1 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
3 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
5 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
6 Use gen_rtx_REG rather than gen_lowpart.
8 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
10 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
11 rather than 0 when creating partial subregs.
13 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
15 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
18 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
21 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
22 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
23 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
24 flags. This restores the settings used before the 2017-07-24.
25 Turning off pre increment/decrement/modify allows IVOPTS to
26 optimize DF/SF loops where the index is an int.
28 2018-01-29 Richard Biener <rguenther@suse.de>
29 Kelvin Nilsen <kelvin@gcc.gnu.org>
32 * tree-vect-stmts.c (vectorizable_call): Don't call
33 targetm.vectorize_builtin_md_vectorized_function if callee is
36 2018-01-22 Carl Love <cel@us.ibm.com>
38 * doc/extend.tex: Fix typo in second arg in
39 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
41 2018-01-29 Richard Biener <rguenther@suse.de>
43 PR tree-optimization/84086
44 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
45 (flush_ssaname_freelist): When SSA names were released reset
48 2018-01-29 Richard Biener <rguenther@suse.de>
50 PR tree-optimization/84057
51 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
52 removed paths when removing edges.
54 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
56 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
57 -mfunction-return=@var{choice}.
59 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
62 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
64 (layout::print_source_line): Likewise.
65 (test_get_line_width_without_trailing_whitespace): Add test cases.
67 2018-01-27 Jakub Jelinek <jakub@redhat.com>
70 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
73 2018-01-26 Jim Wilson <jimw@sifive.com>
75 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
77 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
80 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
82 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
83 and CMP + SUB-immediate -> SUBS.
85 2018-01-26 Martin Sebor <msebor@redhat.com>
87 PR tree-optimization/83896
88 * tree-ssa-strlen.c (get_string_len): Rename...
89 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
90 Avoid assuming length is constant.
91 (handle_char_store): Use HOST_WIDE_INT for string length.
93 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
96 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
97 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
99 2018-01-26 Richard Biener <rguenther@suse.de>
101 PR rtl-optimization/84003
102 * dse.c (record_store): Only record redundant stores when
103 the earlier store aliases at least all accesses the later one does.
105 2018-01-26 Jakub Jelinek <jakub@redhat.com>
107 PR rtl-optimization/83985
108 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
109 REG_CFA_RESTORE insns.
110 (delete_unmarked_insns): Don't ignore separate shrink wrapping
111 REG_CFA_RESTORE insns here.
114 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
115 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
117 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
119 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
120 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
121 (arc_init): Likewise.
122 (arc_override_options): Likewise.
123 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
125 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
127 * config/arc/arc.h (TARGET_DBNZ): Define.
128 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
129 properly set the tune attribute.
130 (dbnz): Use TARGET_DBNZ guard.
131 * config/arc/arc.opt (mtune): Add core3 option.
133 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
135 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
136 recognize new pic like addresses.
137 (arc_delegitimize_address): Clean up.
139 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
141 * config/arc/arc-arches.def: Option mrf16 valid for all
143 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
144 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
145 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
146 * config/arc/arc-tables.opt: Regenerate.
147 * config/arc/arc.c (arc_conditional_register_usage): Handle
148 reduced register file case.
149 (arc_file_start): Set must have build attributes.
150 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
152 * config/arc/arc.opt (mrf16): Add new option.
153 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
154 * config/arc/genmultilib.awk: Handle new mrf16 option.
155 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
156 * config/arc/t-multilib: Regenerate.
157 * doc/invoke.texi (ARC Options): Document mrf16 option.
159 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
161 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
162 * config/arc/arc.c (arc_handle_secure_attribute): New function.
163 (arc_attribute_table): Add 'secure_call' attribute.
164 (arc_print_operand): Print secure call operand.
165 (arc_function_ok_for_sibcall): Don't optimize tail calls when
167 (arc_is_secure_call_p): New function. * config/arc/arc.md
168 (call_i): Add support for sjli instruction.
169 (call_value_i): Likewise.
170 * config/arc/constraints.md (Csc): New constraint.
172 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
173 John Eric Martin <John.Martin@emmicro-us.com>
175 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
176 * config/arc/arc.c (_arc_jli_section): New struct.
177 (arc_jli_section): New type.
178 (rc_jli_sections): New static variable.
179 (arc_handle_jli_attribute): New function.
180 (arc_attribute_table): Add jli_always and jli_fixed attribute.
181 (arc_file_end): New function.
182 (TARGET_ASM_FILE_END): Define.
183 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
184 (arc_add_jli_section): New function.
185 (jli_call_scan): Likewise.
186 (arc_reorg): Call jli_call_scan.
187 (arc_output_addsi): Remove 'S' from printing asm operand.
188 (arc_is_jli_call_p): New function.
189 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
191 (movhi_insn): Likewise.
192 (movsi_insn): Likewise.
193 (movsi_set_cc_insn): Likewise.
194 (loadqi_update): Likewise.
195 (load_zeroextendqisi_update): Likewise.
196 (load_signextendqisi_update): Likewise.
197 (loadhi_update): Likewise.
198 (load_zeroextendhisi_update): Likewise.
199 (load_signextendhisi_update): Likewise.
200 (loadsi_update): Likewise.
201 (loadsf_update): Likewise.
202 (movsicc_insn): Likewise.
203 (bset_insn): Likewise.
204 (bxor_insn): Likewise.
205 (bclr_insn): Likewise.
206 (bmsk_insn): Likewise.
207 (bicsi3_insn): Likewise.
208 (cmpsi_cc_c_insn): Likewise.
209 (movsi_ne): Likewise.
210 (movsi_cond_exec): Likewise.
211 (clrsbsi2): Likewise.
222 (call_i): Remove 'S' asm letter, add jli instruction.
223 (call_value_i): Likewise.
224 * config/arc/arc.op (mjli-always): New option.
225 * config/arc/constraints.md (Cji): New constraint.
226 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
228 (subsf3_fpx): Likewise.
229 (mulsf3_fpx): Likewise.
230 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
232 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
234 * doc/invoke.texi (ARC): Document mjli-always option.
236 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
238 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
239 and use incw and decw where possible
240 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
242 2018-01-26 Richard Biener <rguenther@suse.de>
244 PR tree-optimization/81082
245 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
246 association if it requires casting to unsigned.
247 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
248 from fold_plusminus_mult_expr to catch important cases late when
249 range info is available.
251 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
253 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
254 * configure.ac (hidden_linkonce): New test.
255 * configure: Regenerate.
256 * config.in: Regenerate.
258 2018-01-26 Julia Koval <julia.koval@intel.com>
260 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
261 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
262 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
263 _mm_mask_bitshuffle_epi64_mask): Fix type.
264 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
265 USI_FTYPE_V4DI_V4DI_USI): Remove.
266 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
267 __builtin_ia32_vpshufbitqmb256_mask,
268 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
269 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
270 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
272 2018-01-26 Alan Modra <amodra@gmail.com>
275 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
276 UNSPEC_VBPERMQ. Sort other unspecs.
278 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
280 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
282 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
285 * predict.c (drop_profile): Do not push/pop cfun; update also
287 (handle_missing_profiles): Fix logic looking for zero profiles.
289 2018-01-25 Jakub Jelinek <jakub@redhat.com>
292 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
293 on functions with #pragma omp declare simd or functions with simd
295 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
296 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
297 Remove trailing \n from warning_at calls.
299 2018-01-25 Tom de Vries <tom@codesourcery.com>
302 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
303 for neutered workers.
305 2018-01-24 Joseph Myers <joseph@codesourcery.com>
308 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
309 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
311 2017-01-08 Jeff Law <law@redhat.com>
314 * i386.c (get_probe_interval): Move to earlier point.
315 (ix86_compute_frame_layout): If -fstack-clash-protection and
316 the frame is larger than the probe interval, then use pushes
317 to save registers rather than reg->mem moves.
318 (ix86_expand_prologue): Remove conditional for int_registers_saved
321 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
324 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
325 min/max for never referenced object.
327 2018-01-24 Jakub Jelinek <jakub@redhat.com>
330 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
332 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
333 attributes from DECL_ATTRIBUTES (decl) without affecting
334 DECL_ATTRIBUTES (current_function_decl).
335 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
336 functions with non-NULL DECL_ABSTRACT_ORIGIN.
338 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
340 PR tree-optimization/83979
341 * fold-const.c (fold_comparison): Use constant_boolean_node
342 instead of boolean_{true,false}_node.
344 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
346 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
349 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
351 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
352 Simplify the clause that sets the length attribute.
353 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
354 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
355 clause that sets the length attribute.
356 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
358 2018-01-24 Tom de Vries <tom@codesourcery.com>
361 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
362 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
363 Add strict parameter.
364 (prevent_branch_around_nothing): Insert dummy insn between branch to
365 label and label with no ptx insn inbetween.
366 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
368 2018-01-24 Tom de Vries <tom@codesourcery.com>
371 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
372 for neutered threads in warp.
373 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
375 2018-01-24 Richard Biener <rguenther@suse.de>
377 PR tree-optimization/83176
378 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
381 2018-01-24 Richard Biener <rguenther@suse.de>
383 PR tree-optimization/82819
384 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
385 code generating pluses that are no-ops in the target precision.
387 2018-01-24 Richard Biener <rguenther@suse.de>
390 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
392 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
394 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
395 to merge probabilities.
396 * predict.c (probably_never_executed): Also mark as cold functions
397 with global 0 profile and guessed local profile.
398 * profile-count.c (profile_probability::combine_with_count): New
400 * profile-count.h (profile_probability::operator*,
401 profile_probability::operator*=, profile_probability::operator/,
402 profile_probability::operator/=): Reduce precision to adjusted
403 and set value to guessed on contradictory divisions.
404 (profile_probability::combine_with_freq): Remove.
405 (profile_probability::combine_wiht_count): Declare.
406 (profile_count::force_nonzero):: Set to adjusted.
407 (profile_count::probability_in):: Set quality to adjusted.
408 * tree-ssa-tail-merge.c (replace_block_by): Use
411 2018-01-23 Andrew Waterman <andrew@sifive.com>
412 Jim Wilson <jimw@sifive.com>
414 * config/riscv/riscv.c (riscv_stack_boundary): New.
415 (riscv_option_override): Set riscv_stack_boundary. Handle
416 riscv_preferred_stack_boundary_arg.
417 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
418 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
419 (STACK_BOUNDARY): Set to riscv_stack_boundary.
420 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
421 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
422 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
424 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
427 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
428 of struct ix86_frame.
429 (ix86_expand_epilogue): Likewise. Add a local variable for
430 the reg_save_offset field in struct ix86_frame.
432 2018-01-23 Bin Cheng <bin.cheng@arm.com>
434 PR tree-optimization/82604
435 * tree-loop-distribution.c (enum partition_kind): New enum item
436 PKIND_PARTIAL_MEMSET.
437 (partition_builtin_p): Support above new enum item.
438 (generate_code_for_partition): Ditto.
439 (compute_access_range): Differentiate cases that equality can be
440 proven at all loops, the innermost loops or no loops.
441 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
442 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
443 (finalize_partitions, distribute_loop): Don't fuse partition of
444 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
445 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
448 2018-01-23 Martin Liska <mliska@suse.cz>
450 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
451 order to ignore the predictor.
452 (PRED_POLYMORPHIC_CALL): Likewise.
453 (PRED_RECURSIVE_CALL): Likewise.
455 2018-01-23 Martin Liska <mliska@suse.cz>
457 * tree-profile.c (tree_profiling): Print function header to
458 aware reader which function we are working on.
459 * value-prof.c (gimple_find_values_to_profile): Do not print
460 not interesting value histograms.
462 2018-01-23 Martin Liska <mliska@suse.cz>
464 * profile-count.h (enum profile_quality): Add
465 profile_uninitialized as the first value. Do not number values
466 as they are zero based.
467 (profile_count::verify): Update sanity check.
468 (profile_probability::verify): Likewise.
470 2018-01-23 Nathan Sidwell <nathan@acm.org>
472 * doc/invoke.texi (ffor-scope): Deprecate.
474 2018-01-23 David Malcolm <dmalcolm@redhat.com>
476 PR tree-optimization/83510
477 * domwalk.c (set_all_edges_as_executable): New function.
478 (dom_walker::dom_walker): Convert bool param
479 "skip_unreachable_blocks" to enum reachability. Move setup of
480 edge flags to set_all_edges_as_executable and only do it when
481 reachability is REACHABLE_BLOCKS.
482 * domwalk.h (enum dom_walker::reachability): New enum.
483 (dom_walker::dom_walker): Convert bool param
484 "skip_unreachable_blocks" to enum reachability.
485 (set_all_edges_as_executable): New decl.
486 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
487 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
489 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
490 but converting true to REACHABLE_BLOCKS.
491 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
493 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
494 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
495 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
497 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
498 if check_all_array_refs will be called.
500 2018-01-23 David Malcolm <dmalcolm@redhat.com>
502 * tree.c (selftest::test_location_wrappers): Add more test
505 2018-01-23 David Malcolm <dmalcolm@redhat.com>
507 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
508 (selftest::test_bit_in_range): Likewise.
510 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
513 * doc/sourcebuild.texi (vect_float): Say that the selector
514 only describes the situation when -funsafe-math-optimizations is on.
515 (vect_float_strict): Document.
517 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
519 PR tree-optimization/83965
520 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
521 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
522 instead of checking only for a reduction.
523 (vect_recog_widen_sum_pattern): Likewise.
525 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
527 * predict.c (probably_never_executed): Only use precise profile info.
528 (compute_function_frequency): Skip after inlining hack since we now
529 have quality checking.
531 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
533 * profile-count.h (profile_probability::very_unlikely,
534 profile_probability::unlikely, profile_probability::even): Set
535 precision to guessed.
537 2018-01-23 Richard Biener <rguenther@suse.de>
539 PR tree-optimization/83963
540 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
541 Properly terminate dominator walk when crossing the exit edge not
542 when visiting its source block.
544 2018-01-23 Jakub Jelinek <jakub@redhat.com>
547 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
548 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
550 2018-01-22 Jakub Jelinek <jakub@redhat.com>
552 PR tree-optimization/83957
553 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
554 semicolon after for body surrounded by braces.
556 PR tree-optimization/83081
557 * profile-count.h (profile_probability::split): New method.
558 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
559 Use profile_probability::split.
560 (do_compare_rtx_and_jump): Fix adjustment of probabilities
561 when splitting a single conditional jump into 2.
563 2018-01-22 David Malcolm <dmalcolm@redhat.com>
565 PR tree-optimization/69452
566 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
569 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
571 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
572 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
573 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
575 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
577 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
578 * config/rl78/rl78.md: New define_expand "movdi"
579 * config/rl78/rl78.c: New function definition rl78_split_movdi
581 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
584 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
586 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
587 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
588 128-bit to produce an UNSPEC move to get the double word with the
589 signbit and then a shift directly to do signbit.
590 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
591 implementation with a new version that just does either a direct
592 move or a regular move. Move memory interface to separate insns.
593 Move insns so they are next to the expander.
594 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
595 with signbit move. Split big and little endian case.
596 (signbit<mode>2_dm_mem_le): Likewise.
597 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
598 (signbit<mode>2_dm2): Likewise.
600 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
602 * config/rl78/rl78.md: New define_expand "anddi3".
604 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
606 * config/rl78/rl78.md: New define_expand "umindi3".
608 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
610 * config/rl78/rl78.md: New define_expand "smindi3".
612 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
614 * config/rl78/rl78.md: New define_expand "smaxdi3".
616 2018-01-22 Carl Love <cel@us.ibm.com>
618 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
619 LVX_V1TI): Add macro expansion.
620 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
621 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
622 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
623 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
624 Change check to determine if the instruction is a byte reversing
625 entry. Fix typo in comment.
626 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
627 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
628 Add def_builtin calls for new builtins.
629 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
630 Add define_insn expansion.
632 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
634 * config/rl78/rl78.md: New define_expand "umaxdi3".
636 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
638 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
639 for non-QImode registers
641 2018-01-22 Richard Biener <rguenther@suse.de>
643 PR tree-optimization/83963
644 * graphite-scop-detection.c (scop_detection::get_sese): Delay
645 including the loop exit block.
646 (scop_detection::merge_sese): Likewise.
647 (scop_detection::add_scop): Do it here instead.
649 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
651 * doc/sourcebuild.texi (arm_softfloat): Document.
