct directive::set_precision): Similarly.
113 (format_integer, format_directive, parse_directive): Similarly.
114 (format_none): Accept unnamed vr_values parameter.
115 (format_percent, format_floating, format_character): Similarly.
116 (format_string, format_plain): Similarly.
117 * gimple-ssa-sprintf.c (sprintf_dom_walker::handle_gimple_call): Query
118 the EVRP range analyzer for range data rather than using global data.
119 * gimple-ssa-sprintf.c: Include alloc-pool.h, vr-values.h and
120 gimple-ssa-evrp-analyze.h
121 (class sprintf_dom_walker): Add after_dom_children member function.
122 Add evrp_range_analyzer member.
123 (sprintf_dom_walker::before_dom_children): Call into the EVRP
124 range analyzer as needed.
125 (sprintf_dom_walker::after_dom_children): New member function.
126 * gimple-ssa-evrp-analyze.c (evrp_range_analyzer::enter): Do nothing
128 (evrp_range_analyzer::record_ranges_from_stmt): Likewise.
129 (evrp_range_analyzer::pop_to_marker): Likewise.
131 2018-02-20 Richard Sandiford <richard.sandiford@linaro.org>
133 PR tree-optimization/84419
134 * internal-fn.c (expand_call_mem_ref): Create a TARGET_MEM_REF
135 with the required type if its current type is compatible but
138 2018-02-20 Jakub Jelinek <jakub@redhat.com>
141 * match.pd (pow(C,x) -> exp(log(C)*x)): Delay all folding until
144 2018-02-20 Martin Liska <mliska@suse.cz>
147 * config/aarch64/aarch64.c (aarch64_print_hint_for_core_or_arch): Print
148 possible values if we don't have a hint.
150 2018-02-20 Martin Liska <mliska@suse.cz>
154 * final.c (shorten_branches): Build align_tab array with one
156 * opts.c (finish_options): Add alignment option limit check.
157 (MAX_CODE_ALIGN): Likewise.
158 (MAX_CODE_ALIGN_VALUE): Likewise.
159 * doc/invoke.texi: Document maximum allowed option value for
160 all -falign-* options.
162 2018-02-19 Jakub Jelinek <jakub@redhat.com>
165 * reg-notes.def (REG_CALL_ARG_LOCATION): New reg note.
166 * insn-notes.def (NOTE_INSN_CALL_ARG_LOCATION): Remove.
167 * var-tracking.c (emit_note_insn_var_location): Remove all references
168 to NOTE_INSN_CALL_ARG_LOCATION.
169 (emit_notes_in_bb): Emit arguments as REG_CALL_ARG_LOCATION note on
170 the CALL_INSN rather than separate NOTE_INSN_CALL_ARG_LOCATION note.
171 Use copy_rtx_if_shared.
172 * dwarf2out.c (gen_subprogram_die): Use XEXP with 0 instead of
173 NOTE_VAR_LOCATION on ca_loc->call_arg_loc_note.
174 (dwarf2out_var_location): Remove handling of
175 NOTE_INSN_CALL_ARG_LOCATION, instead handle REG_CALL_ARG_LOCATION note
177 * final.c (final_scan_insn): Remove all references to
178 NOTE_INSN_CALL_ARG_LOCATION.
179 (rest_of_clean_state): Likewise. Remove REG_CALL_ARG_LOCATION notes
180 before dumping final insns.
181 * except.c (emit_note_eh_region_end): Remove all references to
182 NOTE_INSN_CALL_ARG_LOCATION.
183 * config/alpha/alpha.c (alpha_pad_function_end): Likewise.
184 * config/c6x/c6x.c (c6x_gen_bundles): Likewise.
185 * config/arc/arc.c (hwloop_optimize): Likewise.
186 * config/arm/arm.c (create_fix_barrier): Likewise.
187 * config/s390/s390.c (s390_chunkify_start): Likewise.
188 * config/sh/sh.c (find_barrier): Likewise.
189 * config/i386/i386.c (rest_of_insert_endbranch,
190 ix86_seh_fixup_eh_fallthru): Likewise.
191 * config/xtensa/xtensa.c (hwloop_optimize): Likewise.
192 * config/iq2000/iq2000.c (final_prescan_insn): Likewise.
193 * config/frv/frv.c (frv_function_prologue): Likewise.
194 * emit-rtl.c (try_split): Likewise. Copy over REG_CALL_ARG_LOCATION
196 (note_outside_basic_block_p): Remove all references to
197 NOTE_INSN_CALL_ARG_LOCATION.
198 * gengtype.c (adjust_field_rtx_def): Likewise.
199 * print-rtl.c (rtx_writer::print_rtx_operand_code_0, print_insn):
201 * jump.c (cleanup_barriers, delete_related_insns): Likewise.
202 * cfgrtl.c (force_nonfallthru_and_redirect): Likewise.
205 * builtins.c (builtin_mathfn_code): Don't check if CALL_EXPR_FN (t)
208 PR tree-optimization/84452
209 * tree-vect-patterns.c (vect_recog_pow_pattern): Don't call
210 expand_simd_clones if targetm.simd_clone.compute_vecsize_and_simdlen
213 2018-02-19 Martin Liska <mliska@suse.cz>
216 * passes.def: Put pass_sancov_O0 before pass_lower_switch with -O0.
218 2018-02-19 Martin Liska <mliska@suse.cz>
219 Richard Sandiford <richard.sandiford@linaro.org>
221 PR tree-optimization/82491
222 * gimple-fold.c (get_base_constructor): Make earlier bail out
225 2018-02-19 Carl Love <cel@us.ibm.com>
227 * config/rs6000/rs6000-builtin.def: Change NEG macro expansions from
228 BU_ALTIVEC_A to BU_P8V_AV_1 and BU_ALTIVEC_OVERLOAD_1 to
230 * config/rs6000/rs6000-c.c: Change ALTIVEC_BUILTIN_VEC_NEG to
233 2018-02-19 Sebastian Perta <sebastian.perta@renesas.com>
235 * config/rl78/rl78.md (movdf): New define expand.
237 2018-02-19 Martin Liska <mliska@suse.cz>
240 * doc/invoke.texi: Fix typo.
241 * params.def (PARAM_MAX_LOOP_HEADER_INSNS): Likewise.
243 2018-02-18 Segher Boessenkool <segher@kernel.crashing.org>
245 * config/rs6000/rs6000.c (rs6000_option_override_internal): Don't
246 handle rs6000_single_float and rs6000_double_float specially for
249 2018-02-16 Jeff Law <law@redhat.com>
251 * config/rx/rx.c (add_pop_cfi_notes): New function.;
254 2018-02-16 Jakub Jelinek <jakub@redhat.com>
257 * ipa-inline.c (inline_small_functions): Fix a typo.
259 2018-02-16 Nathan Sidwell <nathan@acm.org>
261 * doc/extend.texi (Backwards Compatibility): Americanize 'behaviour'.
263 2018-02-16 Carl Love <cel@us.ibm.com>
265 * config/rs6000/rs6000-builtin.def: Add BU_P8V_VSX_2 macro definition.
266 Change FLOAT2 expansions from BU_VSX_2 to BU_P8V_VSX_2 and
267 from BU_VSX_OVERLOAD_2 to BU_P8V_OVERLOAD_2.
268 * config/rs6000/rs6000-c.c: Changed macro VSX_BUILTIN_VEC_FLOAT2
269 expansion to P8V_BUILTIN_VEC_FLOAT2.
271 2018-02-16 Vladimir Makarov <vmakarov@redhat.com>
273 PR rtl-optimization/70023
274 * lra-constraints.c (inherit_in_ebb): Take hard reg mode of
275 src_regno into account.
277 2018-02-16 Carl Love <cel@us.ibm.com>
279 * config/rs6000/altivec.h: Remove vec_vextract4b and vec_vinsert4b.
280 * config/rs6000/rs6000-builtin.def: Remove macro expansion for
281 VEXTRACT4B, VINSERT4B, VINSERT4B_DI and VEXTRACT4B.
282 * config/rs6000/rs6000.c: Remove case statements for
283 P9V_BUILTIN_VEXTRACT4B, P9V_BUILTIN_VEC_VEXTRACT4B,
284 P9V_BUILTIN_VINSERT4B, P9V_BUILTIN_VINSERT4B_DI,
285 and P9V_BUILTIN_VEC_VINSERT4B.
286 * config/rs6000/rs6000-c.c (altivec_expand_builtin): Remove entries for
287 P9V_BUILTIN_VEC_VEXTRACT4B and P9V_BUILTIN_VEC_VINSERT4B.
288 * config/rs6000/vsx.md:
289 * doc/extend.texi: Remove vec_vextract4b, non ABI definitions for
292 2018-02-16 Carl Love <cel@us.ibm.com>
294 * config/rs6000/altivec.h: Add builtin names vec_extract4b
296 * config/rs6000/rs6000-builtin.def: Add INSERT4B and EXTRACT4B
298 * config/rs6000/rs6000-c.c: Add the definitions for
299 P9V_BUILTIN_VEC_EXTRACT4B and P9V_BUILTIN_VEC_INSERT4B.
300 * config/rs6000/rs6000.c (altivec_expand_builtin): Add
301 P9V_BUILTIN_EXTRACT4B and P9V_BUILTIN_INSERT4B case statements.
302 * config/rs6000/vsx.md: Add define_insn extract4b. Add define_expand
303 definition for insert4b and define insn *insert3b_internal.
304 * doc/extend.texi: Add documentation for vec_extract4b.
306 2018-02-16 Nathan Sidwell <nathan@acm.org>
308 * doc/extend.texi (Backwards Compatibility): Mention friend
309 injection. Note for-scope is deprecated.
310 * doc/invoke.texi (-ffriend-injection): Deprecate.
312 2018-02-16 Segher Boessenkool <segher@kernel.crashing.org>
314 * combine.c (try_combine): When adjusting LOG_LINKS for the destination
315 that moved to I2, also allow destinations that are a paradoxical
316 subreg (instead of a normal reg).
318 2018-02-16 Oleg Endo <olegendo@gcc.gnu.org>
321 * config/rx/rx.c (rx_fuse_in_memory_bitop): Convert shift operand
324 2018-02-16 Richard Biener <rguenther@suse.de>
326 PR tree-optimization/84037
327 PR tree-optimization/84016
329 * config/i386/i386.c (ix86_builtin_vectorization_cost):
330 Adjust vec_construct for the fact we need additional higher latency
331 128bit inserts for AVX256 and AVX512 vector builds.
332 (ix86_add_stmt_cost): Scale vector construction cost for
335 2018-02-16 Richard Biener <rguenther@suse.de>
337 PR tree-optimization/84417
338 * tree-ssa.c (non_rewritable_mem_ref_base): Properly constrain
339 the MEM_REF offset when conversion to BIT_FIELD_REF is desired.
340 (non_rewritable_lvalue_p): Likewise, use poly-ints.
342 2018-02-16 Martin Liska <mliska@suse.cz>
345 * internal-fn.def (ASAN_CHECK): Set proper flags.
346 (ASAN_MARK): Likewise.
348 2018-02-16 Julia Koval <julia.koval@intel.com>
350 * config/i386/i386.c (ix86_option_override_internal): Remove PTA_CLWB
353 2018-02-16 Jakub Jelinek <jakub@redhat.com>
356 * config/aarch64/cortex-a57-fma-steering.c (fma_forest::merge_forest):
357 Use ++iter rather than iter++ for std::list iterators.
358 (func_fma_steering::dfs): Likewise. Don't delete nodes right away,
359 defer deleting them until all nodes in the forest are processed. Do
360 free even leaf nodes. Change to_process into auto_vec.
363 * system.h (BROKEN_VALUE_INITIALIZATION): Define for GCC < 4.3.
364 * vec.h (vec_default_construct): Use memset instead of placement new
365 if BROKEN_VALUE_INITIALIZATION is defined.
366 * hash-table.h (hash_table<Descriptor, Allocator>::empty_slow): Use
367 memset instead of value initialization if BROKEN_VALUE_INITIALIZATION
370 PR rtl-optimization/83723
371 * lra-int.h (lra_substitute_pseudo): Add DEBUG_P argument.
372 * lra.c (lra_substitute_pseudo): Likewise. If true, use
373 gen_rtx_raw_SUBREG instead of gen_rtx_SUBREG. Pass DEBUG_P to
375 (lra_substitute_pseudo_within_insn): Adjust lra_substitute_pseudo
377 * lra-constraints.c (inherit_reload_reg, split_reg): Likewise.
379 2018-02-16 Eric Botcazou <ebotcazou@adacore.com>
381 PR rtl-optimization/81443
382 * rtlanal.c (num_sign_bit_copies1) <SUBREG>: Do not propagate results
383 from inner REGs to paradoxical SUBREGs.
385 2018-02-16 Richard Biener <rguenther@suse.de>
387 PR tree-optimization/84399
388 * graphite-scop-detection.c (scop_detection::stmt_simple_for_scop_p):
389 For operands we can analyze at their definition make sure we can
390 analyze them at each use as well.
392 2018-02-16 Richard Biener <rguenther@suse.de>
394 PR tree-optimization/84190
395 * tree-ssa.c (non_rewritable_mem_ref_base): Do not touch
396 volatile accesses if the decl isn't volatile.
398 2018-02-15 Jason Merrill <jason@redhat.com>
400 PR c++/84314 - ICE with templates and fastcall attribute.
401 * attribs.c (build_type_attribute_qual_variant): Don't clobber
402 TYPE_CANONICAL on an existing type.
404 2018-02-15 Jakub Jelinek <jakub@redhat.com>
406 PR tree-optimization/84383
407 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Don't look at
408 dstoff nor call operand_equal_p if dstbase is NULL.
410 PR tree-optimization/84334
411 * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
412 also a CONSTANT_CLASS_P, punt.
414 2018-02-14 Jim Wilson <jimw@sifive.com>
416 * config/riscv/riscv.c (riscv_first_stack_step): Move locals after
417 first SMALL_OPERAND check. New local min_second_step. Move assert
418 to where locals are set. Add TARGET_RVC support.
419 * config/riscv/riscv.h (C_SxSP_BITS, SWSP_REACH, SDSP_REACH): New.
421 2018-02-14 Indu Bhagat <indu.bhagat@oracle.com>
423 * doc/invoke.texi: Correct -Wformat-overflow code sample.
425 2018-02-14 Martin Sebor <msebor@redhat.com>
427 PR tree-optimization/83698
428 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For
429 arrays constrain the offset range to their bounds.
430 (builtin_access::strcat_overlap): Adjust the bounds of overlap offset.
431 (builtin_access::overlap): Avoid setting the size of overlap if it's
433 (maybe_diag_overlap): Also consider arrays when deciding what values
434 of offsets to include in diagnostics.
436 2018-02-14 Martin Sebor <msebor@redhat.com>
439 * attribs.c (diag_attr_exclusions): Consider the exclusion(s)
440 that correspond to the kind of a declaration.
442 2018-02-14 John David Anglin <danglin@gcc.gnu.org>
445 * config/pa/pa.md: Load address of PIC label using the linkage table
446 if the label is nonlocal.
448 2018-02-14 Kelvin Nilsen <kelvin@gcc.gnu.org>
450 * config/rs6000/rs6000.c (rs6000_option_override_internal): Issue
451 warning message if user requests -maltivec=be.
452 * doc/invoke.texi: Document deprecation of -maltivec=be.
454 2018-02-14 Will Schmidt <will_schmidt@vnet.ibm.com>
457 * config/rs6000/rs6000-c.c: Update definitions for
458 ALTIVEC_BUILTIN_VEC_SLD, ALTIVEC_BUILTIN_VEC_SLDW,
459 VEC_XXSLDWI and ALTIVEC_BUILTIN_VEC_XXPERMDI builtins.
461 2018-02-14 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
464 * config/i386/cetintrin.h: Remove _rdssp[d|q] and
465 add _get_ssp intrinsics. Remove argument from
466 __builtin_ia32_rdssp[d|q].
467 * config/i386/i386-builtin-types.def: Add UINT_FTYPE_VOID.
468 * config/i386/i386-builtin.def: Remove argument from
469 __builtin_ia32_rdssp[d|q].
470 * config/i386/i386.c: Use UINT_FTYPE_VOID. Use
471 ix86_expand_special_args_builtin for _rdssp[d|q].
472 * config/i386/i386.md: Remove argument from rdssp[si|di] insn.
473 Clear register before usage.
474 * doc/extend.texi: Remove argument from __builtin_ia32_rdssp[d|q].
475 Add documentation for new _get_ssp and _inc_ssp intrinsics.
477 2018-02-14 Richard Sandiford <richard.sandiford@linaro.org>
479 PR tree-optimization/84357
480 * tree-data-ref.c (object_address_invariant_in_loop_p): Check
481 operand 1 of an ARRAY_REF too.
483 2018-02-14 Oleg Endo <olegendo@gcc.gnu.org>
486 * config/rx/rx-protos.h (rx_reg_dead_or_unused_after_insn,
487 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
489 (set_of_reg): New struct.
490 (rx_find_set_of_reg, rx_find_use_of_reg): New functions.
491 * config/rx/rx.c (rx_reg_dead_or_unused_after_insn,
492 rx_copy_reg_dead_or_unused_notes, rx_fuse_in_memory_bitop): New
494 * config/rx/rx.md (andsi3, iorsi3, xorsi3): Convert to insn_and_split.
495 Split into bitclr, bitset, bitinvert patterns if appropriate.
496 (*bitset, *bitinvert, *bitclr): Convert to named insn_and_split and
497 use rx_fuse_in_memory_bitop.
498 (*bitset_in_memory, *bitinvert_in_memory, *bitclr_in_memory): Convert
499 to named insn, correct maximum insn length.
501 2018-02-14 Jozef Lawrynowicz <jozefl.gcc@gmail.com>
504 * machmode.def: Define a complex mode for PARTIAL_INT.
505 * genmodes.c (complex_class): Return MODE_COMPLEX_INT for
507 * doc/rtl.texi: Document CSPImode.
508 * config/msp430/msp430.c (msp430_hard_regno_nregs): Add CPSImode
510 (msp430_hard_regno_nregs_with_padding): Likewise.
512 2018-02-13 Peter Bergner <bergner@vnet.ibm.com>
515 * config/rs6000/rs6000.c (mem_operand_gpr): Disallow altivec addresses.
517 2018-02-13 Segher Boessenkool <segher@kernel.crashing.org>
519 PR rtl-optimization/84169
520 * combine.c (try_combine): New variable split_i2i3. Set it to true if
521 we generated a parallel as new i3 and we split that to new i2 and i3
522 instructions. Handle split_i2i3 similar to swap_i2i3: scan the
523 LOG_LINKs of i3 to see which of those need to link to i2 now. Link
524 those to i2, not i1. Partially rewrite this scan code.
526 2018-02-13 Jakub Jelinek <jakub@redhat.com>
529 * stor-layout.c (place_field): For variable length fields, adjust
530 offset_align afterwards not just based on the field's alignment,
531 but also on the size.
534 * match.pd (pow(C,x) -> exp(log(C)*x)): Use exp2s and log2s instead
535 of exps and logs in the use_exp2 case.
537 2018-02-13 Jeff Law <law@redhat.com>
539 * config/rl/rl78.c (rl78_attribute_table): Fix terminator and
542 * config/rl78/rl78.c (rl78_handle_func_attribute): Mark
545 2018-02-13 Alexandre Oliva <aoliva@redhat.com>
549 * common.opt (gas-loc-support, gas-locview-support): New.
550 (ginline-points, ginternal-reset-location-views): New.
551 * doc/invoke.texi: Document them. Use @itemx where intended.
552 (gvariable-location-views): Adjust.
553 * target.def (reset_location_view): New.
554 * doc/tm.texi.in (DWARF2_ASM_VIEW_DEBUG_INFO): New.
555 (TARGET_RESET_LOCATION_VIEW): New.
556 * doc/tm.texi: Rebuilt.
557 * dwarf2out.c (dwarf2out_default_as_loc_support): New.
558 (dwarf2out_default_as_locview_support): New.
559 (output_asm_line_debug_info): Use option variables.
560 (dwarf2out_maybe_output_loclist_view_pair): Likewise.
561 (output_loc_list): Likewise.
562 (add_high_low_attributes): Check option variables.
563 Don't output entry view attribute in strict mode.
564 (gen_inlined_subroutine_die): Check option variables.
565 (dwarf2out_inline_entry): Likewise.
566 (init_sections_and_labels): Likewise.
567 (dwarf2out_early_finish): Likewise.
568 (maybe_reset_location_view): New, from...
569 (dwarf2out_var_location): ... here. Call it.
570 * debug.h (dwarf2out_default_as_loc_support): Declare.
571 (dwarf2out_default_as_locview_support): Declare.
572 * hooks.c (hook_int_rtx_insn_0): New.
573 * hooks.h (hook_int_rtx_insn_0): Declare.
574 * toplev.c (process_options): Take -gas-loc-support and
575 -gas-locview-support from dwarf2out. Enable
576 -gvariable-location-views by default only with locview
577 assembler support. Enable -ginternal-reset-location-views by
578 default only if the target defines the corresponding hook.
579 Enable -ginline-points by default if location views are
580 enabled; force it disabled if statement frontiers are
582 * tree-inline.c (expand_call_inline): Check option variables.
583 * tree-ssa-live.c (remove_unused_scope_block_p): Likewise.
585 2018-02-13 Richard Sandiford <richard.sandiford@linaro.org>
587 PR tree-optimization/84321
588 * tree-vrp.c (intersect_range_with_nonzero_bits): Fix VR_ANTI_RANGE
589 handling. Also check whether the anti-range contains any values
590 that satisfy the mask; switch to a VR_RANGE if not.
592 2018-02-13 Paolo Bonzini <bonzini@gnu.org>
595 * internal-fn.def (ASAN_CHECK, ASAN_MARK): Revert changes to fnspec.
597 2018-02-13 Martin Jambor <mjambor@suse.cz>
600 * ipa-param-manipulation.c (ipa_modify_call_arguments): Use location
601 of call statements, also set location of a load to a temporary.
603 2018-02-13 Sebastian Perta <sebastian.perta@renesas.com>
605 * config/rl78/rl78.c (add_vector_labels): New function.
606 * config/rl78/rl78.c (rl78_handle_vector_attribute): New function.
607 * config/rl78/rl78.c (rl78_start_function): Call add_vector_labels.
608 * config/rl78/rl78.c (rl78_handle_func_attribute): Removed the assert
609 which checks that no arguments are passed.
610 * config/rl78/rl78.c (rl78_attribute_table): Add "vector" attribute.
611 * doc/extend.texi: Documentation for the new attribute.
613 2018-02-13 Andreas Schwab <schwab@suse.de>
615 * config/riscv/linux.h (CPP_SPEC): Define.
617 2018-02-13 Jakub Jelinek <jakub@redhat.com>
620 * config/i386/i386.c (ix86_init_mmx_sse_builtins): Pass
621 OPTION_MASK_ISA_AES | OPTION_MASK_ISA_SSE2 instead of
622 OPTION_MASK_ISA_AES as first argument to def_builtin_const
623 for AES builtins. Pass OPTION_MASK_ISA_PCLMUL | OPTION_MASK_ISA_SSE2
624 instead of OPTION_MASK_ISA_PCLMUL as first argument to
625 def_builtin_const for __builtin_ia32_pclmulqdq128 builtin.
626 * config/i386/wmmintrin.h: If __SSE2__ is not defined, enable it
627 temporarily for AES and PCLMUL builtins.
629 PR tree-optimization/84339
630 * gimple-fold.c (get_range_strlen): Set *FLEXP to true when handling
631 ARRAY_REF where first operand is array_at_struct_end_p COMPONENT_REF.
635 * match.pd (pow(C,x) -> exp(log(C)*x)): Optimize instead into
636 exp2(log2(C)*x) if C is a power of 2 and c99 runtime is available.
637 * generic-match-head.c (canonicalize_math_after_vectorization_p): New
639 * gimple-match-head.c (canonicalize_math_after_vectorization_p): New
641 * omp-simd-clone.h: New file.
642 * omp-simd-clone.c: Include omp-simd-clone.h.
643 (expand_simd_clones): No longer static.
644 * tree-vect-patterns.c: Include fold-const-call.h, attribs.h,
645 cgraph.h and omp-simd-clone.h.
646 (vect_recog_pow_pattern): Optimize pow(C,x) to exp(log(C)*x).
647 (vect_recog_widen_shift_pattern): Formatting fix.
648 (vect_pattern_recog_1): Don't check optab for calls.
651 * config/i386/sse.md (<avx512>_vpermi2var<mode>3_mask): Force
652 operands[2] into a REG before using gen_lowpart on it.
654 2018-02-12 Jeff Law <law@redhat.com>
657 * config/sh/sh.c (find_barrier): Consider a sibling call
660 * cse.c (try_back_substitute_reg): Move any REG_ARGS_SIZE note when
661 successfully back substituting a reg.
663 2018-02-12 Richard Biener <rguenther@suse.de>
665 PR tree-optimization/84037
666 * tree-vect-slp.c (vect_analyze_slp_cost): Add visited
667 parameter, move visited init to caller.
668 (vect_slp_analyze_operations): Separate cost from validity
669 check, initialize visited once for all instances.
670 (vect_schedule_slp): Analyze map to CSE vectorized nodes once
672 * tree-vect-stmts.c (vect_model_simple_cost): Make early
674 (vect_model_promotion_demotion_cost): Likewise.
675 (vectorizable_bswap): Guard cost modeling with !slp_node
676 instead of !PURE_SLP_STMT to avoid double-counting on hybrid
678 (vectorizable_call): Likewise.
679 (vectorizable_conversion): Likewise.
680 (vectorizable_assignment): Likewise.
681 (vectorizable_shift): Likewise.
682 (vectorizable_operation): Likewise.
683 (vectorizable_store): Likewise.
684 (vectorizable_load): Likewise.
685 (vectorizable_condition): Likewise.
686 (vectorizable_comparison): Likewise.
688 2018-02-12 Paolo Bonzini <bonzini@gnu.org>
691 * internal-fn.def (ASAN_CHECK): Fix fnspec to account for return value.
692 (ASAN_MARK): Fix fnspec to account for return value, change pointer
693 argument from 'R' to 'W' so that the pointed-to datum is clobbered.
695 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
698 * params.def (inline-min-speedup): Increase from 8 to 15.
699 (max-inline-insns-auto): Decrease from 40 to 30.
700 * ipa-split.c (consider_split): Add some buffer for function to
701 be considered inlining candidate.
702 * invoke.texi (max-inline-insns-auto, inline-min-speedup): UPdate
705 2018-02-12 Richard Biener <rguenther@suse.de>
707 PR tree-optimization/84037
708 * tree-vect-slp.c (vect_build_slp_tree_2): Try swapping the
709 matched stmts if we cannot swap the non-matched ones.
711 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
713 * config/i386/avx512fintrin.h (_mm_mask_scalef_round_sd,
714 _mm_maskz_scalef_round_sd, _mm_mask_scalef_round_ss,
715 _mm_maskz_scalef_round_ss): New intrinsics.
716 (__builtin_ia32_scalefsd_round, __builtin_ia32_scalefss_round): Fix.
717 * config/i386/i386-builtin.def (__builtin_ia32_scalefsd_round,
718 __builtin_ia32_scalefss_round): Remove.
719 (__builtin_ia32_scalefsd_mask_round,
720 __builtin_ia32_scalefss_mask_round): New intrinsics.
721 * config/i386/sse.md (vmscalef<mode><round_name>): Renamed to ...
722 (vmscalef<mode><mask_scalar_name><round_scalar_name>): ... this.
723 ((match_operand:VF_128 2 "<round_nimm_predicate>"
724 "<round_constraint>")): Changed to ...
725 ((match_operand:VF_128 2 "<round_scalar_nimm_predicate>"
726 "<round_scalar_constraint>")): ... this.
727 ("vscalef<ssescalarmodesuffix>\t{<round_op3>%2, %1, %0|
728 %0, %1, %2<round_op3>}"): Changed to ...
729 ("vscalef<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%2, %1,
730 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %1,
731 %2<round_scalar_mask_op3>}"): ... this.
732 * config/i386/subst.md (round_scalar_nimm_predicate): New.
734 2018-02-12 Olga Makhotina <olga.makhotina@intel.com>
736 * config/i386/avx512fintrin.h (_mm_mask_sqrt_round_sd)
737 (_mm_maskz_sqrt_round_sd, _mm_mask_sqrt_round_ss)
738 (_mm_maskz_sqrt_round_ss): New intrinsics.
739 (__builtin_ia32_sqrtsd_round, __builtin_ia32_sqrtss_round): Remove.
740 (__builtin_ia32_sqrtsd_mask_round)
741 (__builtin_ia32_sqrtss_mask_round): New builtins.
742 * config/i386/i386-builtin.def (__builtin_ia32_sqrtsd_round)
743 (__builtin_ia32_sqrtss_round): Remove.
744 (__builtin_ia32_sqrtsd_mask_round)
745 (__builtin_ia32_sqrtss_mask_round): New builtins.
746 * config/i386/sse.md (vmsqrt<mode>2<round_name>): Renamed to ...
747 (vmsqrt<mode>2<mask_scalar_name><round_scalar_name>): ... this.
748 ((match_operand:VF_128 1 "vector_operand"
749 "xBm,<round_constraint>")): Changed to ...
750 ((match_operand:VF_128 1 "vector_operand"
751 "xBm,<round_scalar_constraint>")): ... this.
752 (vsqrt<ssescalarmodesuffix>\t{<round_op3>%1, %2, %0|
753 %0, %2, %<iptr>1<round_op3>}): Changed to ...
754 (vsqrt<ssescalarmodesuffix>\t{<round_scalar_mask_op3>%1, %2,
755 %0<mask_scalar_operand3>|%0<mask_scalar_operand3>, %2,
756 %<iptr>1<round_scalar_mask_op3>}): ... this.
757 ((set_attr "prefix" "<round_prefix>")): Changed to ...
758 ((set_attr "prefix" "<round_scalar_prefix>")): ... this.
760 2018-02-11 Steven Munroe <munroesj@gcc.gnu.org>
763 * config/rs6000/mmintrin.h (_mm_cmpeq_pi32 [_ARCH_PWR9]):
764 Cast vec_cmpeq result to correct type.
765 * config/rs6000/mmintrin.h (_mm_cmpgt_pi32 [_ARCH_PWR9]):
766 Cast vec_cmpgt result to correct type.
768 2018-02-11 Alexandre Oliva <aoliva@redhat.com>
770 * final.c (final_scan_insn_1): Renamed from...
771 (final_scan_insn): ... this. New wrapper, to recover
772 seen from the outermost call in recursive ones.
773 * config/sparc/sparc.c (output_return): Drop seen from call.