653 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
656 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
657 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
658 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
660 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
661 David Edelsohn <dje.gcc@gmail.com>
664 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
665 Change "crset eq" to "crset 2".
666 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
667 (*call_indirect_aix<mode>_nospec): Likewise.
668 (*call_value_indirect_aix<mode>_nospec): Likewise.
669 (*call_indirect_elfv2<mode>_nospec): Likewise.
670 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
671 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
672 change assembly output from . to $.
673 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
674 (indirect_jump<mode>_nospec): Change assembly output from . to $.
675 (*tablejump<mode>_internal1_nospec): Likewise.
677 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
680 * config/sh/sh_optimize_sett_clrt.cc:
681 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
683 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
685 PR tree-optimization/83940
686 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
687 offset_dt to vect_constant_def rather than vect_unknown_def_type.
688 (vect_check_load_store_mask): Add a mask_dt_out parameter and
689 use it to pass back the definition type.
690 (vect_check_store_rhs): Likewise rhs_dt_out.
691 (vect_build_gather_load_calls): Add a mask_dt argument and use
692 it instead of a call to vect_is_simple_use.
693 (vectorizable_store): Update calls to vect_check_load_store_mask
694 and vect_check_store_rhs. Use the dt returned by the latter instead
695 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
696 instead of calls to vect_is_simple_use. Pass the scalar rather
697 than the vector operand to vect_is_simple_use when handling
698 second and subsequent copies of an rhs value.
699 (vectorizable_load): Update calls to vect_check_load_store_mask
700 and vect_build_gather_load_calls. Use the cached mask_dt and
701 gs_info.offset_dt instead of calls to vect_is_simple_use.
703 2018-01-20 Jakub Jelinek <jakub@redhat.com>
706 * tree-emutls.c: Include gimplify.h.
707 (lower_emutls_2): New function.
708 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
709 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
710 it before further processing.
713 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
714 UINTVAL (trueop1) instead of INTVAL (op1).
716 2018-01-19 Jakub Jelinek <jakub@redhat.com>
720 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
721 INCOMING_FRAME_SP_OFFSET if not defined.
722 (scan_trace): Add ENTRY argument. If true and
723 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
724 emit a note to adjust the CFA offset.
725 (create_cfi_notes): Adjust scan_trace callers.
726 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
727 INCOMING_FRAME_SP_OFFSET in the CIE.
728 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
729 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
731 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
732 * doc/tm.texi: Regenerated.
734 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
736 PR rtl-optimization/83147
737 * lra-constraints.c (remove_inheritance_pseudos): Use
738 lra_substitute_pseudo_within_insn.
740 2018-01-19 Tom de Vries <tom@codesourcery.com>
741 Cesar Philippidis <cesar@codesourcery.com>
744 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
746 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
749 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
750 spaces for function labels.
752 2018-01-19 Martin Liska <mliska@suse.cz>
754 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
755 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
756 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
757 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
758 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
759 (PRED_CONST_RETURN): Change from 69 to 65.
760 (PRED_NULL_RETURN): Change from 91 to 71.
761 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
762 (PRED_LOOP_GUARD): Change from 66 to 73.
764 2018-01-19 Martin Liska <mliska@suse.cz>
766 * predict.c (predict_insn_def): Add new assert.
767 (struct branch_predictor): Change type to signed integer.
768 (test_prediction_value_range): Amend test to cover
770 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
771 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
772 (PRED_LOOP_ITERATIONS_MAX): Likewise.
773 (PRED_LOOP_IV_COMPARE): Likewise.
774 * predict.h (PROB_UNINITIALIZED): Define new constant.
776 2018-01-19 Martin Liska <mliska@suse.cz>
778 * predict.c (dump_prediction): Add new format for
779 analyze_brprob.py script which is enabled with -details
781 * profile-count.h (precise_p): New function.
783 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
785 PR tree-optimization/83922
786 * tree-vect-loop.c (vect_verify_full_masking): Return false if
787 there are no statements that need masking.
788 (vect_active_double_reduction_p): New function.
789 (vect_analyze_loop_operations): Use it when handling phis that
790 are not in the loop header.
792 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
794 PR tree-optimization/83914
795 * tree-vect-loop.c (vectorizable_induction): Don't convert
796 init_expr or apply the peeling adjustment for inductions
797 that are nested within the vectorized loop.
799 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
801 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
804 2018-01-18 Jakub Jelinek <jakub@redhat.com>
808 * function.h (gimplify_parameters): Add gimple_seq * argument.
809 * function.c: Include gimple.h and options.h.
810 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
811 for the added local temporaries if needed.
812 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
813 if there are any parameter cleanups, wrap whole body into a
814 try/finally with the cleanups.
816 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
819 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
820 Use GET_MODE_CLASS for scalar floating point.
822 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
826 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
827 Fix call of call_cgraph_insertion_hooks.
829 2018-01-18 Martin Sebor <msebor@redhat.com>
831 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
833 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
836 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
839 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
842 * common.opt: (-ffile-prefix-map): New option.
843 * opts.c (common_handle_option): Defer it.
844 * opts-global.c (handle_common_deferred_options): Handle it.
845 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
846 * file-prefix-map.h: New file.
847 (remap_debug_filename, add_debug_prefix_map): ...here.
848 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
849 * final.c (debug_prefix_map, add_debug_prefix_map
850 remap_debug_filename): Move to...
851 * file-prefix-map.c: New file.
852 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
853 generalize, get rid of alloca(), use strrchr() instead of strchr().
854 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
855 Implement in terms of add_prefix_map().
856 (remap_macro_filename, remap_debug_filename): Implement in term of
858 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
859 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
860 * dbxout.c: Include file-prefix-map.h.
861 * varasm.c: Likewise.
862 * vmsdbgout.c: Likewise.
863 * xcoffout.c: Likewise.
864 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
865 * doc/cppopts.texi (-fmacro-prefix-map): Document.
866 * doc/invoke.texi (-ffile-prefix-map): Document.
867 (-fdebug-prefix-map): Update description.
869 2018-01-18 Martin Liska <mliska@suse.cz>
871 * config/i386/i386.c (indirect_thunk_name): Document that also
873 (output_indirect_thunk): Document why both instructions
874 (pause and lfence) are generated.
876 2018-01-18 Richard Biener <rguenther@suse.de>
878 PR tree-optimization/83887
879 * graphite-scop-detection.c
880 (scop_detection::get_nearest_dom_with_single_entry): Remove.
881 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
882 (scop_detection::merge_sese): Re-implement with a flood-fill
883 algorithm that properly finds a SESE region if it exists.
885 2018-01-18 Jakub Jelinek <jakub@redhat.com>
888 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
889 pointer_diff optimizations use view_convert instead of convert.
891 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
893 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
894 Generate different code for -mno-speculate-indirect-jumps.
895 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
896 (*call_indirect_aix<mode>): Disable for
897 -mno-speculate-indirect-jumps.
898 (*call_indirect_aix<mode>_nospec): New define_insn.
899 (*call_value_indirect_aix<mode>): Disable for
900 -mno-speculate-indirect-jumps.
901 (*call_value_indirect_aix<mode>_nospec): New define_insn.
902 (*sibcall_nonlocal_sysv<mode>): Generate different code for
903 -mno-speculate-indirect-jumps.
904 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
906 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
908 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
909 long double type, set the flags for noting the default long double
910 type, even if we don't pass or return a long double type.
912 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
915 * ipa-inline.c (flatten_function): Do not overwrite final inlining
918 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
920 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
921 support for merge[hl].
922 (fold_mergehl_helper): New helper function.
923 (tree-vector-builder.h): New #include for tree_vector_builder usage.
924 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
925 (altivec_vmrglw_direct): Add xxmrglw insn.
927 2018-01-17 Andrew Waterman <andrew@sifive.com>
929 * config/riscv/riscv.c (riscv_conditional_register_usage): If
930 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
932 2018-01-17 David Malcolm <dmalcolm@redhat.com>
935 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
936 call the lto_location_cache before reading the
937 DECL_SOURCE_LOCATION of the types.
939 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
940 Richard Sandiford <richard.sandiford@linaro.org>
942 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
943 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
944 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
945 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
946 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
948 * config/aarch64/constraints.md (aarch64_movti_operand):
950 * config/aarch64/predicates.md (Uti): Add new constraint.
952 2018-01-17 Carl Love <cel@us.ibm.com>
953 * config/rs6000/vsx.md (define_expand xl_len_r,
954 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
955 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
957 (define_expand, define_insn): Move the shift left from the
958 define_insn to the define_expand for lxvl and stxvl instructions.
959 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
960 and XL_LEN_R definitions to PURE.
962 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
964 * config/i386/i386.c (indirect_thunk_name): Declare regno
965 as unsigned int. Compare regno with INVALID_REGNUM.
966 (output_indirect_thunk): Ditto.
967 (output_indirect_thunk_function): Ditto.
968 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
969 in the call to output_indirect_thunk_function.
971 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
974 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
975 rather than the size of inner_type to determine the stack slot size
976 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
978 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
981 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
984 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
986 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
987 endian Linux systems to optionally enable multilibs for selecting
988 the long double type if the user configured an explicit type.
989 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
990 have no long double multilibs if not defined.
991 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
992 warn if the user used -mabi={ieee,ibm}longdouble and we built
993 multilibs for long double.
994 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
995 appropriate multilib option.
996 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
998 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
999 for building long double multilibs.
1000 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1002 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1004 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1007 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1009 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1012 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1015 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1018 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1020 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1021 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1023 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1025 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1026 different rtl trees depending on TARGET_64BIT.
1027 (rs6000_gen_lvx): Likewise.
1029 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1031 * config/visium/visium.md (nop): Tweak comment.
1032 (hazard_nop): Likewise.
1034 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1036 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1037 -mspeculate-indirect-jumps.
1038 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1039 for -mno-speculate-indirect-jumps.
1040 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1041 (*call_value_indirect_elfv2<mode>): Disable for
1042 -mno-speculate-indirect-jumps.
1043 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1044 (indirect_jump): Emit different RTL for
1045 -mno-speculate-indirect-jumps.
1046 (*indirect_jump<mode>): Disable for
1047 -mno-speculate-indirect-jumps.
1048 (*indirect_jump<mode>_nospec): New define_insn.
1049 (tablejump): Emit different RTL for
1050 -mno-speculate-indirect-jumps.
1051 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1052 (tablejumpsi_nospec): New define_expand.
1053 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1054 (tablejumpdi_nospec): New define_expand.
1055 (*tablejump<mode>_internal1): Disable for
1056 -mno-speculate-indirect-jumps.
1057 (*tablejump<mode>_internal1_nospec): New define_insn.
1058 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1061 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1063 * caller-save.c (insert_save): Drop unnecessary parameter. All
1066 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1067 Richard Biener <rguenth@suse.de>
1070 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1071 return early, inline manually is_gimple_sizepos. Make sure if we
1072 call gimplify_expr we don't end up with a gimple constant.
1073 * tree.c (variably_modified_type_p): Don't return true for
1074 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1075 * gimplify.h (is_gimple_sizepos): Remove.
1077 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1079 PR tree-optimization/83857
1080 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1081 vectorizable_live_operation for pure SLP statements.
1082 (vectorizable_live_operation): Handle PHIs.
1084 2018-01-16 Richard Biener <rguenther@suse.de>
1086 PR tree-optimization/83867
1087 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1088 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1090 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1093 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1094 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1095 If off is not INTEGER_CST, issue a may not be aligned warning
1096 rather than isn't aligned. Use isn%'t rather than isn't.
1097 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1099 <case MULT_EXPR>: Improve the case when bottom and one of the
1100 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1101 operand, in that case check if the other operand is multiple of
1102 bottom divided by the INTEGER_CST operand.
1104 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1107 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1108 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1109 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1110 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1111 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1112 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1113 (pa_function_arg_size): New function.
1115 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1117 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1118 in a separate statement.
1120 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1122 PR tree-optimization/83847
1123 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1124 group gathers and scatters.
1126 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1128 PR rtl-optimization/86620
1129 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1131 PR rtl-optimization/83213
1132 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1133 to last if both are JUMP_INSNs.
1135 PR tree-optimization/83843
1136 * gimple-ssa-store-merging.c
1137 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1138 store_immediate_info for bswap/nop orig_stores.
1140 2018-01-15 Andrew Waterman <andrew@sifive.com>
1142 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1144 <UDIV>: Increase cost if !TARGET_DIV.
1146 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1148 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1149 (define_attr "cr_logical_3op"): New.
1150 (cceq_ior_compare): Adjust.
1151 (cceq_ior_compare_complement): Adjust.
1152 (*cceq_rev_compare): Adjust.
1153 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1154 (is_cracked_insn): Adjust.
1155 (insn_must_be_first_in_group): Adjust.
1156 * config/rs6000/40x.md: Adjust.
1157 * config/rs6000/440.md: Adjust.
1158 * config/rs6000/476.md: Adjust.
1159 * config/rs6000/601.md: Adjust.
1160 * config/rs6000/603.md: Adjust.
1161 * config/rs6000/6xx.md: Adjust.
1162 * config/rs6000/7450.md: Adjust.
1163 * config/rs6000/7xx.md: Adjust.
1164 * config/rs6000/8540.md: Adjust.
1165 * config/rs6000/cell.md: Adjust.
1166 * config/rs6000/e300c2c3.md: Adjust.
1167 * config/rs6000/e500mc.md: Adjust.
1168 * config/rs6000/e500mc64.md: Adjust.
1169 * config/rs6000/e5500.md: Adjust.
1170 * config/rs6000/e6500.md: Adjust.
1171 * config/rs6000/mpc.md: Adjust.
1172 * config/rs6000/power4.md: Adjust.
1173 * config/rs6000/power5.md: Adjust.
1174 * config/rs6000/power6.md: Adjust.
1175 * config/rs6000/power7.md: Adjust.
1176 * config/rs6000/power8.md: Adjust.
1177 * config/rs6000/power9.md: Adjust.
1178 * config/rs6000/rs64.md: Adjust.
1179 * config/rs6000/titan.md: Adjust.
1181 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1183 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1184 ix86_indirect_branch_register logic.
1186 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1188 * config/i386/constraints.md (Bs): Update
1189 ix86_indirect_branch_register check. Don't check
1190 ix86_indirect_branch_register with GOT_memory_operand.
1192 * config/i386/predicates.md (GOT_memory_operand): Don't check
1193 ix86_indirect_branch_register here.
1194 (GOT32_symbol_operand): Likewise.
1196 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1198 * config/i386/predicates.md (constant_call_address_operand):
1199 Rewrite ix86_indirect_branch_register logic.
1200 (sibcall_insn_operand): Likewise.
1202 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1204 * config/i386/constraints.md (Bs): Replace
1205 ix86_indirect_branch_thunk_register with
1206 ix86_indirect_branch_register.
1208 * config/i386/i386.md (indirect_jump): Likewise.
1209 (tablejump): Likewise.
1210 (*sibcall_memory): Likewise.
1211 (*sibcall_value_memory): Likewise.
1212 Peepholes of indirect call and jump via memory: Likewise.
1213 * config/i386/i386.opt: Likewise.
1214 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1215 (GOT_memory_operand): Likewise.
1216 (call_insn_operand): Likewise.
1217 (sibcall_insn_operand): Likewise.
1218 (GOT32_symbol_operand): Likewise.
1220 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1223 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1224 type rather than type addr's type points to.
1225 (expand_omp_atomic_mutex): Likewise.
1226 (expand_omp_atomic): Likewise.
1228 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1231 * config/i386/i386.c (output_indirect_thunk_function): Use
1232 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1233 for __x86_return_thunk.
1235 2018-01-15 Richard Biener <rguenther@suse.de>
1238 * expmed.c (extract_bit_field_1): Fix typo.
1240 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1243 * config/arm/iterators.md (VF): New mode iterator.
1244 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1245 Remove integer-related logic from pattern.
1246 (neon_vabd<mode>_3): Likewise.
1248 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1251 * common.opt (fstrict-overflow): No longer an alias.
1252 (fwrapv-pointer): New option.
1253 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1254 also for pointer types based on flag_wrapv_pointer.
1255 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1256 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1257 opts->x_flag_wrapv got set.
1258 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1259 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1260 POINTER_TYPE_OVERFLOW_UNDEFINED.
1261 * match.pd: Likewise in address comparison pattern.
1262 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1264 2018-01-15 Richard Biener <rguenther@suse.de>
1267 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1268 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1269 Reset type names to their identifier if their TYPE_DECL doesn't
1270 have linkage (and thus is used for ODR and devirt).
1271 (save_debug_info_for_decl): Remove.