774 (output_sibcall): Likewise.
775 * config/visium/visium.c (output_branch): Likewise.
777 2018-02-10 John David Anglin <danglin@gcc.gnu.org>
779 * config/pa/pa.c (hppa_profile_hook): Mark SYMBOL_REF for _mcount as
782 2018-02-10 Alan Modra <amodra@gmail.com>
785 * config/rs6000/rs6000.md (split_stack_return): Remove (use ..).
786 Specify LR as an input.
788 2018-02-10 Jakub Jelinek <jakub@redhat.com>
791 * omp-low.c (maybe_remove_omp_member_access_dummy_vars,
792 remove_member_access_dummy_vars): New functions.
793 (lower_omp_for, lower_omp_taskreg, lower_omp_target,
794 lower_omp_1, execute_lower_omp): Use them.
796 PR rtl-optimization/84308
797 * shrink-wrap.c (spread_components): Release todo vector.
799 2018-02-09 Vladimir Makarov <vmakarov@redhat.com>
801 PR rtl-optimization/57193
802 * ira-color.c (struct allocno_color_data): Add member
803 conflict_allocno_hard_prefs.
804 (update_conflict_allocno_hard_prefs): New.
805 (bucket_allocno_compare_func): Add a preference based on
806 conflict_allocno_hard_prefs.
807 (push_allocno_to_stack): Update conflict_allocno_hard_prefs.
808 (color_allocnos): Remove a dead code. Initiate
809 conflict_allocno_hard_prefs. Call update_costs_from_prefs.
811 2018-02-09 Jakub Jelinek <jakub@redhat.com>
814 * config/rs6000/vsx.md (p9_xxbrq_v16qi): Change input operand
815 constraint from =wa to wa. Avoid a subreg on the output operand,
816 instead use a pseudo and subreg it in a move.
817 (p9_xxbrd_<mode>): Changed to ...
818 (p9_xxbrd_v2di): ... this insn, without VSX_D iterator.
819 (p9_xxbrd_v2df): New expander.
820 (p9_xxbrw_<mode>): Changed to ...
821 (p9_xxbrw_v4si): ... this insn, without VSX_W iterator.
822 (p9_xxbrw_v4sf): New expander.
824 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
826 * config/rx/rx.md (movsicc): Update expander to be matched by GCC.
828 2018-02-09 Peter Bergner <bergner@vnet.ibm.com>
831 * config/rs6000/vsx.md (vsx_mul_v2di): Handle generating a 64-bit
832 multiply in 32-bit mode.
833 (vsx_div_v2di): Handle generating a 64-bit signed divide in 32-bit mode.
834 (vsx_udiv_v2di): Handle generating a 64-bit unsigned divide in 32-bit
837 2018-02-09 Sebastian Perta <sebastian.perta@renesas.com>
839 * config/rx/constraints.md (CALL_OP_SYMBOL_REF): Added new constraint
840 to allow or block "symbol_ref" depending on the value of TARGET_JSR.
841 * config/rx/rx.md (call_internal): Use CALL_OP_SYMBOL_REF.
842 * config/rx/rx.md (call_value_internal): Use CALL_OP_SYMBOL_REF.
844 2018-02-09 Pierre-Marie de Rodat <derodat@adacore.com>
847 * dwarf2out.c (is_trivial_indirect_ref): New function.
848 (dwarf2out_late_global_decl): Do not generate a location
849 attribute for variables that have a non-trivial DECL_VALUE_EXPR
850 and that are not defined in the current unit.
852 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
854 * optabs.c (prepare_cmp_insn): Try harder to emit a direct comparison
855 instead of a libcall for UNORDERED.
857 2018-02-09 Tamar Christina <tamar.christina@arm.com>
860 * config/arm/arm-c.c (arm_cpu_builtins): Un-define __ARM_FEATURE_LDREX,
861 __ARM_ARCH_PROFILE, __ARM_ARCH_ISA_THUMB, __ARM_FP and __ARM_NEON_FP.
863 2018-02-09 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
866 * config/s390/s390.c (s390_set_current_function): Invoke
867 s390_indirect_branch_settings also if fndecl didn't change.
869 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
871 * config/rs6000/rs6000.md (blockage): Set length to zero.
873 2018-02-09 Eric Botcazou <ebotcazou@adacore.com>
875 * expr.c (optimize_bitfield_assignment_op): Remove obsolete assertion.
877 2018-02-09 Jakub Jelinek <jakub@redhat.com>
880 * gcc.c (STATIC_LIBASAN_LIBS, STATIC_LIBTSAN_LIBS,
881 STATIC_LIBLSAN_LIBS, STATIC_LIBUBSAN_LIBS): Handle -static like
885 * var-tracking.c (vt_add_function_parameter): Punt for non-onepart
886 PARALLEL incoming that failed vt_get_decl_and_offset check.
889 * output.h (bss_initializer_p): Add NAMED argument, defaulted to false.
890 * varasm.c (bss_initializer_p): Add NAMED argument, if true, ignore
892 (get_variable_section): For decls in named .bss* sections pass true as
893 second argument to bss_initializer_p.
895 2018-02-09 Marek Polacek <polacek@redhat.com>
896 Jakub Jelinek <jakub@redhat.com>
899 * fold-const.c (fold_indirect_ref_1): Use VECTOR_TYPE_P macro.
900 Formatting fixes. Verify first that tree_fits_poly_int64_p (op01).
901 Sync some changes from cxx_fold_indirect_ref.
903 2018-02-09 Alexandre Oliva <aoliva@redhat.com>
905 * cfgexpand.c (expand_gimple_basic_block): Handle inline entry
907 * dwarf2out.c (dwarf2_debug_hooks): Enable inline_entry hook.
908 (BLOCK_INLINE_ENTRY_LABEL): New.
909 (dwarf2out_var_location): Disregard inline entry markers.
910 (inline_entry_data): New struct.
911 (inline_entry_data_hasher): New hashtable type.
912 (inline_entry_data_hasher::hash): New.
913 (inline_entry_data_hasher::equal): New.
914 (inline_entry_data_table): New variable.
915 (add_high_low_attributes): Add DW_AT_entry_pc and
916 DW_AT_GNU_entry_view attributes if a pending entry is found
917 in inline_entry_data_table. Add old entry_pc attribute only
918 if debug nonbinding markers are disabled.
919 (gen_inlined_subroutine_die): Set BLOCK_DIE if nonbinding
921 (block_within_block_p, dwarf2out_inline_entry): New.
922 (dwarf2out_finish): Check that no entries remained in
923 inline_entry_data_table.
924 * final.c (reemit_insn_block_notes): Handle inline entry notes.
925 (final_scan_insn, notice_source_line): Likewise.
926 (rest_of_clean_state): Skip inline entry markers.
927 * gimple-pretty-print.c (dump_gimple_debug): Handle inline entry
929 * gimple.c (gimple_build_debug_inline_entry): New.
930 * gimple.h (enum gimple_debug_subcode): Add
931 GIMPLE_DEBUG_INLINE_ENTRY.
932 (gimple_build_debug_inline_entry): Declare.
933 (gimple_debug_inline_entry_p): New.
934 (gimple_debug_nonbind_marker_p): Adjust.
935 * insn-notes.def (INLINE_ENTRY): New.
936 * print-rtl.c (rtx_writer::print_rtx_operand_code_0): Handle
937 inline entry marker notes.
938 (print_insn): Likewise.
939 * rtl.h (NOTE_MARKER_P): Add INLINE_ENTRY support.
940 (INSN_DEBUG_MARKER_KIND): Likewise.
941 (GEN_RTX_DEBUG_MARKER_INLINE_ENTRY_PAT): New.
942 * tree-inline.c (expand_call_inline): Build and insert
943 debug_inline_entry stmt.
944 * tree-ssa-live.c (remove_unused_scope_block_p): Preserve
945 inline entry blocks early, if nonbind markers are enabled.
946 (dump_scope_block): Dump fragment info.
947 * var-tracking.c (reemit_marker_as_note): Handle inline entry note.
948 * doc/gimple.texi (gimple_debug_inline_entry_p): New.
949 (gimple_build_debug_inline_entry): New.
950 * doc/invoke.texi (gstatement-frontiers, gno-statement-frontiers):
951 Enable/disable inline entry points too.
952 * doc/rtl.texi (NOTE_INSN_INLINE_ENTRY): New.
953 (DEBUG_INSN): Describe inline entry markers.
955 * common.opt (gvariable-location-views): New.
956 (gvariable-location-views=incompat5): New.
957 * config.in: Rebuilt.
958 * configure: Rebuilt.
959 * configure.ac: Test assembler for view support.
960 * dwarf2asm.c (dw2_asm_output_symname_uleb128): New.
961 * dwarf2asm.h (dw2_asm_output_symname_uleb128): Declare.
962 * dwarf2out.c (var_loc_view): New typedef.
963 (struct dw_loc_list_struct): Add vl_symbol, vbegin, vend.
964 (dwarf2out_locviews_in_attribute): New.
965 (dwarf2out_locviews_in_loclist): New.
966 (dw_val_equal_p): Compare val_view_list of dw_val_class_view_lists.
967 (enum dw_line_info_opcode): Add LI_adv_address.
968 (struct dw_line_info_table): Add view.
969 (RESET_NEXT_VIEW, RESETTING_VIEW_P): New macros.
970 (DWARF2_ASM_VIEW_DEBUG_INFO): Define default.
971 (zero_view_p): New variable.
972 (ZERO_VIEW_P): New macro.
973 (output_asm_line_debug_info): New.
974 (struct var_loc_node): Add view.
975 (add_AT_view_list, AT_loc_list): New.
976 (add_var_loc_to_decl): Add view param. Test it against last.
977 (new_loc_list): Add view params. Record them.
978 (AT_loc_list_ptr): Handle loc and view lists.
979 (view_list_to_loc_list_val_node): New.
980 (print_dw_val): Handle dw_val_class_view_list.
981 (size_of_die): Likewise.
982 (value_format): Likewise.
983 (loc_list_has_views): New.
984 (gen_llsym): Set vl_symbol too.
985 (maybe_gen_llsym, skip_loc_list_entry): New.
986 (dwarf2out_maybe_output_loclist_view_pair): New.
987 (output_loc_list): Output view list or entries too.
988 (output_view_list_offset): New.
989 (output_die): Handle dw_val_class_view_list.
990 (output_dwarf_version): New.
991 (output_compilation_unit_header): Use it.
992 (output_skeleton_debug_sections): Likewise.
993 (output_rnglists, output_line_info): Likewise.
994 (output_pubnames, output_aranges): Update version comments.
995 (output_one_line_info_table): Output view numbers in asm comments.
996 (dw_loc_list): Determine current endview, pass it to new_loc_list.
997 Call maybe_gen_llsym.
998 (loc_list_from_tree_1): Adjust.
999 (add_AT_location_description): Create view list attribute if
1000 needed, check it's absent otherwise.
1001 (convert_cfa_to_fb_loc_list): Adjust.
1002 (maybe_emit_file): Call output_asm_line_debug_info for test.
1003 (dwarf2out_var_location): Reset views as needed. Precompute
1004 add_var_loc_to_decl args. Call get_attr_min_length only if we have the
1005 attribute. Set view.
1006 (new_line_info_table): Reset next view.
1007 (set_cur_line_info_table): Call output_asm_line_debug_info for test.
1008 (dwarf2out_source_line): Likewise. Output view resets and labels to
1009 the assembler, or select appropriate line info opcodes.
1010 (prune_unused_types_walk_attribs): Handle dw_val_class_view_list.
1011 (optimize_string_length): Catch it. Adjust.
1012 (resolve_addr): Copy vl_symbol along with ll_symbol. Handle
1013 dw_val_class_view_list, and remove it if no longer needed.
1014 (hash_loc_list): Hash view numbers.
1015 (loc_list_hasher::equal): Compare them.
1016 (optimize_location_lists): Check whether a view list symbol is
1017 needed, and whether the locview attribute is present, and
1018 whether they match. Remove the locview attribute if no longer
1020 (index_location_lists): Call skip_loc_list_entry for test.
1021 (dwarf2out_finish): Call output_asm_line_debug_info for test.
1022 Use output_dwarf_version.
1023 * dwarf2out.h (enum dw_val_class): Add dw_val_class_view_list.
1024 (struct dw_val_node): Add val_view_list.
1025 * final.c (SEEN_NEXT_VIEW): New.
1026 (set_next_view_needed): New.
1027 (clear_next_view_needed): New.
1028 (maybe_output_next_view): New.
1029 (final_start_function): Rename to...
1030 (final_start_function_1): ... this. Take pointer to FIRST,
1031 add SEEN parameter. Emit param bindings in the initial view.
1032 (final_start_function): Reintroduce SEEN-less interface.
1033 (final): Rename to...
1034 (final_1): ... this. Take SEEN parameter. Output final pending
1035 next view at the end.
1036 (final): Reintroduce seen-less interface.
1037 (final_scan_insn): Output pending next view before switching
1038 sections or ending a block. Mark the next view as needed when
1039 outputting variable locations. Notify debug backend of section
1040 changes, and of location view changes.
1041 (rest_of_handle_final): Adjust.
1042 * toplev.c (process_options): Autodetect value for debug variable
1043 location views option. Warn on incompat5 without -gdwarf-5.
1044 * doc/invoke.texi (gvariable-location-views): New.
1045 (gvariable-location-views=incompat5): New.
1046 (gno-variable-location-views): New.
1048 2018-02-08 David Malcolm <dmalcolm@redhat.com>
1050 PR tree-optimization/84136
1051 * tree-cfg.c (find_taken_edge_computed_goto): Remove assertion
1052 that the result of find_edge is non-NULL.
1054 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1057 * config/i386/x86-tune-costs.h (skylake_cost): Fix cost of
1058 storing integer register in SImode. Fix cost of 256 and 512
1059 byte aligned SSE register store.
1061 2018-02-08 Sergey Shalnov <sergey.shalnov@intel.com>
1063 * config/i386/i386.c (ix86_multiplication_cost): Fix
1064 multiplication cost for TARGET_AVX512DQ.
1066 2018-02-08 Marek Polacek <polacek@redhat.com>
1068 PR tree-optimization/84238
1069 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Verify the result of
1072 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1074 PR tree-optimization/84265
1075 * tree-vect-stmts.c (vectorizable_store): Don't treat
1076 VMAT_CONTIGUOUS accesses as grouped.
1077 (vectorizable_load): Likewise.
1079 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1081 PR tree-optimization/81635
1082 * wide-int.h (wi::round_down_for_mask, wi::round_up_for_mask): Declare.
1083 * wide-int.cc (wi::round_down_for_mask, wi::round_up_for_mask)
1084 (test_round_for_mask): New functions.
1085 (wide_int_cc_tests): Call test_round_for_mask.
1086 * tree-vrp.h (intersect_range_with_nonzero_bits): Declare.
1087 * tree-vrp.c (intersect_range_with_nonzero_bits): New function.
1088 * tree-data-ref.c (split_constant_offset_1): Use it to refine the
1089 range returned by get_range_info.
1091 2018-02-08 Jan Hubicka <hubicka@ucw.cz>
1094 * cgraph.h (symtab_node::output_to_lto_symbol_table_p): Declare
1095 * symtab.c: Include builtins.h
1096 (symtab_node::output_to_lto_symbol_table_p): Move here
1097 from lto-streamer-out.c:output_symbol_p.
1098 * lto-streamer-out.c (write_symbol): Turn early exit to assert.
1099 (output_symbol_p): Move all logic to symtab.c
1100 (produce_symtab): Update.
1102 2018-02-08 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1104 * config/s390/s390-opts.h (enum indirect_branch): Define.
1105 * config/s390/s390-protos.h (s390_return_addr_from_memory)
1106 (s390_indirect_branch_via_thunk)
1107 (s390_indirect_branch_via_inline_thunk): Add function prototypes.
1108 (enum s390_indirect_branch_type): Define.
1109 * config/s390/s390.c (struct s390_frame_layout, struct
1110 machine_function): Remove.
1111 (indirect_branch_prez10thunk_mask, indirect_branch_z10thunk_mask)
1112 (indirect_branch_table_label_no, indirect_branch_table_name):
1114 (INDIRECT_BRANCH_NUM_OPTIONS): Define macro.
1115 (enum s390_indirect_branch_option): Define.
1116 (s390_return_addr_from_memory): New function.
1117 (s390_handle_string_attribute): New function.
1118 (s390_attribute_table): Add new attribute handler.
1119 (s390_execute_label): Handle UNSPEC_EXECUTE_JUMP patterns.
1120 (s390_indirect_branch_via_thunk): New function.
1121 (s390_indirect_branch_via_inline_thunk): New function.
1122 (s390_function_ok_for_sibcall): When jumping via thunk disallow
1123 sibling call optimization for non z10 compiles.
1124 (s390_emit_call): Force indirect branch target to be a single
1125 register. Add r1 clobber for non-z10 compiles.
1126 (s390_emit_epilogue): Emit return jump via return_use expander.
1127 (s390_reorg): Handle JUMP_INSNs as execute targets.
1128 (s390_option_override_internal): Perform validity checks for the
1129 new command line options.
1130 (s390_indirect_branch_attrvalue): New function.
1131 (s390_indirect_branch_settings): New function.
1132 (s390_set_current_function): Invoke s390_indirect_branch_settings.
1133 (s390_output_indirect_thunk_function): New function.
1134 (s390_code_end): Implement target hook.
1135 (s390_case_values_threshold): Implement target hook.
1136 (TARGET_ASM_CODE_END, TARGET_CASE_VALUES_THRESHOLD): Define target
1138 * config/s390/s390.h (struct s390_frame_layout)
1139 (struct machine_function): Move here from s390.c.
1140 (TARGET_INDIRECT_BRANCH_NOBP_RET)
1141 (TARGET_INDIRECT_BRANCH_NOBP_JUMP)
1142 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_THUNK)
1143 (TARGET_INDIRECT_BRANCH_NOBP_JUMP_INLINE_THUNK)
1144 (TARGET_INDIRECT_BRANCH_NOBP_CALL)
1145 (TARGET_DEFAULT_INDIRECT_BRANCH_TABLE)
1146 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EXRL)
1147 (TARGET_INDIRECT_BRANCH_THUNK_NAME_EX)
1148 (TARGET_INDIRECT_BRANCH_TABLE): Define macros.
1149 * config/s390/s390.md (UNSPEC_EXECUTE_JUMP)
1150 (INDIRECT_BRANCH_THUNK_REGNUM): Define constants.
1151 (mnemonic attribute): Add values which aren't recognized
1153 ("*cjump_long", "*icjump_long", "*basr", "*basr_r"): Disable
1154 pattern for branch conversion. Fix mnemonic attribute.
1155 ("*c<code>", "*sibcall_br", "*sibcall_value_br", "*return"): Emit
1156 indirect branch via thunk if requested.
1157 ("indirect_jump", "<code>"): Expand patterns for branch conversion.
1158 ("*indirect_jump"): Disable for branch conversion using out of
1160 ("indirect_jump_via_thunk<mode>_z10")
1161 ("indirect_jump_via_thunk<mode>")
1162 ("indirect_jump_via_inlinethunk<mode>_z10")
1163 ("indirect_jump_via_inlinethunk<mode>", "*casesi_jump")
1164 ("casesi_jump_via_thunk<mode>_z10", "casesi_jump_via_thunk<mode>")
1165 ("casesi_jump_via_inlinethunk<mode>_z10")
1166 ("casesi_jump_via_inlinethunk<mode>", "*basr_via_thunk<mode>_z10")
1167 ("*basr_via_thunk<mode>", "*basr_r_via_thunk_z10")
1168 ("*basr_r_via_thunk", "return<mode>_prez10"): New pattern.
1169 ("*indirect2_jump"): Disable for branch conversion.
1170 ("casesi_jump"): Turn into expander and expand patterns for branch
1172 ("return_use"): New expander.
1173 ("*return"): Emit return via thunk and rename it to ...
1174 ("*return<mode>"): ... this one.
1175 * config/s390/s390.opt: Add new options and and enum for the
1178 2018-02-08 Richard Sandiford <richard.sandiford@linaro.org>
1180 * lra-constraints.c (match_reload): Unconditionally use
1181 gen_lowpart_SUBREG, rather than selecting between that
1182 and equivalent gen_rtx_SUBREG code.
1184 2018-02-08 Richard Biener <rguenther@suse.de>
1186 PR tree-optimization/84233
1187 * tree-ssa-phiprop.c (propagate_with_phi): Use separate
1188 changed flag instead of boguously re-using phi_inserted.
1190 2018-02-08 Martin Jambor <mjambor@suse.cz>
1192 * hsa-gen.c (get_symbol_for_decl): Set program allocation for
1193 static local variables.
1195 2018-02-08 Richard Biener <rguenther@suse.de>
1197 PR tree-optimization/84278
1198 * tree-vect-stmts.c (vectorizable_store): When looking for
1199 smaller vector types to perform grouped strided loads/stores
1200 make sure the mode is supported by the target.
1201 (vectorizable_load): Likewise.
1203 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1205 * config/aarch64/aarch64.c (aarch64_components_for_bb):
1206 Increase LDP/STP opportunities by adding adjacent callee-saves.
1208 2018-02-08 Wilco Dijkstra <wdijkstr@arm.com>
1210 PR rtl-optimization/84068
1211 PR rtl-optimization/83459
1212 * haifa-sched.c (rank_for_schedule): Fix SCHED_PRESSURE_MODEL sorting.
1214 2018-02-08 Aldy Hernandez <aldyh@redhat.com>
1216 PR tree-optimization/84224
1217 * gimple-ssa-warn-alloca.c (pass_walloca::execute): Remove assert.
1218 * calls.c (gimple_alloca_call_p): Only return TRUE when we have
1221 2018-02-07 Iain Sandoe <iain@codesourcery.com>
1224 * config/rs6000/altivec.md (*restore_world): Remove LR use.
1225 * config/rs6000/predicates.md (restore_world_operation): Adjust op
1226 count, remove one USE.
1228 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
1230 * doc/install.texi (Configuration): Document the
1231 --with-long-double-format={ibm,ieee} PowerPC configuration
1235 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
1236 Convert from define_expand to be define_insn_and_split. Rework
1237 float/double/_Float128 conversions to QI/HI/SImode to work with
1238 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
1239 conversions to QI/HImode types did a store and then a load to
1240 truncate the value. For conversions to VSX registers, don't split
1241 the insn, instead emit the code directly. Use the code iterator
1242 any_fix to combine signed and unsigned conversions.
1243 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
1244 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
1245 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1246 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
1247 (fix_<mode>di2_hw): Likewise.
1248 (fixuns_<mode>di2_hw): Likewise.
1249 (fix_<mode>si2_hw): Likewise.
1250 (fixuns_<mode>si2_hw): Likewise.
1251 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
1252 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
1253 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
1254 fix<uns>_trunc<SFDF:mode>si2_p8.
1255 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
1257 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
1258 (fix<uns>_<mode>_mem): Likewise.
1259 (fctiw<u>z_<mode>_mem): Likewise.
1260 (fix<uns>_<mode>_mem): Likewise.
1261 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
1262 the register allocator from doing a direct move to the GPRs to do
1263 a store, and instead use the ISA 3.0 store byte/half-word from
1264 vector register instruction. For IEEE 128-bit floating point,
1265 also optimize stores of 32-bit ints.
1266 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
1268 2018-02-07 Alan Hayward <alan.hayward@arm.com>
1270 * genextract.c (push_pathstr_operand): New function to support
1272 (walk_rtx): Call push_pathstr_operand.
1273 (print_path): Support [a-zA-Z].
1275 2018-02-07 Richard Biener <rguenther@suse.de>
1277 PR tree-optimization/84037
1278 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
1279 (cse_and_gimplify_to_preheader): Declare.
1280 (vect_get_place_in_interleaving_chain): Likewise.
1281 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1283 (_loop_vec_info::~_loop_vec_info): Delete it.
1284 (cse_and_gimplify_to_preheader): New function.
1285 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
1286 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
1287 (vectorizable_load): Likewise. For grouped stores always base
1288 the IV on the first element.
1289 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
1290 condition before gimplifying.
1292 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1294 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
1295 *DIV_EXPR and *MOD_EXPR.
1297 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
1300 * config/i386/i386.c (ix86_option_override_internal): Mask out
1301 the CF_SET bit when checking -fcf-protection.
1303 2018-02-07 Tom de Vries <tom@codesourcery.com>
1306 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
1309 2018-02-07 Richard Biener <rguenther@suse.de>
1311 PR tree-optimization/84204
1312 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
1315 PR tree-optimization/84205
1316 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
1317 special-case isl_ast_op_zdiv_r.
1319 PR tree-optimization/84223
1320 * graphite-scop-detection.c (gather_bbs::before_dom_children):
1321 Only add conditions from within the region.
1322 (gather_bbs::after_dom_children): Adjust.
1324 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
1327 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
1328 * config/avr/avr.md: Only post-reload split REG-REG moves if
1329 either register is REGERAL_REG_P.
1331 2018-02-07 Jakub Jelinek <jakub@redhat.com>
1333 PR tree-optimization/84235
1334 * tree-ssa-scopedtables.c
1335 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
1336 if the subtraction is performed in floating point type where NaNs are
1337 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
1338 build 1. Formatting fix.
1340 2018-02-06 Jakub Jelinek <jakub@redhat.com>
1343 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
1344 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
1345 and skip it regardless of bb boundaries. Use CALL_P macro,
1346 don't test INSN_P (insn) together with CALL_P or JUMP_P check
1347 unnecessarily, formatting fix.
1349 2018-02-06 Michael Collison <michael.collison@arm.com>
1351 * config/arm/thumb2.md:
1352 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
1353 (*thumb_mov_notscc): Ditto.
1355 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
1358 * config/rs6000/rs6000.md (su code attribute): Use "u" for
1359 unsigned_fix, not "s".
1361 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1363 * configure.ac (gcc_fn_eh_frame_ro): New function.
1364 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
1365 correct .eh_frame permissions.
1366 * configure: Regenerate.
1368 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
1370 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
1373 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1375 * config/rs6000/rs6000.c (rs6000_option_override_internal):
1376 Display warning message for -mno-speculate-indirect-jumps.
1378 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
1380 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
1382 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
1384 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
1386 PR tree-optimization/84225
1387 * tree-eh.c (find_trapping_overflow): Only call
1388 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
1390 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1393 * config/i386/i386.c: Reimplement the check of possible options
1394 -mibt/-mshstk conbination. Change error messages.
1395 * doc/invoke.texi: Fix a typo: remove extra '='.
1397 2018-02-06 Marek Polacek <polacek@redhat.com>
1399 PR tree-optimization/84228
1400 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
1402 2018-02-06 Tamar Christina <tamar.christina@arm.com>
1405 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
1406 emitted arch directives.
1407 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
1408 __ARM_FEATURE_COPROC before changing architectures.
1410 2018-02-06 Richard Biener <rguenther@suse.de>
1412 * config/i386/i386.c (print_reg): Fix typo.
1413 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
1415 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
1417 * configure: Regenerate.
1419 2018-02-05 Martin Sebor <msebor@redhat.com>
1421 PR tree-optimization/83369
1422 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
1425 2018-02-05 Martin Liska <mliska@suse.cz>
1427 * doc/invoke.texi: Cherry-pick upstream r323995.
1429 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
1431 * ira.c (ira_init_register_move_cost): Adjust comment.
1433 2018-02-05 Martin Liska <mliska@suse.cz>
1435 PR gcov-profile/84137
1436 * doc/gcov.texi: Fix typo in documentation.
1438 2018-02-05 Martin Liska <mliska@suse.cz>
1440 PR gcov-profile/83879
1441 * doc/gcov.texi: Document necessity of --dynamic-list-data when
1442 using dlopen functionality.
1444 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
1446 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
1447 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
1448 _mm_maskz_range_ss, _mm_mask_range_round_ss,
1449 _mm_maskz_range_round_ss): New intrinsics.
1450 (__builtin_ia32_rangesd128_round)
1451 (__builtin_ia32_rangess128_round): Remove.
1452 (__builtin_ia32_rangesd128_mask_round,
1453 __builtin_ia32_rangess128_mask_round): New builtins.
1454 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
1455 __builtin_ia32_rangess128_round): Remove.
1456 (__builtin_ia32_rangesd128_mask_round,
1457 __builtin_ia32_rangess128_mask_round): New builtins.
1458 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
1459 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
1460 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
1461 "<round_saeonly_constraint>")): Changed to ...
1462 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
1463 "<round_saeonly_scalar_constraint>")): ... this.
1464 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
1465 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
1466 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
1467 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
1468 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
1470 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
1472 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
1474 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
1475 Remove all values except native, 8540 and 8548.
1477 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
1479 * config/i386/i386.c (ix86_output_function_return): Pass
1480 INVALID_REGNUM, instead of -1, as invalid register number to
1481 indirect_thunk_name and output_indirect_thunk.
1483 2018-02-02 Julia Koval <julia.koval@intel.com>
1485 * config.gcc: Add -march=icelake.
1486 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
1487 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
1488 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
1489 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
1490 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
1491 (processor_target_table): Add icelake.
1492 (ix86_option_override_internal): Handle new PTAs.
1493 (get_builtin_code_for_version): Handle icelake.
1494 (M_INTEL_COREI7_ICELAKE): New.
1495 (fold_builtin_cpu): Handle icelake.
1496 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
1497 * doc/invoke.texi: Add -march=icelake.
1499 2018-02-02 Julia Koval <julia.koval@intel.com>
1501 * config/i386/i386.c (ix86_option_override_internal): Change flags type
1502 to wide_int_bitmask.
1503 * wide-int-bitmask.h: New.
1505 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
1508 * config/i386/i386.md: Replace Pmode with word_mode in
1509 builtin_setjmp_setup and builtin_longjmp to support x32.
1511 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
1515 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
1517 (rs6000_supported_cpu_names): New static variable.
1518 (linux_cpu_translation_table): Likewise.
1519 (elf_platform) <cpu>: Define new static variable and use it.
1520 Translate kernel AT_PLATFORM name to canonical name if needed.
1521 Error if platform name is unknown.
1523 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
1526 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
1528 2018-02-01 Jeff Law <law@redhat.com>
1531 * config/i386/i386.c (release_scratch_register_on_entry): Add new
1532 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
1533 the scratch if RELEASE_VIA_POP is false.
1534 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
1535 If we have to save a temporary register, decrement SIZE appropriately.
1536 Pass new arguments to release_scratch_register_on_entry.
1537 (ix86_adjust_stack_and_probe): Likewise.
1538 (ix86_emit_probe_stack_range): Pass new arguments to
1539 release_scratch_register_on_entry.
1541 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
1543 PR rtl-optimization/84157
1544 * combine.c (change_zero_ext): Use REG_P predicate in
1545 front of HARD_REGISTER_P predicate.
1547 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
1549 * config/avr/avr.c (avr_option_override): Move disabling of
1550 -fdelete-null-pointer-checks to...
1551 * common/config/avr/avr-common.c (avr_option_optimization_table):
1554 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1556 PR tree-optimization/81635
1557 * tree-data-ref.c (split_constant_offset_1): For types that
1558 wrap on overflow, try to use range info to prove that wrapping
1561 2018-02-01 Renlin Li <renlin.li@arm.com>
1564 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
1566 (aarch64_register_move_cost): Likewise.
1567 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
1569 (REG_CLASS_NAMES): Likewise.
1570 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
1571 TAILCALL_ADDR_REGS. Remove IP registers.
1572 * config/aarch64/aarch64.md (Ucs): Update register constraint.
1574 2018-02-01 Richard Biener <rguenther@suse.de>
1576 * domwalk.h (dom_walker::dom_walker): Add additional constructor
1577 for specifying RPO order and allow NULL for that.
1578 * domwalk.c (dom_walker::dom_walker): Likewise.
1579 (dom_walker::walk): Handle NULL RPO order.
1580 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
1582 (rewrite_update_dom_walker): Likewise.
1583 (mark_def_dom_walker): Likewise.
1585 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1587 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
1588 (aarch64_maybe_expand_sve_subreg_move): Declare.
1589 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
1590 * config/aarch64/predicates.md (aarch64_any_register_operand): New
1592 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
1593 that are semantically a reverse operation.
1594 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
1595 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
1596 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
1598 (aarch64_can_change_mode_class): For big-endian, forbid changes
1599 between two SVE modes if they have different element sizes.
1601 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1603 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
1604 the TImode handling for big-endian targets.
1606 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1608 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
1609 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
1611 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
1612 Remove BSWAP handing for big-endian targets and use the form of
1613 LD1RQ appropariate for the mode.
1615 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1617 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
1618 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
1621 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
1624 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
1625 check for operands that need to go through aarch64_sve_reload_be.
1627 2018-02-01 Jakub Jelinek <jakub@redhat.com>
1629 PR tree-optimization/81661
1630 PR tree-optimization/84117
1631 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
1632 * tree-eh.c: Include gimplify.h.
1633 (find_trapping_overflow, replace_trapping_overflow,
1634 rewrite_to_non_trapping_overflow): New functions.
1635 * tree-vect-loop.c: Include tree-eh.h.
1636 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
1637 * tree-data-ref.c: Include tree-eh.h.
1638 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
1640 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
1642 PR rtl-optimization/84123
1643 * combine.c (change_zero_ext): Check if hard register satisfies
1644 can_change_dest_mode before calling gen_lowpart_SUBREG.
1646 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
1649 * ira.c (ira_init_register_move_cost): Remove assert.
1651 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1653 PR rtl-optimization/84071
1654 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
1655 * doc/tm.texi: Regenerate.
1657 2018-01-31 Richard Biener <rguenther@suse.de>
1659 PR tree-optimization/84132
1660 * tree-data-ref.c (analyze_miv_subscript): Properly
1661 check whether evolution_function_is_affine_multivariate_p
1662 before calling gcd_of_steps_may_divide_p.
1664 2018-01-31 Julia Koval <julia.koval@intel.com>
1667 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
1668 * config/i386/i386.md (rdpid_rex64) New.
1669 (rdpid): Make 32bit only.
1671 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
1674 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
1675 an IDENTIFIER_NODE for FUNCTION_TYPE's.
1677 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1680 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
1682 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
1684 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
1686 PR rtl-optimization/84071
1687 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
1688 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
1690 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1692 * config/arc/arc.c (arc_handle_aux_attribute): New function.
1693 (arc_attribute_table): Add 'aux' attribute.
1694 (arc_in_small_data_p): Consider aux like variables.
1695 (arc_is_aux_reg_p): New function.
1696 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
1697 (arc_get_aux_arg): New function.
1698 (prepare_move_operands): Handle aux-register access.
1699 (arc_handle_aux_attribute): New function.
1700 * doc/extend.texi (ARC Variable attributes): Add subsection.
1702 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
1704 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
1705 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
1706 (arc_attribute_table): Add 'uncached' attribute.
1707 (arc_print_operand): Print '.di' flag for uncached memory
1709 (arc_in_small_data_p): Do not consider for small data the uncached
1711 (arc_is_uncached_mem_p): New function.
1712 * config/arc/predicates.md (compact_store_memory_operand): Check
1713 for uncached memory accesses.
1714 (nonvol_nonimm_operand): Likewise.
1715 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
1717 2018-01-31 Jakub Jelinek <jakub@redhat.com>
1720 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
1721 falign-loops=): Add Optimization flag.
1723 2018-01-30 Jeff Law <law@redhat.com>
1726 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
1727 INT_REGISTERS_SAVED. Check it prior to calling
1728 get_scratch_register_on_entry.
1729 (ix86_adjust_stack_and_probe): Similarly.
1730 (ix86_emit_probe_stack_range): Similarly.
1731 (ix86_expand_prologue): Corresponding changes.
1733 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1736 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
1737 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
1739 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
1742 * lra-constraints.c (curr_insn_transform): Process AND in the
1745 2018-01-30 Jakub Jelinek <jakub@redhat.com>
1747 PR rtl-optimization/83986
1748 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
1749 dependence against last_pending_memory_flush in addition to
1752 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
1754 PR tree-optimization/81611
1755 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
1758 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
1761 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
1764 2018-01-30 Richard Biener <rguenther@suse.de>
1765 Jakub Jelinek <jakub@redhat.com>
1767 PR tree-optimization/84111
1768 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
1769 inner loops added during recursion, as they don't have up-to-date
1772 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1775 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
1776 (can_inline_edge_by_limits_p): ... here.
1777 (can_early_inline_edge_p, check_callers,
1778 update_caller_keys, update_callee_keys, recursive_inlining,
1779 add_new_edges_to_heap, speculation_useful_p,
1780 inline_small_functions,
1781 inline_small_functions, flatten_function,
1782 inline_to_all_callers_1): Update.
1784 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
1786 * profile-count.c (profile_count::combine_with_ipa_count): Handle
1789 2018-01-30 Richard Biener <rguenther@suse.de>
1791 PR tree-optimization/83008
1792 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
1793 invariant and constant vector uses in stmts when they need
1796 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
1799 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
1800 * configure: Regenerate.
1802 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1804 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
1806 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
1807 Use gen_rtx_REG rather than gen_lowpart.
1809 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1811 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
1812 rather than 0 when creating partial subregs.
1814 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
1816 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
1819 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
1822 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
1823 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
1824 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
1825 flags. This restores the settings used before the 2017-07-24.
1826 Turning off pre increment/decrement/modify allows IVOPTS to
1827 optimize DF/SF loops where the index is an int.
1829 2018-01-29 Richard Biener <rguenther@suse.de>
1830 Kelvin Nilsen <kelvin@gcc.gnu.org>
1833 * tree-vect-stmts.c (vectorizable_call): Don't call
1834 targetm.vectorize_builtin_md_vectorized_function if callee is
1837 2018-01-22 Carl Love <cel@us.ibm.com>
1839 * doc/extend.tex: Fix typo in second arg in
1840 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
1842 2018-01-29 Richard Biener <rguenther@suse.de>
1844 PR tree-optimization/84086
1845 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
1846 (flush_ssaname_freelist): When SSA names were released reset
1847 the SCEV hash table.
1849 2018-01-29 Richard Biener <rguenther@suse.de>
1851 PR tree-optimization/84057
1852 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
1853 removed paths when removing edges.
1855 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
1857 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
1858 -mfunction-return=@var{choice}.
1860 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
1863 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
1865 (layout::print_source_line): Likewise.
1866 (test_get_line_width_without_trailing_whitespace): Add test cases.
1868 2018-01-27 Jakub Jelinek <jakub@redhat.com>
1871 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
1874 2018-01-26 Jim Wilson <jimw@sifive.com>
1876 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
1878 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
1881 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1883 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
1884 and CMP + SUB-immediate -> SUBS.
1886 2018-01-26 Martin Sebor <msebor@redhat.com>
1888 PR tree-optimization/83896
1889 * tree-ssa-strlen.c (get_string_len): Rename...
1890 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
1891 Avoid assuming length is constant.
1892 (handle_char_store): Use HOST_WIDE_INT for string length.
1894 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
1897 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
1898 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
1900 2018-01-26 Richard Biener <rguenther@suse.de>
1902 PR rtl-optimization/84003
1903 * dse.c (record_store): Only record redundant stores when
1904 the earlier store aliases at least all accesses the later one does.
1906 2018-01-26 Jakub Jelinek <jakub@redhat.com>
1908 PR rtl-optimization/83985
1909 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
1910 REG_CFA_RESTORE insns.
1911 (delete_unmarked_insns): Don't ignore separate shrink wrapping
1912 REG_CFA_RESTORE insns here.
1915 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
1916 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
1918 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1920 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
1921 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
1922 (arc_init): Likewise.
1923 (arc_override_options): Likewise.
1924 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
1926 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
1928 * config/arc/arc.h (TARGET_DBNZ): Define.
1929 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
1930 properly set the tune attribute.
1931 (dbnz): Use TARGET_DBNZ guard.
1932 * config/arc/arc.opt (mtune): Add core3 option.
1934 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1936 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
1937 recognize new pic like addresses.
1938 (arc_delegitimize_address): Clean up.
1940 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1942 * config/arc/arc-arches.def: Option mrf16 valid for all
1944 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
1945 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
1946 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
1947 * config/arc/arc-tables.opt: Regenerate.
1948 * config/arc/arc.c (arc_conditional_register_usage): Handle
1949 reduced register file case.
1950 (arc_file_start): Set must have build attributes.
1951 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
1953 * config/arc/arc.opt (mrf16): Add new option.
1954 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
1955 * config/arc/genmultilib.awk: Handle new mrf16 option.
1956 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
1957 * config/arc/t-multilib: Regenerate.
1958 * doc/invoke.texi (ARC Options): Document mrf16 option.
1960 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1962 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
1963 * config/arc/arc.c (arc_handle_secure_attribute): New function.
1964 (arc_attribute_table): Add 'secure_call' attribute.
1965 (arc_print_operand): Print secure call operand.
1966 (arc_function_ok_for_sibcall): Don't optimize tail calls when
1968 (arc_is_secure_call_p): New function. * config/arc/arc.md
1969 (call_i): Add support for sjli instruction.
1970 (call_value_i): Likewise.
1971 * config/arc/constraints.md (Csc): New constraint.
1973 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
1974 John Eric Martin <John.Martin@emmicro-us.com>
1976 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
1977 * config/arc/arc.c (_arc_jli_section): New struct.
1978 (arc_jli_section): New type.
1979 (rc_jli_sections): New static variable.
1980 (arc_handle_jli_attribute): New function.
1981 (arc_attribute_table): Add jli_always and jli_fixed attribute.
1982 (arc_file_end): New function.
1983 (TARGET_ASM_FILE_END): Define.
1984 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
1985 (arc_add_jli_section): New function.
1986 (jli_call_scan): Likewise.
1987 (arc_reorg): Call jli_call_scan.
1988 (arc_output_addsi): Remove 'S' from printing asm operand.
1989 (arc_is_jli_call_p): New function.
1990 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
1992 (movhi_insn): Likewise.
1993 (movsi_insn): Likewise.
1994 (movsi_set_cc_insn): Likewise.
1995 (loadqi_update): Likewise.
1996 (load_zeroextendqisi_update): Likewise.
1997 (load_signextendqisi_update): Likewise.
1998 (loadhi_update): Likewise.
1999 (load_zeroextendhisi_update): Likewise.
2000 (load_signextendhisi_update): Likewise.
2001 (loadsi_update): Likewise.
2002 (loadsf_update): Likewise.
2003 (movsicc_insn): Likewise.
2004 (bset_insn): Likewise.
2005 (bxor_insn): Likewise.
2006 (bclr_insn): Likewise.
2007 (bmsk_insn): Likewise.
2008 (bicsi3_insn): Likewise.
2009 (cmpsi_cc_c_insn): Likewise.
2010 (movsi_ne): Likewise.
2011 (movsi_cond_exec): Likewise.
2012 (clrsbsi2): Likewise.
2023 (call_i): Remove 'S' asm letter, add jli instruction.
2024 (call_value_i): Likewise.
2025 * config/arc/arc.op (mjli-always): New option.
2026 * config/arc/constraints.md (Cji): New constraint.
2027 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
2029 (subsf3_fpx): Likewise.
2030 (mulsf3_fpx): Likewise.
2031 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
2033 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
2035 * doc/invoke.texi (ARC): Document mjli-always option.
2037 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
2039 * config/rl78/rl78.c (rl78_addsi3_internal): If operand 2 is const
2040 avoid addition with 0 and use incw and decw where possible.
2042 2018-01-26 Richard Biener <rguenther@suse.de>
2044 PR tree-optimization/81082
2045 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
2046 association if it requires casting to unsigned.
2047 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
2048 from fold_plusminus_mult_expr to catch important cases late when
2049 range info is available.
2051 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2053 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
2054 * configure.ac (hidden_linkonce): New test.
2055 * configure: Regenerate.
2056 * config.in: Regenerate.
2058 2018-01-26 Julia Koval <julia.koval@intel.com>
2060 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
2061 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
2062 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
2063 _mm_mask_bitshuffle_epi64_mask): Fix type.
2064 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
2065 USI_FTYPE_V4DI_V4DI_USI): Remove.
2066 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
2067 __builtin_ia32_vpshufbitqmb256_mask,
2068 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
2069 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
2070 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
2072 2018-01-26 Alan Modra <amodra@gmail.com>
2075 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
2076 UNSPEC_VBPERMQ. Sort other unspecs.
2078 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
2080 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
2082 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
2085 * predict.c (drop_profile): Do not push/pop cfun; update also
2087 (handle_missing_profiles): Fix logic looking for zero profiles.
2089 2018-01-25 Jakub Jelinek <jakub@redhat.com>
2092 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
2093 on functions with #pragma omp declare simd or functions with simd
2095 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
2096 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
2097 Remove trailing \n from warning_at calls.
2099 2018-01-25 Tom de Vries <tom@codesourcery.com>
2102 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2103 for neutered workers.
2105 2018-01-24 Joseph Myers <joseph@codesourcery.com>
2108 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
2109 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
2111 2018-01-24 Jeff Law <law@redhat.com>
2114 * i386.c (get_probe_interval): Move to earlier point.
2115 (ix86_compute_frame_layout): If -fstack-clash-protection and
2116 the frame is larger than the probe interval, then use pushes
2117 to save registers rather than reg->mem moves.
2118 (ix86_expand_prologue): Remove conditional for int_registers_saved
2121 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
2124 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
2125 min/max for never referenced object.
2127 2018-01-24 Jakub Jelinek <jakub@redhat.com>
2130 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
2132 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
2133 attributes from DECL_ATTRIBUTES (decl) without affecting
2134 DECL_ATTRIBUTES (current_function_decl).
2135 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
2136 functions with non-NULL DECL_ABSTRACT_ORIGIN.
2138 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
2140 PR tree-optimization/83979
2141 * fold-const.c (fold_comparison): Use constant_boolean_node
2142 instead of boolean_{true,false}_node.
2144 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
2146 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
2149 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2151 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2152 Simplify the clause that sets the length attribute.
2153 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2154 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
2155 clause that sets the length attribute.
2156 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2158 2018-01-24 Tom de Vries <tom@codesourcery.com>
2161 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
2162 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
2163 Add strict parameter.
2164 (prevent_branch_around_nothing): Insert dummy insn between branch to
2165 label and label with no ptx insn inbetween.
2166 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
2168 2018-01-24 Tom de Vries <tom@codesourcery.com>
2171 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
2172 for neutered threads in warp.
2173 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
2175 2018-01-24 Richard Biener <rguenther@suse.de>
2177 PR tree-optimization/83176
2178 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
2181 2018-01-24 Richard Biener <rguenther@suse.de>
2183 PR tree-optimization/82819
2184 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
2185 code generating pluses that are no-ops in the target precision.
2187 2018-01-24 Richard Biener <rguenther@suse.de>
2190 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
2192 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2194 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
2195 to merge probabilities.
2196 * predict.c (probably_never_executed): Also mark as cold functions
2197 with global 0 profile and guessed local profile.
2198 * profile-count.c (profile_probability::combine_with_count): New
2200 * profile-count.h (profile_probability::operator*,
2201 profile_probability::operator*=, profile_probability::operator/,
2202 profile_probability::operator/=): Reduce precision to adjusted
2203 and set value to guessed on contradictory divisions.
2204 (profile_probability::combine_with_freq): Remove.
2205 (profile_probability::combine_wiht_count): Declare.
2206 (profile_count::force_nonzero):: Set to adjusted.
2207 (profile_count::probability_in):: Set quality to adjusted.
2208 * tree-ssa-tail-merge.c (replace_block_by): Use
2211 2018-01-23 Andrew Waterman <andrew@sifive.com>
2212 Jim Wilson <jimw@sifive.com>
2214 * config/riscv/riscv.c (riscv_stack_boundary): New.
2215 (riscv_option_override): Set riscv_stack_boundary. Handle
2216 riscv_preferred_stack_boundary_arg.
2217 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
2218 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
2219 (STACK_BOUNDARY): Set to riscv_stack_boundary.
2220 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
2221 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
2222 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
2224 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
2227 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
2228 of struct ix86_frame.
2229 (ix86_expand_epilogue): Likewise. Add a local variable for
2230 the reg_save_offset field in struct ix86_frame.
2232 2018-01-23 Bin Cheng <bin.cheng@arm.com>
2234 PR tree-optimization/82604
2235 * tree-loop-distribution.c (enum partition_kind): New enum item
2236 PKIND_PARTIAL_MEMSET.
2237 (partition_builtin_p): Support above new enum item.
2238 (generate_code_for_partition): Ditto.
2239 (compute_access_range): Differentiate cases that equality can be
2240 proven at all loops, the innermost loops or no loops.
2241 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
2242 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
2243 (finalize_partitions, distribute_loop): Don't fuse partition of
2244 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
2245 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
2248 2018-01-23 Martin Liska <mliska@suse.cz>
2250 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
2251 order to ignore the predictor.
2252 (PRED_POLYMORPHIC_CALL): Likewise.
2253 (PRED_RECURSIVE_CALL): Likewise.
2255 2018-01-23 Martin Liska <mliska@suse.cz>
2257 * tree-profile.c (tree_profiling): Print function header to
2258 aware reader which function we are working on.
2259 * value-prof.c (gimple_find_values_to_profile): Do not print
2260 not interesting value histograms.
2262 2018-01-23 Martin Liska <mliska@suse.cz>
2264 * profile-count.h (enum profile_quality): Add
2265 profile_uninitialized as the first value. Do not number values
2266 as they are zero based.
2267 (profile_count::verify): Update sanity check.
2268 (profile_probability::verify): Likewise.
2270 2018-01-23 Nathan Sidwell <nathan@acm.org>
2272 * doc/invoke.texi (ffor-scope): Deprecate.
2274 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2276 PR tree-optimization/83510
2277 * domwalk.c (set_all_edges_as_executable): New function.
2278 (dom_walker::dom_walker): Convert bool param
2279 "skip_unreachable_blocks" to enum reachability. Move setup of
2280 edge flags to set_all_edges_as_executable and only do it when
2281 reachability is REACHABLE_BLOCKS.
2282 * domwalk.h (enum dom_walker::reachability): New enum.
2283 (dom_walker::dom_walker): Convert bool param
2284 "skip_unreachable_blocks" to enum reachability.
2285 (set_all_edges_as_executable): New decl.
2286 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
2287 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
2289 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
2290 but converting true to REACHABLE_BLOCKS.
2291 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
2293 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
2294 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
2295 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
2297 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
2298 if check_all_array_refs will be called.
2300 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2302 * tree.c (selftest::test_location_wrappers): Add more test
2305 2018-01-23 David Malcolm <dmalcolm@redhat.com>
2307 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
2308 (selftest::test_bit_in_range): Likewise.
2310 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2313 * doc/sourcebuild.texi (vect_float): Say that the selector
2314 only describes the situation when -funsafe-math-optimizations is on.
2315 (vect_float_strict): Document.
2317 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
2319 PR tree-optimization/83965
2320 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
2321 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
2322 instead of checking only for a reduction.
2323 (vect_recog_widen_sum_pattern): Likewise.
2325 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2327 * predict.c (probably_never_executed): Only use precise profile info.
2328 (compute_function_frequency): Skip after inlining hack since we now
2329 have quality checking.
2331 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
2333 * profile-count.h (profile_probability::very_unlikely,
2334 profile_probability::unlikely, profile_probability::even): Set
2335 precision to guessed.
2337 2018-01-23 Richard Biener <rguenther@suse.de>
2339 PR tree-optimization/83963
2340 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
2341 Properly terminate dominator walk when crossing the exit edge not
2342 when visiting its source block.
2344 2018-01-23 Jakub Jelinek <jakub@redhat.com>
2347 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
2348 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
2350 2018-01-22 Jakub Jelinek <jakub@redhat.com>
2352 PR tree-optimization/83957
2353 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
2354 semicolon after for body surrounded by braces.
2356 PR tree-optimization/83081
2357 * profile-count.h (profile_probability::split): New method.
2358 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
2359 Use profile_probability::split.
2360 (do_compare_rtx_and_jump): Fix adjustment of probabilities
2361 when splitting a single conditional jump into 2.
2363 2018-01-22 David Malcolm <dmalcolm@redhat.com>
2365 PR tree-optimization/69452
2366 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
2369 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2371 * config/rl78/rl78-expand.md (bswaphi2): New define_expand.
2372 * config/rl78/rl78-virt.md (*bswaphi2_virt): New define_insn.
2373 * config/rl78/rl78-real.md (*bswaphi2_real): New define_insn.
2375 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2377 * config/rl78/rl78-protos.h (rl78_split_movdi): New function declaration.
2378 * config/rl78/rl78.md (movdi): New define_expand.
2379 * config/rl78/rl78.c (rl78_split_movdi): New function.
2381 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
2384 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
2386 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
2387 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
2388 128-bit to produce an UNSPEC move to get the double word with the
2389 signbit and then a shift directly to do signbit.
2390 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
2391 implementation with a new version that just does either a direct
2392 move or a regular move. Move memory interface to separate insns.
2393 Move insns so they are next to the expander.
2394 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
2395 with signbit move. Split big and little endian case.
2396 (signbit<mode>2_dm_mem_le): Likewise.
2397 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
2398 (signbit<mode>2_dm2): Likewise.
2400 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2402 * config/rl78/rl78.md (anddi3): New define_expand.
2404 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2406 * config/rl78/rl78.md (umindi3): New define_expand.
2408 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2410 * config/rl78/rl78.md (smindi3): New define_expand.
2412 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2414 * config/rl78/rl78.md (smaxdi3): New define_expand.
2416 2018-01-22 Carl Love <cel@us.ibm.com>
2418 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
2419 LVX_V1TI): Add macro expansion.
2420 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
2421 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
2422 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
2423 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
2424 Change check to determine if the instruction is a byte reversing
2425 entry. Fix typo in comment.
2426 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
2427 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
2428 Add def_builtin calls for new builtins.
2429 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
2430 Add define_insn expansion.
2432 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2434 * config/rl78/rl78.md (umaxdi3): New define_expand.
2436 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
2438 * config/rl78/rl78.c (rl78_note_reg_set): Fixed dead reg check
2439 for non-QImode registers.
2441 2018-01-22 Richard Biener <rguenther@suse.de>
2443 PR tree-optimization/83963
2444 * graphite-scop-detection.c (scop_detection::get_sese): Delay
2445 including the loop exit block.
2446 (scop_detection::merge_sese): Likewise.
2447 (scop_detection::add_scop): Do it here instead.
2449 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2451 * doc/sourcebuild.texi (arm_softfloat): Document.
2453 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
2456 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
2457 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
2458 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
2460 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2461 David Edelsohn <dje.gcc@gmail.com>
2464 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2465 Change "crset eq" to "crset 2".
2466 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2467 (*call_indirect_aix<mode>_nospec): Likewise.
2468 (*call_value_indirect_aix<mode>_nospec): Likewise.
2469 (*call_indirect_elfv2<mode>_nospec): Likewise.
2470 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
2471 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
2472 change assembly output from . to $.
2473 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2474 (indirect_jump<mode>_nospec): Change assembly output from . to $.
2475 (*tablejump<mode>_internal1_nospec): Likewise.
2477 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
2480 * config/sh/sh_optimize_sett_clrt.cc:
2481 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
2483 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
2485 PR tree-optimization/83940
2486 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
2487 offset_dt to vect_constant_def rather than vect_unknown_def_type.
2488 (vect_check_load_store_mask): Add a mask_dt_out parameter and
2489 use it to pass back the definition type.
2490 (vect_check_store_rhs): Likewise rhs_dt_out.
2491 (vect_build_gather_load_calls): Add a mask_dt argument and use
2492 it instead of a call to vect_is_simple_use.
2493 (vectorizable_store): Update calls to vect_check_load_store_mask
2494 and vect_check_store_rhs. Use the dt returned by the latter instead
2495 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
2496 instead of calls to vect_is_simple_use. Pass the scalar rather
2497 than the vector operand to vect_is_simple_use when handling
2498 second and subsequent copies of an rhs value.
2499 (vectorizable_load): Update calls to vect_check_load_store_mask
2500 and vect_build_gather_load_calls. Use the cached mask_dt and
2501 gs_info.offset_dt instead of calls to vect_is_simple_use.
2503 2018-01-20 Jakub Jelinek <jakub@redhat.com>
2506 * tree-emutls.c: Include gimplify.h.
2507 (lower_emutls_2): New function.
2508 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
2509 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
2510 it before further processing.
2513 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
2514 UINTVAL (trueop1) instead of INTVAL (op1).
2516 2018-01-19 Jakub Jelinek <jakub@redhat.com>
2520 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
2521 INCOMING_FRAME_SP_OFFSET if not defined.
2522 (scan_trace): Add ENTRY argument. If true and
2523 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
2524 emit a note to adjust the CFA offset.
2525 (create_cfi_notes): Adjust scan_trace callers.
2526 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
2527 INCOMING_FRAME_SP_OFFSET in the CIE.
2528 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
2529 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
2531 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
2532 * doc/tm.texi: Regenerated.
2534 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
2536 PR rtl-optimization/83147
2537 * lra-constraints.c (remove_inheritance_pseudos): Use
2538 lra_substitute_pseudo_within_insn.
2540 2018-01-19 Tom de Vries <tom@codesourcery.com>
2541 Cesar Philippidis <cesar@codesourcery.com>
2544 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
2546 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
2549 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
2550 spaces for function labels.
2552 2018-01-19 Martin Liska <mliska@suse.cz>
2554 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
2555 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
2556 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
2557 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
2558 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
2559 (PRED_CONST_RETURN): Change from 69 to 65.
2560 (PRED_NULL_RETURN): Change from 91 to 71.
2561 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
2562 (PRED_LOOP_GUARD): Change from 66 to 73.
2564 2018-01-19 Martin Liska <mliska@suse.cz>
2566 * predict.c (predict_insn_def): Add new assert.
2567 (struct branch_predictor): Change type to signed integer.
2568 (test_prediction_value_range): Amend test to cover
2570 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
2571 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
2572 (PRED_LOOP_ITERATIONS_MAX): Likewise.
2573 (PRED_LOOP_IV_COMPARE): Likewise.
2574 * predict.h (PROB_UNINITIALIZED): Define new constant.
2576 2018-01-19 Martin Liska <mliska@suse.cz>
2578 * predict.c (dump_prediction): Add new format for
2579 analyze_brprob.py script which is enabled with -details
2581 * profile-count.h (precise_p): New function.
2583 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2585 PR tree-optimization/83922
2586 * tree-vect-loop.c (vect_verify_full_masking): Return false if
2587 there are no statements that need masking.
2588 (vect_active_double_reduction_p): New function.
2589 (vect_analyze_loop_operations): Use it when handling phis that
2590 are not in the loop header.
2592 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
2594 PR tree-optimization/83914
2595 * tree-vect-loop.c (vectorizable_induction): Don't convert
2596 init_expr or apply the peeling adjustment for inductions
2597 that are nested within the vectorized loop.
2599 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2601 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
2604 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2608 * function.h (gimplify_parameters): Add gimple_seq * argument.
2609 * function.c: Include gimple.h and options.h.
2610 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
2611 for the added local temporaries if needed.
2612 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
2613 if there are any parameter cleanups, wrap whole body into a
2614 try/finally with the cleanups.
2616 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
2619 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
2620 Use GET_MODE_CLASS for scalar floating point.
2622 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2626 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
2627 Fix call of call_cgraph_insertion_hooks.
2629 2018-01-18 Martin Sebor <msebor@redhat.com>
2631 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
2633 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
2636 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
2639 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
2642 * common.opt: (-ffile-prefix-map): New option.
2643 * opts.c (common_handle_option): Defer it.
2644 * opts-global.c (handle_common_deferred_options): Handle it.
2645 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
2646 * file-prefix-map.h: New file.
2647 (remap_debug_filename, add_debug_prefix_map): ...here.
2648 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
2649 * final.c (debug_prefix_map, add_debug_prefix_map
2650 remap_debug_filename): Move to...
2651 * file-prefix-map.c: New file.
2652 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
2653 generalize, get rid of alloca(), use strrchr() instead of strchr().
2654 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
2655 Implement in terms of add_prefix_map().
2656 (remap_macro_filename, remap_debug_filename): Implement in term of
2658 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
2659 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
2660 * dbxout.c: Include file-prefix-map.h.
2661 * varasm.c: Likewise.
2662 * vmsdbgout.c: Likewise.
2663 * xcoffout.c: Likewise.
2664 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
2665 * doc/cppopts.texi (-fmacro-prefix-map): Document.
2666 * doc/invoke.texi (-ffile-prefix-map): Document.
2667 (-fdebug-prefix-map): Update description.
2669 2018-01-18 Martin Liska <mliska@suse.cz>
2671 * config/i386/i386.c (indirect_thunk_name): Document that also
2673 (output_indirect_thunk): Document why both instructions
2674 (pause and lfence) are generated.
2676 2018-01-18 Richard Biener <rguenther@suse.de>
2678 PR tree-optimization/83887
2679 * graphite-scop-detection.c
2680 (scop_detection::get_nearest_dom_with_single_entry): Remove.
2681 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
2682 (scop_detection::merge_sese): Re-implement with a flood-fill
2683 algorithm that properly finds a SESE region if it exists.
2685 2018-01-18 Jakub Jelinek <jakub@redhat.com>
2688 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
2689 pointer_diff optimizations use view_convert instead of convert.