1272 (save_debug_info_for_type): Likewise.
1273 (add_tree_to_fld_list): Adjust.
1274 * tree-pretty-print.c (dump_generic_node): Make dumping of
1275 type names more robust.
1277 2018-01-15 Richard Biener <rguenther@suse.de>
1279 * BASE-VER: Bump to 8.0.1.
1281 2018-01-14 Martin Sebor <msebor@redhat.com>
1284 * builtins.c (check_access): Avoid warning when the no-warning bit
1287 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1289 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1290 * ira-color (allocno_hard_regs_compare): Likewise.
1292 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1295 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1296 Use .pushsection/.popsection.
1298 2018-01-14 Martin Sebor <msebor@redhat.com>
1301 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1303 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1305 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1306 entry from extra_headers.
1307 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1308 extra_headers, make the list bitwise identical to the i?86-*-* one.
1310 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1312 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1313 -mcmodel=large with -mindirect-branch=thunk,
1314 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1315 -mfunction-return=thunk-extern.
1316 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1317 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1318 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1320 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1322 * config/i386/i386.c (print_reg): Print the name of the full
1323 integer register without '%'.
1324 (ix86_print_operand): Handle 'V'.
1325 * doc/extend.texi: Document 'V' modifier.
1327 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1329 * config/i386/constraints.md (Bs): Disallow memory operand for
1330 -mindirect-branch-register.
1332 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1333 (GOT_memory_operand): Likewise.
1334 (call_insn_operand): Likewise.
1335 (sibcall_insn_operand): Likewise.
1336 (GOT32_symbol_operand): Likewise.
1337 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1338 for -mindirect-branch-register.
1339 (tablejump): Likewise.
1340 (*sibcall_memory): Likewise.
1341 (*sibcall_value_memory): Likewise.
1342 Disallow peepholes of indirect call and jump via memory for
1343 -mindirect-branch-register.
1344 (*call_pop): Replace m with Bw.
1345 (*call_value_pop): Likewise.
1346 (*sibcall_pop_memory): Replace m with Bs.
1347 * config/i386/i386.opt (mindirect-branch-register): New option.
1348 * doc/invoke.texi: Document -mindirect-branch-register option.
1350 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1352 * config/i386/i386-protos.h (ix86_output_function_return): New.
1353 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1354 set function_return_type.
1355 (indirect_thunk_name): Add ret_p to indicate thunk for function
1357 (output_indirect_thunk_function): Pass false to
1358 indirect_thunk_name.
1359 (ix86_output_indirect_branch_via_reg): Likewise.
1360 (ix86_output_indirect_branch_via_push): Likewise.
1361 (output_indirect_thunk_function): Create alias for function
1362 return thunk if regno < 0.
1363 (ix86_output_function_return): New function.
1364 (ix86_handle_fndecl_attribute): Handle function_return.
1365 (ix86_attribute_table): Add function_return.
1366 * config/i386/i386.h (machine_function): Add
1367 function_return_type.
1368 * config/i386/i386.md (simple_return_internal): Use
1369 ix86_output_function_return.
1370 (simple_return_internal_long): Likewise.
1371 * config/i386/i386.opt (mfunction-return=): New option.
1372 (indirect_branch): Mention -mfunction-return=.
1373 * doc/extend.texi: Document function_return function attribute.
1374 * doc/invoke.texi: Document -mfunction-return= option.
1376 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1378 * config/i386/i386-opts.h (indirect_branch): New.
1379 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1380 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1381 with local indirect jump when converting indirect call and jump.
1382 (ix86_set_indirect_branch_type): New.
1383 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1384 (indirectlabelno): New.
1385 (indirect_thunk_needed): Likewise.
1386 (indirect_thunk_bnd_needed): Likewise.
1387 (indirect_thunks_used): Likewise.
1388 (indirect_thunks_bnd_used): Likewise.
1389 (INDIRECT_LABEL): Likewise.
1390 (indirect_thunk_name): Likewise.
1391 (output_indirect_thunk): Likewise.
1392 (output_indirect_thunk_function): Likewise.
1393 (ix86_output_indirect_branch_via_reg): Likewise.
1394 (ix86_output_indirect_branch_via_push): Likewise.
1395 (ix86_output_indirect_branch): Likewise.
1396 (ix86_output_indirect_jmp): Likewise.
1397 (ix86_code_end): Call output_indirect_thunk_function if needed.
1398 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1400 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1401 (ix86_attribute_table): Add indirect_branch.
1402 * config/i386/i386.h (machine_function): Add indirect_branch_type
1403 and has_local_indirect_jump.
1404 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1406 (tablejump): Likewise.
1407 (*indirect_jump): Use ix86_output_indirect_jmp.
1408 (*tablejump_1): Likewise.
1409 (simple_return_indirect_internal): Likewise.
1410 * config/i386/i386.opt (mindirect-branch=): New option.
1411 (indirect_branch): New.
1414 (thunk-inline): Likewise.
1415 (thunk-extern): Likewise.
1416 * doc/extend.texi: Document indirect_branch function attribute.
1417 * doc/invoke.texi: Document -mindirect-branch= option.
1419 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1422 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1424 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1426 * ipa-inline.c (want_inline_small_function_p): Return false if
1427 inlining has already failed with CIF_FINAL_ERROR.
1428 (update_caller_keys): Call want_inline_small_function_p before
1430 (update_callee_keys): Likewise.
1432 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1434 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1436 (rs6000_quadword_masked_address_p): Likewise.
1437 (quad_aligned_load_p): Likewise.
1438 (quad_aligned_store_p): Likewise.
1439 (const_load_sequence_p): Add comment to describe the outer-most loop.
1440 (mimic_memory_attributes_and_flags): New function.
1441 (rs6000_gen_stvx): Likewise.
1442 (replace_swapped_aligned_store): Likewise.
1443 (rs6000_gen_lvx): Likewise.
1444 (replace_swapped_aligned_load): Likewise.
1445 (replace_swapped_load_constant): Capitalize argument name in
1446 comment describing this function.
1447 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1448 and stores that access quad-word aligned addresses and replace
1449 with stvx or lvx instructions when appropriate.
1450 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1451 New function prototype.
1452 (rs6000_quadword_masked_address_p): Likewise.
1453 (rs6000_gen_lvx): Likewise.
1454 (rs6000_gen_stvx): Likewise.
1455 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1456 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1457 when memory address is aligned.
1458 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1459 this split to select lvx instruction when memory address is aligned.
1460 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1461 instruction when memory address is aligned.
1462 (*vsx_le_perm_load_v16qi): Likewise.
1463 (four unnamed splitters): Modify to select the stvx instruction
1464 when memory is aligned.
1466 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1468 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1469 which appears in the queue multiple times.
1471 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1472 Alan Hayward <alan.hayward@arm.com>
1473 David Sherwood <david.sherwood@arm.com>
1475 * tree-vectorizer.h (vec_lower_bound): New structure.
1476 (_loop_vec_info): Add check_nonzero and lower_bounds.
1477 (LOOP_VINFO_CHECK_NONZERO): New macro.
1478 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1479 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1480 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1481 fields. Make seg_len the distance travelled, not including the
1483 (dr_direction_indicator): Declare.
1484 (dr_zero_step_indicator): Likewise.
1485 (dr_known_forward_stride_p): Likewise.
1486 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1488 (runtime_alias_check_p): Allow runtime alias checks with
1490 (operator ==): Compare access_size and align.
1491 (prune_runtime_alias_test_list): Rework for new distinction between
1492 the access_size and seg_len.
1493 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1495 (get_segment_min_max): New function.
1496 (create_intersect_range_checks): Use it.
1497 (dr_step_indicator): New function.
1498 (dr_direction_indicator): Likewise.
1499 (dr_zero_step_indicator): Likewise.
1500 (dr_known_forward_stride_p): Likewise.
1501 * tree-loop-distribution.c (data_ref_segment_size): Return
1502 DR_STEP * (niters - 1).
1503 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1505 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1506 (vect_preserves_scalar_order_p): New function, split out from...
1507 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1508 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1509 (vect_vfa_access_size): New function.
1510 (vect_vfa_align): Likewise.
1511 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1512 (dump_lower_bound): New function.
1513 (vect_check_lower_bound): Likewise.
1514 (vect_small_gap_p): Likewise.
1515 (vectorizable_with_step_bound_p): Likewise.
1516 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1517 depencies if the vectorization factor is 1. Convert the checks
1518 for nonzero steps into checks on the bounds of DR_STEP. Try using
1519 a bunds check for variable steps if the minimum required step is
1520 relatively small. Update calls to the dr_with_seg_len
1521 constructor and to vect_compile_time_alias.
1522 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1524 (vect_loop_versioning): Call it.
1525 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1527 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1529 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1530 Alan Hayward <alan.hayward@arm.com>
1531 David Sherwood <david.sherwood@arm.com>
1533 * doc/sourcebuild.texi (vect_scatter_store): Document.
1534 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1536 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1538 * genopinit.c (main): Add supports_vec_scatter_store and
1539 supports_vec_scatter_store_cached to target_optabs.
1540 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1541 IFN_MASK_SCATTER_STORE.
1542 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1544 * internal-fn.h (internal_store_fn_p): Declare.
1545 (internal_fn_stored_value_index): Likewise.
1546 * internal-fn.c (scatter_store_direct): New macro.
1547 (expand_scatter_store_optab_fn): New function.
1548 (direct_scatter_store_optab_supported_p): New macro.
1549 (internal_store_fn_p): New function.
1550 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1551 IFN_MASK_SCATTER_STORE.
1552 (internal_fn_mask_index): Likewise.
1553 (internal_fn_stored_value_index): New function.
1554 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1556 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1557 * optabs-query.c (supports_vec_scatter_store_p): New function.
1558 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1559 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1560 true for scatter stores.
1561 (vect_gather_scatter_fn_p): Handle scatter stores too.
1562 (vect_check_gather_scatter): Consider using scatter stores if
1563 supports_vec_scatter_store_p.
1564 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1566 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1567 internal_fn_stored_value_index.
1568 (check_load_store_masking): Handle scatter stores too.
1569 (vect_get_store_rhs): Make public.
1570 (vectorizable_call): Use internal_store_fn_p.
1571 (vectorizable_store): Handle scatter store internal functions.
1572 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1573 when deciding whether the end of the group has been reached.
1574 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1575 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1576 (mask_scatter_store<mode>): New insns.
1578 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1579 Alan Hayward <alan.hayward@arm.com>
1580 David Sherwood <david.sherwood@arm.com>
1582 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1583 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1584 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1586 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1587 Use vect_truncate_gather_scatter_offset if we can't treat the
1588 operation as a normal gather load or scatter store.
1589 (get_group_load_store_type): Take the gather_scatter_info
1590 as argument. Try using a gather load or scatter store for
1591 single-element groups.
1592 (get_load_store_type): Update calls to get_group_load_store_type
1593 and vect_use_strided_gather_scatters_p.
1595 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1596 Alan Hayward <alan.hayward@arm.com>
1597 David Sherwood <david.sherwood@arm.com>
1599 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1600 optional tree argument.
1601 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1603 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1604 but continue to use the current value as a fallback.
1605 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1606 to compare the updates.
1607 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1608 (get_load_store_type): Use it when handling a strided access.
1609 (vect_get_strided_load_store_ops): New function.
1610 (vect_get_data_ptr_increment): Likewise.
1611 (vectorizable_load): Handle strided gather loads. Always pass
1612 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1614 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1615 Alan Hayward <alan.hayward@arm.com>
1616 David Sherwood <david.sherwood@arm.com>
1618 * doc/md.texi (gather_load@var{m}): Document.
1619 (mask_gather_load@var{m}): Likewise.
1620 * genopinit.c (main): Add supports_vec_gather_load and
1621 supports_vec_gather_load_cached to target_optabs.
1622 * optabs-tree.c (init_tree_optimization_optabs): Use
1623 ggc_cleared_alloc to allocate target_optabs.
1624 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1625 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1627 * internal-fn.h (internal_load_fn_p): Declare.
1628 (internal_gather_scatter_fn_p): Likewise.
1629 (internal_fn_mask_index): Likewise.
1630 (internal_gather_scatter_fn_supported_p): Likewise.
1631 * internal-fn.c (gather_load_direct): New macro.
1632 (expand_gather_load_optab_fn): New function.
1633 (direct_gather_load_optab_supported_p): New macro.
1634 (direct_internal_fn_optab): New function.
1635 (internal_load_fn_p): Likewise.
1636 (internal_gather_scatter_fn_p): Likewise.
1637 (internal_fn_mask_index): Likewise.
1638 (internal_gather_scatter_fn_supported_p): Likewise.
1639 * optabs-query.c (supports_at_least_one_mode_p): New function.
1640 (supports_vec_gather_load_p): Likewise.
1641 * optabs-query.h (supports_vec_gather_load_p): Declare.
1642 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1643 and memory_type field.
1644 (NUM_PATTERNS): Bump to 15.
1645 * tree-vect-data-refs.c: Include internal-fn.h.
1646 (vect_gather_scatter_fn_p): New function.
1647 (vect_describe_gather_scatter_call): Likewise.
1648 (vect_check_gather_scatter): Try using internal functions for
1649 gather loads. Recognize existing calls to a gather load function.
1650 (vect_analyze_data_refs): Consider using gather loads if
1651 supports_vec_gather_load_p.
1652 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1653 (vect_get_gather_scatter_offset_type): Likewise.
1654 (vect_convert_mask_for_vectype): Likewise.
1655 (vect_add_conversion_to_patterm): Likewise.
1656 (vect_try_gather_scatter_pattern): Likewise.
1657 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1658 (vect_vect_recog_func_ptrs): Add it.
1659 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1660 internal_fn_mask_index and internal_gather_scatter_fn_p.
1661 (check_load_store_masking): Take the gather_scatter_info as an
1662 argument and handle gather loads.
1663 (vect_get_gather_scatter_ops): New function.
1664 (vectorizable_call): Check internal_load_fn_p.
1665 (vectorizable_load): Likewise. Handle gather load internal
1667 (vectorizable_store): Update call to check_load_store_masking.
1668 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1669 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1670 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1671 (aarch64_gather_scale_operand_d): New predicates.
1672 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1673 (mask_gather_load<mode>): New insns.
1675 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1676 Alan Hayward <alan.hayward@arm.com>
1677 David Sherwood <david.sherwood@arm.com>
1679 * optabs.def (fold_left_plus_optab): New optab.
1680 * doc/md.texi (fold_left_plus_@var{m}): Document.
1681 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1682 * internal-fn.c (fold_left_direct): Define.
1683 (expand_fold_left_optab_fn): Likewise.
1684 (direct_fold_left_optab_supported_p): Likewise.
1685 * fold-const-call.c (fold_const_fold_left): New function.
1686 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1687 * tree-parloops.c (valid_reduction_p): New function.
1688 (gather_scalar_reductions): Use it.
1689 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1690 (vect_finish_replace_stmt): Declare.
1691 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1692 (needs_fold_left_reduction_p): New function, split out from...
1693 (vect_is_simple_reduction): ...here. Accept reductions that
1694 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1695 (vect_force_simple_reduction): Also store the reduction type in
1696 the assignment's STMT_VINFO_REDUC_TYPE.
1697 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1698 (merge_with_identity): New function.
1699 (vect_expand_fold_left): Likewise.
1700 (vectorize_fold_left_reduction): Likewise.
1701 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1702 scalar phi in place for it. Check for target support and reject
1703 cases that would reassociate the operation. Defer the transform
1704 phase to vectorize_fold_left_reduction.
1705 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1706 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1707 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1709 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1711 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1712 call to ifc_temp_var.
1714 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1715 Alan Hayward <alan.hayward@arm.com>
1716 David Sherwood <david.sherwood@arm.com>
1718 * target.def (legitimize_address_displacement): Take the original
1719 offset as a poly_int.
1720 * targhooks.h (default_legitimize_address_displacement): Update
1722 * targhooks.c (default_legitimize_address_displacement): Likewise.
1723 * doc/tm.texi: Regenerate.
1724 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1725 as an argument, moving assert of ad->disp == ad->disp_term to...
1726 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1727 Try calling targetm.legitimize_address_displacement before expanding
1728 the address rather than afterwards, and adjust for the new interface.
1729 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1730 Match the new hook interface. Handle SVE addresses.
1731 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1734 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1736 * Makefile.in (OBJS): Add early-remat.o.
1737 * target.def (select_early_remat_modes): New hook.