2691 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2693 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
2694 Generate different code for -mno-speculate-indirect-jumps.
2695 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
2696 (*call_indirect_aix<mode>): Disable for
2697 -mno-speculate-indirect-jumps.
2698 (*call_indirect_aix<mode>_nospec): New define_insn.
2699 (*call_value_indirect_aix<mode>): Disable for
2700 -mno-speculate-indirect-jumps.
2701 (*call_value_indirect_aix<mode>_nospec): New define_insn.
2702 (*sibcall_nonlocal_sysv<mode>): Generate different code for
2703 -mno-speculate-indirect-jumps.
2704 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
2706 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
2708 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
2709 long double type, set the flags for noting the default long double
2710 type, even if we don't pass or return a long double type.
2712 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
2715 * ipa-inline.c (flatten_function): Do not overwrite final inlining
2718 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
2720 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
2721 support for merge[hl].
2722 (fold_mergehl_helper): New helper function.
2723 (tree-vector-builder.h): New #include for tree_vector_builder usage.
2724 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
2725 (altivec_vmrglw_direct): Add xxmrglw insn.
2727 2018-01-17 Andrew Waterman <andrew@sifive.com>
2729 * config/riscv/riscv.c (riscv_conditional_register_usage): If
2730 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
2732 2018-01-17 David Malcolm <dmalcolm@redhat.com>
2735 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
2736 call the lto_location_cache before reading the
2737 DECL_SOURCE_LOCATION of the types.
2739 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
2740 Richard Sandiford <richard.sandiford@linaro.org>
2742 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
2743 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
2744 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
2745 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
2746 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
2748 * config/aarch64/constraints.md (aarch64_movti_operand):
2750 * config/aarch64/predicates.md (Uti): Add new constraint.
2752 2018-01-17 Carl Love <cel@us.ibm.com>
2754 * config/rs6000/vsx.md (define_expand xl_len_r,
2755 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
2756 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
2758 (define_expand, define_insn): Move the shift left from the
2759 define_insn to the define_expand for lxvl and stxvl instructions.
2760 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
2761 and XL_LEN_R definitions to PURE.
2763 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
2765 * config/i386/i386.c (indirect_thunk_name): Declare regno
2766 as unsigned int. Compare regno with INVALID_REGNUM.
2767 (output_indirect_thunk): Ditto.
2768 (output_indirect_thunk_function): Ditto.
2769 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
2770 in the call to output_indirect_thunk_function.
2772 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
2775 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
2776 rather than the size of inner_type to determine the stack slot size
2777 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
2779 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
2782 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
2785 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
2787 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
2788 endian Linux systems to optionally enable multilibs for selecting
2789 the long double type if the user configured an explicit type.
2790 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
2791 have no long double multilibs if not defined.
2792 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
2793 warn if the user used -mabi={ieee,ibm}longdouble and we built
2794 multilibs for long double.
2795 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
2796 appropriate multilib option.
2797 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
2799 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
2800 for building long double multilibs.
2801 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
2803 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
2805 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
2808 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
2810 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
2813 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
2816 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
2819 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2821 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
2822 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
2824 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
2826 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
2827 different rtl trees depending on TARGET_64BIT.
2828 (rs6000_gen_lvx): Likewise.
2830 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
2832 * config/visium/visium.md (nop): Tweak comment.
2833 (hazard_nop): Likewise.
2835 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2837 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
2838 -mspeculate-indirect-jumps.
2839 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
2840 for -mno-speculate-indirect-jumps.
2841 (*call_indirect_elfv2<mode>_nospec): New define_insn.
2842 (*call_value_indirect_elfv2<mode>): Disable for
2843 -mno-speculate-indirect-jumps.
2844 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
2845 (indirect_jump): Emit different RTL for
2846 -mno-speculate-indirect-jumps.
2847 (*indirect_jump<mode>): Disable for
2848 -mno-speculate-indirect-jumps.
2849 (*indirect_jump<mode>_nospec): New define_insn.
2850 (tablejump): Emit different RTL for
2851 -mno-speculate-indirect-jumps.
2852 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
2853 (tablejumpsi_nospec): New define_expand.
2854 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
2855 (tablejumpdi_nospec): New define_expand.
2856 (*tablejump<mode>_internal1): Disable for
2857 -mno-speculate-indirect-jumps.
2858 (*tablejump<mode>_internal1_nospec): New define_insn.
2859 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
2862 2018-01-16 Artyom Skrobov tyomitch@gmail.com
2864 * caller-save.c (insert_save): Drop unnecessary parameter. All
2867 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2868 Richard Biener <rguenth@suse.de>
2871 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
2872 return early, inline manually is_gimple_sizepos. Make sure if we
2873 call gimplify_expr we don't end up with a gimple constant.
2874 * tree.c (variably_modified_type_p): Don't return true for
2875 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
2876 * gimplify.h (is_gimple_sizepos): Remove.
2878 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2880 PR tree-optimization/83857
2881 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
2882 vectorizable_live_operation for pure SLP statements.
2883 (vectorizable_live_operation): Handle PHIs.
2885 2018-01-16 Richard Biener <rguenther@suse.de>
2887 PR tree-optimization/83867
2888 * tree-vect-stmts.c (vect_transform_stmt): Precompute
2889 nested_in_vect_loop_p since the scalar stmt may get invalidated.
2891 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2894 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
2895 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
2896 If off is not INTEGER_CST, issue a may not be aligned warning
2897 rather than isn't aligned. Use isn%'t rather than isn't.
2898 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
2900 <case MULT_EXPR>: Improve the case when bottom and one of the
2901 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
2902 operand, in that case check if the other operand is multiple of
2903 bottom divided by the INTEGER_CST operand.
2905 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2908 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
2909 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
2910 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
2911 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
2912 * config/pa/pa.c (pa_function_arg_advance): Likewise.
2913 (pa_function_arg, pa_arg_partial_bytes): Likewise.
2914 (pa_function_arg_size): New function.
2916 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2918 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
2919 in a separate statement.
2921 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
2923 PR tree-optimization/83847
2924 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
2925 group gathers and scatters.
2927 2018-01-16 Jakub Jelinek <jakub@redhat.com>
2929 PR rtl-optimization/86620
2930 * params.def (max-sched-ready-insns): Bump minimum value to 1.
2932 PR rtl-optimization/83213
2933 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
2934 to last if both are JUMP_INSNs.
2936 PR tree-optimization/83843
2937 * gimple-ssa-store-merging.c
2938 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
2939 store_immediate_info for bswap/nop orig_stores.
2941 2018-01-15 Andrew Waterman <andrew@sifive.com>
2943 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
2945 <UDIV>: Increase cost if !TARGET_DIV.
2947 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
2949 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
2950 (define_attr "cr_logical_3op"): New.
2951 (cceq_ior_compare): Adjust.
2952 (cceq_ior_compare_complement): Adjust.
2953 (*cceq_rev_compare): Adjust.
2954 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
2955 (is_cracked_insn): Adjust.
2956 (insn_must_be_first_in_group): Adjust.
2957 * config/rs6000/40x.md: Adjust.
2958 * config/rs6000/440.md: Adjust.
2959 * config/rs6000/476.md: Adjust.
2960 * config/rs6000/601.md: Adjust.
2961 * config/rs6000/603.md: Adjust.
2962 * config/rs6000/6xx.md: Adjust.
2963 * config/rs6000/7450.md: Adjust.
2964 * config/rs6000/7xx.md: Adjust.
2965 * config/rs6000/8540.md: Adjust.
2966 * config/rs6000/cell.md: Adjust.
2967 * config/rs6000/e300c2c3.md: Adjust.
2968 * config/rs6000/e500mc.md: Adjust.
2969 * config/rs6000/e500mc64.md: Adjust.
2970 * config/rs6000/e5500.md: Adjust.
2971 * config/rs6000/e6500.md: Adjust.
2972 * config/rs6000/mpc.md: Adjust.
2973 * config/rs6000/power4.md: Adjust.
2974 * config/rs6000/power5.md: Adjust.
2975 * config/rs6000/power6.md: Adjust.
2976 * config/rs6000/power7.md: Adjust.
2977 * config/rs6000/power8.md: Adjust.
2978 * config/rs6000/power9.md: Adjust.
2979 * config/rs6000/rs64.md: Adjust.
2980 * config/rs6000/titan.md: Adjust.
2982 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2984 * config/i386/predicates.md (indirect_branch_operand): Rewrite
2985 ix86_indirect_branch_register logic.
2987 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2989 * config/i386/constraints.md (Bs): Update
2990 ix86_indirect_branch_register check. Don't check
2991 ix86_indirect_branch_register with GOT_memory_operand.
2993 * config/i386/predicates.md (GOT_memory_operand): Don't check
2994 ix86_indirect_branch_register here.
2995 (GOT32_symbol_operand): Likewise.
2997 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
2999 * config/i386/predicates.md (constant_call_address_operand):
3000 Rewrite ix86_indirect_branch_register logic.
3001 (sibcall_insn_operand): Likewise.
3003 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3005 * config/i386/constraints.md (Bs): Replace
3006 ix86_indirect_branch_thunk_register with
3007 ix86_indirect_branch_register.
3009 * config/i386/i386.md (indirect_jump): Likewise.
3010 (tablejump): Likewise.
3011 (*sibcall_memory): Likewise.
3012 (*sibcall_value_memory): Likewise.
3013 Peepholes of indirect call and jump via memory: Likewise.
3014 * config/i386/i386.opt: Likewise.
3015 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3016 (GOT_memory_operand): Likewise.
3017 (call_insn_operand): Likewise.
3018 (sibcall_insn_operand): Likewise.
3019 (GOT32_symbol_operand): Likewise.
3021 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3024 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
3025 type rather than type addr's type points to.
3026 (expand_omp_atomic_mutex): Likewise.
3027 (expand_omp_atomic): Likewise.
3029 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
3032 * config/i386/i386.c (output_indirect_thunk_function): Use
3033 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
3034 for __x86_return_thunk.
3036 2018-01-15 Richard Biener <rguenther@suse.de>
3039 * expmed.c (extract_bit_field_1): Fix typo.
3041 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3044 * config/arm/iterators.md (VF): New mode iterator.
3045 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
3046 Remove integer-related logic from pattern.
3047 (neon_vabd<mode>_3): Likewise.
3049 2018-01-15 Jakub Jelinek <jakub@redhat.com>
3052 * common.opt (fstrict-overflow): No longer an alias.
3053 (fwrapv-pointer): New option.
3054 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
3055 also for pointer types based on flag_wrapv_pointer.
3056 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
3057 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
3058 opts->x_flag_wrapv got set.
3059 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
3060 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
3061 POINTER_TYPE_OVERFLOW_UNDEFINED.
3062 * match.pd: Likewise in address comparison pattern.
3063 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
3065 2018-01-15 Richard Biener <rguenther@suse.de>
3068 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
3069 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
3070 Reset type names to their identifier if their TYPE_DECL doesn't
3071 have linkage (and thus is used for ODR and devirt).
3072 (save_debug_info_for_decl): Remove.
3073 (save_debug_info_for_type): Likewise.
3074 (add_tree_to_fld_list): Adjust.
3075 * tree-pretty-print.c (dump_generic_node): Make dumping of
3076 type names more robust.
3078 2018-01-15 Richard Biener <rguenther@suse.de>
3080 * BASE-VER: Bump to 8.0.1.
3082 2018-01-14 Martin Sebor <msebor@redhat.com>
3085 * builtins.c (check_access): Avoid warning when the no-warning bit
3088 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
3090 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
3091 * ira-color (allocno_hard_regs_compare): Likewise.
3093 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
3096 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
3097 Use .pushsection/.popsection.
3099 2018-01-14 Martin Sebor <msebor@redhat.com>
3102 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
3104 2018-01-14 Jakub Jelinek <jakub@redhat.com>
3106 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
3107 entry from extra_headers.
3108 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
3109 extra_headers, make the list bitwise identical to the i?86-*-* one.
3111 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3113 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
3114 -mcmodel=large with -mindirect-branch=thunk,
3115 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
3116 -mfunction-return=thunk-extern.
3117 * doc/invoke.texi: Document -mcmodel=large is incompatible with
3118 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
3119 -mfunction-return=thunk and -mfunction-return=thunk-extern.
3121 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3123 * config/i386/i386.c (print_reg): Print the name of the full
3124 integer register without '%'.
3125 (ix86_print_operand): Handle 'V'.
3126 * doc/extend.texi: Document 'V' modifier.
3128 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3130 * config/i386/constraints.md (Bs): Disallow memory operand for
3131 -mindirect-branch-register.
3133 * config/i386/predicates.md (indirect_branch_operand): Likewise.
3134 (GOT_memory_operand): Likewise.
3135 (call_insn_operand): Likewise.
3136 (sibcall_insn_operand): Likewise.
3137 (GOT32_symbol_operand): Likewise.
3138 * config/i386/i386.md (indirect_jump): Call convert_memory_address
3139 for -mindirect-branch-register.
3140 (tablejump): Likewise.
3141 (*sibcall_memory): Likewise.
3142 (*sibcall_value_memory): Likewise.
3143 Disallow peepholes of indirect call and jump via memory for
3144 -mindirect-branch-register.
3145 (*call_pop): Replace m with Bw.
3146 (*call_value_pop): Likewise.
3147 (*sibcall_pop_memory): Replace m with Bs.
3148 * config/i386/i386.opt (mindirect-branch-register): New option.
3149 * doc/invoke.texi: Document -mindirect-branch-register option.
3151 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3153 * config/i386/i386-protos.h (ix86_output_function_return): New.
3154 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
3155 set function_return_type.
3156 (indirect_thunk_name): Add ret_p to indicate thunk for function
3158 (output_indirect_thunk_function): Pass false to
3159 indirect_thunk_name.
3160 (ix86_output_indirect_branch_via_reg): Likewise.
3161 (ix86_output_indirect_branch_via_push): Likewise.
3162 (output_indirect_thunk_function): Create alias for function
3163 return thunk if regno < 0.
3164 (ix86_output_function_return): New function.
3165 (ix86_handle_fndecl_attribute): Handle function_return.
3166 (ix86_attribute_table): Add function_return.
3167 * config/i386/i386.h (machine_function): Add
3168 function_return_type.
3169 * config/i386/i386.md (simple_return_internal): Use
3170 ix86_output_function_return.
3171 (simple_return_internal_long): Likewise.
3172 * config/i386/i386.opt (mfunction-return=): New option.
3173 (indirect_branch): Mention -mfunction-return=.
3174 * doc/extend.texi: Document function_return function attribute.
3175 * doc/invoke.texi: Document -mfunction-return= option.
3177 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
3179 * config/i386/i386-opts.h (indirect_branch): New.
3180 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
3181 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
3182 with local indirect jump when converting indirect call and jump.
3183 (ix86_set_indirect_branch_type): New.
3184 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
3185 (indirectlabelno): New.
3186 (indirect_thunk_needed): Likewise.
3187 (indirect_thunk_bnd_needed): Likewise.
3188 (indirect_thunks_used): Likewise.
3189 (indirect_thunks_bnd_used): Likewise.
3190 (INDIRECT_LABEL): Likewise.
3191 (indirect_thunk_name): Likewise.
3192 (output_indirect_thunk): Likewise.
3193 (output_indirect_thunk_function): Likewise.
3194 (ix86_output_indirect_branch_via_reg): Likewise.
3195 (ix86_output_indirect_branch_via_push): Likewise.
3196 (ix86_output_indirect_branch): Likewise.
3197 (ix86_output_indirect_jmp): Likewise.
3198 (ix86_code_end): Call output_indirect_thunk_function if needed.
3199 (ix86_output_call_insn): Call ix86_output_indirect_branch if
3201 (ix86_handle_fndecl_attribute): Handle indirect_branch.
3202 (ix86_attribute_table): Add indirect_branch.
3203 * config/i386/i386.h (machine_function): Add indirect_branch_type
3204 and has_local_indirect_jump.
3205 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
3207 (tablejump): Likewise.
3208 (*indirect_jump): Use ix86_output_indirect_jmp.
3209 (*tablejump_1): Likewise.
3210 (simple_return_indirect_internal): Likewise.
3211 * config/i386/i386.opt (mindirect-branch=): New option.
3212 (indirect_branch): New.
3215 (thunk-inline): Likewise.
3216 (thunk-extern): Likewise.
3217 * doc/extend.texi: Document indirect_branch function attribute.
3218 * doc/invoke.texi: Document -mindirect-branch= option.
3220 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
3223 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
3225 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
3227 * ipa-inline.c (want_inline_small_function_p): Return false if
3228 inlining has already failed with CIF_FINAL_ERROR.
3229 (update_caller_keys): Call want_inline_small_function_p before
3231 (update_callee_keys): Likewise.
3233 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
3235 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
3237 (rs6000_quadword_masked_address_p): Likewise.
3238 (quad_aligned_load_p): Likewise.
3239 (quad_aligned_store_p): Likewise.
3240 (const_load_sequence_p): Add comment to describe the outer-most loop.
3241 (mimic_memory_attributes_and_flags): New function.
3242 (rs6000_gen_stvx): Likewise.
3243 (replace_swapped_aligned_store): Likewise.
3244 (rs6000_gen_lvx): Likewise.
3245 (replace_swapped_aligned_load): Likewise.
3246 (replace_swapped_load_constant): Capitalize argument name in
3247 comment describing this function.
3248 (rs6000_analyze_swaps): Add a third pass to search for vector loads
3249 and stores that access quad-word aligned addresses and replace
3250 with stvx or lvx instructions when appropriate.
3251 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
3252 New function prototype.
3253 (rs6000_quadword_masked_address_p): Likewise.
3254 (rs6000_gen_lvx): Likewise.
3255 (rs6000_gen_stvx): Likewise.
3256 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
3257 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
3258 when memory address is aligned.
3259 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
3260 this split to select lvx instruction when memory address is aligned.
3261 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
3262 instruction when memory address is aligned.
3263 (*vsx_le_perm_load_v16qi): Likewise.
3264 (four unnamed splitters): Modify to select the stvx instruction
3265 when memory is aligned.
3267 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
3269 * predict.c (determine_unlikely_bbs): Handle correctly BBs
3270 which appears in the queue multiple times.
3272 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3273 Alan Hayward <alan.hayward@arm.com>
3274 David Sherwood <david.sherwood@arm.com>
3276 * tree-vectorizer.h (vec_lower_bound): New structure.
3277 (_loop_vec_info): Add check_nonzero and lower_bounds.
3278 (LOOP_VINFO_CHECK_NONZERO): New macro.
3279 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
3280 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
3281 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
3282 fields. Make seg_len the distance travelled, not including the
3284 (dr_direction_indicator): Declare.
3285 (dr_zero_step_indicator): Likewise.
3286 (dr_known_forward_stride_p): Likewise.
3287 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
3289 (runtime_alias_check_p): Allow runtime alias checks with
3291 (operator ==): Compare access_size and align.
3292 (prune_runtime_alias_test_list): Rework for new distinction between
3293 the access_size and seg_len.
3294 (create_intersect_range_checks_index): Likewise. Cope with polynomial
3296 (get_segment_min_max): New function.
3297 (create_intersect_range_checks): Use it.
3298 (dr_step_indicator): New function.
3299 (dr_direction_indicator): Likewise.
3300 (dr_zero_step_indicator): Likewise.
3301 (dr_known_forward_stride_p): Likewise.
3302 * tree-loop-distribution.c (data_ref_segment_size): Return
3303 DR_STEP * (niters - 1).
3304 (compute_alias_check_pairs): Update call to the dr_with_seg_len
3306 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
3307 (vect_preserves_scalar_order_p): New function, split out from...
3308 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
3309 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
3310 (vect_vfa_access_size): New function.
3311 (vect_vfa_align): Likewise.
3312 (vect_compile_time_alias): Take access_size_a and access_b arguments.
3313 (dump_lower_bound): New function.
3314 (vect_check_lower_bound): Likewise.
3315 (vect_small_gap_p): Likewise.
3316 (vectorizable_with_step_bound_p): Likewise.
3317 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
3318 depencies if the vectorization factor is 1. Convert the checks
3319 for nonzero steps into checks on the bounds of DR_STEP. Try using
3320 a bunds check for variable steps if the minimum required step is
3321 relatively small. Update calls to the dr_with_seg_len
3322 constructor and to vect_compile_time_alias.
3323 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
3325 (vect_loop_versioning): Call it.
3326 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
3328 (vect_estimate_min_profitable_iters): Account for any bounds checks.
3330 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3331 Alan Hayward <alan.hayward@arm.com>
3332 David Sherwood <david.sherwood@arm.com>
3334 * doc/sourcebuild.texi (vect_scatter_store): Document.
3335 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
3337 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
3339 * genopinit.c (main): Add supports_vec_scatter_store and
3340 supports_vec_scatter_store_cached to target_optabs.
3341 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
3342 IFN_MASK_SCATTER_STORE.
3343 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
3345 * internal-fn.h (internal_store_fn_p): Declare.
3346 (internal_fn_stored_value_index): Likewise.
3347 * internal-fn.c (scatter_store_direct): New macro.
3348 (expand_scatter_store_optab_fn): New function.
3349 (direct_scatter_store_optab_supported_p): New macro.
3350 (internal_store_fn_p): New function.
3351 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
3352 IFN_MASK_SCATTER_STORE.
3353 (internal_fn_mask_index): Likewise.
3354 (internal_fn_stored_value_index): New function.
3355 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
3357 * optabs-query.h (supports_vec_scatter_store_p): Declare.
3358 * optabs-query.c (supports_vec_scatter_store_p): New function.
3359 * tree-vectorizer.h (vect_get_store_rhs): Declare.
3360 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
3361 true for scatter stores.
3362 (vect_gather_scatter_fn_p): Handle scatter stores too.
3363 (vect_check_gather_scatter): Consider using scatter stores if
3364 supports_vec_scatter_store_p.
3365 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
3367 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3368 internal_fn_stored_value_index.
3369 (check_load_store_masking): Handle scatter stores too.
3370 (vect_get_store_rhs): Make public.
3371 (vectorizable_call): Use internal_store_fn_p.
3372 (vectorizable_store): Handle scatter store internal functions.
3373 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
3374 when deciding whether the end of the group has been reached.
3375 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
3376 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
3377 (mask_scatter_store<mode>): New insns.
3379 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3380 Alan Hayward <alan.hayward@arm.com>
3381 David Sherwood <david.sherwood@arm.com>
3383 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
3384 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
3385 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
3387 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
3388 Use vect_truncate_gather_scatter_offset if we can't treat the
3389 operation as a normal gather load or scatter store.
3390 (get_group_load_store_type): Take the gather_scatter_info
3391 as argument. Try using a gather load or scatter store for
3392 single-element groups.
3393 (get_load_store_type): Update calls to get_group_load_store_type
3394 and vect_use_strided_gather_scatters_p.
3396 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3397 Alan Hayward <alan.hayward@arm.com>
3398 David Sherwood <david.sherwood@arm.com>
3400 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
3401 optional tree argument.
3402 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
3404 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
3405 but continue to use the current value as a fallback.
3406 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
3407 to compare the updates.
3408 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
3409 (get_load_store_type): Use it when handling a strided access.
3410 (vect_get_strided_load_store_ops): New function.
3411 (vect_get_data_ptr_increment): Likewise.
3412 (vectorizable_load): Handle strided gather loads. Always pass
3413 a step to vect_create_data_ref_ptr and bump_vector_ptr.
3415 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3416 Alan Hayward <alan.hayward@arm.com>
3417 David Sherwood <david.sherwood@arm.com>
3419 * doc/md.texi (gather_load@var{m}): Document.
3420 (mask_gather_load@var{m}): Likewise.
3421 * genopinit.c (main): Add supports_vec_gather_load and
3422 supports_vec_gather_load_cached to target_optabs.
3423 * optabs-tree.c (init_tree_optimization_optabs): Use
3424 ggc_cleared_alloc to allocate target_optabs.
3425 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
3426 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
3428 * internal-fn.h (internal_load_fn_p): Declare.
3429 (internal_gather_scatter_fn_p): Likewise.
3430 (internal_fn_mask_index): Likewise.
3431 (internal_gather_scatter_fn_supported_p): Likewise.
3432 * internal-fn.c (gather_load_direct): New macro.
3433 (expand_gather_load_optab_fn): New function.
3434 (direct_gather_load_optab_supported_p): New macro.
3435 (direct_internal_fn_optab): New function.
3436 (internal_load_fn_p): Likewise.
3437 (internal_gather_scatter_fn_p): Likewise.
3438 (internal_fn_mask_index): Likewise.
3439 (internal_gather_scatter_fn_supported_p): Likewise.
3440 * optabs-query.c (supports_at_least_one_mode_p): New function.
3441 (supports_vec_gather_load_p): Likewise.
3442 * optabs-query.h (supports_vec_gather_load_p): Declare.
3443 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
3444 and memory_type field.
3445 (NUM_PATTERNS): Bump to 15.
3446 * tree-vect-data-refs.c: Include internal-fn.h.
3447 (vect_gather_scatter_fn_p): New function.
3448 (vect_describe_gather_scatter_call): Likewise.
3449 (vect_check_gather_scatter): Try using internal functions for
3450 gather loads. Recognize existing calls to a gather load function.
3451 (vect_analyze_data_refs): Consider using gather loads if
3452 supports_vec_gather_load_p.
3453 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
3454 (vect_get_gather_scatter_offset_type): Likewise.
3455 (vect_convert_mask_for_vectype): Likewise.
3456 (vect_add_conversion_to_patterm): Likewise.
3457 (vect_try_gather_scatter_pattern): Likewise.
3458 (vect_recog_gather_scatter_pattern): New pattern recognizer.
3459 (vect_vect_recog_func_ptrs): Add it.
3460 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
3461 internal_fn_mask_index and internal_gather_scatter_fn_p.
3462 (check_load_store_masking): Take the gather_scatter_info as an
3463 argument and handle gather loads.
3464 (vect_get_gather_scatter_ops): New function.
3465 (vectorizable_call): Check internal_load_fn_p.
3466 (vectorizable_load): Likewise. Handle gather load internal
3468 (vectorizable_store): Update call to check_load_store_masking.
3469 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
3470 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
3471 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
3472 (aarch64_gather_scale_operand_d): New predicates.
3473 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
3474 (mask_gather_load<mode>): New insns.
3476 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3477 Alan Hayward <alan.hayward@arm.com>
3478 David Sherwood <david.sherwood@arm.com>
3480 * optabs.def (fold_left_plus_optab): New optab.
3481 * doc/md.texi (fold_left_plus_@var{m}): Document.
3482 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
3483 * internal-fn.c (fold_left_direct): Define.
3484 (expand_fold_left_optab_fn): Likewise.
3485 (direct_fold_left_optab_supported_p): Likewise.
3486 * fold-const-call.c (fold_const_fold_left): New function.
3487 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
3488 * tree-parloops.c (valid_reduction_p): New function.
3489 (gather_scalar_reductions): Use it.
3490 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
3491 (vect_finish_replace_stmt): Declare.
3492 * tree-vect-loop.c (fold_left_reduction_fn): New function.
3493 (needs_fold_left_reduction_p): New function, split out from...
3494 (vect_is_simple_reduction): ...here. Accept reductions that
3495 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
3496 (vect_force_simple_reduction): Also store the reduction type in
3497 the assignment's STMT_VINFO_REDUC_TYPE.
3498 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
3499 (merge_with_identity): New function.
3500 (vect_expand_fold_left): Likewise.
3501 (vectorize_fold_left_reduction): Likewise.
3502 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
3503 scalar phi in place for it. Check for target support and reject
3504 cases that would reassociate the operation. Defer the transform
3505 phase to vectorize_fold_left_reduction.
3506 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
3507 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
3508 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
3510 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3512 * tree-if-conv.c (predicate_mem_writes): Remove redundant
3513 call to ifc_temp_var.
3515 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3516 Alan Hayward <alan.hayward@arm.com>
3517 David Sherwood <david.sherwood@arm.com>
3519 * target.def (legitimize_address_displacement): Take the original
3520 offset as a poly_int.
3521 * targhooks.h (default_legitimize_address_displacement): Update
3523 * targhooks.c (default_legitimize_address_displacement): Likewise.
3524 * doc/tm.texi: Regenerate.
3525 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
3526 as an argument, moving assert of ad->disp == ad->disp_term to...
3527 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
3528 Try calling targetm.legitimize_address_displacement before expanding
3529 the address rather than afterwards, and adjust for the new interface.
3530 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
3531 Match the new hook interface. Handle SVE addresses.
3532 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
3535 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3537 * Makefile.in (OBJS): Add early-remat.o.
3538 * target.def (select_early_remat_modes): New hook.
3539 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
3540 * doc/tm.texi: Regenerate.
3541 * targhooks.h (default_select_early_remat_modes): Declare.
3542 * targhooks.c (default_select_early_remat_modes): New function.
3543 * timevar.def (TV_EARLY_REMAT): New timevar.
3544 * passes.def (pass_early_remat): New pass.
3545 * tree-pass.h (make_pass_early_remat): Declare.
3546 * early-remat.c: New file.
3547 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
3549 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
3551 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3552 Alan Hayward <alan.hayward@arm.com>
3553 David Sherwood <david.sherwood@arm.com>
3555 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
3556 vfm1 with a bound_epilog parameter.
3557 (vect_do_peeling): Update calls accordingly, and move the prologue
3558 call earlier in the function. Treat the base bound_epilog as 0 for
3559 fully-masked loops and retain vf - 1 for other loops. Add 1 to
3560 this base when peeling for gaps.
3561 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
3562 with fully-masked loops.
3563 (vect_estimate_min_profitable_iters): Handle the single peeled
3564 iteration in that case.
3566 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3567 Alan Hayward <alan.hayward@arm.com>
3568 David Sherwood <david.sherwood@arm.com>
3570 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
3571 single-element interleaving even if the size is not a power of 2.
3572 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
3573 accesses for single-element interleaving if the group size is
3576 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3577 Alan Hayward <alan.hayward@arm.com>
3578 David Sherwood <david.sherwood@arm.com>
3580 * doc/md.texi (fold_extract_last_@var{m}): Document.
3581 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
3582 * optabs.def (fold_extract_last_optab): New optab.
3583 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
3584 * internal-fn.c (fold_extract_direct): New macro.
3585 (expand_fold_extract_optab_fn): Likewise.
3586 (direct_fold_extract_optab_supported_p): Likewise.
3587 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
3588 * tree-vect-loop.c (vect_model_reduction_cost): Handle
3589 EXTRACT_LAST_REDUCTION.
3590 (get_initial_def_for_reduction): Do not create an initial vector
3591 for EXTRACT_LAST_REDUCTION reductions.
3592 (vectorizable_reduction): Leave the scalar phi in place for
3593 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
3594 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
3595 epilogue code for EXTRACT_LAST_REDUCTION and defer the
3596 transform phase to vectorizable_condition.
3597 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
3599 (vect_finish_stmt_generation): ...here.
3600 (vect_finish_replace_stmt): New function.
3601 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
3602 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
3604 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
3606 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3607 Alan Hayward <alan.hayward@arm.com>
3608 David Sherwood <david.sherwood@arm.com>
3610 * doc/md.texi (extract_last_@var{m}): Document.
3611 * optabs.def (extract_last_optab): New optab.
3612 * internal-fn.def (EXTRACT_LAST): New internal function.
3613 * internal-fn.c (cond_unary_direct): New macro.
3614 (expand_cond_unary_optab_fn): Likewise.
3615 (direct_cond_unary_optab_supported_p): Likewise.
3616 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
3617 loops using EXTRACT_LAST.
3618 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
3619 (extract_last_<mode>): ...this optab.
3620 (vec_extract<mode><Vel>): Update accordingly.
3622 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3623 Alan Hayward <alan.hayward@arm.com>
3624 David Sherwood <david.sherwood@arm.com>
3626 * target.def (empty_mask_is_expensive): New hook.
3627 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
3628 * doc/tm.texi: Regenerate.
3629 * targhooks.h (default_empty_mask_is_expensive): Declare.
3630 * targhooks.c (default_empty_mask_is_expensive): New function.
3631 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
3632 if the target says that empty masks are expensive.
3633 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
3635 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
3637 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3638 Alan Hayward <alan.hayward@arm.com>
3639 David Sherwood <david.sherwood@arm.com>
3641 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
3642 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
3643 (vect_use_loop_mask_for_alignment_p): New function.
3644 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
3645 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
3646 niters_skip argument. Make sure that the first niters_skip elements
3647 of the first iteration are inactive.
3648 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
3649 Update call to vect_set_loop_masks_directly.
3650 (get_misalign_in_elems): New function, split out from...
3651 (vect_gen_prolog_loop_niters): ...here.
3652 (vect_update_init_of_dr): Take a code argument that specifies whether
3653 the adjustment should be added or subtracted.
3654 (vect_update_init_of_drs): Likewise.
3655 (vect_prepare_for_masked_peels): New function.
3656 (vect_do_peeling): Skip prologue peeling if we're using a mask
3657 instead. Update call to vect_update_inits_of_drs.
3658 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3660 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
3661 alignment. Do not include the number of peeled iterations in
3662 the minimum threshold in that case.
3663 (vectorizable_induction): Adjust the start value down by
3664 LOOP_VINFO_MASK_SKIP_NITERS iterations.
3665 (vect_transform_loop): Call vect_prepare_for_masked_peels.
3666 Take the number of skipped iterations into account when calculating
3668 * tree-vect-stmts.c (vect_gen_while_not): New function.
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 * doc/sourcebuild.texi (vect_fully_masked): Document.
3675 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
3677 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
3679 (vect_analyze_loop_2): ...here. Don't check the vectorization
3680 factor against the number of loop iterations if the loop is
3683 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3684 Alan Hayward <alan.hayward@arm.com>
3685 David Sherwood <david.sherwood@arm.com>
3687 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
3688 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
3689 (dump_groups): Update accordingly.
3690 (iv_use::mem_type): New member variable.
3691 (address_p): New function.
3692 (record_use): Add a mem_type argument and initialize the new
3694 (record_group_use): Add a mem_type argument. Use address_p.
3695 Remove obsolete null checks of base_object. Update call to record_use.
3696 (find_interesting_uses_op): Update call to record_group_use.
3697 (find_interesting_uses_cond): Likewise.
3698 (find_interesting_uses_address): Likewise.
3699 (get_mem_type_for_internal_fn): New function.
3700 (find_address_like_use): Likewise.
3701 (find_interesting_uses_stmt): Try find_address_like_use before
3702 calling find_interesting_uses_op.
3703 (addr_offset_valid_p): Use the iv mem_type field as the type
3704 of the addressed memory.
3705 (add_autoinc_candidates): Likewise.
3706 (get_address_cost): Likewise.
3707 (split_small_address_groups_p): Use address_p.
3708 (split_address_groups): Likewise.
3709 (add_iv_candidate_for_use): Likewise.
3710 (autoinc_possible_for_pair): Likewise.
3711 (rewrite_groups): Likewise.
3712 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
3713 (determine_group_iv_cost): Update after split of USE_ADDRESS.
3714 (get_alias_ptr_type_for_ptr_address): New function.
3715 (rewrite_use_address): Rewrite address uses in calls that were
3716 identified by find_address_like_use.
3718 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3719 Alan Hayward <alan.hayward@arm.com>
3720 David Sherwood <david.sherwood@arm.com>
3722 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
3724 * gimple-expr.h (is_gimple_addressable: Likewise.
3725 * gimple-expr.c (is_gimple_address): Likewise.
3726 * internal-fn.c (expand_call_mem_ref): New function.
3727 (expand_mask_load_optab_fn): Use it.
3728 (expand_mask_store_optab_fn): Likewise.
3730 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3731 Alan Hayward <alan.hayward@arm.com>
3732 David Sherwood <david.sherwood@arm.com>
3734 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
3735 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
3736 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
3737 (cond_umax@var{mode}): Document.
3738 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
3739 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
3740 (cond_umin_optab, cond_umax_optab): New optabs.
3741 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
3742 (COND_IOR, COND_XOR): New internal functions.
3743 * internal-fn.h (get_conditional_internal_fn): Declare.
3744 * internal-fn.c (cond_binary_direct): New macro.
3745 (expand_cond_binary_optab_fn): Likewise.
3746 (direct_cond_binary_optab_supported_p): Likewise.
3747 (get_conditional_internal_fn): New function.
3748 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
3749 Cope with reduction statements that are vectorized as calls rather
3751 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
3752 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
3753 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
3754 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
3755 (UNSPEC_COND_EOR): New unspecs.
3756 (optab): Add mappings for them.
3757 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
3758 (sve_int_op, sve_fp_op): New int attributes.
3760 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3761 Alan Hayward <alan.hayward@arm.com>
3762 David Sherwood <david.sherwood@arm.com>
3764 * optabs.def (while_ult_optab): New optab.
3765 * doc/md.texi (while_ult@var{m}@var{n}): Document.
3766 * internal-fn.def (WHILE_ULT): New internal function.
3767 * internal-fn.h (direct_internal_fn_supported_p): New override
3768 that takes two types as argument.
3769 * internal-fn.c (while_direct): New macro.
3770 (expand_while_optab_fn): New function.
3771 (convert_optab_supported_p): Likewise.
3772 (direct_while_optab_supported_p): New macro.
3773 * wide-int.h (wi::udiv_ceil): New function.
3774 * tree-vectorizer.h (rgroup_masks): New structure.
3775 (vec_loop_masks): New typedef.
3776 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
3778 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
3779 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
3780 (vect_max_vf): New function.
3781 (slpeel_make_loop_iterate_ntimes): Delete.
3782 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
3783 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
3784 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
3785 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
3786 internal-fn.h, stor-layout.h and optabs-query.h.
3787 (vect_set_loop_mask): New function.
3788 (add_preheader_seq): Likewise.
3789 (add_header_seq): Likewise.
3790 (interleave_supported_p): Likewise.
3791 (vect_maybe_permute_loop_masks): Likewise.
3792 (vect_set_loop_masks_directly): Likewise.
3793 (vect_set_loop_condition_masked): Likewise.
3794 (vect_set_loop_condition_unmasked): New function, split out from
3795 slpeel_make_loop_iterate_ntimes.
3796 (slpeel_make_loop_iterate_ntimes): Rename to..
3797 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
3798 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
3799 (vect_do_peeling): Update call accordingly.
3800 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
3802 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
3803 mask_compare_type, can_fully_mask_p and fully_masked_p.
3804 (release_vec_loop_masks): New function.
3805 (_loop_vec_info): Use it to free the loop masks.
3806 (can_produce_all_loop_masks_p): New function.
3807 (vect_get_max_nscalars_per_iter): Likewise.
3808 (vect_verify_full_masking): Likewise.
3809 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
3810 retries, and free the mask rgroups before retrying. Check loop-wide
3811 reasons for disallowing fully-masked loops. Make the final decision
3812 about whether use a fully-masked loop or not.
3813 (vect_estimate_min_profitable_iters): Do not assume that peeling
3814 for the number of iterations will be needed for fully-masked loops.
3815 (vectorizable_reduction): Disable fully-masked loops.
3816 (vectorizable_live_operation): Likewise.
3817 (vect_halve_mask_nunits): New function.
3818 (vect_double_mask_nunits): Likewise.
3819 (vect_record_loop_mask): Likewise.
3820 (vect_get_loop_mask): Likewise.
3821 (vect_transform_loop): Handle the case in which the final loop
3822 iteration might handle a partial vector. Call vect_set_loop_condition
3823 instead of slpeel_make_loop_iterate_ntimes.
3824 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
3825 (check_load_store_masking): New function.
3826 (prepare_load_store_mask): Likewise.
3827 (vectorizable_store): Handle fully-masked loops.
3828 (vectorizable_load): Likewise.
3829 (supportable_widening_operation): Use vect_halve_mask_nunits for
3831 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
3832 (vect_gen_while): New function.
3833 * config/aarch64/aarch64.md (umax<mode>3): New expander.
3834 (aarch64_uqdec<mode>): New insn.
3836 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3837 Alan Hayward <alan.hayward@arm.com>
3838 David Sherwood <david.sherwood@arm.com>
3840 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
3841 (reduc_xor_scal_optab): New optabs.
3842 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
3843 (reduc_xor_scal_@var{m}): Document.
3844 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
3845 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
3847 * fold-const-call.c (fold_const_call): Handle them.
3848 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
3849 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
3850 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
3851 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
3852 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
3853 (UNSPEC_XORV): New unspecs.
3854 (optab): Add entries for them.
3855 (BITWISEV): New int iterator.
3856 (bit_reduc_op): New int attributes.
3858 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3859 Alan Hayward <alan.hayward@arm.com>
3860 David Sherwood <david.sherwood@arm.com>
3862 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
3863 * internal-fn.def (VEC_SHL_INSERT): New internal function.
3864 * optabs.def (vec_shl_insert_optab): New optab.
3865 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
3866 (duplicate_and_interleave): Likewise.
3867 * tree-vect-loop.c: Include internal-fn.h.
3868 (neutral_op_for_slp_reduction): New function, split out from
3869 get_initial_defs_for_reduction.
3870 (get_initial_def_for_reduction): Handle option 2 for variable-length
3871 vectors by loading the neutral value into a vector and then shifting
3872 the initial value into element 0.
3873 (get_initial_defs_for_reduction): Replace the code argument with
3874 the neutral value calculated by neutral_op_for_slp_reduction.
3875 Use gimple_build_vector for constant-length vectors.
3876 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
3877 but the first group_size elements have a neutral value.
3878 Use duplicate_and_interleave otherwise.
3879 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
3880 Update call to get_initial_defs_for_reduction. Handle SLP
3881 reductions for variable-length vectors by creating one vector
3882 result for each scalar result, with the elements associated
3883 with other scalar results stubbed out with the neutral value.
3884 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
3885 Require IFN_VEC_SHL_INSERT for double reductions on
3886 variable-length vectors, or SLP reductions that have
3887 a neutral value. Require can_duplicate_and_interleave_p
3888 support for variable-length unchained SLP reductions if there
3889 is no neutral value, such as for MIN/MAX reductions. Also require
3890 the number of vector elements to be a multiple of the number of
3891 SLP statements when doing variable-length unchained SLP reductions.
3892 Update call to vect_create_epilog_for_reduction.
3893 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
3894 and remove initial values.
3895 (duplicate_and_interleave): Make public.
3896 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
3897 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
3899 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3900 Alan Hayward <alan.hayward@arm.com>
3901 David Sherwood <david.sherwood@arm.com>
3903 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
3904 (can_duplicate_and_interleave_p): New function.
3905 (vect_get_and_check_slp_defs): Take the vector of statements
3906 rather than just the current one. Remove excess parentheses.
3907 Restriction rejectinon of vect_constant_def and vect_external_def
3908 for variable-length vectors to boolean types, or types for which
3909 can_duplicate_and_interleave_p is false.
3910 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
3911 (duplicate_and_interleave): New function.
3912 (vect_get_constant_vectors): Use gimple_build_vector for
3913 constant-length vectors and suitable variable-length constant
3914 vectors. Use duplicate_and_interleave for other variable-length
3915 vectors. Don't defer the update when inserting new statements.
3917 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3918 Alan Hayward <alan.hayward@arm.com>
3919 David Sherwood <david.sherwood@arm.com>
3921 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
3922 min_profitable_iters doesn't go negative.
3924 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3925 Alan Hayward <alan.hayward@arm.com>
3926 David Sherwood <david.sherwood@arm.com>
3928 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
3929 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
3930 * optabs.def (vec_mask_load_lanes_optab): New optab.
3931 (vec_mask_store_lanes_optab): Likewise.
3932 * internal-fn.def (MASK_LOAD_LANES): New internal function.
3933 (MASK_STORE_LANES): Likewise.
3934 * internal-fn.c (mask_load_lanes_direct): New macro.
3935 (mask_store_lanes_direct): Likewise.
3936 (expand_mask_load_optab_fn): Handle masked operations.
3937 (expand_mask_load_lanes_optab_fn): New macro.
3938 (expand_mask_store_optab_fn): Handle masked operations.
3939 (expand_mask_store_lanes_optab_fn): New macro.
3940 (direct_mask_load_lanes_optab_supported_p): Likewise.
3941 (direct_mask_store_lanes_optab_supported_p): Likewise.
3942 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
3944 (vect_load_lanes_supported): Likewise.
3945 * tree-vect-data-refs.c (strip_conversion): New function.
3946 (can_group_stmts_p): Likewise.
3947 (vect_analyze_data_ref_accesses): Use it instead of checking
3948 for a pair of assignments.
3949 (vect_store_lanes_supported): Take a masked_p parameter.
3950 (vect_load_lanes_supported): Likewise.
3951 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
3952 vect_store_lanes_supported and vect_load_lanes_supported.
3953 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
3954 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
3955 parameter. Don't allow gaps for masked accesses.
3956 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
3957 and vect_load_lanes_supported.
3958 (get_load_store_type): Take a masked_p parameter and update
3959 call to get_group_load_store_type.
3960 (vectorizable_store): Update call to get_load_store_type.
3961 Handle IFN_MASK_STORE_LANES.
3962 (vectorizable_load): Update call to get_load_store_type.
3963 Handle IFN_MASK_LOAD_LANES.
3965 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3966 Alan Hayward <alan.hayward@arm.com>
3967 David Sherwood <david.sherwood@arm.com>
3969 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
3971 * config/aarch64/aarch64-protos.h
3972 (aarch64_sve_struct_memory_operand_p): Declare.
3973 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
3974 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
3975 (VPRED, vpred): Handle SVE structure modes.
3976 * config/aarch64/constraints.md (Utx): New constraint.
3977 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
3978 (aarch64_sve_struct_nonimmediate_operand): New predicates.
3979 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
3980 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
3981 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
3982 structure modes. Split into pieces after RA.
3983 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
3984 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
3986 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
3987 SVE structure modes.
3988 (aarch64_classify_address): Likewise.
3989 (sizetochar): Move earlier in file.
3990 (aarch64_print_operand): Handle SVE register lists.
3991 (aarch64_array_mode): New function.
3992 (aarch64_sve_struct_memory_operand_p): Likewise.
3993 (TARGET_ARRAY_MODE): Redefine.
3995 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3996 Alan Hayward <alan.hayward@arm.com>
3997 David Sherwood <david.sherwood@arm.com>
3999 * target.def (array_mode): New target hook.
4000 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
4001 * doc/tm.texi: Regenerate.
4002 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
4003 * hooks.c (hook_optmode_mode_uhwi_none): New function.
4004 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
4006 * stor-layout.c (mode_for_array): Likewise. Support polynomial
4009 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4010 Alan Hayward <alan.hayward@arm.com>
4011 David Sherwood <david.sherwood@arm.com>
4013 * fold-const.c (fold_binary_loc): Check the argument types
4014 rather than the result type when testing for a vector operation.
4016 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4018 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
4019 * doc/tm.texi: Regenerate.
4021 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4022 Alan Hayward <alan.hayward@arm.com>
4023 David Sherwood <david.sherwood@arm.com>
4025 * doc/invoke.texi (-msve-vector-bits=): Document new option.
4026 (sve): Document new AArch64 extension.
4027 * doc/md.texi (w): Extend the description of the AArch64
4028 constraint to include SVE vectors.
4029 (Upl, Upa): Document new AArch64 predicate constraints.
4030 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
4032 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
4033 (msve-vector-bits=): New option.
4034 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
4035 SVE when these are disabled.
4036 (sve): New extension.
4037 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
4038 modes. Adjust their number of units based on aarch64_sve_vg.
4039 (MAX_BITSIZE_MODE_ANY_MODE): Define.
4040 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
4041 aarch64_addr_query_type.
4042 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
4043 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
4044 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
4045 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
4046 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
4047 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
4048 (aarch64_simd_imm_zero_p): Delete.
4049 (aarch64_check_zero_based_sve_index_immediate): Declare.
4050 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4051 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4052 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4053 (aarch64_sve_float_mul_immediate_p): Likewise.
4054 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4056 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
4057 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
4058 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
4059 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
4060 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
4061 (aarch64_regmode_natural_size): Likewise.
4062 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
4063 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
4065 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
4066 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
4067 for VG and the SVE predicate registers.
4068 (V_ALIASES): Add a "z"-prefixed alias.
4069 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
4070 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
4071 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
4072 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
4073 (REG_CLASS_NAMES): Add entries for them.
4074 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
4075 and the predicate registers.
4076 (aarch64_sve_vg): Declare.
4077 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
4078 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
4079 (REGMODE_NATURAL_SIZE): Define.
4080 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
4082 * config/aarch64/aarch64.c: Include cfgrtl.h.
4083 (simd_immediate_info): Add a constructor for series vectors,
4084 and an associated step field.
4085 (aarch64_sve_vg): New variable.
4086 (aarch64_dbx_register_number): Handle VG and the predicate registers.
4087 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
4088 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
4089 (VEC_ANY_DATA, VEC_STRUCT): New constants.
4090 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
4091 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
4092 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
4093 (aarch64_get_mask_mode): New functions.
4094 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
4095 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
4096 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
4097 predicate modes and predicate registers. Explicitly restrict
4098 GPRs to modes of 16 bytes or smaller. Only allow FP registers
4099 to store a vector mode if it is recognized by
4100 aarch64_classify_vector_mode.
4101 (aarch64_regmode_natural_size): New function.
4102 (aarch64_hard_regno_caller_save_mode): Return the original mode
4104 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
4105 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
4106 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
4107 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
4109 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
4110 does not overlap dest if the function is frame-related. Handle
4112 (aarch64_split_add_offset): New function.
4113 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
4114 them aarch64_add_offset.
4115 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
4116 and update call to aarch64_sub_sp.
4117 (aarch64_add_cfa_expression): New function.
4118 (aarch64_expand_prologue): Pass extra temporary registers to the
4119 functions above. Handle the case in which we need to emit new
4120 DW_CFA_expressions for registers that were originally saved
4121 relative to the stack pointer, but now have to be expressed
4122 relative to the frame pointer.
4123 (aarch64_output_mi_thunk): Pass extra temporary registers to the
4125 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
4126 IP0 and IP1 values for SVE frames.
4127 (aarch64_expand_vec_series): New function.
4128 (aarch64_expand_sve_widened_duplicate): Likewise.
4129 (aarch64_expand_sve_const_vector): Likewise.
4130 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
4131 Handle SVE constants. Use emit_move_insn to move a force_const_mem
4132 into the register, rather than emitting a SET directly.
4133 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
4134 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
4135 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
4136 (offset_9bit_signed_scaled_p): New functions.
4137 (aarch64_replicate_bitmask_imm): New function.
4138 (aarch64_bitmask_imm): Use it.
4139 (aarch64_cannot_force_const_mem): Reject expressions involving
4140 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
4141 (aarch64_classify_index): Handle SVE indices, by requiring
4142 a plain register index with a scale that matches the element size.
4143 (aarch64_classify_address): Handle SVE addresses. Assert that
4144 the mode of the address is VOIDmode or an integer mode.
4145 Update call to aarch64_classify_symbol.
4146 (aarch64_classify_symbolic_expression): Update call to
4147 aarch64_classify_symbol.
4148 (aarch64_const_vec_all_in_range_p): New function.
4149 (aarch64_print_vector_float_operand): Likewise.
4150 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
4151 "vN" for FP registers with SVE modes. Handle (const ...) vectors
4152 and the FP immediates 1.0 and 0.5.
4153 (aarch64_print_address_internal): Handle SVE addresses.
4154 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
4155 (aarch64_regno_regclass): Handle predicate registers.
4156 (aarch64_secondary_reload): Handle big-endian reloads of SVE
4158 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
4159 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
4160 (aarch64_convert_sve_vector_bits): New function.
4161 (aarch64_override_options): Use it to handle -msve-vector-bits=.
4162 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
4164 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
4165 Handle SVE vector and predicate modes. Accept VL-based constants
4166 that need only one temporary register, and VL offsets that require
4167 no temporary registers.
4168 (aarch64_conditional_register_usage): Mark the predicate registers
4169 as fixed if SVE isn't available.
4170 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
4171 Return true for SVE vector and predicate modes.
4172 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
4173 rather than an unsigned int. Handle SVE modes.
4174 (aarch64_preferred_simd_mode): Update call accordingly. Handle
4176 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
4178 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
4179 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
4180 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
4181 (aarch64_sve_float_mul_immediate_p): New functions.
4182 (aarch64_sve_valid_immediate): New function.
4183 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
4184 Explicitly reject structure modes. Check for INDEX constants.
4185 Handle PTRUE and PFALSE constants.
4186 (aarch64_check_zero_based_sve_index_immediate): New function.
4187 (aarch64_simd_imm_zero_p): Delete.
4188 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
4189 vector modes. Accept constants in the range of CNT[BHWD].
4190 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
4191 ask for an Advanced SIMD mode.
4192 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
4193 (aarch64_simd_vector_alignment): Handle SVE predicates.
4194 (aarch64_vectorize_preferred_vector_alignment): New function.
4195 (aarch64_simd_vector_alignment_reachable): Use it instead of
4197 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
4198 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
4200 (MAX_VECT_LEN): Delete.
4201 (expand_vec_perm_d): Add a vec_flags field.
4202 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
4203 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
4204 (aarch64_evpc_ext): Don't apply a big-endian lane correction
4206 (aarch64_evpc_rev): Rename to...
4207 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
4208 (aarch64_evpc_rev_global): New function.
4209 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
4210 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
4212 (aarch64_evpc_sve_tbl): New function.
4213 (aarch64_expand_vec_perm_const_1): Update after rename of
4214 aarch64_evpc_rev. Handle SVE permutes too, trying
4215 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
4216 than aarch64_evpc_tbl.
4217 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
4218 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
4219 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
4220 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
4221 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
4222 (aarch64_expand_sve_vcond): New functions.
4223 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
4224 of aarch64_vector_mode_p.
4225 (aarch64_dwarf_poly_indeterminate_value): New function.
4226 (aarch64_compute_pressure_classes): Likewise.
4227 (aarch64_can_change_mode_class): Likewise.
4228 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
4229 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
4230 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
4231 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
4232 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
4233 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
4234 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
4235 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
4237 (Dn, Dl, Dr): Accept const as well as const_vector.
4238 (Dz): Likewise. Compare against CONST0_RTX.
4239 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
4240 of "vector" where appropriate.
4241 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
4242 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
4243 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
4244 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
4245 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
4246 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
4247 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
4248 (v_int_equiv): Extend to SVE modes.
4249 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
4251 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
4252 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
4253 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
4254 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
4255 (SVE_COND_FP_CMP): New int iterators.
4256 (perm_hilo): Handle the new unpack unspecs.
4257 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
4259 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
4260 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
4261 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
4262 (aarch64_equality_operator, aarch64_constant_vector_operand)
4263 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
4264 (aarch64_sve_nonimmediate_operand): Likewise.
4265 (aarch64_sve_general_operand): Likewise.
4266 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
4267 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
4268 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
4269 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
4270 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
4271 (aarch64_sve_float_arith_immediate): Likewise.
4272 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
4273 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
4274 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
4275 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
4276 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
4277 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
4278 (aarch64_sve_float_arith_operand): Likewise.
4279 (aarch64_sve_float_arith_with_sub_operand): Likewise.
4280 (aarch64_sve_float_mul_operand): Likewise.
4281 (aarch64_sve_vec_perm_operand): Likewise.
4282 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
4283 (aarch64_mov_operand): Accept const_poly_int and const_vector.
4284 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
4285 as well as const_vector.
4286 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
4287 in file. Use CONST0_RTX and CONSTM1_RTX.
4288 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
4289 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
4290 Use aarch64_simd_imm_zero.
4291 * config/aarch64/aarch64-sve.md: New file.
4292 * config/aarch64/aarch64.md: Include it.
4293 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
4294 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
4295 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
4296 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
4297 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
4298 (sve): New attribute.
4299 (enabled): Disable instructions with the sve attribute unless
4301 (movqi, movhi): Pass CONST_POLY_INT operaneds through
4302 aarch64_expand_mov_immediate.
4303 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
4304 CNT[BHSD] immediates.
4305 (movti): Split CONST_POLY_INT moves into two halves.
4306 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
4307 Split additions that need a temporary here if the destination
4308 is the stack pointer.
4309 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
4310 (*add<mode>3_poly_1): New instruction.
4311 (set_clobber_cc): New expander.
4313 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4315 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
4316 parameter and use it instead of GET_MODE_SIZE (innermode). Use
4317 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
4318 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
4319 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
4320 Change innermode from fixed_mode_size to machine_mode.
4321 (simplify_subreg): Update call accordingly. Handle a constant-sized
4322 subreg of a variable-length CONST_VECTOR.
4324 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
4325 Alan Hayward <alan.hayward@arm.com>
4326 David Sherwood <david.sherwood@arm.com>
4328 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
4329 (add_offset_to_base): New function, split out from...
4330 (create_mem_ref): ...here. When handling a scale other than 1,
4331 check first whether the address is valid without the offset.
4332 Add it into the base if so, leaving the index and scale as-is.
4334 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4337 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
4338 fold_for_warn before checking if arg2 is INTEGER_CST.
4340 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
4342 * config/rs6000/predicates.md (load_multiple_operation): Delete.
4343 (store_multiple_operation): Delete.
4344 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
4345 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
4346 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
4347 guarded by TARGET_STRING.
4348 (rs6000_output_load_multiple): Delete.
4349 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
4350 OPTION_MASK_STRING / TARGET_STRING handling.
4351 (print_operand) <'N', 'O'>: Add comment that these are unused now.
4352 (const rs6000_opt_masks) <"string">: Change mask to 0.
4353 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
4354 (MASK_STRING): Delete.
4355 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
4357 (load_multiple): Delete.
4364 (store_multiple): Delete.
4371 (movmemsi_8reg): Delete.
4372 (corresponding unnamed define_insn): Delete.
4373 (movmemsi_6reg): Delete.
4374 (corresponding unnamed define_insn): Delete.
4375 (movmemsi_4reg): Delete.
4376 (corresponding unnamed define_insn): Delete.
4377 (movmemsi_2reg): Delete.
4378 (corresponding unnamed define_insn): Delete.
4379 (movmemsi_1reg): Delete.
4380 (corresponding unnamed define_insn): Delete.
4381 * config/rs6000/rs6000.opt (mno-string): New.
4382 (mstring): Replace by deprecation warning stub.
4383 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
4385 2018-01-12 Jakub Jelinek <jakub@redhat.com>
4387 * regrename.c (regrename_do_replace): If replacing the same
4388 reg multiple times, try to reuse last created gen_raw_REG.
4391 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
4392 main to workaround a bug in GDB.
4394 2018-01-12 Tom de Vries <tom@codesourcery.com>
4397 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
4399 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
4401 PR rtl-optimization/80481
4402 * ira-color.c (get_cap_member): New function.
4403 (allocnos_conflict_by_live_ranges_p): Use it.
4404 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
4405 (setup_slot_coalesced_allocno_live_ranges): Ditto.
4407 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
4410 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
4411 (*saddl_se_1): Ditto.
4413 (*ssubl_se_1): Ditto.
4415 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4417 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
4418 rather than wi::to_widest for DR_INITs.
4419 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
4420 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
4421 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
4423 (vect_analyze_group_access_1): Note that here.
4425 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4427 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
4428 polynomial type sizes.
4430 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
4432 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
4433 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
4434 (gimple_add_tmp_var): Likewise.
4436 2018-01-12 Martin Liska <mliska@suse.cz>
4438 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
4439 (gimple_alloc_sizes): Likewise.
4440 (dump_gimple_statistics): Use PRIu64 in printf format.
4441 * gimple.h: Change uint64_t to int.
4443 2018-01-12 Martin Liska <mliska@suse.cz>
4445 * tree-core.h: Use uint64_t instead of int.
4446 * tree.c (tree_node_counts): Likewise.
4447 (tree_node_sizes): Likewise.
4448 (dump_tree_statistics): Use PRIu64 in printf format.
4450 2018-01-12 Martin Liska <mliska@suse.cz>
4452 * Makefile.in: As qsort_chk is implemented in vec.c, add
4453 vec.o to linkage of gencfn-macros.
4454 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
4455 passing the info to record_node_allocation_statistics.
4456 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
4458 * ggc-common.c (struct ggc_usage): Add operator== and use
4459 it in operator< and compare function.
4460 * mem-stats.h (struct mem_usage): Likewise.
4461 * vec.c (struct vec_usage): Remove operator< and compare
4462 function. Can be simply inherited.
4464 2018-01-12 Martin Jambor <mjambor@suse.cz>
4467 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
4468 * tree-ssa-math-opts.c: Include domwalk.h.
4469 (convert_mult_to_fma_1): New function.
4470 (fma_transformation_info): New type.
4471 (fma_deferring_state): Likewise.
4472 (cancel_fma_deferring): New function.
4473 (result_of_phi): Likewise.
4474 (last_fma_candidate_feeds_initial_phi): Likewise.
4475 (convert_mult_to_fma): Added deferring logic, split actual
4476 transformation to convert_mult_to_fma_1.
4477 (math_opts_dom_walker): New type.
4478 (math_opts_dom_walker::after_dom_children): New method, body moved
4479 here from pass_optimize_widening_mul::execute, added deferring logic
4481 (pass_optimize_widening_mul::execute): Moved most of code to
4482 math_opts_dom_walker::after_dom_children.