1738 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1739 * doc/tm.texi: Regenerate.
1740 * targhooks.h (default_select_early_remat_modes): Declare.
1741 * targhooks.c (default_select_early_remat_modes): New function.
1742 * timevar.def (TV_EARLY_REMAT): New timevar.
1743 * passes.def (pass_early_remat): New pass.
1744 * tree-pass.h (make_pass_early_remat): Declare.
1745 * early-remat.c: New file.
1746 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1748 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1750 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1751 Alan Hayward <alan.hayward@arm.com>
1752 David Sherwood <david.sherwood@arm.com>
1754 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1755 vfm1 with a bound_epilog parameter.
1756 (vect_do_peeling): Update calls accordingly, and move the prologue
1757 call earlier in the function. Treat the base bound_epilog as 0 for
1758 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1759 this base when peeling for gaps.
1760 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1761 with fully-masked loops.
1762 (vect_estimate_min_profitable_iters): Handle the single peeled
1763 iteration in that case.
1765 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1766 Alan Hayward <alan.hayward@arm.com>
1767 David Sherwood <david.sherwood@arm.com>
1769 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1770 single-element interleaving even if the size is not a power of 2.
1771 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1772 accesses for single-element interleaving if the group size is
1775 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1776 Alan Hayward <alan.hayward@arm.com>
1777 David Sherwood <david.sherwood@arm.com>
1779 * doc/md.texi (fold_extract_last_@var{m}): Document.
1780 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1781 * optabs.def (fold_extract_last_optab): New optab.
1782 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1783 * internal-fn.c (fold_extract_direct): New macro.
1784 (expand_fold_extract_optab_fn): Likewise.
1785 (direct_fold_extract_optab_supported_p): Likewise.
1786 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1787 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1788 EXTRACT_LAST_REDUCTION.
1789 (get_initial_def_for_reduction): Do not create an initial vector
1790 for EXTRACT_LAST_REDUCTION reductions.
1791 (vectorizable_reduction): Leave the scalar phi in place for
1792 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1793 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1794 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1795 transform phase to vectorizable_condition.
1796 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1798 (vect_finish_stmt_generation): ...here.
1799 (vect_finish_replace_stmt): New function.
1800 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1801 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1803 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1805 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1806 Alan Hayward <alan.hayward@arm.com>
1807 David Sherwood <david.sherwood@arm.com>
1809 * doc/md.texi (extract_last_@var{m}): Document.
1810 * optabs.def (extract_last_optab): New optab.
1811 * internal-fn.def (EXTRACT_LAST): New internal function.
1812 * internal-fn.c (cond_unary_direct): New macro.
1813 (expand_cond_unary_optab_fn): Likewise.
1814 (direct_cond_unary_optab_supported_p): Likewise.
1815 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1816 loops using EXTRACT_LAST.
1817 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1818 (extract_last_<mode>): ...this optab.
1819 (vec_extract<mode><Vel>): Update accordingly.
1821 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1822 Alan Hayward <alan.hayward@arm.com>
1823 David Sherwood <david.sherwood@arm.com>
1825 * target.def (empty_mask_is_expensive): New hook.
1826 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
1827 * doc/tm.texi: Regenerate.
1828 * targhooks.h (default_empty_mask_is_expensive): Declare.
1829 * targhooks.c (default_empty_mask_is_expensive): New function.
1830 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
1831 if the target says that empty masks are expensive.
1832 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
1834 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
1836 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1837 Alan Hayward <alan.hayward@arm.com>
1838 David Sherwood <david.sherwood@arm.com>
1840 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
1841 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
1842 (vect_use_loop_mask_for_alignment_p): New function.
1843 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
1844 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
1845 niters_skip argument. Make sure that the first niters_skip elements
1846 of the first iteration are inactive.
1847 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
1848 Update call to vect_set_loop_masks_directly.
1849 (get_misalign_in_elems): New function, split out from...
1850 (vect_gen_prolog_loop_niters): ...here.
1851 (vect_update_init_of_dr): Take a code argument that specifies whether
1852 the adjustment should be added or subtracted.
1853 (vect_update_init_of_drs): Likewise.
1854 (vect_prepare_for_masked_peels): New function.
1855 (vect_do_peeling): Skip prologue peeling if we're using a mask
1856 instead. Update call to vect_update_inits_of_drs.
1857 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1859 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
1860 alignment. Do not include the number of peeled iterations in
1861 the minimum threshold in that case.
1862 (vectorizable_induction): Adjust the start value down by
1863 LOOP_VINFO_MASK_SKIP_NITERS iterations.
1864 (vect_transform_loop): Call vect_prepare_for_masked_peels.
1865 Take the number of skipped iterations into account when calculating
1867 * tree-vect-stmts.c (vect_gen_while_not): New function.
1869 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1870 Alan Hayward <alan.hayward@arm.com>
1871 David Sherwood <david.sherwood@arm.com>
1873 * doc/sourcebuild.texi (vect_fully_masked): Document.
1874 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
1876 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
1878 (vect_analyze_loop_2): ...here. Don't check the vectorization
1879 factor against the number of loop iterations if the loop is
1882 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1883 Alan Hayward <alan.hayward@arm.com>
1884 David Sherwood <david.sherwood@arm.com>
1886 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
1887 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
1888 (dump_groups): Update accordingly.
1889 (iv_use::mem_type): New member variable.
1890 (address_p): New function.
1891 (record_use): Add a mem_type argument and initialize the new
1893 (record_group_use): Add a mem_type argument. Use address_p.
1894 Remove obsolete null checks of base_object. Update call to record_use.
1895 (find_interesting_uses_op): Update call to record_group_use.
1896 (find_interesting_uses_cond): Likewise.
1897 (find_interesting_uses_address): Likewise.
1898 (get_mem_type_for_internal_fn): New function.
1899 (find_address_like_use): Likewise.
1900 (find_interesting_uses_stmt): Try find_address_like_use before
1901 calling find_interesting_uses_op.
1902 (addr_offset_valid_p): Use the iv mem_type field as the type
1903 of the addressed memory.
1904 (add_autoinc_candidates): Likewise.
1905 (get_address_cost): Likewise.
1906 (split_small_address_groups_p): Use address_p.
1907 (split_address_groups): Likewise.
1908 (add_iv_candidate_for_use): Likewise.
1909 (autoinc_possible_for_pair): Likewise.
1910 (rewrite_groups): Likewise.
1911 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
1912 (determine_group_iv_cost): Update after split of USE_ADDRESS.
1913 (get_alias_ptr_type_for_ptr_address): New function.
1914 (rewrite_use_address): Rewrite address uses in calls that were
1915 identified by find_address_like_use.
1917 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1918 Alan Hayward <alan.hayward@arm.com>
1919 David Sherwood <david.sherwood@arm.com>
1921 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
1923 * gimple-expr.h (is_gimple_addressable: Likewise.
1924 * gimple-expr.c (is_gimple_address): Likewise.
1925 * internal-fn.c (expand_call_mem_ref): New function.
1926 (expand_mask_load_optab_fn): Use it.
1927 (expand_mask_store_optab_fn): Likewise.
1929 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1930 Alan Hayward <alan.hayward@arm.com>
1931 David Sherwood <david.sherwood@arm.com>
1933 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
1934 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
1935 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
1936 (cond_umax@var{mode}): Document.
1937 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
1938 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
1939 (cond_umin_optab, cond_umax_optab): New optabs.
1940 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
1941 (COND_IOR, COND_XOR): New internal functions.
1942 * internal-fn.h (get_conditional_internal_fn): Declare.
1943 * internal-fn.c (cond_binary_direct): New macro.
1944 (expand_cond_binary_optab_fn): Likewise.
1945 (direct_cond_binary_optab_supported_p): Likewise.
1946 (get_conditional_internal_fn): New function.
1947 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
1948 Cope with reduction statements that are vectorized as calls rather
1950 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
1951 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
1952 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
1953 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
1954 (UNSPEC_COND_EOR): New unspecs.
1955 (optab): Add mappings for them.
1956 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
1957 (sve_int_op, sve_fp_op): New int attributes.
1959 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1960 Alan Hayward <alan.hayward@arm.com>
1961 David Sherwood <david.sherwood@arm.com>
1963 * optabs.def (while_ult_optab): New optab.
1964 * doc/md.texi (while_ult@var{m}@var{n}): Document.
1965 * internal-fn.def (WHILE_ULT): New internal function.
1966 * internal-fn.h (direct_internal_fn_supported_p): New override
1967 that takes two types as argument.
1968 * internal-fn.c (while_direct): New macro.
1969 (expand_while_optab_fn): New function.
1970 (convert_optab_supported_p): Likewise.
1971 (direct_while_optab_supported_p): New macro.
1972 * wide-int.h (wi::udiv_ceil): New function.
1973 * tree-vectorizer.h (rgroup_masks): New structure.
1974 (vec_loop_masks): New typedef.
1975 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
1977 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
1978 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
1979 (vect_max_vf): New function.
1980 (slpeel_make_loop_iterate_ntimes): Delete.
1981 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
1982 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
1983 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
1984 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
1985 internal-fn.h, stor-layout.h and optabs-query.h.
1986 (vect_set_loop_mask): New function.
1987 (add_preheader_seq): Likewise.
1988 (add_header_seq): Likewise.
1989 (interleave_supported_p): Likewise.
1990 (vect_maybe_permute_loop_masks): Likewise.
1991 (vect_set_loop_masks_directly): Likewise.
1992 (vect_set_loop_condition_masked): Likewise.
1993 (vect_set_loop_condition_unmasked): New function, split out from
1994 slpeel_make_loop_iterate_ntimes.
1995 (slpeel_make_loop_iterate_ntimes): Rename to..
1996 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
1997 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
1998 (vect_do_peeling): Update call accordingly.
1999 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2001 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2002 mask_compare_type, can_fully_mask_p and fully_masked_p.
2003 (release_vec_loop_masks): New function.
2004 (_loop_vec_info): Use it to free the loop masks.
2005 (can_produce_all_loop_masks_p): New function.
2006 (vect_get_max_nscalars_per_iter): Likewise.
2007 (vect_verify_full_masking): Likewise.
2008 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2009 retries, and free the mask rgroups before retrying. Check loop-wide
2010 reasons for disallowing fully-masked loops. Make the final decision
2011 about whether use a fully-masked loop or not.
2012 (vect_estimate_min_profitable_iters): Do not assume that peeling
2013 for the number of iterations will be needed for fully-masked loops.
2014 (vectorizable_reduction): Disable fully-masked loops.
2015 (vectorizable_live_operation): Likewise.
2016 (vect_halve_mask_nunits): New function.
2017 (vect_double_mask_nunits): Likewise.
2018 (vect_record_loop_mask): Likewise.
2019 (vect_get_loop_mask): Likewise.
2020 (vect_transform_loop): Handle the case in which the final loop
2021 iteration might handle a partial vector. Call vect_set_loop_condition
2022 instead of slpeel_make_loop_iterate_ntimes.
2023 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2024 (check_load_store_masking): New function.
2025 (prepare_load_store_mask): Likewise.
2026 (vectorizable_store): Handle fully-masked loops.
2027 (vectorizable_load): Likewise.
2028 (supportable_widening_operation): Use vect_halve_mask_nunits for
2030 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2031 (vect_gen_while): New function.
2032 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2033 (aarch64_uqdec<mode>): New insn.
2035 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2036 Alan Hayward <alan.hayward@arm.com>
2037 David Sherwood <david.sherwood@arm.com>
2039 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2040 (reduc_xor_scal_optab): New optabs.
2041 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2042 (reduc_xor_scal_@var{m}): Document.
2043 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2044 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2046 * fold-const-call.c (fold_const_call): Handle them.
2047 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2048 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2049 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2050 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2051 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2052 (UNSPEC_XORV): New unspecs.
2053 (optab): Add entries for them.
2054 (BITWISEV): New int iterator.
2055 (bit_reduc_op): New int attributes.
2057 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2058 Alan Hayward <alan.hayward@arm.com>
2059 David Sherwood <david.sherwood@arm.com>
2061 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2062 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2063 * optabs.def (vec_shl_insert_optab): New optab.
2064 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2065 (duplicate_and_interleave): Likewise.
2066 * tree-vect-loop.c: Include internal-fn.h.
2067 (neutral_op_for_slp_reduction): New function, split out from
2068 get_initial_defs_for_reduction.
2069 (get_initial_def_for_reduction): Handle option 2 for variable-length
2070 vectors by loading the neutral value into a vector and then shifting
2071 the initial value into element 0.
2072 (get_initial_defs_for_reduction): Replace the code argument with
2073 the neutral value calculated by neutral_op_for_slp_reduction.
2074 Use gimple_build_vector for constant-length vectors.
2075 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2076 but the first group_size elements have a neutral value.
2077 Use duplicate_and_interleave otherwise.
2078 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2079 Update call to get_initial_defs_for_reduction. Handle SLP
2080 reductions for variable-length vectors by creating one vector
2081 result for each scalar result, with the elements associated
2082 with other scalar results stubbed out with the neutral value.
2083 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2084 Require IFN_VEC_SHL_INSERT for double reductions on
2085 variable-length vectors, or SLP reductions that have
2086 a neutral value. Require can_duplicate_and_interleave_p
2087 support for variable-length unchained SLP reductions if there
2088 is no neutral value, such as for MIN/MAX reductions. Also require
2089 the number of vector elements to be a multiple of the number of
2090 SLP statements when doing variable-length unchained SLP reductions.
2091 Update call to vect_create_epilog_for_reduction.
2092 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2093 and remove initial values.
2094 (duplicate_and_interleave): Make public.
2095 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2096 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2098 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2099 Alan Hayward <alan.hayward@arm.com>
2100 David Sherwood <david.sherwood@arm.com>
2102 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2103 (can_duplicate_and_interleave_p): New function.
2104 (vect_get_and_check_slp_defs): Take the vector of statements
2105 rather than just the current one. Remove excess parentheses.
2106 Restriction rejectinon of vect_constant_def and vect_external_def
2107 for variable-length vectors to boolean types, or types for which
2108 can_duplicate_and_interleave_p is false.
2109 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2110 (duplicate_and_interleave): New function.
2111 (vect_get_constant_vectors): Use gimple_build_vector for
2112 constant-length vectors and suitable variable-length constant
2113 vectors. Use duplicate_and_interleave for other variable-length
2114 vectors. Don't defer the update when inserting new statements.
2116 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2117 Alan Hayward <alan.hayward@arm.com>
2118 David Sherwood <david.sherwood@arm.com>
2120 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2121 min_profitable_iters doesn't go negative.
2123 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2124 Alan Hayward <alan.hayward@arm.com>
2125 David Sherwood <david.sherwood@arm.com>
2127 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2128 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2129 * optabs.def (vec_mask_load_lanes_optab): New optab.
2130 (vec_mask_store_lanes_optab): Likewise.
2131 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2132 (MASK_STORE_LANES): Likewise.
2133 * internal-fn.c (mask_load_lanes_direct): New macro.
2134 (mask_store_lanes_direct): Likewise.
2135 (expand_mask_load_optab_fn): Handle masked operations.
2136 (expand_mask_load_lanes_optab_fn): New macro.
2137 (expand_mask_store_optab_fn): Handle masked operations.
2138 (expand_mask_store_lanes_optab_fn): New macro.
2139 (direct_mask_load_lanes_optab_supported_p): Likewise.
2140 (direct_mask_store_lanes_optab_supported_p): Likewise.
2141 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2143 (vect_load_lanes_supported): Likewise.
2144 * tree-vect-data-refs.c (strip_conversion): New function.
2145 (can_group_stmts_p): Likewise.
2146 (vect_analyze_data_ref_accesses): Use it instead of checking
2147 for a pair of assignments.
2148 (vect_store_lanes_supported): Take a masked_p parameter.
2149 (vect_load_lanes_supported): Likewise.
2150 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2151 vect_store_lanes_supported and vect_load_lanes_supported.
2152 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2153 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2154 parameter. Don't allow gaps for masked accesses.
2155 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2156 and vect_load_lanes_supported.
2157 (get_load_store_type): Take a masked_p parameter and update
2158 call to get_group_load_store_type.
2159 (vectorizable_store): Update call to get_load_store_type.
2160 Handle IFN_MASK_STORE_LANES.
2161 (vectorizable_load): Update call to get_load_store_type.
2162 Handle IFN_MASK_LOAD_LANES.
2164 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2165 Alan Hayward <alan.hayward@arm.com>
2166 David Sherwood <david.sherwood@arm.com>
2168 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2170 * config/aarch64/aarch64-protos.h
2171 (aarch64_sve_struct_memory_operand_p): Declare.