4483 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
4484 * config/i386/i386.c (ix86_option_override_internal): Added
4485 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
4487 2018-01-12 Richard Biener <rguenther@suse.de>
4490 * dwarf2out.c (gen_variable_die): Do not reset old_die for
4491 inline instance vars.
4493 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
4496 * config/rx/rx.c (rx_is_restricted_memory_address):
4499 2018-01-12 Richard Biener <rguenther@suse.de>
4501 PR tree-optimization/80846
4502 * target.def (split_reduction): New target hook.
4503 * targhooks.c (default_split_reduction): New function.
4504 * targhooks.h (default_split_reduction): Declare.
4505 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
4506 target requests first reduce vectors by combining low and high
4508 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
4509 (get_vectype_for_scalar_type_and_size): Export.
4510 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
4511 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
4512 * doc/tm.texi: Regenerate.
4513 * config/i386/i386.c (ix86_split_reduction): Implement
4514 TARGET_VECTORIZE_SPLIT_REDUCTION.
4516 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4519 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
4520 in PIC mode except for TARGET_VXWORKS_RTP.
4521 * config/sparc/sparc.c: Include cfgrtl.h.
4522 (TARGET_INIT_PIC_REG): Define.
4523 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
4524 (sparc_pic_register_p): New predicate.
4525 (sparc_legitimate_address_p): Use it.
4526 (sparc_legitimize_pic_address): Likewise.
4527 (sparc_delegitimize_address): Likewise.
4528 (sparc_mode_dependent_address_p): Likewise.
4529 (gen_load_pcrel_sym): Remove 4th parameter.
4530 (load_got_register): Adjust call to above. Remove obsolete stuff.
4531 (sparc_expand_prologue): Do not call load_got_register here.
4532 (sparc_flat_expand_prologue): Likewise.
4533 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
4534 (sparc_use_pseudo_pic_reg): New function.
4535 (sparc_init_pic_reg): Likewise.
4536 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
4537 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
4539 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
4541 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
4542 Add item for branch_cost.
4544 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
4546 PR rtl-optimization/83565
4547 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
4548 not extend the result to a larger mode for rotate operations.
4549 (num_sign_bit_copies1): Likewise.
4551 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4554 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
4556 Use values-Xc.o for -pedantic.
4557 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
4559 2018-01-12 Martin Liska <mliska@suse.cz>
4562 * ipa-devirt.c (final_warning_record::grow_type_warnings):
4564 (possible_polymorphic_call_targets): Use it.
4565 (ipa_devirt): Likewise.
4567 2018-01-12 Martin Liska <mliska@suse.cz>
4569 * profile-count.h (enum profile_quality): Use 0 as invalid
4570 enum value of profile_quality.
4572 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
4574 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
4575 -mext-string options.
4577 2018-01-12 Richard Biener <rguenther@suse.de>
4579 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
4580 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
4581 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
4583 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
4585 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
4587 * configure.ac (--with-long-double-format): Add support for the
4588 configuration option to change the default long double format on
4590 * config.gcc (powerpc*-linux*-*): Likewise.
4591 * configure: Regenerate.
4592 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
4593 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
4594 used without modification.
4596 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4598 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
4599 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
4600 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
4601 MISC_BUILTIN_SPEC_BARRIER.
4602 (rs6000_init_builtins): Likewise.
4603 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
4605 (speculation_barrier): New define_insn.
4606 * doc/extend.texi: Document __builtin_speculation_barrier.
4608 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4611 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
4612 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
4613 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
4615 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
4616 integral modes instead of "ss" and "sd".
4617 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
4618 vectors with 32-bit and 64-bit elements.
4619 (vecdupssescalarmodesuffix): New mode attribute.
4620 (vec_dup<mode>): Use it.
4622 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
4625 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
4626 frame if argument is passed on stack.
4628 2018-01-11 Jakub Jelinek <jakub@redhat.com>
4631 * ree.c (combine_reaching_defs): Optimize also
4632 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
4633 reg2=any_extend(exp); reg1=reg2;, formatting fix.
4635 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4638 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
4640 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
4643 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
4644 after they are computed.
4646 2018-01-11 Bin Cheng <bin.cheng@arm.com>
4648 PR tree-optimization/83695
4649 * gimple-loop-linterchange.cc
4650 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
4651 reset cached scev information after interchange.
4652 (pass_linterchange::execute): Remove call to scev_reset_htab.
4654 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4656 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
4657 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
4658 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
4659 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
4660 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
4661 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
4662 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
4663 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
4664 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
4665 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
4666 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
4667 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
4668 (V_lane_reg): Likewise.
4669 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
4671 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
4672 (vfmal_lane_low<mode>_intrinsic,
4673 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
4674 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
4675 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
4676 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
4677 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
4678 vfmsl_lane_high<mode>_intrinsic): New define_insns.
4680 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4682 * config/arm/arm-cpus.in (fp16fml): New feature.
4683 (ALL_SIMD): Add fp16fml.
4684 (armv8.2-a): Add fp16fml as an option.
4685 (armv8.3-a): Likewise.
4686 (armv8.4-a): Add fp16fml as part of fp16.
4687 * config/arm/arm.h (TARGET_FP16FML): Define.
4688 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
4690 * config/arm/arm-modes.def (V2HF): Define.
4691 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
4692 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
4693 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
4694 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
4695 vfmsl_low, vfmsl_high): New set of builtins.
4696 * config/arm/iterators.md (PLUSMINUS): New code iterator.
4697 (vfml_op): New code attribute.
4698 (VFMLHALVES): New int iterator.
4699 (VFML, VFMLSEL): New mode attributes.
4700 (V_reg): Define mapping for V2HF.
4701 (V_hi, V_lo): New mode attributes.
4702 (VF_constraint): Likewise.
4703 (vfml_half, vfml_half_selector): New int attributes.
4704 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
4706 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
4707 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
4709 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
4710 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
4711 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
4712 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
4714 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
4715 Document new effective target and option set.
4717 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4719 * config/arm/arm-cpus.in (armv8_4): New feature.
4720 (ARMv8_4a): New fgroup.
4721 (armv8.4-a): New arch.
4722 * config/arm/arm-tables.opt: Regenerate.
4723 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
4724 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
4725 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
4726 Add matching rules for -march=armv8.4-a and extensions.
4727 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
4729 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
4732 * config/rx/rx.md (BW): New mode attribute.
4733 (sync_lock_test_and_setsi): Add mode suffix to insn output.
4735 2018-01-11 Richard Biener <rguenther@suse.de>
4737 PR tree-optimization/83435
4738 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
4739 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
4740 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
4742 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4743 Alan Hayward <alan.hayward@arm.com>
4744 David Sherwood <david.sherwood@arm.com>
4746 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
4748 (aarch64_classify_address): Initialize it. Track polynomial offsets.
4749 (aarch64_print_address_internal): Use it to check for a zero offset.
4751 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4752 Alan Hayward <alan.hayward@arm.com>
4753 David Sherwood <david.sherwood@arm.com>
4755 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
4756 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
4757 Return a poly_int64 rather than a HOST_WIDE_INT.
4758 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
4759 rather than a HOST_WIDE_INT.
4760 * config/aarch64/aarch64.h (aarch64_frame): Protect with
4761 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
4762 hard_fp_offset, frame_size, initial_adjust, callee_offset and
4763 final_offset from HOST_WIDE_INT to poly_int64.
4764 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
4765 to_constant when getting the number of units in an Advanced SIMD
4767 (aarch64_builtin_vectorized_function): Check for a constant number
4769 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
4771 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
4772 attribute instead of GET_MODE_NUNITS.
4773 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
4774 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
4775 GET_MODE_SIZE for fixed-size registers.
4776 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
4777 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
4778 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
4779 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
4780 (aarch64_print_operand, aarch64_print_address_internal)
4781 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
4782 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
4783 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
4784 Handle polynomial GET_MODE_SIZE.
4785 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
4786 wider than SImode without modification.
4787 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
4788 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
4789 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
4790 passing and returning SVE modes.
4791 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
4792 rather than GEN_INT.
4793 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
4794 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
4795 (aarch64_allocate_and_probe_stack_space): Likewise.
4796 (aarch64_layout_frame): Cope with polynomial offsets.
4797 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
4798 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
4800 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
4801 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
4802 poly_int64 rather than a HOST_WIDE_INT.
4803 (aarch64_get_separate_components, aarch64_process_components)
4804 (aarch64_expand_prologue, aarch64_expand_epilogue)
4805 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
4806 (aarch64_anchor_offset): New function, split out from...
4807 (aarch64_legitimize_address): ...here.
4808 (aarch64_builtin_vectorization_cost): Handle polynomial
4809 TYPE_VECTOR_SUBPARTS.
4810 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
4812 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
4813 number of elements from the PARALLEL rather than the mode.
4814 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
4815 rather than GET_MODE_BITSIZE.
4816 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
4817 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
4818 (aarch64_expand_vec_perm_const_1): Handle polynomial
4819 d->perm.length () and d->perm elements.
4820 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
4821 Apply to_constant to d->perm elements.
4822 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
4823 polynomial CONST_VECTOR_NUNITS.
4824 (aarch64_move_pointer): Take amount as a poly_int64 rather
4826 (aarch64_progress_pointer): Avoid temporary variable.
4827 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
4828 the mode attribute instead of GET_MODE.
4830 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4831 Alan Hayward <alan.hayward@arm.com>
4832 David Sherwood <david.sherwood@arm.com>
4834 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
4835 x exists before using it.
4836 (aarch64_add_constant_internal): Rename to...
4837 (aarch64_add_offset_1): ...this. Replace regnum with separate
4838 src and dest rtxes. Handle the case in which they're different,
4839 including when the offset is zero. Replace scratchreg with an rtx.
4840 Use 2 additions if there is no spare register into which we can
4841 move a 16-bit constant.
4842 (aarch64_add_constant): Delete.
4843 (aarch64_add_offset): Replace reg with separate src and dest
4844 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
4845 Use aarch64_add_offset_1.
4846 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
4847 an rtx rather than an int. Take the delta as a poly_int64
4848 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
4849 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
4850 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
4851 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
4852 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
4854 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
4855 aarch64_add_constant.
4857 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4859 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
4860 Use scalar_float_mode.
4862 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4864 * config/aarch64/aarch64-simd.md
4865 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
4866 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
4867 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
4868 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
4869 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
4870 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
4871 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
4872 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
4873 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
4874 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
4876 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4879 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
4880 targ_options->x_arm_arch_string is non NULL.
4882 2018-01-11 Tamar Christina <tamar.christina@arm.com>
4884 * config/aarch64/aarch64.h
4885 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
4887 2018-01-11 Sudakshina Das <sudi.das@arm.com>
4890 * expmed.c (emit_store_flag_force): Swap if const op0
4891 and change VOIDmode to mode of op0.
4893 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
4895 PR rtl-optimization/83761
4896 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
4897 than bytes to mode_for_size.
4899 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4902 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
4903 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
4906 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4909 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
4910 when in layout mode.
4911 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
4912 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
4915 2018-01-10 Michael Collison <michael.collison@arm.com>
4917 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
4918 * config/aarch64/aarch64-option-extension.def: Add
4919 AARCH64_OPT_EXTENSION of 'fp16fml'.
4920 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
4921 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
4922 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
4923 * config/aarch64/constraints.md (Ui7): New constraint.
4924 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
4925 (VFMLA_SEL_W): Ditto.
4928 (VFMLA16_LOW): New int iterator.
4929 (VFMLA16_HIGH): Ditto.
4930 (UNSPEC_FMLAL): New unspec.
4931 (UNSPEC_FMLSL): Ditto.
4932 (UNSPEC_FMLAL2): Ditto.
4933 (UNSPEC_FMLSL2): Ditto.
4934 (f16mac): New code attribute.
4935 * config/aarch64/aarch64-simd-builtins.def
4936 (aarch64_fmlal_lowv2sf): Ditto.
4937 (aarch64_fmlsl_lowv2sf): Ditto.
4938 (aarch64_fmlalq_lowv4sf): Ditto.
4939 (aarch64_fmlslq_lowv4sf): Ditto.
4940 (aarch64_fmlal_highv2sf): Ditto.
4941 (aarch64_fmlsl_highv2sf): Ditto.
4942 (aarch64_fmlalq_highv4sf): Ditto.
4943 (aarch64_fmlslq_highv4sf): Ditto.
4944 (aarch64_fmlal_lane_lowv2sf): Ditto.
4945 (aarch64_fmlsl_lane_lowv2sf): Ditto.
4946 (aarch64_fmlal_laneq_lowv2sf): Ditto.
4947 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
4948 (aarch64_fmlalq_lane_lowv4sf): Ditto.
4949 (aarch64_fmlsl_lane_lowv4sf): Ditto.
4950 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
4951 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
4952 (aarch64_fmlal_lane_highv2sf): Ditto.
4953 (aarch64_fmlsl_lane_highv2sf): Ditto.
4954 (aarch64_fmlal_laneq_highv2sf): Ditto.
4955 (aarch64_fmlsl_laneq_highv2sf): Ditto.
4956 (aarch64_fmlalq_lane_highv4sf): Ditto.
4957 (aarch64_fmlsl_lane_highv4sf): Ditto.
4958 (aarch64_fmlalq_laneq_highv4sf): Ditto.
4959 (aarch64_fmlsl_laneq_highv4sf): Ditto.
4960 * config/aarch64/aarch64-simd.md:
4961 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
4962 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4963 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
4964 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
4965 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
4966 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
4967 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
4968 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
4969 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
4970 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
4971 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
4972 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
4973 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
4974 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
4975 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
4976 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
4977 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
4978 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
4979 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
4980 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
4981 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
4982 (vfmlsl_low_u32): Ditto.
4983 (vfmlalq_low_u32): Ditto.
4984 (vfmlslq_low_u32): Ditto.
4985 (vfmlal_high_u32): Ditto.
4986 (vfmlsl_high_u32): Ditto.
4987 (vfmlalq_high_u32): Ditto.
4988 (vfmlslq_high_u32): Ditto.
4989 (vfmlal_lane_low_u32): Ditto.
4990 (vfmlsl_lane_low_u32): Ditto.
4991 (vfmlal_laneq_low_u32): Ditto.
4992 (vfmlsl_laneq_low_u32): Ditto.
4993 (vfmlalq_lane_low_u32): Ditto.
4994 (vfmlslq_lane_low_u32): Ditto.
4995 (vfmlalq_laneq_low_u32): Ditto.
4996 (vfmlslq_laneq_low_u32): Ditto.
4997 (vfmlal_lane_high_u32): Ditto.
4998 (vfmlsl_lane_high_u32): Ditto.
4999 (vfmlal_laneq_high_u32): Ditto.
5000 (vfmlsl_laneq_high_u32): Ditto.
5001 (vfmlalq_lane_high_u32): Ditto.
5002 (vfmlslq_lane_high_u32): Ditto.
5003 (vfmlalq_laneq_high_u32): Ditto.
5004 (vfmlslq_laneq_high_u32): Ditto.
5005 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
5006 (AARCH64_FL_FOR_ARCH8_4): New.
5007 (AARCH64_ISA_F16FML): New ISA flag.
5008 (TARGET_F16FML): New feature flag for fp16fml.
5009 (doc/invoke.texi): Document new fp16fml option.
5011 2018-01-10 Michael Collison <michael.collison@arm.com>
5013 * config/aarch64/aarch64-builtins.c:
5014 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
5015 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5016 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
5017 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
5018 (AARCH64_ISA_SHA3): New ISA flag.
5019 (TARGET_SHA3): New feature flag for sha3.
5020 * config/aarch64/iterators.md (sha512_op): New int attribute.
5021 (CRYPTO_SHA512): New int iterator.
5022 (UNSPEC_SHA512H): New unspec.
5023 (UNSPEC_SHA512H2): Ditto.
5024 (UNSPEC_SHA512SU0): Ditto.
5025 (UNSPEC_SHA512SU1): Ditto.
5026 * config/aarch64/aarch64-simd-builtins.def
5027 (aarch64_crypto_sha512hqv2di): New builtin.
5028 (aarch64_crypto_sha512h2qv2di): Ditto.
5029 (aarch64_crypto_sha512su0qv2di): Ditto.
5030 (aarch64_crypto_sha512su1qv2di): Ditto.
5031 (aarch64_eor3qv8hi): Ditto.
5032 (aarch64_rax1qv2di): Ditto.
5033 (aarch64_xarqv2di): Ditto.
5034 (aarch64_bcaxqv8hi): Ditto.
5035 * config/aarch64/aarch64-simd.md:
5036 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
5037 (aarch64_crypto_sha512su0qv2di): Ditto.
5038 (aarch64_crypto_sha512su1qv2di): Ditto.
5039 (aarch64_eor3qv8hi): Ditto.
5040 (aarch64_rax1qv2di): Ditto.
5041 (aarch64_xarqv2di): Ditto.
5042 (aarch64_bcaxqv8hi): Ditto.
5043 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
5044 (vsha512h2q_u64): Ditto.
5045 (vsha512su0q_u64): Ditto.
5046 (vsha512su1q_u64): Ditto.
5047 (veor3q_u16): Ditto.
5048 (vrax1q_u64): Ditto.
5050 (vbcaxq_u16): Ditto.
5051 * config/arm/types.md (crypto_sha512): New type attribute.
5052 (crypto_sha3): Ditto.
5053 (doc/invoke.texi): Document new sha3 option.
5055 2018-01-10 Michael Collison <michael.collison@arm.com>
5057 * config/aarch64/aarch64-builtins.c:
5058 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
5059 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5060 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
5061 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
5062 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
5063 (AARCH64_ISA_SM4): New ISA flag.
5064 (TARGET_SM4): New feature flag for sm4.
5065 * config/aarch64/aarch64-simd-builtins.def
5066 (aarch64_sm3ss1qv4si): Ditto.
5067 (aarch64_sm3tt1aq4si): Ditto.
5068 (aarch64_sm3tt1bq4si): Ditto.
5069 (aarch64_sm3tt2aq4si): Ditto.
5070 (aarch64_sm3tt2bq4si): Ditto.
5071 (aarch64_sm3partw1qv4si): Ditto.
5072 (aarch64_sm3partw2qv4si): Ditto.
5073 (aarch64_sm4eqv4si): Ditto.
5074 (aarch64_sm4ekeyqv4si): Ditto.
5075 * config/aarch64/aarch64-simd.md:
5076 (aarch64_sm3ss1qv4si): Ditto.
5077 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
5078 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
5079 (aarch64_sm4eqv4si): Ditto.
5080 (aarch64_sm4ekeyqv4si): Ditto.
5081 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
5082 (sm3part_op): Ditto.
5083 (CRYPTO_SM3TT): Ditto.
5084 (CRYPTO_SM3PART): Ditto.
5085 (UNSPEC_SM3SS1): New unspec.
5086 (UNSPEC_SM3TT1A): Ditto.
5087 (UNSPEC_SM3TT1B): Ditto.
5088 (UNSPEC_SM3TT2A): Ditto.
5089 (UNSPEC_SM3TT2B): Ditto.
5090 (UNSPEC_SM3PARTW1): Ditto.
5091 (UNSPEC_SM3PARTW2): Ditto.
5092 (UNSPEC_SM4E): Ditto.
5093 (UNSPEC_SM4EKEY): Ditto.
5094 * config/aarch64/constraints.md (Ui2): New constraint.
5095 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
5096 * config/arm/types.md (crypto_sm3): New type attribute.
5097 (crypto_sm4): Ditto.
5098 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
5099 (vsm3tt1aq_u32): Ditto.
5100 (vsm3tt1bq_u32): Ditto.
5101 (vsm3tt2aq_u32): Ditto.
5102 (vsm3tt2bq_u32): Ditto.
5103 (vsm3partw1q_u32): Ditto.
5104 (vsm3partw2q_u32): Ditto.
5105 (vsm4eq_u32): Ditto.
5106 (vsm4ekeyq_u32): Ditto.
5107 (doc/invoke.texi): Document new sm4 option.
5109 2018-01-10 Michael Collison <michael.collison@arm.com>
5111 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
5112 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
5113 (AARCH64_FL_FOR_ARCH8_4): New.
5114 (AARCH64_FL_V8_4): New flag.
5115 (doc/invoke.texi): Document new armv8.4-a option.
5117 2018-01-10 Michael Collison <michael.collison@arm.com>
5119 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
5120 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
5121 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
5122 * config/aarch64/aarch64-option-extension.def: Add
5123 AARCH64_OPT_EXTENSION of 'sha2'.
5124 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
5125 (crypto): Disable sha2 and aes if crypto disabled.
5126 (crypto): Enable aes and sha2 if enabled.
5127 (simd): Disable sha2 and aes if simd disabled.
5128 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
5130 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
5131 (TARGET_SHA2): New feature flag for sha2.
5132 (TARGET_AES): New feature flag for aes.
5133 * config/aarch64/aarch64-simd.md:
5134 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
5135 conditional on TARGET_AES.
5136 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
5137 (aarch64_crypto_sha1hsi): Make pattern conditional
5139 (aarch64_crypto_sha1hv4si): Ditto.
5140 (aarch64_be_crypto_sha1hv4si): Ditto.
5141 (aarch64_crypto_sha1su1v4si): Ditto.
5142 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
5143 (aarch64_crypto_sha1su0v4si): Ditto.
5144 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
5145 (aarch64_crypto_sha256su0v4si): Ditto.
5146 (aarch64_crypto_sha256su1v4si): Ditto.
5147 (doc/invoke.texi): Document new aes and sha2 options.
5149 2018-01-10 Martin Sebor <msebor@redhat.com>
5151 PR tree-optimization/83781
5152 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
5155 2018-01-11 Martin Sebor <msebor@gmail.com>
5156 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5158 PR tree-optimization/83501
5159 PR tree-optimization/81703
5161 * tree-ssa-strlen.c (get_string_cst): Rename...
5162 (get_string_len): ...to this. Handle global constants.
5163 (handle_char_store): Adjust.
5165 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
5166 Jim Wilson <jimw@sifive.com>
5168 * config/riscv/riscv-protos.h (riscv_output_return): New.
5169 * config/riscv/riscv.c (struct machine_function): New naked_p field.
5170 (riscv_attribute_table, riscv_output_return),
5171 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
5172 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
5173 (riscv_compute_frame_info): Only compute frame->mask if not a naked
5175 (riscv_expand_prologue): Add early return for naked function.
5176 (riscv_expand_epilogue): Likewise.
5177 (riscv_function_ok_for_sibcall): Return false for naked function.
5178 (riscv_set_current_function): New.
5179 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
5180 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
5181 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
5182 * doc/extend.texi (RISC-V Function Attributes): New.
5184 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
5186 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
5187 check for 128-bit long double before checking TCmode.
5188 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
5189 128-bit long doubles before checking TFmode or TCmode.
5190 (FLOAT128_IBM_P): Likewise.
5192 2018-01-10 Martin Sebor <msebor@redhat.com>
5194 PR tree-optimization/83671
5195 * builtins.c (c_strlen): Unconditionally return zero for the empty
5197 Use -Warray-bounds for warnings.
5198 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
5199 for non-constant array indices with COMPONENT_REF, arrays of
5200 arrays, and pointers to arrays.
5201 (gimple_fold_builtin_strlen): Determine and set length range for
5202 non-constant character arrays.
5204 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
5207 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
5210 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
5212 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
5214 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5217 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
5218 VECTOR_MEM_ALTIVEC_OR_VSX_P.
5219 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
5220 indexed_or_indirect_operand predicate.
5221 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
5222 (*vsx_le_perm_load_v8hi): Likewise.
5223 (*vsx_le_perm_load_v16qi): Likewise.
5224 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
5225 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
5226 (*vsx_le_perm_store_v8hi): Likewise.
5227 (*vsx_le_perm_store_v16qi): Likewise.
5228 (eight unnamed splitters): Likewise.
5230 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
5232 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
5233 * config/rs6000/emmintrin.h: Likewise.
5234 * config/rs6000/mmintrin.h: Likewise.
5235 * config/rs6000/xmmintrin.h: Likewise.
5237 2018-01-10 David Malcolm <dmalcolm@redhat.com>
5240 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
5242 * tree.c (tree_nop_conversion): Return true for location wrapper
5244 (maybe_wrap_with_location): New function.
5245 (selftest::check_strip_nops): New function.
5246 (selftest::test_location_wrappers): New function.
5247 (selftest::tree_c_tests): Call it.
5248 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
5249 (maybe_wrap_with_location): New decl.
5250 (EXPR_LOCATION_WRAPPER_P): New macro.
5251 (location_wrapper_p): New inline function.
5252 (tree_strip_any_location_wrapper): New inline function.
5254 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
5257 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
5258 stack_realign_offset for the largest alignment of stack slot
5260 (ix86_find_max_used_stack_alignment): New function.
5261 (ix86_finalize_stack_frame_flags): Use it. Set
5262 max_used_stack_alignment if we don't realign stack.
5263 * config/i386/i386.h (machine_function): Add
5264 max_used_stack_alignment.
5266 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
5268 * config/arm/arm.opt (-mbranch-cost): New option.
5269 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
5272 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
5275 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
5276 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
5278 2018-01-10 Richard Biener <rguenther@suse.de>
5281 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
5282 early out so it also covers the case where we have a non-NULL
5285 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5287 PR tree-optimization/83753
5288 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
5289 for non-strided grouped accesses if the number of elements is 1.
5291 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
5294 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
5295 * i386.h (TARGET_USE_GATHER): Define.
5296 * x86-tune.def (X86_TUNE_USE_GATHER): New.
5298 2018-01-10 Martin Liska <mliska@suse.cz>
5301 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
5302 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
5304 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
5305 CLEANUP_NO_PARTITIONING is not set.
5307 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
5309 * doc/rtl.texi: Remove documentation of (const ...) wrappers
5310 for vectors, as a partial revert of r254296.
5311 * rtl.h (const_vec_p): Delete.
5312 (const_vec_duplicate_p): Don't test for vector CONSTs.
5313 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
5314 * expmed.c (make_tree): Likewise.
5317 * common.md (E, F): Use CONSTANT_P instead of checking for
5319 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
5320 checking for CONST_VECTOR.
5322 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5325 * predict.c (force_edge_cold): Handle in more sane way edges
5328 2018-01-09 Carl Love <cel@us.ibm.com>
5330 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
5332 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
5333 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
5334 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
5335 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
5336 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
5337 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
5338 * config/rs6000/rs6000-protos.h: Add extern defition for
5339 rs6000_generate_float2_double_code.
5340 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
5342 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
5343 (float2_v2df): Add define_expand.
5345 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
5348 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
5349 op_mode in the force_to_mode call.
5351 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5353 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
5354 instead of checking each element individually.
5355 (aarch64_evpc_uzp): Likewise.
5356 (aarch64_evpc_zip): Likewise.
5357 (aarch64_evpc_ext): Likewise.
5358 (aarch64_evpc_rev): Likewise.
5359 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
5360 instead of checking each element individually. Return true without
5362 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
5363 whether all selected elements come from the same input, instead of
5364 checking each element individually. Remove calls to gen_rtx_REG,
5365 start_sequence and end_sequence and instead assert that no rtl is
5368 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5370 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
5371 order of HIGH and CONST checks.
5373 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
5375 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
5376 if the destination isn't an SSA_NAME.
5378 2018-01-09 Richard Biener <rguenther@suse.de>
5380 PR tree-optimization/83668
5381 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
5383 (canonicalize_loop_form): ... here, renamed from ...
5384 (canonicalize_loop_closed_ssa_form): ... this and amended to
5385 swap successor edges for loop exit blocks to make us use
5386 the RPO order we need for initial schedule generation.
5388 2018-01-09 Joseph Myers <joseph@codesourcery.com>
5390 PR tree-optimization/64811
5391 * match.pd: When optimizing comparisons with Inf, avoid
5392 introducing or losing exceptions from comparisons with NaN.
5394 2018-01-09 Martin Liska <mliska@suse.cz>
5397 * asan.c (shadow_mem_size): Add gcc_assert.
5399 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
5401 Don't save registers in main().
5404 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
5405 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
5406 * config/avr/avr.c (avr_set_current_function): Don't error if
5407 naked, OS_task or OS_main are specified at the same time.
5408 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
5410 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
5412 * common/config/avr/avr-common.c (avr_option_optimization_table):
5413 Switch on -mmain-is-OS_task for optimizing compilations.
5415 2018-01-09 Richard Biener <rguenther@suse.de>
5417 PR tree-optimization/83572
5418 * graphite.c: Include cfganal.h.
5419 (graphite_transform_loops): Connect infinite loops to exit
5420 and remove fake edges at the end.
5422 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5424 * ipa-inline.c (edge_badness): Revert accidental checkin.
5426 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
5429 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
5430 symbols; not inline clones.
5432 2018-01-09 Jakub Jelinek <jakub@redhat.com>
5435 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
5436 hard registers. Formatting fixes.
5438 PR preprocessor/83722
5439 * gcc.c (try_generate_repro): Pass
5440 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
5441 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
5444 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
5445 Kito Cheng <kito.cheng@gmail.com>
5447 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
5448 (riscv_leaf_function_p): Delete.
5449 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
5451 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5453 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
5455 (do_ifelse): New function.
5456 (do_isel): New function.
5457 (do_sub3): New function.
5458 (do_add3): New function.
5459 (do_load_mask_compare): New function.
5460 (do_overlap_load_compare): New function.
5461 (expand_compare_loop): New function.
5462 (expand_block_compare): Call expand_compare_loop() when appropriate.
5463 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
5465 (-mblock-compare-inline-loop-limit): New option.
5467 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5470 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
5471 Reverse order of second and third operands in first alternative.
5472 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
5473 of first and second elements in UNSPEC_VPERMR vector.
5474 (altivec_expand_vec_perm_le): Likewise.
5476 2018-01-08 Jeff Law <law@redhat.com>
5478 PR rtl-optimizatin/81308
5479 * tree-switch-conversion.c (cfg_altered): New file scoped static.
5480 (process_switch): If group_case_labels makes a change, then set
5482 (pass_convert_switch::execute): If a switch is converted, then
5483 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
5485 PR rtl-optimization/81308
5486 * recog.c (split_all_insns): Conditionally cleanup the CFG after
5489 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
5491 PR target/83663 - Revert r255946
5492 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
5493 generation for cases where splatting a value is not useful.
5494 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
5495 across a vec_duplicate and a paradoxical subreg forming a vector
5496 mode to a vec_concat.
5498 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
5500 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
5501 -march=armv8.3-a variants.
5502 * config/arm/t-multilib: Likewise.
5503 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
5505 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5507 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
5509 (cceq_ior_compare_complement): Give it a name so I can use it, and
5510 change boolean_or_operator predicate to boolean_operator so it can
5511 be used to generate a crand.