2172 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2173 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2174 (VPRED, vpred): Handle SVE structure modes.
2175 * config/aarch64/constraints.md (Utx): New constraint.
2176 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2177 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2178 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2179 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2180 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2181 structure modes. Split into pieces after RA.
2182 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2183 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2185 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2186 SVE structure modes.
2187 (aarch64_classify_address): Likewise.
2188 (sizetochar): Move earlier in file.
2189 (aarch64_print_operand): Handle SVE register lists.
2190 (aarch64_array_mode): New function.
2191 (aarch64_sve_struct_memory_operand_p): Likewise.
2192 (TARGET_ARRAY_MODE): Redefine.
2194 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2195 Alan Hayward <alan.hayward@arm.com>
2196 David Sherwood <david.sherwood@arm.com>
2198 * target.def (array_mode): New target hook.
2199 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2200 * doc/tm.texi: Regenerate.
2201 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2202 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2203 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2205 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2208 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2209 Alan Hayward <alan.hayward@arm.com>
2210 David Sherwood <david.sherwood@arm.com>
2212 * fold-const.c (fold_binary_loc): Check the argument types
2213 rather than the result type when testing for a vector operation.
2215 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2217 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2218 * doc/tm.texi: Regenerate.
2220 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2221 Alan Hayward <alan.hayward@arm.com>
2222 David Sherwood <david.sherwood@arm.com>
2224 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2225 (sve): Document new AArch64 extension.
2226 * doc/md.texi (w): Extend the description of the AArch64
2227 constraint to include SVE vectors.
2228 (Upl, Upa): Document new AArch64 predicate constraints.
2229 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2231 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2232 (msve-vector-bits=): New option.
2233 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2234 SVE when these are disabled.
2235 (sve): New extension.
2236 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2237 modes. Adjust their number of units based on aarch64_sve_vg.
2238 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2239 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2240 aarch64_addr_query_type.
2241 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2242 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2243 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2244 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2245 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2246 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2247 (aarch64_simd_imm_zero_p): Delete.
2248 (aarch64_check_zero_based_sve_index_immediate): Declare.
2249 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2250 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2251 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2252 (aarch64_sve_float_mul_immediate_p): Likewise.
2253 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2255 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2256 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2257 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2258 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2259 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2260 (aarch64_regmode_natural_size): Likewise.
2261 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2262 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2264 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2265 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2266 for VG and the SVE predicate registers.
2267 (V_ALIASES): Add a "z"-prefixed alias.
2268 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2269 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2270 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2271 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2272 (REG_CLASS_NAMES): Add entries for them.
2273 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2274 and the predicate registers.
2275 (aarch64_sve_vg): Declare.
2276 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2277 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2278 (REGMODE_NATURAL_SIZE): Define.
2279 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2281 * config/aarch64/aarch64.c: Include cfgrtl.h.
2282 (simd_immediate_info): Add a constructor for series vectors,
2283 and an associated step field.
2284 (aarch64_sve_vg): New variable.
2285 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2286 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2287 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2288 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2289 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2290 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2291 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2292 (aarch64_get_mask_mode): New functions.
2293 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2294 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2295 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2296 predicate modes and predicate registers. Explicitly restrict
2297 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2298 to store a vector mode if it is recognized by
2299 aarch64_classify_vector_mode.
2300 (aarch64_regmode_natural_size): New function.
2301 (aarch64_hard_regno_caller_save_mode): Return the original mode
2303 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2304 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2305 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2306 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2308 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2309 does not overlap dest if the function is frame-related. Handle
2311 (aarch64_split_add_offset): New function.
2312 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2313 them aarch64_add_offset.
2314 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2315 and update call to aarch64_sub_sp.
2316 (aarch64_add_cfa_expression): New function.
2317 (aarch64_expand_prologue): Pass extra temporary registers to the
2318 functions above. Handle the case in which we need to emit new
2319 DW_CFA_expressions for registers that were originally saved
2320 relative to the stack pointer, but now have to be expressed
2321 relative to the frame pointer.
2322 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2324 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2325 IP0 and IP1 values for SVE frames.
2326 (aarch64_expand_vec_series): New function.
2327 (aarch64_expand_sve_widened_duplicate): Likewise.
2328 (aarch64_expand_sve_const_vector): Likewise.
2329 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2330 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2331 into the register, rather than emitting a SET directly.
2332 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2333 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2334 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2335 (offset_9bit_signed_scaled_p): New functions.
2336 (aarch64_replicate_bitmask_imm): New function.
2337 (aarch64_bitmask_imm): Use it.
2338 (aarch64_cannot_force_const_mem): Reject expressions involving
2339 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2340 (aarch64_classify_index): Handle SVE indices, by requiring
2341 a plain register index with a scale that matches the element size.
2342 (aarch64_classify_address): Handle SVE addresses. Assert that
2343 the mode of the address is VOIDmode or an integer mode.
2344 Update call to aarch64_classify_symbol.
2345 (aarch64_classify_symbolic_expression): Update call to
2346 aarch64_classify_symbol.
2347 (aarch64_const_vec_all_in_range_p): New function.
2348 (aarch64_print_vector_float_operand): Likewise.
2349 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2350 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2351 and the FP immediates 1.0 and 0.5.
2352 (aarch64_print_address_internal): Handle SVE addresses.
2353 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2354 (aarch64_regno_regclass): Handle predicate registers.
2355 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2357 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2358 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2359 (aarch64_convert_sve_vector_bits): New function.
2360 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2361 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2363 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2364 Handle SVE vector and predicate modes. Accept VL-based constants
2365 that need only one temporary register, and VL offsets that require
2366 no temporary registers.
2367 (aarch64_conditional_register_usage): Mark the predicate registers
2368 as fixed if SVE isn't available.
2369 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2370 Return true for SVE vector and predicate modes.
2371 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2372 rather than an unsigned int. Handle SVE modes.
2373 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2375 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2377 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2378 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2379 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2380 (aarch64_sve_float_mul_immediate_p): New functions.
2381 (aarch64_sve_valid_immediate): New function.
2382 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2383 Explicitly reject structure modes. Check for INDEX constants.
2384 Handle PTRUE and PFALSE constants.
2385 (aarch64_check_zero_based_sve_index_immediate): New function.
2386 (aarch64_simd_imm_zero_p): Delete.
2387 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2388 vector modes. Accept constants in the range of CNT[BHWD].
2389 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2390 ask for an Advanced SIMD mode.
2391 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2392 (aarch64_simd_vector_alignment): Handle SVE predicates.
2393 (aarch64_vectorize_preferred_vector_alignment): New function.
2394 (aarch64_simd_vector_alignment_reachable): Use it instead of
2396 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2397 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2399 (MAX_VECT_LEN): Delete.
2400 (expand_vec_perm_d): Add a vec_flags field.
2401 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2402 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2403 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2405 (aarch64_evpc_rev): Rename to...
2406 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2407 (aarch64_evpc_rev_global): New function.
2408 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2409 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2411 (aarch64_evpc_sve_tbl): New function.
2412 (aarch64_expand_vec_perm_const_1): Update after rename of
2413 aarch64_evpc_rev. Handle SVE permutes too, trying
2414 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2415 than aarch64_evpc_tbl.
2416 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2417 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2418 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2419 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2420 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2421 (aarch64_expand_sve_vcond): New functions.
2422 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2423 of aarch64_vector_mode_p.
2424 (aarch64_dwarf_poly_indeterminate_value): New function.
2425 (aarch64_compute_pressure_classes): Likewise.
2426 (aarch64_can_change_mode_class): Likewise.
2427 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2428 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2429 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2430 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2431 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2432 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2433 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2434 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2436 (Dn, Dl, Dr): Accept const as well as const_vector.
2437 (Dz): Likewise. Compare against CONST0_RTX.
2438 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2439 of "vector" where appropriate.
2440 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2441 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2442 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2443 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2444 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2445 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2446 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2447 (v_int_equiv): Extend to SVE modes.
2448 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2450 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2451 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2452 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2453 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2454 (SVE_COND_FP_CMP): New int iterators.
2455 (perm_hilo): Handle the new unpack unspecs.
2456 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2458 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2459 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2460 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2461 (aarch64_equality_operator, aarch64_constant_vector_operand)
2462 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2463 (aarch64_sve_nonimmediate_operand): Likewise.
2464 (aarch64_sve_general_operand): Likewise.
2465 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2466 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2467 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2468 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2469 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2470 (aarch64_sve_float_arith_immediate): Likewise.
2471 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2472 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2473 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2474 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2475 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2476 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2477 (aarch64_sve_float_arith_operand): Likewise.
2478 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2479 (aarch64_sve_float_mul_operand): Likewise.
2480 (aarch64_sve_vec_perm_operand): Likewise.
2481 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2482 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2483 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2484 as well as const_vector.
2485 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2486 in file. Use CONST0_RTX and CONSTM1_RTX.
2487 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2488 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2489 Use aarch64_simd_imm_zero.
2490 * config/aarch64/aarch64-sve.md: New file.
2491 * config/aarch64/aarch64.md: Include it.
2492 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2493 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2494 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2495 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2496 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2497 (sve): New attribute.
2498 (enabled): Disable instructions with the sve attribute unless
2500 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2501 aarch64_expand_mov_immediate.
2502 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2503 CNT[BHSD] immediates.
2504 (movti): Split CONST_POLY_INT moves into two halves.
2505 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2506 Split additions that need a temporary here if the destination
2507 is the stack pointer.
2508 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2509 (*add<mode>3_poly_1): New instruction.
2510 (set_clobber_cc): New expander.
2512 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2514 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2515 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2516 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2517 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2518 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2519 Change innermode from fixed_mode_size to machine_mode.
2520 (simplify_subreg): Update call accordingly. Handle a constant-sized
2521 subreg of a variable-length CONST_VECTOR.
2523 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2524 Alan Hayward <alan.hayward@arm.com>
2525 David Sherwood <david.sherwood@arm.com>
2527 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2528 (add_offset_to_base): New function, split out from...
2529 (create_mem_ref): ...here. When handling a scale other than 1,
2530 check first whether the address is valid without the offset.
2531 Add it into the base if so, leaving the index and scale as-is.
2533 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2536 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2537 fold_for_warn before checking if arg2 is INTEGER_CST.
2539 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2541 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2542 (store_multiple_operation): Delete.
2543 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2544 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2545 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2546 guarded by TARGET_STRING.
2547 (rs6000_output_load_multiple): Delete.
2548 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2549 OPTION_MASK_STRING / TARGET_STRING handling.
2550 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2551 (const rs6000_opt_masks) <"string">: Change mask to 0.
2552 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2553 (MASK_STRING): Delete.
2554 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2556 (load_multiple): Delete.
2563 (store_multiple): Delete.
2570 (movmemsi_8reg): Delete.
2571 (corresponding unnamed define_insn): Delete.
2572 (movmemsi_6reg): Delete.
2573 (corresponding unnamed define_insn): Delete.
2574 (movmemsi_4reg): Delete.
2575 (corresponding unnamed define_insn): Delete.
2576 (movmemsi_2reg): Delete.
2577 (corresponding unnamed define_insn): Delete.
2578 (movmemsi_1reg): Delete.
2579 (corresponding unnamed define_insn): Delete.
2580 * config/rs6000/rs6000.opt (mno-string): New.
2581 (mstring): Replace by deprecation warning stub.
2582 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2584 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2586 * regrename.c (regrename_do_replace): If replacing the same
2587 reg multiple times, try to reuse last created gen_raw_REG.
2590 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2591 main to workaround a bug in GDB.
2593 2018-01-12 Tom de Vries <tom@codesourcery.com>
2596 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2598 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2600 PR rtl-optimization/80481
2601 * ira-color.c (get_cap_member): New function.
2602 (allocnos_conflict_by_live_ranges_p): Use it.
2603 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2604 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2606 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2609 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2610 (*saddl_se_1): Ditto.
2612 (*ssubl_se_1): Ditto.
2614 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2616 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2617 rather than wi::to_widest for DR_INITs.
2618 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2619 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2620 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2622 (vect_analyze_group_access_1): Note that here.
2624 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2626 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2627 polynomial type sizes.
2629 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2631 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2632 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2633 (gimple_add_tmp_var): Likewise.
2635 2018-01-12 Martin Liska <mliska@suse.cz>
2637 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2638 (gimple_alloc_sizes): Likewise.
2639 (dump_gimple_statistics): Use PRIu64 in printf format.
2640 * gimple.h: Change uint64_t to int.
2642 2018-01-12 Martin Liska <mliska@suse.cz>
2644 * tree-core.h: Use uint64_t instead of int.
2645 * tree.c (tree_node_counts): Likewise.
2646 (tree_node_sizes): Likewise.
2647 (dump_tree_statistics): Use PRIu64 in printf format.
2649 2018-01-12 Martin Liska <mliska@suse.cz>
2651 * Makefile.in: As qsort_chk is implemented in vec.c, add
2652 vec.o to linkage of gencfn-macros.
2653 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2654 passing the info to record_node_allocation_statistics.
2655 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2657 * ggc-common.c (struct ggc_usage): Add operator== and use
2658 it in operator< and compare function.
2659 * mem-stats.h (struct mem_usage): Likewise.
2660 * vec.c (struct vec_usage): Remove operator< and compare
2661 function. Can be simply inherited.
2663 2018-01-12 Martin Jambor <mjambor@suse.cz>
2666 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2667 * tree-ssa-math-opts.c: Include domwalk.h.
2668 (convert_mult_to_fma_1): New function.
2669 (fma_transformation_info): New type.
2670 (fma_deferring_state): Likewise.
2671 (cancel_fma_deferring): New function.
2672 (result_of_phi): Likewise.
2673 (last_fma_candidate_feeds_initial_phi): Likewise.
2674 (convert_mult_to_fma): Added deferring logic, split actual
2675 transformation to convert_mult_to_fma_1.
2676 (math_opts_dom_walker): New type.
2677 (math_opts_dom_walker::after_dom_children): New method, body moved
2678 here from pass_optimize_widening_mul::execute, added deferring logic
2680 (pass_optimize_widening_mul::execute): Moved most of code to
2681 math_opts_dom_walker::after_dom_children.
2682 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2683 * config/i386/i386.c (ix86_option_override_internal): Added
2684 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2686 2018-01-12 Richard Biener <rguenther@suse.de>
2689 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2690 inline instance vars.
2692 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2695 * config/rx/rx.c (rx_is_restricted_memory_address):
2698 2018-01-12 Richard Biener <rguenther@suse.de>
2700 PR tree-optimization/80846
2701 * target.def (split_reduction): New target hook.
2702 * targhooks.c (default_split_reduction): New function.
2703 * targhooks.h (default_split_reduction): Declare.
2704 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2705 target requests first reduce vectors by combining low and high
2707 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2708 (get_vectype_for_scalar_type_and_size): Export.
2709 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2710 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2711 * doc/tm.texi: Regenerate.
2712 * config/i386/i386.c (ix86_split_reduction): Implement
2713 TARGET_VECTORIZE_SPLIT_REDUCTION.
2715 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2718 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2719 in PIC mode except for TARGET_VXWORKS_RTP.
2720 * config/sparc/sparc.c: Include cfgrtl.h.
2721 (TARGET_INIT_PIC_REG): Define.
2722 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2723 (sparc_pic_register_p): New predicate.
2724 (sparc_legitimate_address_p): Use it.
2725 (sparc_legitimize_pic_address): Likewise.
2726 (sparc_delegitimize_address): Likewise.
2727 (sparc_mode_dependent_address_p): Likewise.
2728 (gen_load_pcrel_sym): Remove 4th parameter.
2729 (load_got_register): Adjust call to above. Remove obsolete stuff.
2730 (sparc_expand_prologue): Do not call load_got_register here.
2731 (sparc_flat_expand_prologue): Likewise.
2732 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2733 (sparc_use_pseudo_pic_reg): New function.
2734 (sparc_init_pic_reg): Likewise.
2735 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2736 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2738 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2740 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2741 Add item for branch_cost.
2743 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2745 PR rtl-optimization/83565
2746 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2747 not extend the result to a larger mode for rotate operations.
2748 (num_sign_bit_copies1): Likewise.
2750 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2753 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2755 Use values-Xc.o for -pedantic.
2756 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2758 2018-01-12 Martin Liska <mliska@suse.cz>
2761 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2763 (possible_polymorphic_call_targets): Use it.