5512 (eqne): New code iterator.
5513 (bd/bd_neg): New code_attrs.
5514 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
5515 a single define_insn.
5516 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
5517 decrement (bdnzt/bdnzf/bdzt/bdzf).
5518 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
5519 with the new names of the branch decrement patterns, and added the
5520 names of the branch decrement conditional patterns.
5522 2018-01-08 Richard Biener <rguenther@suse.de>
5524 PR tree-optimization/83563
5525 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
5528 2018-01-08 Richard Biener <rguenther@suse.de>
5531 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
5533 2018-01-08 Richard Biener <rguenther@suse.de>
5535 PR tree-optimization/83685
5536 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
5537 references to abnormals.
5539 2018-01-08 Richard Biener <rguenther@suse.de>
5542 * dwarf2out.c (output_indirect_strings): Handle empty
5543 skeleton_debug_str_hash.
5544 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
5546 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5548 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
5549 (emit_store_direct): Likewise.
5550 (arc_trampoline_adjust_address): Likewise.
5551 (arc_asm_trampoline_template): New function.
5552 (arc_initialize_trampoline): Use asm_trampoline_template.
5553 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
5554 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
5555 * config/arc/arc.md (flush_icache): Delete pattern.
5557 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
5559 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
5560 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
5563 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5566 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
5567 by not USED_FOR_TARGET.
5568 (make_pass_resolve_sw_modes): Likewise.
5570 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
5572 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
5575 2018-01-08 Richard Biener <rguenther@suse.de>
5578 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
5580 2018-01-08 Richard Biener <rguenther@suse.de>
5583 * match.pd ((t * 2) / 2) -> t): Add missing :c.
5585 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
5588 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
5589 basic blocks with a small number of successors.
5590 (convert_control_dep_chain_into_preds): Improve handling of
5592 (dump_predicates): Split apart into...
5593 (dump_pred_chain): ...here...
5594 (dump_pred_info): ...and here.
5595 (can_one_predicate_be_invalidated_p): Add debugging printfs.
5596 (can_chain_union_be_invalidated_p): Improve check for invalidation
5598 (uninit_uses_cannot_happen): Avoid unnecessary if
5599 convert_control_dep_chain_into_preds yielded nothing.
5601 2018-01-06 Martin Sebor <msebor@redhat.com>
5603 PR tree-optimization/83640
5604 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
5605 subtracting negative offset from size.
5606 (builtin_access::overlap): Adjust offset bounds of the access to fall
5607 within the size of the object if possible.
5609 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
5611 PR rtl-optimization/83699
5612 * expmed.c (extract_bit_field_1): Restrict the vector usage of
5613 extract_bit_field_as_subreg to cases in which the extracted
5614 value is also a vector.
5616 * lra-constraints.c (process_alt_operands): Test for the equivalence
5617 substitutions when detecting a possible reload cycle.
5619 2018-01-06 Jakub Jelinek <jakub@redhat.com>
5622 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
5623 by default if flag_selective_schedling{,2}. Formatting fixes.
5625 PR rtl-optimization/83682
5626 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
5627 if it has non-VECTOR_MODE element mode.
5628 (vec_duplicate_p): Likewise.
5631 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
5632 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
5634 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5637 * config/i386/i386-builtin.def
5638 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
5639 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
5640 Require also OPTION_MASK_ISA_AVX512F in addition to
5641 OPTION_MASK_ISA_GFNI.
5642 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
5643 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
5644 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
5645 to OPTION_MASK_ISA_GFNI.
5646 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
5647 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
5648 OPTION_MASK_ISA_AVX512BW.
5649 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
5650 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
5651 addition to OPTION_MASK_ISA_GFNI.
5652 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
5653 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
5654 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
5655 to OPTION_MASK_ISA_GFNI.
5656 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
5657 a requirement for all ISAs rather than any of them with a few
5659 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
5661 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
5662 bitmasks to be enabled with 3 exceptions, instead of requiring any
5663 enabled ISA with lots of exceptions.
5664 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
5665 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
5666 Change avx512bw in isa attribute to avx512f.
5667 * config/i386/sgxintrin.h: Add license boilerplate.
5668 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
5669 to __AVX512F__ and __AVX512VL to __AVX512VL__.
5670 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
5671 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
5673 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
5674 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
5675 temporarily sse2 rather than sse if not enabled already.
5678 * config/i386/sse.md (VI248_VLBW): Rename to ...
5679 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
5680 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
5681 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
5682 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
5683 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
5684 mode iterator instead of VI248_VLBW.
5686 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
5688 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
5689 (record_modified): Skip clobbers; add debug output.
5690 (param_change_prob): Use sreal frequencies.
5692 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5694 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
5695 punt for user-aligned variables.
5697 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
5699 * tree-chrec.c (chrec_contains_symbols): Return true for
5702 2018-01-05 Sudakshina Das <sudi.das@arm.com>
5705 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
5706 of (x|y) == x for BICS pattern.
5708 2018-01-05 Jakub Jelinek <jakub@redhat.com>
5710 PR tree-optimization/83605
5711 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
5712 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
5715 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
5717 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
5718 * config/epiphany/rtems.h: New file.
5720 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5721 Uros Bizjak <ubizjak@gmail.com>
5724 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
5725 QIreg_operand instead of register_operand predicate.
5726 * config/i386/i386.c (ix86_rop_should_change_byte_p,
5727 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
5728 comments instead of -fmitigate[-_]rop.
5730 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
5733 * cgraphunit.c (symbol_table::compile): Switch to text_section
5734 before calling assembly_start debug hook.
5735 * run-rtl-passes.c (run_rtl_passes): Likewise.
5738 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5740 * tree-vrp.c (extract_range_from_binary_expr_1): Check
5741 range_int_cst_p rather than !symbolic_range_p before calling
5742 extract_range_from_multiplicative_op_1.
5744 2018-01-04 Jeff Law <law@redhat.com>
5746 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
5747 redundant test in assertion.
5749 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5751 * doc/rtl.texi: Document machine_mode wrapper classes.
5753 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5755 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
5758 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5760 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
5761 the VEC_PERM_EXPR fold to fail.
5763 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5766 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
5767 to switched_sections.
5769 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5772 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
5775 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
5778 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
5779 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
5781 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5784 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
5785 is BLKmode and bitpos not zero or mode change is needed.
5787 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
5790 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
5793 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
5796 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
5797 instead of MULT rtx. Update all corresponding splitters.
5799 (*ssub<modesuffix>): Ditto.
5801 (*cmp_sadd_di): Update split patterns.
5802 (*cmp_sadd_si): Ditto.
5803 (*cmp_sadd_sidi): Ditto.
5804 (*cmp_ssub_di): Ditto.
5805 (*cmp_ssub_si): Ditto.
5806 (*cmp_ssub_sidi): Ditto.
5807 * config/alpha/predicates.md (const23_operand): New predicate.
5808 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
5809 Look for ASHIFT, not MULT inner operand.
5810 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
5812 2018-01-04 Martin Liska <mliska@suse.cz>
5814 PR gcov-profile/83669
5815 * gcov.c (output_intermediate_file): Add version to intermediate
5817 * doc/gcov.texi: Document new field 'version' in intermediate
5818 file format. Fix location of '-k' option of gcov command.
5820 2018-01-04 Martin Liska <mliska@suse.cz>
5823 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
5825 2018-01-04 Jakub Jelinek <jakub@redhat.com>
5827 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
5829 2018-01-03 Martin Sebor <msebor@redhat.com>
5831 PR tree-optimization/83655
5832 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
5833 checking calls with invalid arguments.
5835 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5837 * tree-vect-stmts.c (vect_get_store_rhs): New function.
5838 (vectorizable_mask_load_store): Delete.
5839 (vectorizable_call): Return false for masked loads and stores.
5840 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
5841 instead of gimple_assign_rhs1.
5842 (vectorizable_load): Handle IFN_MASK_LOAD.
5843 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
5845 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5847 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
5849 (vectorizable_mask_load_store): ...here.
5850 (vectorizable_load): ...and here.
5852 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5854 * tree-vect-stmts.c (vect_build_all_ones_mask)
5855 (vect_build_zero_merge_argument): New functions, split out from...
5856 (vectorizable_load): ...here.
5858 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5860 * tree-vect-stmts.c (vect_check_store_rhs): New function,
5862 (vectorizable_mask_load_store): ...here.
5863 (vectorizable_store): ...and here.
5865 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5867 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
5869 (vectorizable_mask_load_store): ...here.
5871 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5873 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
5874 (vect_model_store_cost): Take a vec_load_store_type instead of a
5876 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
5877 (vect_model_store_cost): Take a vec_load_store_type instead of a
5879 (vectorizable_mask_load_store): Update accordingly.
5880 (vectorizable_store): Likewise.
5881 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
5883 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5885 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
5886 IFN_MASK_LOAD calls here rather than...
5887 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
5889 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5890 Alan Hayward <alan.hayward@arm.com>
5891 David Sherwood <david.sherwood@arm.com>
5893 * expmed.c (extract_bit_field_1): For vector extracts,
5894 fall back to extract_bit_field_as_subreg if vec_extract
5897 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5898 Alan Hayward <alan.hayward@arm.com>
5899 David Sherwood <david.sherwood@arm.com>
5901 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
5902 they are variable or constant sized.
5903 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
5904 slots for constant-sized data.
5906 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5907 Alan Hayward <alan.hayward@arm.com>
5908 David Sherwood <david.sherwood@arm.com>
5910 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
5911 handling COND_EXPRs with boolean comparisons, try to find a better
5912 basis for the mask type than the boolean itself.
5914 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5916 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
5917 is calculated and how it can be overridden.
5918 * genmodes.c (max_bitsize_mode_any_mode): New variable.
5919 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
5921 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
5924 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5925 Alan Hayward <alan.hayward@arm.com>
5926 David Sherwood <david.sherwood@arm.com>
5928 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
5929 Remove the mode argument.
5930 (aarch64_simd_valid_immediate): Remove the mode and inverse
5932 * config/aarch64/iterators.md (bitsize): New iterator.
5933 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
5934 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
5935 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
5936 aarch64_simd_valid_immediate.
5937 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
5938 (aarch64_reg_or_bic_imm): Likewise.
5939 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
5940 with an insn_type enum and msl with a modifier_type enum.
5941 Replace element_width with a scalar_mode. Change the shift
5942 to unsigned int. Add constructors for scalar_float_mode and
5943 scalar_int_mode elements.
5944 (aarch64_vect_float_const_representable_p): Delete.
5945 (aarch64_can_const_movi_rtx_p)
5946 (aarch64_simd_scalar_immediate_valid_for_move)
5947 (aarch64_simd_make_constant): Update call to
5948 aarch64_simd_valid_immediate.
5949 (aarch64_advsimd_valid_immediate_hs): New function.
5950 (aarch64_advsimd_valid_immediate): Likewise.
5951 (aarch64_simd_valid_immediate): Remove mode and inverse
5952 arguments. Rewrite to use the above. Use const_vec_duplicate_p
5953 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
5954 and aarch64_float_const_representable_p on the result.
5955 (aarch64_output_simd_mov_immediate): Remove mode argument.
5956 Update call to aarch64_simd_valid_immediate and use of
5957 simd_immediate_info.
5958 (aarch64_output_scalar_simd_mov_immediate): Update call
5961 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5962 Alan Hayward <alan.hayward@arm.com>
5963 David Sherwood <david.sherwood@arm.com>
5965 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
5966 (mode_nunits): Likewise CONST_MODE_NUNITS.
5967 * machmode.def (ADJUST_NUNITS): Document.
5968 * genmodes.c (mode_data::need_nunits_adj): New field.
5969 (blank_mode): Update accordingly.
5970 (adj_nunits): New variable.
5971 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
5973 (emit_mode_size_inline): Set need_bytesize_adj for all modes
5974 listed in adj_nunits.
5975 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
5976 listed in adj_nunits. Don't emit case statements for such modes.
5977 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
5978 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
5979 nothing if adj_nunits is nonnull.
5980 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
5981 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
5982 (emit_mode_fbit): Update use of print_maybe_const_decl.
5983 (emit_move_size): Likewise. Treat the array as non-const
5985 (emit_mode_adjustments): Handle adj_nunits.
5987 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5989 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
5990 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
5991 (VECTOR_MODES): Use it.
5992 (make_vector_modes): Take the prefix as an argument.
5994 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5995 Alan Hayward <alan.hayward@arm.com>
5996 David Sherwood <david.sherwood@arm.com>
5998 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
5999 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
6000 for MODE_VECTOR_BOOL.
6001 * machmode.def (VECTOR_BOOL_MODE): Document.
6002 * genmodes.c (VECTOR_BOOL_MODE): New macro.
6003 (make_vector_bool_mode): New function.
6004 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
6006 * lto-streamer-in.c (lto_input_mode_table): Likewise.
6007 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
6009 * stor-layout.c (int_mode_for_mode): Likewise.
6010 * tree.c (build_vector_type_for_mode): Likewise.
6011 * varasm.c (output_constant_pool_2): Likewise.
6012 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
6013 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
6014 for MODE_VECTOR_BOOL.
6015 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
6016 of mode class checks.
6017 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
6018 instead of a list of mode class checks.
6019 (expand_vector_scalar_condition): Likewise.
6020 (type_for_widest_vector_mode): Handle BImode as an inner mode.
6022 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6023 Alan Hayward <alan.hayward@arm.com>
6024 David Sherwood <david.sherwood@arm.com>
6026 * machmode.h (mode_size): Change from unsigned short to
6028 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
6029 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6030 or if measurement_type is not polynomial.
6031 (fixed_size_mode::includes_p): Check for constant-sized modes.
6032 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
6033 return a poly_uint16 rather than an unsigned short.
6034 (emit_mode_size): Change the type of mode_size from unsigned short
6035 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
6036 (emit_mode_adjustments): Cope with polynomial vector sizes.
6037 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6039 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6041 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
6042 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
6043 * caller-save.c (setup_save_areas): Likewise.
6044 (replace_reg_with_saved_mem): Likewise.
6045 * calls.c (emit_library_call_value_1): Likewise.
6046 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
6047 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
6048 (gen_lowpart_for_combine): Likewise.
6049 * convert.c (convert_to_integer_1): Likewise.
6050 * cse.c (equiv_constant, cse_insn): Likewise.
6051 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
6052 (cselib_subst_to_values): Likewise.
6053 * dce.c (word_dce_process_block): Likewise.
6054 * df-problems.c (df_word_lr_mark_ref): Likewise.
6055 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
6056 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
6057 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
6058 (rtl_for_decl_location): Likewise.
6059 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
6060 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
6061 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
6062 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
6063 (expand_expr_real_1): Likewise.
6064 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
6065 (pad_below): Likewise.
6066 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6067 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
6068 * ira.c (get_subreg_tracking_sizes): Likewise.
6069 * ira-build.c (ira_create_allocno_objects): Likewise.
6070 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
6071 (ira_sort_regnos_for_alter_reg): Likewise.
6072 * ira-costs.c (record_operand_costs): Likewise.
6073 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
6074 (resolve_simple_move): Likewise.
6075 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
6076 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
6077 (lra_constraints): Likewise.
6078 (CONST_POOL_OK_P): Reject variable-sized modes.
6079 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
6080 (add_pseudo_to_slot, lra_spill): Likewise.
6081 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6082 * optabs-query.c (get_best_extraction_insn): Likewise.
6083 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6084 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
6085 (expand_mult_highpart, valid_multiword_target_p): Likewise.
6086 * recog.c (offsettable_address_addr_space_p): Likewise.
6087 * regcprop.c (maybe_mode_change): Likewise.
6088 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
6089 * regrename.c (build_def_use): Likewise.
6090 * regstat.c (dump_reg_info): Likewise.
6091 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
6092 (find_reloads, find_reloads_subreg_address): Likewise.
6093 * reload1.c (eliminate_regs_1): Likewise.
6094 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
6095 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
6096 (simplify_binary_operation_1, simplify_subreg): Likewise.
6097 * targhooks.c (default_function_arg_padding): Likewise.
6098 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
6099 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
6100 (verify_gimple_assign_ternary): Likewise.
6101 * tree-inline.c (estimate_move_cost): Likewise.
6102 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6103 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
6104 (get_address_cost_ainc): Likewise.
6105 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
6106 (vect_supportable_dr_alignment): Likewise.
6107 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6108 (vectorizable_reduction): Likewise.
6109 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
6110 (vectorizable_operation, vectorizable_load): Likewise.
6111 * tree.c (build_same_sized_truth_vector_type): Likewise.
6112 * valtrack.c (cleanup_auto_inc_dec): Likewise.
6113 * var-tracking.c (emit_note_insn_var_location): Likewise.
6114 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
6115 (ADDR_VEC_ALIGN): Likewise.
6117 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6118 Alan Hayward <alan.hayward@arm.com>
6119 David Sherwood <david.sherwood@arm.com>
6121 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
6123 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
6124 or if measurement_type is polynomial.
6125 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
6126 * combine.c (make_extraction): Likewise.
6127 * dse.c (find_shift_sequence): Likewise.
6128 * dwarf2out.c (mem_loc_descriptor): Likewise.
6129 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
6130 (extract_bit_field, extract_low_bits): Likewise.
6131 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
6132 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
6133 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
6134 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
6135 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
6136 * reload.c (find_reloads): Likewise.
6137 * reload1.c (alter_reg): Likewise.
6138 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
6139 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
6140 * tree-if-conv.c (predicate_mem_writes): Likewise.
6141 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
6142 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
6143 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
6144 * valtrack.c (dead_debug_insert_temp): Likewise.
6145 * varasm.c (mergeable_constant_section): Likewise.
6146 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
6148 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6149 Alan Hayward <alan.hayward@arm.com>
6150 David Sherwood <david.sherwood@arm.com>
6152 * expr.c (expand_assignment): Cope with polynomial mode sizes
6153 when assigning to a CONCAT.
6155 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6156 Alan Hayward <alan.hayward@arm.com>
6157 David Sherwood <david.sherwood@arm.com>
6159 * machmode.h (mode_precision): Change from unsigned short to
6161 (mode_to_precision): Return a poly_uint16 rather than an unsigned
6163 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
6164 or if measurement_type is not polynomial.
6165 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
6166 in which the mode is already known to be a scalar_int_mode.
6167 * genmodes.c (emit_mode_precision): Change the type of mode_precision
6168 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
6170 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6171 for GET_MODE_PRECISION.
6172 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6173 for GET_MODE_PRECISION.
6174 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
6176 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
6177 (expand_field_assignment, make_extraction): Likewise.
6178 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
6179 (get_last_value): Likewise.
6180 * convert.c (convert_to_integer_1): Likewise.
6181 * cse.c (cse_insn): Likewise.
6182 * expr.c (expand_expr_real_1): Likewise.
6183 * lra-constraints.c (simplify_operand_subreg): Likewise.
6184 * optabs-query.c (can_atomic_load_p): Likewise.
6185 * optabs.c (expand_atomic_load): Likewise.
6186 (expand_atomic_store): Likewise.
6187 * ree.c (combine_reaching_defs): Likewise.
6188 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
6189 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
6190 * tree.h (type_has_mode_precision_p): Likewise.
6191 * ubsan.c (instrument_si_overflow): Likewise.
6193 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6194 Alan Hayward <alan.hayward@arm.com>
6195 David Sherwood <david.sherwood@arm.com>
6197 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
6198 polynomial numbers of units.
6199 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
6200 (valid_vector_subparts_p): New function.
6201 (build_vector_type): Remove temporary shim and take the number
6202 of units as a poly_uint64 rather than an int.
6203 (build_opaque_vector_type): Take the number of units as a
6204 poly_uint64 rather than an int.
6205 * tree.c (build_vector_from_ctor): Handle polynomial
6206 TYPE_VECTOR_SUBPARTS.
6207 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
6208 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
6209 (build_vector_from_val): If the number of units is variable,
6210 use build_vec_duplicate_cst for constant operands and
6211 VEC_DUPLICATE_EXPR otherwise.
6212 (make_vector_type): Remove temporary is_constant ().
6213 (build_vector_type, build_opaque_vector_type): Take the number of
6214 units as a poly_uint64 rather than an int.
6215 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
6217 * cfgexpand.c (expand_debug_expr): Likewise.
6218 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
6219 (store_constructor, expand_expr_real_1): Likewise.
6220 (const_scalar_mask_from_tree): Likewise.
6221 * fold-const-call.c (fold_const_reduction): Likewise.
6222 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
6223 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
6224 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
6225 (fold_relational_const): Likewise.
6226 (native_interpret_vector): Likewise. Change the size from an
6227 int to an unsigned int.
6228 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
6229 TYPE_VECTOR_SUBPARTS.
6230 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
6231 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
6232 duplicating a non-constant operand into a variable-length vector.
6233 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
6234 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
6235 * ipa-icf.c (sem_variable::equals): Likewise.
6236 * match.pd: Likewise.
6237 * omp-simd-clone.c (simd_clone_subparts): Likewise.
6238 * print-tree.c (print_node): Likewise.
6239 * stor-layout.c (layout_type): Likewise.
6240 * targhooks.c (default_builtin_vectorization_cost): Likewise.
6241 * tree-cfg.c (verify_gimple_comparison): Likewise.
6242 (verify_gimple_assign_binary): Likewise.
6243 (verify_gimple_assign_ternary): Likewise.
6244 (verify_gimple_assign_single): Likewise.
6245 * tree-pretty-print.c (dump_generic_node): Likewise.
6246 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6247 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
6248 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
6249 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
6250 (vect_shift_permute_load_chain): Likewise.
6251 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
6252 (expand_vector_condition, optimize_vector_constructor): Likewise.
6253 (lower_vec_perm, get_compute_type): Likewise.
6254 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
6255 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
6256 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
6257 (vect_recog_mask_conversion_pattern): Likewise.
6258 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
6259 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
6260 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6261 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
6262 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
6263 (vectorizable_shift, vectorizable_operation, vectorizable_store)
6264 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
6265 (supportable_widening_operation): Likewise.
6266 (supportable_narrowing_operation): Likewise.
6267 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
6269 * varasm.c (output_constant): Likewise.
6271 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6272 Alan Hayward <alan.hayward@arm.com>
6273 David Sherwood <david.sherwood@arm.com>
6275 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
6276 so that both the length == 3 and length != 3 cases set up their
6277 own permute vectors. Add comments explaining why we know the
6278 number of elements is constant.
6279 (vect_permute_load_chain): Likewise.
6281 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6282 Alan Hayward <alan.hayward@arm.com>
6283 David Sherwood <david.sherwood@arm.com>
6285 * machmode.h (mode_nunits): Change from unsigned char to
6287 (ONLY_FIXED_SIZE_MODES): New macro.
6288 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
6289 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
6290 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
6292 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
6293 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
6294 or if measurement_type is not polynomial.
6295 * genmodes.c (ZERO_COEFFS): New macro.
6296 (emit_mode_nunits_inline): Make mode_nunits_inline return a
6298 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
6299 Use ZERO_COEFFS when emitting initializers.
6300 * data-streamer.h (bp_pack_poly_value): New function.
6301 (bp_unpack_poly_value): Likewise.
6302 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
6303 for GET_MODE_NUNITS.
6304 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
6305 for GET_MODE_NUNITS.
6306 * tree.c (make_vector_type): Remove temporary shim and make
6307 the real function take the number of units as a poly_uint64
6309 (build_vector_type_for_mode): Handle polynomial nunits.
6310 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
6311 * emit-rtl.c (const_vec_series_p_1): Likewise.
6312 (gen_rtx_CONST_VECTOR): Likewise.
6313 * fold-const.c (test_vec_duplicate_folding): Likewise.
6314 * genrecog.c (validate_pattern): Likewise.
6315 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
6316 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
6317 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
6318 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
6319 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
6320 * rtlanal.c (subreg_get_info): Likewise.
6321 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6322 (vect_grouped_load_supported): Likewise.
6323 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
6324 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
6325 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
6326 (simplify_const_unary_operation, simplify_binary_operation_1)
6327 (simplify_const_binary_operation, simplify_ternary_operation)
6328 (test_vector_ops_duplicate, test_vector_ops): Likewise.
6329 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
6330 instead of CONST_VECTOR_NUNITS.
6331 * varasm.c (output_constant_pool_2): Likewise.
6332 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
6333 explicit-encoded elements in the XVEC for variable-length vectors.
6335 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6337 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
6339 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6340 Alan Hayward <alan.hayward@arm.com>
6341 David Sherwood <david.sherwood@arm.com>
6343 * coretypes.h (fixed_size_mode): Declare.
6344 (fixed_size_mode_pod): New typedef.
6345 * builtins.h (target_builtins::x_apply_args_mode)
6346 (target_builtins::x_apply_result_mode): Change type to
6347 fixed_size_mode_pod.
6348 * builtins.c (apply_args_size, apply_result_size, result_vector)
6349 (expand_builtin_apply_args_1, expand_builtin_apply)
6350 (expand_builtin_return): Update accordingly.
6352 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6354 * cse.c (hash_rtx_cb): Hash only the encoded elements.
6355 * cselib.c (cselib_hash_rtx): Likewise.
6356 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
6357 CONST_VECTOR encoding.
6359 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6360 Jeff Law <law@redhat.com>
6363 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
6364 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
6365 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
6366 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
6369 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
6370 explicitly probe *sp in a noreturn function if there were any callee
6371 register saves or frame pointer is needed.
6373 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6376 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
6377 BLKmode for ternary, binary or unary expressions.
6380 * var-tracking.c (delete_vta_debug_insn): New inline function.
6381 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
6382 insns from get_insns () to NULL instead of each bb separately.
6383 Use delete_vta_debug_insn. No longer static.
6384 (vt_debug_insns_local, variable_tracking_main_1): Adjust
6385 delete_vta_debug_insns callers.
6386 * rtl.h (delete_vta_debug_insns): Declare.
6387 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
6388 instead of variable_tracking_main.
6390 2018-01-03 Martin Sebor <msebor@redhat.com>
6392 PR tree-optimization/83603
6393 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
6394 arguments past the endof the argument list in functions declared
6395 without a prototype.
6396 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
6397 Avoid checking when arguments are null.
6399 2018-01-03 Martin Sebor <msebor@redhat.com>
6402 * doc/extend.texi (attribute const): Fix a typo.
6403 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
6404 issuing -Wsuggest-attribute for void functions.
6406 2018-01-03 Martin Sebor <msebor@redhat.com>
6408 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
6409 offset_int::from instead of wide_int::to_shwi.
6410 (maybe_diag_overlap): Remove assertion.
6411 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
6412 * gimple-ssa-sprintf.c (format_directive): Same.
6413 (parse_directive): Same.
6414 (sprintf_dom_walker::compute_format_length): Same.
6415 (try_substitute_return_value): Same.
6417 2018-01-03 Jeff Law <law@redhat.com>
6420 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
6421 non-constant residual for zero at runtime and avoid probing in
6422 that case. Reorganize code for trailing problem to mirror handling
6425 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
6427 PR tree-optimization/83501
6428 * tree-ssa-strlen.c (get_string_cst): New.
6429 (handle_char_store): Call get_string_cst.
6431 2018-01-03 Martin Liska <mliska@suse.cz>
6433 PR tree-optimization/83593
6434 * tree-ssa-strlen.c: Include tree-cfg.h.
6435 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
6436 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
6437 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
6439 (strlen_dom_walker::before_dom_children): Call
6440 gimple_purge_dead_eh_edges. Dump tranformation with details
6442 (strlen_dom_walker::before_dom_children): Update call by adding
6443 new argument cleanup_eh.
6444 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
6446 2018-01-03 Martin Liska <mliska@suse.cz>
6449 * cif-code.def (VARIADIC_THUNK): New enum value.
6450 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
6453 2018-01-03 Jan Beulich <jbeulich@suse.com>
6455 * sse.md (mov<mode>_internal): Tighten condition for when to use
6456 vmovdqu<ssescalarsize> for TI and OI modes.
6458 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6460 Update copyright years.
6462 2018-01-03 Martin Liska <mliska@suse.cz>
6465 * ipa-visibility.c (function_and_variable_visibility): Skip
6466 functions with noipa attribure.
6468 2018-01-03 Jakub Jelinek <jakub@redhat.com>
6470 * gcc.c (process_command): Update copyright notice dates.
6471 * gcov-dump.c (print_version): Ditto.
6472 * gcov.c (print_version): Ditto.
6473 * gcov-tool.c (print_version): Ditto.
6474 * gengtype.c (create_file): Ditto.
6475 * doc/cpp.texi: Bump @copying's copyright year.
6476 * doc/cppinternals.texi: Ditto.
6477 * doc/gcc.texi: Ditto.
6478 * doc/gccint.texi: Ditto.
6479 * doc/gcov.texi: Ditto.
6480 * doc/install.texi: Ditto.
6481 * doc/invoke.texi: Ditto.
6483 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6485 * vector-builder.h (vector_builder::m_full_nelts): Change from
6486 unsigned int to poly_uint64.
6487 (vector_builder::full_nelts): Update prototype accordingly.
6488 (vector_builder::new_vector): Likewise.
6489 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
6490 (vector_builder::operator ==): Likewise.
6491 (vector_builder::finalize): Likewise.
6492 * int-vector-builder.h (int_vector_builder::int_vector_builder):
6493 Take the number of elements as a poly_uint64 rather than an
6495 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
6496 from unsigned int to poly_uint64.
6497 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
6498 (vec_perm_indices::new_vector): Likewise.
6499 (vec_perm_indices::length): Likewise.
6500 (vec_perm_indices::nelts_per_input): Likewise.
6501 (vec_perm_indices::input_nelts): Likewise.
6502 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
6503 number of elements per input as a poly_uint64 rather than an
6504 unsigned int. Use the original encoding for variable-length
6505 vectors, rather than clamping each individual element.
6506 For the second and subsequent elements in each pattern,
6507 clamp the step and base before clamping their sum.
6508 (vec_perm_indices::series_p): Handle polynomial element counts.
6509 (vec_perm_indices::all_in_range_p): Likewise.
6510 (vec_perm_indices_to_tree): Likewise.
6511 (vec_perm_indices_to_rtx): Likewise.
6512 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
6513 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
6514 (tree_vector_builder::new_binary_operation): Handle polynomial
6515 element counts. Return false if we need to know the number
6516 of elements at compile time.
6517 * fold-const.c (fold_vec_perm): Punt if the number of elements
6518 isn't known at compile time.
6520 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6522 * vec-perm-indices.h (vec_perm_builder): Change element type
6523 from HOST_WIDE_INT to poly_int64.
6524 (vec_perm_indices::element_type): Update accordingly.
6525 (vec_perm_indices::clamp): Handle polynomial element_types.
6526 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
6527 (vec_perm_indices::all_in_range_p): Likewise.
6528 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
6530 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
6531 polynomial vec_perm_indices element types.
6532 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
6533 * fold-const.c (fold_vec_perm): Likewise.
6534 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
6535 * tree-vect-generic.c (lower_vec_perm): Likewise.
6536 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
6537 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
6538 element type to HOST_WIDE_INT.
6540 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6541 Alan Hayward <alan.hayward@arm.com>
6542 David Sherwood <david.sherwood@arm.com>
6544 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
6545 rather than an int. Use plus_constant.
6546 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
6547 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
6549 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6550 Alan Hayward <alan.hayward@arm.com>
6551 David Sherwood <david.sherwood@arm.com>
6553 * calls.c (emit_call_1, expand_call): Change struct_value_size from
6554 a HOST_WIDE_INT to a poly_int64.
6556 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6557 Alan Hayward <alan.hayward@arm.com>
6558 David Sherwood <david.sherwood@arm.com>
6560 * calls.c (load_register_parameters): Cope with polynomial
6561 mode sizes. Require a constant size for BLKmode parameters
6562 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
6563 forces a parameter to be padded at the lsb end in order to
6564 fill a complete number of words, require the parameter size
6565 to be ordered wrt UNITS_PER_WORD.
6567 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6568 Alan Hayward <alan.hayward@arm.com>
6569 David Sherwood <david.sherwood@arm.com>
6571 * reload1.c (spill_stack_slot_width): Change element type
6572 from unsigned int to poly_uint64_pod.
6573 (alter_reg): Treat mode sizes as polynomial.
6575 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6576 Alan Hayward <alan.hayward@arm.com>
6577 David Sherwood <david.sherwood@arm.com>
6579 * reload.c (complex_word_subreg_p): New function.
6580 (reload_inner_reg_of_subreg, push_reload): Use it.
6582 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6583 Alan Hayward <alan.hayward@arm.com>
6584 David Sherwood <david.sherwood@arm.com>
6586 * lra-constraints.c (process_alt_operands): Reject matched
6587 operands whose sizes aren't ordered.
6588 (match_reload): Refer to this check here.
6590 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6591 Alan Hayward <alan.hayward@arm.com>
6592 David Sherwood <david.sherwood@arm.com>
6594 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
6595 that the mode size is in the set {1, 2, 4, 8, 16}.
6597 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6598 Alan Hayward <alan.hayward@arm.com>
6599 David Sherwood <david.sherwood@arm.com>
6601 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
6602 Use plus_constant instead of gen_rtx_PLUS.
6604 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6605 Alan Hayward <alan.hayward@arm.com>
6606 David Sherwood <david.sherwood@arm.com>
6608 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
6609 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
6610 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
6611 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
6612 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
6613 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
6614 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
6615 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
6616 * config/i386/i386.c (ix86_push_rounding): ...this new function.
6617 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
6619 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
6620 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
6621 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
6622 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
6623 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
6624 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
6625 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
6626 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
6627 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
6628 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
6630 * expr.c (emit_move_resolve_push): Treat the input and result
6631 of PUSH_ROUNDING as a poly_int64.
6632 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
6633 (emit_push_insn): Likewise.
6634 * lra-eliminations.c (mark_not_eliminable): Likewise.
6635 * recog.c (push_operand): Likewise.
6636 * reload1.c (elimination_effects): Likewise.
6637 * rtlanal.c (nonzero_bits1): Likewise.
6638 * calls.c (store_one_arg): Likewise. Require the padding to be
6639 known at compile time.
6641 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6642 Alan Hayward <alan.hayward@arm.com>
6643 David Sherwood <david.sherwood@arm.com>
6645 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
6646 Use plus_constant instead of gen_rtx_PLUS.
6648 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6649 Alan Hayward <alan.hayward@arm.com>
6650 David Sherwood <david.sherwood@arm.com>
6652 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
6655 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6656 Alan Hayward <alan.hayward@arm.com>
6657 David Sherwood <david.sherwood@arm.com>
6659 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
6660 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
6661 via stack temporaries. Treat the mode size as polynomial too.
6663 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6664 Alan Hayward <alan.hayward@arm.com>
6665 David Sherwood <david.sherwood@arm.com>
6667 * expr.c (expand_expr_real_2): When handling conversions involving
6668 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
6669 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
6670 as a poly_uint64 too.
6672 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6673 Alan Hayward <alan.hayward@arm.com>
6674 David Sherwood <david.sherwood@arm.com>
6676 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
6678 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6679 Alan Hayward <alan.hayward@arm.com>
6680 David Sherwood <david.sherwood@arm.com>
6682 * combine.c (can_change_dest_mode): Handle polynomial
6683 REGMODE_NATURAL_SIZE.
6684 * expmed.c (store_bit_field_1): Likewise.
6685 * expr.c (store_constructor): Likewise.
6686 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
6687 and polynomial REGMODE_NATURAL_SIZE.
6688 (gen_lowpart_common): Likewise.
6689 * reginfo.c (record_subregs_of_mode): Likewise.
6690 * rtlanal.c (read_modify_subreg_p): Likewise.
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 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
6697 numbers of elements.
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 * match.pd: Cope with polynomial numbers of vector elements.
6705 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6706 Alan Hayward <alan.hayward@arm.com>
6707 David Sherwood <david.sherwood@arm.com>
6709 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
6710 in a POINTER_PLUS_EXPR.
6712 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6713 Alan Hayward <alan.hayward@arm.com>
6714 David Sherwood <david.sherwood@arm.com>
6716 * omp-simd-clone.c (simd_clone_subparts): New function.
6717 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
6718 (ipa_simd_modify_function_body): Likewise.
6720 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6721 Alan Hayward <alan.hayward@arm.com>
6722 David Sherwood <david.sherwood@arm.com>
6724 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
6725 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
6726 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
6727 (expand_vector_condition, vector_element): Likewise.
6728 (subparts_gt): New function.
6729 (get_compute_type): Use subparts_gt.
6730 (count_type_subparts): Delete.
6731 (expand_vector_operations_1): Use subparts_gt instead of
6732 count_type_subparts.
6734 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6735 Alan Hayward <alan.hayward@arm.com>
6736 David Sherwood <david.sherwood@arm.com>
6738 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
6739 (vect_compile_time_alias): ...this new function. Do the calculation
6740 on poly_ints rather than trees.
6741 (vect_prune_runtime_alias_test_list): Update call accordingly.
6743 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6744 Alan Hayward <alan.hayward@arm.com>
6745 David Sherwood <david.sherwood@arm.com>
6747 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
6749 (vect_schedule_slp_instance): Likewise.
6751 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6752 Alan Hayward <alan.hayward@arm.com>
6753 David Sherwood <david.sherwood@arm.com>
6755 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
6756 constant and extern definitions for variable-length vectors.
6757 (vect_get_constant_vectors): Note that the number of units
6758 is known to be constant.
6760 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6761 Alan Hayward <alan.hayward@arm.com>
6762 David Sherwood <david.sherwood@arm.com>
6764 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
6765 of units as polynomial. Choose between WIDE and NARROW based
6768 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6769 Alan Hayward <alan.hayward@arm.com>
6770 David Sherwood <david.sherwood@arm.com>
6772 * tree-vect-stmts.c (simd_clone_subparts): New function.
6773 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
6775 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6776 Alan Hayward <alan.hayward@arm.com>
6777 David Sherwood <david.sherwood@arm.com>
6779 * tree-vect-stmts.c (vectorizable_call): Treat the number of
6780 vectors as polynomial. Use build_index_vector for
6783 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6784 Alan Hayward <alan.hayward@arm.com>
6785 David Sherwood <david.sherwood@arm.com>
6787 * tree-vect-stmts.c (get_load_store_type): Treat the number of
6788 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
6789 for variable-length vectors.
6790 (vectorizable_mask_load_store): Treat the number of units as
6791 polynomial, asserting that it is constant if the condition has
6792 already been enforced.
6793 (vectorizable_store, vectorizable_load): Likewise.
6795 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6796 Alan Hayward <alan.hayward@arm.com>
6797 David Sherwood <david.sherwood@arm.com>
6799 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
6800 of units as polynomial. Punt if we can't tell at compile time
6801 which vector contains the final result.
6803 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6804 Alan Hayward <alan.hayward@arm.com>
6805 David Sherwood <david.sherwood@arm.com>
6807 * tree-vect-loop.c (vectorizable_induction): Treat the number
6808 of units as polynomial. Punt on SLP inductions. Use an integer
6809 VEC_SERIES_EXPR for variable-length integer reductions. Use a
6810 cast of such a series for variable-length floating-point
6813 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6814 Alan Hayward <alan.hayward@arm.com>
6815 David Sherwood <david.sherwood@arm.com>
6817 * tree.h (build_index_vector): Declare.
6818 * tree.c (build_index_vector): New function.
6819 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
6820 of units as polynomial, forcibly converting it to a constant if
6821 vectorizable_reduction has already enforced the condition.
6822 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
6823 to create a {1,2,3,...} vector.
6824 (vectorizable_reduction): Treat the number of units as polynomial.
6825 Choose vectype_in based on the largest scalar element size rather
6826 than the smallest number of units. Enforce the restrictions
6829 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6830 Alan Hayward <alan.hayward@arm.com>
6831 David Sherwood <david.sherwood@arm.com>
6833 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
6834 number of units as polynomial.
6836 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6837 Alan Hayward <alan.hayward@arm.com>
6838 David Sherwood <david.sherwood@arm.com>
6840 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
6841 * target.def (autovectorize_vector_sizes): Return the vector sizes
6842 by pointer, using vector_sizes rather than a bitmask.
6843 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
6844 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
6845 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
6847 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
6848 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
6849 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
6850 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
6851 * omp-general.c (omp_max_vf): Likewise.
6852 * omp-low.c (omp_clause_aligned_alignment): Likewise.
6853 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
6854 * tree-vect-loop.c (vect_analyze_loop): Likewise.
6855 * tree-vect-slp.c (vect_slp_bb): Likewise.
6856 * doc/tm.texi: Regenerate.
6857 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
6859 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
6860 the vector size as a poly_uint64 rather than an unsigned int.
6861 (current_vector_size): Change from an unsigned int to a poly_uint64.
6862 (get_vectype_for_scalar_type): Update accordingly.
6863 * tree.h (build_truth_vector_type): Take the size and number of
6864 units as a poly_uint64 rather than an unsigned int.
6865 (build_vector_type): Add a temporary overload that takes
6866 the number of units as a poly_uint64 rather than an unsigned int.
6867 * tree.c (make_vector_type): Likewise.
6868 (build_truth_vector_type): Take the number of units as a poly_uint64
6869 rather than an unsigned int.
6871 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6872 Alan Hayward <alan.hayward@arm.com>
6873 David Sherwood <david.sherwood@arm.com>
6875 * target.def (get_mask_mode): Take the number of units and length
6876 as poly_uint64s rather than unsigned ints.
6877 * targhooks.h (default_get_mask_mode): Update accordingly.
6878 * targhooks.c (default_get_mask_mode): Likewise.
6879 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
6880 * doc/tm.texi: Regenerate.
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 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
6887 * omp-general.c (omp_max_vf): Likewise.
6888 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
6889 (expand_omp_simd): Handle polynomial safelen.
6890 * omp-low.c (omplow_simd_context): Add a default constructor.
6891 (omplow_simd_context::max_vf): Change from int to poly_uint64.
6892 (lower_rec_simd_input_clauses): Update accordingly.
6893 (lower_rec_input_clauses): Likewise.
6895 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6896 Alan Hayward <alan.hayward@arm.com>
6897 David Sherwood <david.sherwood@arm.com>
6899 * tree-vectorizer.h (vect_nunits_for_cost): New function.
6900 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
6901 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
6902 (vect_analyze_slp_cost): Likewise.
6903 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
6904 (vect_model_load_cost): Likewise.
6906 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6907 Alan Hayward <alan.hayward@arm.com>
6908 David Sherwood <david.sherwood@arm.com>
6910 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
6911 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
6912 from an unsigned int * to a poly_uint64_pod *.
6913 (calculate_unrolling_factor): New function.
6914 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
6916 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6917 Alan Hayward <alan.hayward@arm.com>
6918 David Sherwood <david.sherwood@arm.com>
6920 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
6921 from an unsigned int to a poly_uint64.
6922 (_loop_vec_info::slp_unrolling_factor): Likewise.
6923 (_loop_vec_info::vectorization_factor): Change from an int
6925 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
6926 (vect_get_num_vectors): New function.
6927 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
6928 (vect_get_num_copies): Use vect_get_num_vectors.
6929 (vect_analyze_data_ref_dependences): Change max_vf from an int *
6930 to an unsigned int *.
6931 (vect_analyze_data_refs): Change min_vf from an int * to a
6933 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6934 than an unsigned HOST_WIDE_INT.
6935 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
6936 (vect_analyze_data_ref_dependence): Change max_vf from an int *
6937 to an unsigned int *.
6938 (vect_analyze_data_ref_dependences): Likewise.
6939 (vect_compute_data_ref_alignment): Handle polynomial vf.
6940 (vect_enhance_data_refs_alignment): Likewise.
6941 (vect_prune_runtime_alias_test_list): Likewise.
6942 (vect_shift_permute_load_chain): Likewise.
6943 (vect_supportable_dr_alignment): Likewise.
6944 (dependence_distance_ge_vf): Take the vectorization factor as a
6945 poly_uint64 rather than an unsigned HOST_WIDE_INT.
6946 (vect_analyze_data_refs): Change min_vf from an int * to a
6948 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
6949 vfm1 as a poly_uint64 rather than an int. Make the same change
6950 for the returned bound_scalar.
6951 (vect_gen_vector_loop_niters): Handle polynomial vf.
6952 (vect_do_peeling): Likewise. Update call to
6953 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
6954 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
6956 * tree-vect-loop.c (vect_determine_vectorization_factor)
6957 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
6958 (vect_get_known_peeling_cost): Likewise.
6959 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
6960 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
6961 (vect_transform_loop): Likewise. Use the lowest possible VF when
6962 updating the upper bounds of the loop.
6963 (vect_min_worthwhile_factor): Make static. Return an unsigned int
6965 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
6966 polynomial unroll factors.
6967 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
6968 (vect_make_slp_decision): Likewise.
6969 (vect_supported_load_permutation_p): Likewise, and polynomial
6971 (vect_analyze_slp_cost): Handle polynomial vf.
6972 (vect_slp_analyze_node_operations): Likewise.
6973 (vect_slp_analyze_bb_1): Likewise.
6974 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
6975 than an unsigned HOST_WIDE_INT.
6976 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
6977 (vectorizable_load): Handle polynomial vf.
6978 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
6980 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
6982 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6983 Alan Hayward <alan.hayward@arm.com>
6984 David Sherwood <david.sherwood@arm.com>
6986 * match.pd: Handle bit operations involving three constants
6987 and try to fold one pair.
6989 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
6991 * tree-vect-loop-manip.c: Include gimple-fold.h.
6992 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
6993 niters_maybe_zero parameters. Handle other cases besides a step of 1.
6994 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
6995 Add a path that uses a step of VF instead of 1, but disable it
6997 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
6998 and niters_no_overflow parameters. Update calls to
6999 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
7000 Create a new SSA name if the latter choses to use a ste other
7001 than zero, and return it via niters_vector_mult_vf_var.
7002 * tree-vect-loop.c (vect_transform_loop): Update calls to
7003 vect_do_peeling, vect_gen_vector_loop_niters and
7004 slpeel_make_loop_iterate_ntimes.
7005 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
7006 (vect_gen_vector_loop_niters): Update declarations after above changes.
7008 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
7010 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
7011 128-bit round to integer instructions.
7012 (ceil<mode>2): Likewise.
7013 (btrunc<mode>2): Likewise.
7014 (round<mode>2): Likewise.
7016 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7018 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
7019 unaligned VSX load/store on P8/P9.
7020 (expand_block_clear): Allow the use of unaligned VSX
7021 load/store on P8/P9.
7023 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
7025 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
7027 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
7028 swap associated with both a load and a store.
7030 2018-01-02 Andrew Waterman <andrew@sifive.com>
7032 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
7033 * config/riscv/riscv.md (clear_cache): Use it.
7035 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
7037 * web.c: Remove out-of-date comment.
7039 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7041 * expr.c (fixup_args_size_notes): Check that any existing
7042 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
7043 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
7044 (emit_single_push_insn): ...here.
7046 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7048 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
7049 (const_vector_encoded_nelts): New function.
7050 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
7051 (const_vector_int_elt, const_vector_elt): Declare.
7052 * emit-rtl.c (const_vector_int_elt_1): New function.
7053 (const_vector_elt): Likewise.
7054 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
7055 of CONST_VECTOR_ELT.
7057 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7059 * expr.c: Include rtx-vector-builder.h.
7060 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
7061 directly on the tree encoding.
7062 (const_vector_from_tree): Likewise.
7063 * optabs.c: Include rtx-vector-builder.h.
7064 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
7065 sequence of "u" values.
7066 * vec-perm-indices.c: Include rtx-vector-builder.h.
7067 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
7068 directly on the vec_perm_indices encoding.
7070 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7072 * doc/rtl.texi (const_vector): Describe new encoding scheme.
7073 * Makefile.in (OBJS): Add rtx-vector-builder.o.
7074 * rtx-vector-builder.h: New file.
7075 * rtx-vector-builder.c: Likewise.
7076 * rtl.h (rtx_def::u2): Add a const_vector field.
7077 (CONST_VECTOR_NPATTERNS): New macro.
7078 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
7079 (CONST_VECTOR_DUPLICATE_P): Likewise.
7080 (CONST_VECTOR_STEPPED_P): Likewise.
7081 (CONST_VECTOR_ENCODED_ELT): Likewise.
7082 (const_vec_duplicate_p): Check for a duplicated vector encoding.
7083 (unwrap_const_vec_duplicate): Likewise.
7084 (const_vec_series_p): Check for a non-duplicated vector encoding.
7085 Say that the function only returns true for integer vectors.
7086 * emit-rtl.c: Include rtx-vector-builder.h.
7087 (gen_const_vec_duplicate_1): Delete.
7088 (gen_const_vector): Call gen_const_vec_duplicate instead of
7089 gen_const_vec_duplicate_1.
7090 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
7091 (gen_const_vec_duplicate): Use rtx_vector_builder.
7092 (gen_const_vec_series): Likewise.
7093 (gen_rtx_CONST_VECTOR): Likewise.
7094 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
7095 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7096 Build a new vector rather than modifying a CONST_VECTOR in-place.
7097 (handle_special_swappables): Update call accordingly.
7098 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
7099 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
7100 Build a new vector rather than modifying a CONST_VECTOR in-place.
7101 (handle_special_swappables): Update call accordingly.
7103 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7105 * simplify-rtx.c (simplify_const_binary_operation): Use
7106 CONST_VECTOR_ELT instead of XVECEXP.
7108 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7110 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
7111 the selector elements to be different from the data elements
7112 if the selector is a VECTOR_CST.
7113 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
7114 ssizetype for the selector.
7116 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7118 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
7119 before testing each element individually.
7120 * tree-vect-generic.c (lower_vec_perm): Likewise.
7122 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7124 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
7125 * selftest-run-tests.c (selftest::run_tests): Call it.
7126 * vector-builder.h (vector_builder::operator ==): New function.
7127 (vector_builder::operator !=): Likewise.
7128 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
7129 (vec_perm_indices::all_from_input_p): New function.
7130 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
7131 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
7132 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
7133 instead of reading the VECTOR_CST directly. Detect whether both
7134 vector inputs are the same before constructing the vec_perm_indices,
7135 and update the number of inputs argument accordingly. Use the
7136 utility functions added above. Only construct sel2 if we need to.
7138 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7140 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
7141 the broadcast of the low byte.
7142 (expand_mult_highpart): Use an explicit encoding for the permutes.
7143 * optabs-query.c (can_mult_highpart_p): Likewise.
7144 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
7145 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7146 (vectorizable_bswap): Likewise.
7147 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
7148 explicit encoding for the power-of-2 permutes.
7149 (vect_permute_store_chain): Likewise.
7150 (vect_grouped_load_supported): Likewise.
7151 (vect_permute_load_chain): Likewise.
7153 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7155 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
7156 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
7157 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
7158 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
7159 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
7160 (vect_gen_perm_mask_any): Likewise.
7162 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7164 * int-vector-builder.h: New file.
7165 * vec-perm-indices.h: Include int-vector-builder.h.
7166 (vec_perm_indices): Redefine as an int_vector_builder.
7167 (auto_vec_perm_indices): Delete.
7168 (vec_perm_builder): Redefine as a stand-alone class.
7169 (vec_perm_indices::vec_perm_indices): New function.
7170 (vec_perm_indices::clamp): Likewise.
7171 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
7172 (vec_perm_indices::new_vector): New function.
7173 (vec_perm_indices::new_expanded_vector): Update for new
7174 vec_perm_indices class.
7175 (vec_perm_indices::rotate_inputs): New function.
7176 (vec_perm_indices::all_in_range_p): Operate directly on the
7177 encoded form, without computing elided elements.
7178 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
7179 encoding. Update for new vec_perm_indices class.
7180 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
7181 the given vec_perm_builder.
7182 (expand_vec_perm_var): Update vec_perm_builder constructor.
7183 (expand_mult_highpart): Use vec_perm_builder instead of
7184 auto_vec_perm_indices.
7185 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
7186 vec_perm_indices instead of auto_vec_perm_indices. Use a single
7187 or double series encoding as appropriate.
7188 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
7189 vec_perm_indices instead of auto_vec_perm_indices.
7190 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7191 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7192 (vect_permute_store_chain): Likewise.
7193 (vect_grouped_load_supported): Likewise.
7194 (vect_permute_load_chain): Likewise.
7195 (vect_shift_permute_load_chain): Likewise.
7196 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7197 (vect_transform_slp_perm_load): Likewise.
7198 (vect_schedule_slp_instance): Likewise.
7199 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7200 (vectorizable_mask_load_store): Likewise.
7201 (vectorizable_bswap): Likewise.
7202 (vectorizable_store): Likewise.
7203 (vectorizable_load): Likewise.
7204 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
7205 vec_perm_indices instead of auto_vec_perm_indices. Use
7206 tree_to_vec_perm_builder to read the vector from a tree.
7207 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
7208 vec_perm_builder instead of a vec_perm_indices.
7209 (have_whole_vector_shift): Use vec_perm_builder and
7210 vec_perm_indices instead of auto_vec_perm_indices. Leave the
7211 truncation to calc_vec_perm_mask_for_shift.
7212 (vect_create_epilog_for_reduction): Likewise.
7213 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
7214 from auto_vec_perm_indices to vec_perm_indices.
7215 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7216 instead of changing individual elements.
7217 (aarch64_vectorize_vec_perm_const): Use new_vector to install
7218 the vector in d.perm.
7219 * config/arm/arm.c (expand_vec_perm_d::perm): Change
7220 from auto_vec_perm_indices to vec_perm_indices.
7221 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
7222 instead of changing individual elements.
7223 (arm_vectorize_vec_perm_const): Use new_vector to install
7224 the vector in d.perm.
7225 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
7226 Update vec_perm_builder constructor.
7227 (rs6000_expand_interleave): Likewise.
7228 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
7229 (rs6000_expand_interleave): Likewise.
7231 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7233 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
7234 to qimode could truncate the indices.
7235 * optabs.c (expand_vec_perm_var): Likewise.
7237 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7239 * Makefile.in (OBJS): Add vec-perm-indices.o.
7240 * vec-perm-indices.h: New file.
7241 * vec-perm-indices.c: Likewise.
7242 * target.h (vec_perm_indices): Replace with a forward class
7244 (auto_vec_perm_indices): Move to vec-perm-indices.h.
7245 * optabs.h: Include vec-perm-indices.h.
7246 (expand_vec_perm): Delete.
7247 (selector_fits_mode_p, expand_vec_perm_var): Declare.
7248 (expand_vec_perm_const): Declare.
7249 * target.def (vec_perm_const_ok): Replace with...
7250 (vec_perm_const): ...this new hook.
7251 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
7252 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
7253 * doc/tm.texi: Regenerate.
7254 * optabs.def (vec_perm_const): Delete.
7255 * doc/md.texi (vec_perm_const): Likewise.
7256 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
7257 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
7258 expand_vec_perm for constant permutation vectors. Assert that
7259 the mode of variable permutation vectors is the integer equivalent
7260 of the mode that is being permuted.
7261 * optabs-query.h (selector_fits_mode_p): Declare.
7262 * optabs-query.c: Include vec-perm-indices.h.
7263 (selector_fits_mode_p): New function.
7264 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
7265 is defined, instead of checking whether the vec_perm_const_optab
7266 exists. Use targetm.vectorize.vec_perm_const instead of
7267 targetm.vectorize.vec_perm_const_ok. Check whether the indices
7268 fit in the vector mode before using a variable permute.
7269 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
7270 vec_perm_indices instead of an rtx.
7271 (expand_vec_perm): Replace with...
7272 (expand_vec_perm_const): ...this new function. Take the selector
7273 as a vec_perm_indices rather than an rtx. Also take the mode of
7274 the selector. Update call to shift_amt_for_vec_perm_mask.
7275 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
7276 Use vec_perm_indices::new_expanded_vector to expand the original
7277 selector into bytes. Check whether the indices fit in the vector
7278 mode before using a variable permute.
7279 (expand_vec_perm_var): Make global.
7280 (expand_mult_highpart): Use expand_vec_perm_const.
7281 * fold-const.c: Includes vec-perm-indices.h.
7282 * tree-ssa-forwprop.c: Likewise.
7283 * tree-vect-data-refs.c: Likewise.
7284 * tree-vect-generic.c: Likewise.
7285 * tree-vect-loop.c: Likewise.
7286 * tree-vect-slp.c: Likewise.
7287 * tree-vect-stmts.c: Likewise.
7288 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
7290 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
7291 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
7292 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
7293 (aarch64_vectorize_vec_perm_const): ...this new function.
7294 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7295 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7296 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
7297 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
7298 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7299 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7300 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
7302 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
7303 check for NEON modes.
7304 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
7305 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
7306 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
7307 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
7309 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
7310 the old VEC_PERM_CONST conditions.
7311 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
7312 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
7313 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
7314 (ia64_vectorize_vec_perm_const_ok): Merge into...
7315 (ia64_vectorize_vec_perm_const): ...this new function.
7316 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
7317 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
7318 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
7319 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
7320 * config/mips/mips.c (mips_expand_vec_perm_const)
7321 (mips_vectorize_vec_perm_const_ok): Merge into...
7322 (mips_vectorize_vec_perm_const): ...this new function.
7323 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
7324 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
7325 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
7326 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
7327 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
7328 (rs6000_expand_vec_perm_const): Delete.
7329 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
7331 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7332 (altivec_expand_vec_perm_const_le): Take each operand individually.
7333 Operate on constant selectors rather than rtxes.
7334 (altivec_expand_vec_perm_const): Likewise. Update call to
7335 altivec_expand_vec_perm_const_le.
7336 (rs6000_expand_vec_perm_const): Delete.
7337 (rs6000_vectorize_vec_perm_const_ok): Delete.
7338 (rs6000_vectorize_vec_perm_const): New function.
7339 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7340 an element count and rtx array.
7341 (rs6000_expand_extract_even): Update call accordingly.
7342 (rs6000_expand_interleave): Likewise.
7343 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
7344 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
7345 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
7346 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
7347 (rs6000_expand_vec_perm_const): Delete.
7348 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
7349 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7350 (altivec_expand_vec_perm_const_le): Take each operand individually.
7351 Operate on constant selectors rather than rtxes.
7352 (altivec_expand_vec_perm_const): Likewise. Update call to
7353 altivec_expand_vec_perm_const_le.
7354 (rs6000_expand_vec_perm_const): Delete.
7355 (rs6000_vectorize_vec_perm_const_ok): Delete.
7356 (rs6000_vectorize_vec_perm_const): New function. Remove stray
7357 reference to the SPE evmerge intructions.
7358 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
7359 an element count and rtx array.
7360 (rs6000_expand_extract_even): Update call accordingly.
7361 (rs6000_expand_interleave): Likewise.
7362 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
7363 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
7365 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
7367 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7369 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
7370 vector mode and that that mode matches the mode of the data
7372 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
7373 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
7374 directly using expand_vec_perm_1 when forcing selectors into
7376 (expand_vec_perm_var): New function, split out from expand_vec_perm.
7378 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7380 * optabs-query.h (can_vec_perm_p): Delete.
7381 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
7382 * optabs-query.c (can_vec_perm_p): Split into...
7383 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
7384 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
7385 particular selector is valid.
7386 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
7387 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
7388 (vect_grouped_load_supported): Likewise.
7389 (vect_shift_permute_load_chain): Likewise.
7390 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
7391 (vect_transform_slp_perm_load): Likewise.
7392 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
7393 (vectorizable_bswap): Likewise.
7394 (vect_gen_perm_mask_checked): Likewise.
7395 * fold-const.c (fold_ternary_loc): Likewise. Don't take
7396 implementations of variable permutation vectors into account
7397 when deciding which selector to use.
7398 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
7399 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
7400 with a false third argument.
7401 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
7402 to test whether the constant selector is valid and can_vec_perm_var_p
7403 to test whether a variable selector is valid.
7405 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7407 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
7408 * optabs-query.c (can_vec_perm_p): Likewise.
7409 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
7410 instead of vec_perm_indices.
7411 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
7412 (vect_gen_perm_mask_checked): Likewise,
7413 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
7414 (vect_gen_perm_mask_checked): Likewise,
7416 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
7418 * optabs-query.h (qimode_for_vec_perm): Declare.
7419 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
7420 (qimode_for_vec_perm): ...this new function.
7421 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
7423 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
7425 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
7426 does not have a conditional at the top.
7428 2018-01-02 Richard Biener <rguenther@suse.de>
7430 * ipa-inline.c (big_speedup_p): Fix expression.
7432 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7435 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
7438 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
7442 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
7443 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
7444 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
7445 cond_taken_branch_cost 3->4.
7447 2018-01-01 Jakub Jelinek <jakub@redhat.com>
7449 PR tree-optimization/83581
7450 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
7451 TODO_cleanup_cfg if any changes have been made.
7454 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
7455 convert_modes if target mode has the right side, but different mode
7459 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
7460 last argument when extracting from CONCAT. If either from_real or
7461 from_imag is NULL, use expansion through memory. If result is not
7462 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
7463 the parts directly to inner mode, if even that fails, use expansion
7467 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
7468 check for bswap in mode rather than HImode and use that in expand_unop
7471 Copyright (C) 2018 Free Software Foundation, Inc.
7473 Copying and distribution of this file, with or without modification,
7474 are permitted in any medium without royalty provided the copyright
7475 notice and this notice are preserved.