2764 (ipa_devirt): Likewise.
2766 2018-01-12 Martin Liska <mliska@suse.cz>
2768 * profile-count.h (enum profile_quality): Use 0 as invalid
2769 enum value of profile_quality.
2771 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2773 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2774 -mext-string options.
2776 2018-01-12 Richard Biener <rguenther@suse.de>
2778 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2779 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2780 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2782 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2784 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2786 * configure.ac (--with-long-double-format): Add support for the
2787 configuration option to change the default long double format on
2789 * config.gcc (powerpc*-linux*-*): Likewise.
2790 * configure: Regenerate.
2791 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2792 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2793 used without modification.
2795 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2797 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2798 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2799 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2800 MISC_BUILTIN_SPEC_BARRIER.
2801 (rs6000_init_builtins): Likewise.
2802 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2804 (speculation_barrier): New define_insn.
2805 * doc/extend.texi: Document __builtin_speculation_barrier.
2807 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2810 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2811 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2812 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2814 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2815 integral modes instead of "ss" and "sd".
2816 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2817 vectors with 32-bit and 64-bit elements.
2818 (vecdupssescalarmodesuffix): New mode attribute.
2819 (vec_dup<mode>): Use it.
2821 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2824 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2825 frame if argument is passed on stack.
2827 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2830 * ree.c (combine_reaching_defs): Optimize also
2831 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
2832 reg2=any_extend(exp); reg1=reg2;, formatting fix.
2834 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2837 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
2839 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2842 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
2843 after they are computed.
2845 2018-01-11 Bin Cheng <bin.cheng@arm.com>
2847 PR tree-optimization/83695
2848 * gimple-loop-linterchange.cc
2849 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
2850 reset cached scev information after interchange.
2851 (pass_linterchange::execute): Remove call to scev_reset_htab.
2853 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2855 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
2856 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
2857 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
2858 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
2859 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
2860 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
2861 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
2862 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
2863 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
2864 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
2865 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
2866 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
2867 (V_lane_reg): Likewise.
2868 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
2870 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
2871 (vfmal_lane_low<mode>_intrinsic,
2872 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
2873 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
2874 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
2875 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
2876 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
2877 vfmsl_lane_high<mode>_intrinsic): New define_insns.
2879 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2881 * config/arm/arm-cpus.in (fp16fml): New feature.
2882 (ALL_SIMD): Add fp16fml.
2883 (armv8.2-a): Add fp16fml as an option.
2884 (armv8.3-a): Likewise.
2885 (armv8.4-a): Add fp16fml as part of fp16.
2886 * config/arm/arm.h (TARGET_FP16FML): Define.
2887 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
2889 * config/arm/arm-modes.def (V2HF): Define.
2890 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
2891 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
2892 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
2893 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
2894 vfmsl_low, vfmsl_high): New set of builtins.
2895 * config/arm/iterators.md (PLUSMINUS): New code iterator.
2896 (vfml_op): New code attribute.
2897 (VFMLHALVES): New int iterator.
2898 (VFML, VFMLSEL): New mode attributes.
2899 (V_reg): Define mapping for V2HF.
2900 (V_hi, V_lo): New mode attributes.
2901 (VF_constraint): Likewise.
2902 (vfml_half, vfml_half_selector): New int attributes.
2903 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
2905 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
2906 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
2908 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
2909 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
2910 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
2911 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
2913 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
2914 Document new effective target and option set.
2916 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2918 * config/arm/arm-cpus.in (armv8_4): New feature.
2919 (ARMv8_4a): New fgroup.
2920 (armv8.4-a): New arch.
2921 * config/arm/arm-tables.opt: Regenerate.
2922 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
2923 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
2924 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
2925 Add matching rules for -march=armv8.4-a and extensions.
2926 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
2928 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
2931 * config/rx/rx.md (BW): New mode attribute.
2932 (sync_lock_test_and_setsi): Add mode suffix to insn output.
2934 2018-01-11 Richard Biener <rguenther@suse.de>
2936 PR tree-optimization/83435
2937 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
2938 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
2939 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
2941 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2942 Alan Hayward <alan.hayward@arm.com>
2943 David Sherwood <david.sherwood@arm.com>
2945 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
2947 (aarch64_classify_address): Initialize it. Track polynomial offsets.
2948 (aarch64_print_address_internal): Use it to check for a zero offset.
2950 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2951 Alan Hayward <alan.hayward@arm.com>
2952 David Sherwood <david.sherwood@arm.com>
2954 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
2955 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
2956 Return a poly_int64 rather than a HOST_WIDE_INT.
2957 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
2958 rather than a HOST_WIDE_INT.
2959 * config/aarch64/aarch64.h (aarch64_frame): Protect with
2960 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
2961 hard_fp_offset, frame_size, initial_adjust, callee_offset and
2962 final_offset from HOST_WIDE_INT to poly_int64.
2963 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
2964 to_constant when getting the number of units in an Advanced SIMD
2966 (aarch64_builtin_vectorized_function): Check for a constant number
2968 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
2970 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
2971 attribute instead of GET_MODE_NUNITS.
2972 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
2973 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
2974 GET_MODE_SIZE for fixed-size registers.
2975 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
2976 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
2977 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
2978 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
2979 (aarch64_print_operand, aarch64_print_address_internal)
2980 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
2981 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
2982 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
2983 Handle polynomial GET_MODE_SIZE.
2984 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
2985 wider than SImode without modification.
2986 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
2987 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
2988 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
2989 passing and returning SVE modes.
2990 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
2991 rather than GEN_INT.
2992 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
2993 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
2994 (aarch64_allocate_and_probe_stack_space): Likewise.
2995 (aarch64_layout_frame): Cope with polynomial offsets.
2996 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
2997 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
2999 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3000 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3001 poly_int64 rather than a HOST_WIDE_INT.
3002 (aarch64_get_separate_components, aarch64_process_components)
3003 (aarch64_expand_prologue, aarch64_expand_epilogue)
3004 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3005 (aarch64_anchor_offset): New function, split out from...
3006 (aarch64_legitimize_address): ...here.
3007 (aarch64_builtin_vectorization_cost): Handle polynomial
3008 TYPE_VECTOR_SUBPARTS.
3009 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3011 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3012 number of elements from the PARALLEL rather than the mode.
3013 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3014 rather than GET_MODE_BITSIZE.
3015 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3016 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3017 (aarch64_expand_vec_perm_const_1): Handle polynomial
3018 d->perm.length () and d->perm elements.
3019 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3020 Apply to_constant to d->perm elements.
3021 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3022 polynomial CONST_VECTOR_NUNITS.
3023 (aarch64_move_pointer): Take amount as a poly_int64 rather
3025 (aarch64_progress_pointer): Avoid temporary variable.
3026 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3027 the mode attribute instead of GET_MODE.
3029 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3030 Alan Hayward <alan.hayward@arm.com>
3031 David Sherwood <david.sherwood@arm.com>
3033 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3034 x exists before using it.
3035 (aarch64_add_constant_internal): Rename to...
3036 (aarch64_add_offset_1): ...this. Replace regnum with separate
3037 src and dest rtxes. Handle the case in which they're different,
3038 including when the offset is zero. Replace scratchreg with an rtx.
3039 Use 2 additions if there is no spare register into which we can
3040 move a 16-bit constant.
3041 (aarch64_add_constant): Delete.
3042 (aarch64_add_offset): Replace reg with separate src and dest
3043 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3044 Use aarch64_add_offset_1.
3045 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3046 an rtx rather than an int. Take the delta as a poly_int64
3047 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3048 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3049 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3050 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3051 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3053 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3054 aarch64_add_constant.
3056 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3058 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3059 Use scalar_float_mode.
3061 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3063 * config/aarch64/aarch64-simd.md
3064 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3065 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3066 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3067 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3068 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3069 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3070 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3071 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3072 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3073 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3075 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3078 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3079 targ_options->x_arm_arch_string is non NULL.
3081 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3083 * config/aarch64/aarch64.h
3084 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3086 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3089 * expmed.c (emit_store_flag_force): Swap if const op0
3090 and change VOIDmode to mode of op0.
3092 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3094 PR rtl-optimization/83761
3095 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3096 than bytes to mode_for_size.
3098 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3101 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3102 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3105 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3108 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3109 when in layout mode.
3110 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3111 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3114 2018-01-10 Michael Collison <michael.collison@arm.com>
3116 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3117 * config/aarch64/aarch64-option-extension.def: Add
3118 AARCH64_OPT_EXTENSION of 'fp16fml'.
3119 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3120 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3121 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3122 * config/aarch64/constraints.md (Ui7): New constraint.
3123 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3124 (VFMLA_SEL_W): Ditto.
3127 (VFMLA16_LOW): New int iterator.
3128 (VFMLA16_HIGH): Ditto.
3129 (UNSPEC_FMLAL): New unspec.
3130 (UNSPEC_FMLSL): Ditto.
3131 (UNSPEC_FMLAL2): Ditto.
3132 (UNSPEC_FMLSL2): Ditto.
3133 (f16mac): New code attribute.
3134 * config/aarch64/aarch64-simd-builtins.def
3135 (aarch64_fmlal_lowv2sf): Ditto.
3136 (aarch64_fmlsl_lowv2sf): Ditto.
3137 (aarch64_fmlalq_lowv4sf): Ditto.
3138 (aarch64_fmlslq_lowv4sf): Ditto.
3139 (aarch64_fmlal_highv2sf): Ditto.
3140 (aarch64_fmlsl_highv2sf): Ditto.
3141 (aarch64_fmlalq_highv4sf): Ditto.
3142 (aarch64_fmlslq_highv4sf): Ditto.
3143 (aarch64_fmlal_lane_lowv2sf): Ditto.
3144 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3145 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3146 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3147 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3148 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3149 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3150 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3151 (aarch64_fmlal_lane_highv2sf): Ditto.
3152 (aarch64_fmlsl_lane_highv2sf): Ditto.
3153 (aarch64_fmlal_laneq_highv2sf): Ditto.
3154 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3155 (aarch64_fmlalq_lane_highv4sf): Ditto.
3156 (aarch64_fmlsl_lane_highv4sf): Ditto.
3157 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3158 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3159 * config/aarch64/aarch64-simd.md:
3160 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3161 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3162 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3163 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3164 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3165 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3166 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3167 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3168 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3169 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3170 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3171 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3172 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3173 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3174 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3175 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3176 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3177 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3178 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3179 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3180 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3181 (vfmlsl_low_u32): Ditto.
3182 (vfmlalq_low_u32): Ditto.
3183 (vfmlslq_low_u32): Ditto.
3184 (vfmlal_high_u32): Ditto.
3185 (vfmlsl_high_u32): Ditto.
3186 (vfmlalq_high_u32): Ditto.
3187 (vfmlslq_high_u32): Ditto.
3188 (vfmlal_lane_low_u32): Ditto.
3189 (vfmlsl_lane_low_u32): Ditto.
3190 (vfmlal_laneq_low_u32): Ditto.
3191 (vfmlsl_laneq_low_u32): Ditto.
3192 (vfmlalq_lane_low_u32): Ditto.
3193 (vfmlslq_lane_low_u32): Ditto.
3194 (vfmlalq_laneq_low_u32): Ditto.
3195 (vfmlslq_laneq_low_u32): Ditto.
3196 (vfmlal_lane_high_u32): Ditto.
3197 (vfmlsl_lane_high_u32): Ditto.
3198 (vfmlal_laneq_high_u32): Ditto.
3199 (vfmlsl_laneq_high_u32): Ditto.
3200 (vfmlalq_lane_high_u32): Ditto.
3201 (vfmlslq_lane_high_u32): Ditto.
3202 (vfmlalq_laneq_high_u32): Ditto.
3203 (vfmlslq_laneq_high_u32): Ditto.
3204 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3205 (AARCH64_FL_FOR_ARCH8_4): New.
3206 (AARCH64_ISA_F16FML): New ISA flag.
3207 (TARGET_F16FML): New feature flag for fp16fml.
3208 (doc/invoke.texi): Document new fp16fml option.
3210 2018-01-10 Michael Collison <michael.collison@arm.com>
3212 * config/aarch64/aarch64-builtins.c:
3213 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3214 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3215 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3216 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3217 (AARCH64_ISA_SHA3): New ISA flag.
3218 (TARGET_SHA3): New feature flag for sha3.
3219 * config/aarch64/iterators.md (sha512_op): New int attribute.
3220 (CRYPTO_SHA512): New int iterator.
3221 (UNSPEC_SHA512H): New unspec.
3222 (UNSPEC_SHA512H2): Ditto.
3223 (UNSPEC_SHA512SU0): Ditto.
3224 (UNSPEC_SHA512SU1): Ditto.
3225 * config/aarch64/aarch64-simd-builtins.def
3226 (aarch64_crypto_sha512hqv2di): New builtin.
3227 (aarch64_crypto_sha512h2qv2di): Ditto.
3228 (aarch64_crypto_sha512su0qv2di): Ditto.
3229 (aarch64_crypto_sha512su1qv2di): Ditto.
3230 (aarch64_eor3qv8hi): Ditto.
3231 (aarch64_rax1qv2di): Ditto.
3232 (aarch64_xarqv2di): Ditto.
3233 (aarch64_bcaxqv8hi): Ditto.
3234 * config/aarch64/aarch64-simd.md:
3235 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3236 (aarch64_crypto_sha512su0qv2di): Ditto.
3237 (aarch64_crypto_sha512su1qv2di): Ditto.
3238 (aarch64_eor3qv8hi): Ditto.
3239 (aarch64_rax1qv2di): Ditto.
3240 (aarch64_xarqv2di): Ditto.
3241 (aarch64_bcaxqv8hi): Ditto.
3242 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3243 (vsha512h2q_u64): Ditto.
3244 (vsha512su0q_u64): Ditto.
3245 (vsha512su1q_u64): Ditto.
3246 (veor3q_u16): Ditto.
3247 (vrax1q_u64): Ditto.
3249 (vbcaxq_u16): Ditto.
3250 * config/arm/types.md (crypto_sha512): New type attribute.
3251 (crypto_sha3): Ditto.
3252 (doc/invoke.texi): Document new sha3 option.
3254 2018-01-10 Michael Collison <michael.collison@arm.com>
3256 * config/aarch64/aarch64-builtins.c:
3257 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3258 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3259 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3260 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3261 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3262 (AARCH64_ISA_SM4): New ISA flag.
3263 (TARGET_SM4): New feature flag for sm4.
3264 * config/aarch64/aarch64-simd-builtins.def
3265 (aarch64_sm3ss1qv4si): Ditto.
3266 (aarch64_sm3tt1aq4si): Ditto.
3267 (aarch64_sm3tt1bq4si): Ditto.
3268 (aarch64_sm3tt2aq4si): Ditto.
3269 (aarch64_sm3tt2bq4si): Ditto.
3270 (aarch64_sm3partw1qv4si): Ditto.
3271 (aarch64_sm3partw2qv4si): Ditto.
3272 (aarch64_sm4eqv4si): Ditto.
3273 (aarch64_sm4ekeyqv4si): Ditto.
3274 * config/aarch64/aarch64-simd.md:
3275 (aarch64_sm3ss1qv4si): Ditto.
3276 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3277 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3278 (aarch64_sm4eqv4si): Ditto.
3279 (aarch64_sm4ekeyqv4si): Ditto.
3280 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3281 (sm3part_op): Ditto.
3282 (CRYPTO_SM3TT): Ditto.
3283 (CRYPTO_SM3PART): Ditto.
3284 (UNSPEC_SM3SS1): New unspec.
3285 (UNSPEC_SM3TT1A): Ditto.
3286 (UNSPEC_SM3TT1B): Ditto.
3287 (UNSPEC_SM3TT2A): Ditto.
3288 (UNSPEC_SM3TT2B): Ditto.
3289 (UNSPEC_SM3PARTW1): Ditto.
3290 (UNSPEC_SM3PARTW2): Ditto.
3291 (UNSPEC_SM4E): Ditto.
3292 (UNSPEC_SM4EKEY): Ditto.
3293 * config/aarch64/constraints.md (Ui2): New constraint.
3294 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3295 * config/arm/types.md (crypto_sm3): New type attribute.
3296 (crypto_sm4): Ditto.
3297 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3298 (vsm3tt1aq_u32): Ditto.
3299 (vsm3tt1bq_u32): Ditto.
3300 (vsm3tt2aq_u32): Ditto.
3301 (vsm3tt2bq_u32): Ditto.
3302 (vsm3partw1q_u32): Ditto.
3303 (vsm3partw2q_u32): Ditto.
3304 (vsm4eq_u32): Ditto.
3305 (vsm4ekeyq_u32): Ditto.
3306 (doc/invoke.texi): Document new sm4 option.
3308 2018-01-10 Michael Collison <michael.collison@arm.com>
3310 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3311 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3312 (AARCH64_FL_FOR_ARCH8_4): New.
3313 (AARCH64_FL_V8_4): New flag.
3314 (doc/invoke.texi): Document new armv8.4-a option.
3316 2018-01-10 Michael Collison <michael.collison@arm.com>
3318 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3319 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3320 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3321 * config/aarch64/aarch64-option-extension.def: Add
3322 AARCH64_OPT_EXTENSION of 'sha2'.
3323 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3324 (crypto): Disable sha2 and aes if crypto disabled.
3325 (crypto): Enable aes and sha2 if enabled.
3326 (simd): Disable sha2 and aes if simd disabled.
3327 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3329 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3330 (TARGET_SHA2): New feature flag for sha2.
3331 (TARGET_AES): New feature flag for aes.
3332 * config/aarch64/aarch64-simd.md:
3333 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3334 conditional on TARGET_AES.
3335 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3336 (aarch64_crypto_sha1hsi): Make pattern conditional
3338 (aarch64_crypto_sha1hv4si): Ditto.
3339 (aarch64_be_crypto_sha1hv4si): Ditto.
3340 (aarch64_crypto_sha1su1v4si): Ditto.
3341 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3342 (aarch64_crypto_sha1su0v4si): Ditto.
3343 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3344 (aarch64_crypto_sha256su0v4si): Ditto.
3345 (aarch64_crypto_sha256su1v4si): Ditto.
3346 (doc/invoke.texi): Document new aes and sha2 options.
3348 2018-01-10 Martin Sebor <msebor@redhat.com>
3350 PR tree-optimization/83781
3351 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3354 2018-01-11 Martin Sebor <msebor@gmail.com>
3355 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3357 PR tree-optimization/83501
3358 PR tree-optimization/81703
3360 * tree-ssa-strlen.c (get_string_cst): Rename...
3361 (get_string_len): ...to this. Handle global constants.
3362 (handle_char_store): Adjust.
3364 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3365 Jim Wilson <jimw@sifive.com>
3367 * config/riscv/riscv-protos.h (riscv_output_return): New.
3368 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3369 (riscv_attribute_table, riscv_output_return),
3370 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3371 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3372 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3374 (riscv_expand_prologue): Add early return for naked function.
3375 (riscv_expand_epilogue): Likewise.
3376 (riscv_function_ok_for_sibcall): Return false for naked function.
3377 (riscv_set_current_function): New.
3378 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3379 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3380 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3381 * doc/extend.texi (RISC-V Function Attributes): New.
3383 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3385 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3386 check for 128-bit long double before checking TCmode.
3387 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3388 128-bit long doubles before checking TFmode or TCmode.
3389 (FLOAT128_IBM_P): Likewise.
3391 2018-01-10 Martin Sebor <msebor@redhat.com>
3393 PR tree-optimization/83671
3394 * builtins.c (c_strlen): Unconditionally return zero for the empty
3396 Use -Warray-bounds for warnings.
3397 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3398 for non-constant array indices with COMPONENT_REF, arrays of
3399 arrays, and pointers to arrays.
3400 (gimple_fold_builtin_strlen): Determine and set length range for
3401 non-constant character arrays.
3403 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3406 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3409 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3411 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3413 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3416 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3417 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3418 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3419 indexed_or_indirect_operand predicate.
3420 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3421 (*vsx_le_perm_load_v8hi): Likewise.
3422 (*vsx_le_perm_load_v16qi): Likewise.
3423 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3424 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3425 (*vsx_le_perm_store_v8hi): Likewise.
3426 (*vsx_le_perm_store_v16qi): Likewise.
3427 (eight unnamed splitters): Likewise.
3429 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3431 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3432 * config/rs6000/emmintrin.h: Likewise.
3433 * config/rs6000/mmintrin.h: Likewise.
3434 * config/rs6000/xmmintrin.h: Likewise.
3436 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3439 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3441 * tree.c (tree_nop_conversion): Return true for location wrapper
3443 (maybe_wrap_with_location): New function.
3444 (selftest::check_strip_nops): New function.
3445 (selftest::test_location_wrappers): New function.
3446 (selftest::tree_c_tests): Call it.
3447 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3448 (maybe_wrap_with_location): New decl.
3449 (EXPR_LOCATION_WRAPPER_P): New macro.
3450 (location_wrapper_p): New inline function.
3451 (tree_strip_any_location_wrapper): New inline function.
3453 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3456 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3457 stack_realign_offset for the largest alignment of stack slot
3459 (ix86_find_max_used_stack_alignment): New function.
3460 (ix86_finalize_stack_frame_flags): Use it. Set
3461 max_used_stack_alignment if we don't realign stack.
3462 * config/i386/i386.h (machine_function): Add
3463 max_used_stack_alignment.
3465 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3467 * config/arm/arm.opt (-mbranch-cost): New option.
3468 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3471 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3474 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3475 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3477 2018-01-10 Richard Biener <rguenther@suse.de>
3480 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3481 early out so it also covers the case where we have a non-NULL
3484 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3486 PR tree-optimization/83753
3487 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3488 for non-strided grouped accesses if the number of elements is 1.
3490 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3493 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3494 * i386.h (TARGET_USE_GATHER): Define.
3495 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3497 2018-01-10 Martin Liska <mliska@suse.cz>
3500 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3501 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3503 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3504 CLEANUP_NO_PARTITIONING is not set.
3506 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3508 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3509 for vectors, as a partial revert of r254296.
3510 * rtl.h (const_vec_p): Delete.
3511 (const_vec_duplicate_p): Don't test for vector CONSTs.
3512 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3513 * expmed.c (make_tree): Likewise.
3516 * common.md (E, F): Use CONSTANT_P instead of checking for
3518 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3519 checking for CONST_VECTOR.
3521 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3524 * predict.c (force_edge_cold): Handle in more sane way edges
3527 2018-01-09 Carl Love <cel@us.ibm.com>
3529 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3531 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3532 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3533 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3534 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3535 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3536 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3537 * config/rs6000/rs6000-protos.h: Add extern defition for
3538 rs6000_generate_float2_double_code.
3539 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3541 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3542 (float2_v2df): Add define_expand.
3544 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3547 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3548 op_mode in the force_to_mode call.
3550 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3552 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3553 instead of checking each element individually.
3554 (aarch64_evpc_uzp): Likewise.
3555 (aarch64_evpc_zip): Likewise.
3556 (aarch64_evpc_ext): Likewise.
3557 (aarch64_evpc_rev): Likewise.
3558 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3559 instead of checking each element individually. Return true without
3561 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3562 whether all selected elements come from the same input, instead of
3563 checking each element individually. Remove calls to gen_rtx_REG,
3564 start_sequence and end_sequence and instead assert that no rtl is
3567 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3569 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3570 order of HIGH and CONST checks.
3572 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3574 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3575 if the destination isn't an SSA_NAME.
3577 2018-01-09 Richard Biener <rguenther@suse.de>
3579 PR tree-optimization/83668
3580 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3582 (canonicalize_loop_form): ... here, renamed from ...
3583 (canonicalize_loop_closed_ssa_form): ... this and amended to
3584 swap successor edges for loop exit blocks to make us use
3585 the RPO order we need for initial schedule generation.
3587 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3589 PR tree-optimization/64811
3590 * match.pd: When optimizing comparisons with Inf, avoid
3591 introducing or losing exceptions from comparisons with NaN.
3593 2018-01-09 Martin Liska <mliska@suse.cz>
3596 * asan.c (shadow_mem_size): Add gcc_assert.
3598 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3600 Don't save registers in main().
3603 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3604 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3605 * config/avr/avr.c (avr_set_current_function): Don't error if
3606 naked, OS_task or OS_main are specified at the same time.
3607 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3609 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3611 * common/config/avr/avr-common.c (avr_option_optimization_table):
3612 Switch on -mmain-is-OS_task for optimizing compilations.
3614 2018-01-09 Richard Biener <rguenther@suse.de>
3616 PR tree-optimization/83572
3617 * graphite.c: Include cfganal.h.
3618 (graphite_transform_loops): Connect infinite loops to exit
3619 and remove fake edges at the end.
3621 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3623 * ipa-inline.c (edge_badness): Revert accidental checkin.
3625 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3628 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3629 symbols; not inline clones.
3631 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3634 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3635 hard registers. Formatting fixes.
3637 PR preprocessor/83722
3638 * gcc.c (try_generate_repro): Pass
3639 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3640 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3643 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3644 Kito Cheng <kito.cheng@gmail.com>
3646 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3647 (riscv_leaf_function_p): Delete.
3648 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3650 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3652 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3654 (do_ifelse): New function.
3655 (do_isel): New function.
3656 (do_sub3): New function.
3657 (do_add3): New function.
3658 (do_load_mask_compare): New function.
3659 (do_overlap_load_compare): New function.
3660 (expand_compare_loop): New function.
3661 (expand_block_compare): Call expand_compare_loop() when appropriate.
3662 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3664 (-mblock-compare-inline-loop-limit): New option.
3666 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3669 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3670 Reverse order of second and third operands in first alternative.
3671 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3672 of first and second elements in UNSPEC_VPERMR vector.
3673 (altivec_expand_vec_perm_le): Likewise.
3675 2017-01-08 Jeff Law <law@redhat.com>
3677 PR rtl-optimizatin/81308
3678 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3679 (process_switch): If group_case_labels makes a change, then set
3681 (pass_convert_switch::execute): If a switch is converted, then
3682 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3684 PR rtl-optimization/81308
3685 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3688 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3690 PR target/83663 - Revert r255946
3691 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3692 generation for cases where splatting a value is not useful.
3693 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3694 across a vec_duplicate and a paradoxical subreg forming a vector
3695 mode to a vec_concat.
3697 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3699 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3700 -march=armv8.3-a variants.
3701 * config/arm/t-multilib: Likewise.
3702 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3704 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3706 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3708 (cceq_ior_compare_complement): Give it a name so I can use it, and
3709 change boolean_or_operator predicate to boolean_operator so it can
3710 be used to generate a crand.
3711 (eqne): New code iterator.
3712 (bd/bd_neg): New code_attrs.
3713 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3714 a single define_insn.
3715 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3716 decrement (bdnzt/bdnzf/bdzt/bdzf).
3717 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3718 with the new names of the branch decrement patterns, and added the
3719 names of the branch decrement conditional patterns.
3721 2018-01-08 Richard Biener <rguenther@suse.de>
3723 PR tree-optimization/83563
3724 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3727 2018-01-08 Richard Biener <rguenther@suse.de>
3730 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3732 2018-01-08 Richard Biener <rguenther@suse.de>
3734 PR tree-optimization/83685
3735 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3736 references to abnormals.
3738 2018-01-08 Richard Biener <rguenther@suse.de>
3741 * dwarf2out.c (output_indirect_strings): Handle empty
3742 skeleton_debug_str_hash.
3743 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3745 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3747 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3748 (emit_store_direct): Likewise.
3749 (arc_trampoline_adjust_address): Likewise.
3750 (arc_asm_trampoline_template): New function.
3751 (arc_initialize_trampoline): Use asm_trampoline_template.
3752 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3753 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3754 * config/arc/arc.md (flush_icache): Delete pattern.
3756 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3758 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3759 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3762 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3765 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3766 by not USED_FOR_TARGET.
3767 (make_pass_resolve_sw_modes): Likewise.
3769 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3771 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3774 2018-01-08 Richard Biener <rguenther@suse.de>
3777 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3779 2018-01-08 Richard Biener <rguenther@suse.de>
3782 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3784 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3787 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3788 basic blocks with a small number of successors.
3789 (convert_control_dep_chain_into_preds): Improve handling of
3791 (dump_predicates): Split apart into...
3792 (dump_pred_chain): ...here...
3793 (dump_pred_info): ...and here.
3794 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3795 (can_chain_union_be_invalidated_p): Improve check for invalidation
3797 (uninit_uses_cannot_happen): Avoid unnecessary if
3798 convert_control_dep_chain_into_preds yielded nothing.
3800 2018-01-06 Martin Sebor <msebor@redhat.com>
3802 PR tree-optimization/83640
3803 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3804 subtracting negative offset from size.
3805 (builtin_access::overlap): Adjust offset bounds of the access to fall
3806 within the size of the object if possible.
3808 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3810 PR rtl-optimization/83699
3811 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3812 extract_bit_field_as_subreg to cases in which the extracted
3813 value is also a vector.
3815 * lra-constraints.c (process_alt_operands): Test for the equivalence
3816 substitutions when detecting a possible reload cycle.
3818 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3821 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3822 by default if flag_selective_schedling{,2}. Formatting fixes.
3824 PR rtl-optimization/83682
3825 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
3826 if it has non-VECTOR_MODE element mode.
3827 (vec_duplicate_p): Likewise.
3830 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
3831 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
3833 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3836 * config/i386/i386-builtin.def
3837 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
3838 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
3839 Require also OPTION_MASK_ISA_AVX512F in addition to
3840 OPTION_MASK_ISA_GFNI.
3841 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
3842 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
3843 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
3844 to OPTION_MASK_ISA_GFNI.
3845 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
3846 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
3847 OPTION_MASK_ISA_AVX512BW.
3848 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
3849 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
3850 addition to OPTION_MASK_ISA_GFNI.
3851 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
3852 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
3853 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
3854 to OPTION_MASK_ISA_GFNI.
3855 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
3856 a requirement for all ISAs rather than any of them with a few
3858 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
3860 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
3861 bitmasks to be enabled with 3 exceptions, instead of requiring any
3862 enabled ISA with lots of exceptions.
3863 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
3864 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
3865 Change avx512bw in isa attribute to avx512f.
3866 * config/i386/sgxintrin.h: Add license boilerplate.
3867 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
3868 to __AVX512F__ and __AVX512VL to __AVX512VL__.
3869 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
3870 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
3872 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
3873 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
3874 temporarily sse2 rather than sse if not enabled already.
3877 * config/i386/sse.md (VI248_VLBW): Rename to ...
3878 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
3879 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
3880 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
3881 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
3882 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
3883 mode iterator instead of VI248_VLBW.
3885 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
3887 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
3888 (record_modified): Skip clobbers; add debug output.
3889 (param_change_prob): Use sreal frequencies.
3891 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3893 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
3894 punt for user-aligned variables.
3896 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3898 * tree-chrec.c (chrec_contains_symbols): Return true for
3901 2018-01-05 Sudakshina Das <sudi.das@arm.com>
3904 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
3905 of (x|y) == x for BICS pattern.
3907 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3909 PR tree-optimization/83605
3910 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
3911 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
3914 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
3916 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
3917 * config/epiphany/rtems.h: New file.
3919 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3920 Uros Bizjak <ubizjak@gmail.com>
3923 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
3924 QIreg_operand instead of register_operand predicate.
3925 * config/i386/i386.c (ix86_rop_should_change_byte_p,
3926 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
3927 comments instead of -fmitigate[-_]rop.
3929 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3932 * cgraphunit.c (symbol_table::compile): Switch to text_section
3933 before calling assembly_start debug hook.
3934 * run-rtl-passes.c (run_rtl_passes): Likewise.
3937 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3939 * tree-vrp.c (extract_range_from_binary_expr_1): Check
3940 range_int_cst_p rather than !symbolic_range_p before calling
3941 extract_range_from_multiplicative_op_1.
3943 2017-01-04 Jeff Law <law@redhat.com>
3945 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
3946 redundant test in assertion.
3948 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3950 * doc/rtl.texi: Document machine_mode wrapper classes.
3952 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3954 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
3957 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3959 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
3960 the VEC_PERM_EXPR fold to fail.
3962 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3965 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
3966 to switched_sections.
3968 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3971 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
3974 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
3977 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
3978 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
3980 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3983 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
3984 is BLKmode and bitpos not zero or mode change is needed.
3986 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3989 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
3992 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
3995 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
3996 instead of MULT rtx. Update all corresponding splitters.
3998 (*ssub<modesuffix>): Ditto.
4000 (*cmp_sadd_di): Update split patterns.
4001 (*cmp_sadd_si): Ditto.
4002 (*cmp_sadd_sidi): Ditto.
4003 (*cmp_ssub_di): Ditto.
4004 (*cmp_ssub_si): Ditto.
4005 (*cmp_ssub_sidi): Ditto.
4006 * config/alpha/predicates.md (const23_operand): New predicate.
4007 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4008 Look for ASHIFT, not MULT inner operand.
4009 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4011 2018-01-04 Martin Liska <mliska@suse.cz>
4013 PR gcov-profile/83669
4014 * gcov.c (output_intermediate_file): Add version to intermediate
4016 * doc/gcov.texi: Document new field 'version' in intermediate
4017 file format. Fix location of '-k' option of gcov command.
4019 2018-01-04 Martin Liska <mliska@suse.cz>
4022 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4024 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4026 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4028 2018-01-03 Martin Sebor <msebor@redhat.com>
4030 PR tree-optimization/83655
4031 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4032 checking calls with invalid arguments.
4034 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4036 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4037 (vectorizable_mask_load_store): Delete.
4038 (vectorizable_call): Return false for masked loads and stores.
4039 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4040 instead of gimple_assign_rhs1.
4041 (vectorizable_load): Handle IFN_MASK_LOAD.
4042 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4044 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4046 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4048 (vectorizable_mask_load_store): ...here.
4049 (vectorizable_load): ...and here.
4051 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4053 * tree-vect-stmts.c (vect_build_all_ones_mask)
4054 (vect_build_zero_merge_argument): New functions, split out from...
4055 (vectorizable_load): ...here.
4057 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4059 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4061 (vectorizable_mask_load_store): ...here.
4062 (vectorizable_store): ...and here.
4064 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4066 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4068 (vectorizable_mask_load_store): ...here.
4070 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4072 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4073 (vect_model_store_cost): Take a vec_load_store_type instead of a
4075 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4076 (vect_model_store_cost): Take a vec_load_store_type instead of a
4078 (vectorizable_mask_load_store): Update accordingly.
4079 (vectorizable_store): Likewise.
4080 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4082 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4084 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4085 IFN_MASK_LOAD calls here rather than...
4086 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4088 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4089 Alan Hayward <alan.hayward@arm.com>
4090 David Sherwood <david.sherwood@arm.com>
4092 * expmed.c (extract_bit_field_1): For vector extracts,
4093 fall back to extract_bit_field_as_subreg if vec_extract
4096 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4097 Alan Hayward <alan.hayward@arm.com>
4098 David Sherwood <david.sherwood@arm.com>
4100 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4101 they are variable or constant sized.
4102 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4103 slots for constant-sized data.
4105 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4106 Alan Hayward <alan.hayward@arm.com>
4107 David Sherwood <david.sherwood@arm.com>
4109 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4110 handling COND_EXPRs with boolean comparisons, try to find a better
4111 basis for the mask type than the boolean itself.
4113 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4115 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4116 is calculated and how it can be overridden.
4117 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4118 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4120 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4123 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4124 Alan Hayward <alan.hayward@arm.com>
4125 David Sherwood <david.sherwood@arm.com>
4127 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4128 Remove the mode argument.
4129 (aarch64_simd_valid_immediate): Remove the mode and inverse
4131 * config/aarch64/iterators.md (bitsize): New iterator.
4132 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4133 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4134 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4135 aarch64_simd_valid_immediate.
4136 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4137 (aarch64_reg_or_bic_imm): Likewise.
4138 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4139 with an insn_type enum and msl with a modifier_type enum.
4140 Replace element_width with a scalar_mode. Change the shift
4141 to unsigned int. Add constructors for scalar_float_mode and
4142 scalar_int_mode elements.
4143 (aarch64_vect_float_const_representable_p): Delete.
4144 (aarch64_can_const_movi_rtx_p)
4145 (aarch64_simd_scalar_immediate_valid_for_move)
4146 (aarch64_simd_make_constant): Update call to
4147 aarch64_simd_valid_immediate.
4148 (aarch64_advsimd_valid_immediate_hs): New function.
4149 (aarch64_advsimd_valid_immediate): Likewise.
4150 (aarch64_simd_valid_immediate): Remove mode and inverse
4151 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4152 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4153 and aarch64_float_const_representable_p on the result.
4154 (aarch64_output_simd_mov_immediate): Remove mode argument.
4155 Update call to aarch64_simd_valid_immediate and use of
4156 simd_immediate_info.
4157 (aarch64_output_scalar_simd_mov_immediate): Update call
4160 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4161 Alan Hayward <alan.hayward@arm.com>
4162 David Sherwood <david.sherwood@arm.com>
4164 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4165 (mode_nunits): Likewise CONST_MODE_NUNITS.
4166 * machmode.def (ADJUST_NUNITS): Document.
4167 * genmodes.c (mode_data::need_nunits_adj): New field.
4168 (blank_mode): Update accordingly.
4169 (adj_nunits): New variable.
4170 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4172 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4173 listed in adj_nunits.
4174 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4175 listed in adj_nunits. Don't emit case statements for such modes.
4176 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4177 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4178 nothing if adj_nunits is nonnull.
4179 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4180 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4181 (emit_mode_fbit): Update use of print_maybe_const_decl.
4182 (emit_move_size): Likewise. Treat the array as non-const
4184 (emit_mode_adjustments): Handle adj_nunits.
4186 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4188 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4189 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4190 (VECTOR_MODES): Use it.
4191 (make_vector_modes): Take the prefix as an argument.
4193 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4194 Alan Hayward <alan.hayward@arm.com>
4195 David Sherwood <david.sherwood@arm.com>
4197 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4198 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4199 for MODE_VECTOR_BOOL.
4200 * machmode.def (VECTOR_BOOL_MODE): Document.
4201 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4202 (make_vector_bool_mode): New function.
4203 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4205 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4206 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4208 * stor-layout.c (int_mode_for_mode): Likewise.
4209 * tree.c (build_vector_type_for_mode): Likewise.
4210 * varasm.c (output_constant_pool_2): Likewise.
4211 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4212 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4213 for MODE_VECTOR_BOOL.
4214 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4215 of mode class checks.
4216 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4217 instead of a list of mode class checks.
4218 (expand_vector_scalar_condition): Likewise.
4219 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4221 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4222 Alan Hayward <alan.hayward@arm.com>
4223 David Sherwood <david.sherwood@arm.com>
4225 * machmode.h (mode_size): Change from unsigned short to
4227 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4228 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4229 or if measurement_type is not polynomial.
4230 (fixed_size_mode::includes_p): Check for constant-sized modes.
4231 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4232 return a poly_uint16 rather than an unsigned short.
4233 (emit_mode_size): Change the type of mode_size from unsigned short
4234 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4235 (emit_mode_adjustments): Cope with polynomial vector sizes.
4236 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4238 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4240 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4241 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4242 * caller-save.c (setup_save_areas): Likewise.
4243 (replace_reg_with_saved_mem): Likewise.
4244 * calls.c (emit_library_call_value_1): Likewise.
4245 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4246 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4247 (gen_lowpart_for_combine): Likewise.
4248 * convert.c (convert_to_integer_1): Likewise.
4249 * cse.c (equiv_constant, cse_insn): Likewise.
4250 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4251 (cselib_subst_to_values): Likewise.
4252 * dce.c (word_dce_process_block): Likewise.
4253 * df-problems.c (df_word_lr_mark_ref): Likewise.
4254 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4255 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4256 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4257 (rtl_for_decl_location): Likewise.
4258 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4259 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4260 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4261 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4262 (expand_expr_real_1): Likewise.
4263 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4264 (pad_below): Likewise.
4265 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4266 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4267 * ira.c (get_subreg_tracking_sizes): Likewise.
4268 * ira-build.c (ira_create_allocno_objects): Likewise.
4269 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4270 (ira_sort_regnos_for_alter_reg): Likewise.
4271 * ira-costs.c (record_operand_costs): Likewise.
4272 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4273 (resolve_simple_move): Likewise.
4274 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4275 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4276 (lra_constraints): Likewise.
4277 (CONST_POOL_OK_P): Reject variable-sized modes.
4278 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4279 (add_pseudo_to_slot, lra_spill): Likewise.
4280 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4281 * optabs-query.c (get_best_extraction_insn): Likewise.
4282 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4283 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4284 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4285 * recog.c (offsettable_address_addr_space_p): Likewise.
4286 * regcprop.c (maybe_mode_change): Likewise.
4287 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4288 * regrename.c (build_def_use): Likewise.
4289 * regstat.c (dump_reg_info): Likewise.
4290 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4291 (find_reloads, find_reloads_subreg_address): Likewise.
4292 * reload1.c (eliminate_regs_1): Likewise.
4293 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4294 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4295 (simplify_binary_operation_1, simplify_subreg): Likewise.
4296 * targhooks.c (default_function_arg_padding): Likewise.
4297 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4298 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4299 (verify_gimple_assign_ternary): Likewise.
4300 * tree-inline.c (estimate_move_cost): Likewise.
4301 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4302 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4303 (get_address_cost_ainc): Likewise.
4304 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4305 (vect_supportable_dr_alignment): Likewise.
4306 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4307 (vectorizable_reduction): Likewise.
4308 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4309 (vectorizable_operation, vectorizable_load): Likewise.
4310 * tree.c (build_same_sized_truth_vector_type): Likewise.
4311 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4312 * var-tracking.c (emit_note_insn_var_location): Likewise.
4313 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4314 (ADDR_VEC_ALIGN): Likewise.
4316 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4317 Alan Hayward <alan.hayward@arm.com>
4318 David Sherwood <david.sherwood@arm.com>
4320 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4322 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4323 or if measurement_type is polynomial.
4324 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4325 * combine.c (make_extraction): Likewise.
4326 * dse.c (find_shift_sequence): Likewise.
4327 * dwarf2out.c (mem_loc_descriptor): Likewise.
4328 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4329 (extract_bit_field, extract_low_bits): Likewise.
4330 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4331 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4332 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4333 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4334 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4335 * reload.c (find_reloads): Likewise.
4336 * reload1.c (alter_reg): Likewise.
4337 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4338 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4339 * tree-if-conv.c (predicate_mem_writes): Likewise.
4340 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4341 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4342 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4343 * valtrack.c (dead_debug_insert_temp): Likewise.
4344 * varasm.c (mergeable_constant_section): Likewise.
4345 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4347 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4348 Alan Hayward <alan.hayward@arm.com>
4349 David Sherwood <david.sherwood@arm.com>
4351 * expr.c (expand_assignment): Cope with polynomial mode sizes
4352 when assigning to a CONCAT.
4354 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4355 Alan Hayward <alan.hayward@arm.com>
4356 David Sherwood <david.sherwood@arm.com>
4358 * machmode.h (mode_precision): Change from unsigned short to
4360 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4362 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4363 or if measurement_type is not polynomial.
4364 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4365 in which the mode is already known to be a scalar_int_mode.
4366 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4367 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4369 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4370 for GET_MODE_PRECISION.
4371 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4372 for GET_MODE_PRECISION.
4373 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4375 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4376 (expand_field_assignment, make_extraction): Likewise.
4377 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4378 (get_last_value): Likewise.
4379 * convert.c (convert_to_integer_1): Likewise.
4380 * cse.c (cse_insn): Likewise.
4381 * expr.c (expand_expr_real_1): Likewise.
4382 * lra-constraints.c (simplify_operand_subreg): Likewise.
4383 * optabs-query.c (can_atomic_load_p): Likewise.
4384 * optabs.c (expand_atomic_load): Likewise.
4385 (expand_atomic_store): Likewise.
4386 * ree.c (combine_reaching_defs): Likewise.
4387 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4388 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4389 * tree.h (type_has_mode_precision_p): Likewise.
4390 * ubsan.c (instrument_si_overflow): Likewise.
4392 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4393 Alan Hayward <alan.hayward@arm.com>
4394 David Sherwood <david.sherwood@arm.com>
4396 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4397 polynomial numbers of units.
4398 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4399 (valid_vector_subparts_p): New function.
4400 (build_vector_type): Remove temporary shim and take the number
4401 of units as a poly_uint64 rather than an int.
4402 (build_opaque_vector_type): Take the number of units as a
4403 poly_uint64 rather than an int.
4404 * tree.c (build_vector_from_ctor): Handle polynomial
4405 TYPE_VECTOR_SUBPARTS.
4406 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4407 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4408 (build_vector_from_val): If the number of units is variable,
4409 use build_vec_duplicate_cst for constant operands and
4410 VEC_DUPLICATE_EXPR otherwise.
4411 (make_vector_type): Remove temporary is_constant ().
4412 (build_vector_type, build_opaque_vector_type): Take the number of
4413 units as a poly_uint64 rather than an int.
4414 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4416 * cfgexpand.c (expand_debug_expr): Likewise.
4417 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4418 (store_constructor, expand_expr_real_1): Likewise.
4419 (const_scalar_mask_from_tree): Likewise.
4420 * fold-const-call.c (fold_const_reduction): Likewise.
4421 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4422 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4423 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4424 (fold_relational_const): Likewise.
4425 (native_interpret_vector): Likewise. Change the size from an
4426 int to an unsigned int.
4427 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4428 TYPE_VECTOR_SUBPARTS.
4429 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4430 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4431 duplicating a non-constant operand into a variable-length vector.
4432 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4433 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4434 * ipa-icf.c (sem_variable::equals): Likewise.
4435 * match.pd: Likewise.
4436 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4437 * print-tree.c (print_node): Likewise.
4438 * stor-layout.c (layout_type): Likewise.
4439 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4440 * tree-cfg.c (verify_gimple_comparison): Likewise.
4441 (verify_gimple_assign_binary): Likewise.
4442 (verify_gimple_assign_ternary): Likewise.
4443 (verify_gimple_assign_single): Likewise.
4444 * tree-pretty-print.c (dump_generic_node): Likewise.
4445 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4446 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4447 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4448 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4449 (vect_shift_permute_load_chain): Likewise.
4450 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4451 (expand_vector_condition, optimize_vector_constructor): Likewise.
4452 (lower_vec_perm, get_compute_type): Likewise.
4453 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4454 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4455 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4456 (vect_recog_mask_conversion_pattern): Likewise.
4457 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4458 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4459 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4460 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4461 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4462 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4463 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4464 (supportable_widening_operation): Likewise.
4465 (supportable_narrowing_operation): Likewise.
4466 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4468 * varasm.c (output_constant): Likewise.
4470 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4471 Alan Hayward <alan.hayward@arm.com>
4472 David Sherwood <david.sherwood@arm.com>
4474 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4475 so that both the length == 3 and length != 3 cases set up their
4476 own permute vectors. Add comments explaining why we know the
4477 number of elements is constant.
4478 (vect_permute_load_chain): Likewise.
4480 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4481 Alan Hayward <alan.hayward@arm.com>
4482 David Sherwood <david.sherwood@arm.com>
4484 * machmode.h (mode_nunits): Change from unsigned char to
4486 (ONLY_FIXED_SIZE_MODES): New macro.
4487 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4488 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4489 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4491 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4492 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4493 or if measurement_type is not polynomial.
4494 * genmodes.c (ZERO_COEFFS): New macro.
4495 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4497 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4498 Use ZERO_COEFFS when emitting initializers.
4499 * data-streamer.h (bp_pack_poly_value): New function.
4500 (bp_unpack_poly_value): Likewise.
4501 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4502 for GET_MODE_NUNITS.
4503 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4504 for GET_MODE_NUNITS.
4505 * tree.c (make_vector_type): Remove temporary shim and make
4506 the real function take the number of units as a poly_uint64
4508 (build_vector_type_for_mode): Handle polynomial nunits.
4509 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4510 * emit-rtl.c (const_vec_series_p_1): Likewise.
4511 (gen_rtx_CONST_VECTOR): Likewise.
4512 * fold-const.c (test_vec_duplicate_folding): Likewise.
4513 * genrecog.c (validate_pattern): Likewise.
4514 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4515 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4516 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4517 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4518 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4519 * rtlanal.c (subreg_get_info): Likewise.
4520 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4521 (vect_grouped_load_supported): Likewise.
4522 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4523 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4524 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4525 (simplify_const_unary_operation, simplify_binary_operation_1)
4526 (simplify_const_binary_operation, simplify_ternary_operation)
4527 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4528 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4529 instead of CONST_VECTOR_NUNITS.
4530 * varasm.c (output_constant_pool_2): Likewise.
4531 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4532 explicit-encoded elements in the XVEC for variable-length vectors.
4534 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4536 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4538 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4539 Alan Hayward <alan.hayward@arm.com>
4540 David Sherwood <david.sherwood@arm.com>
4542 * coretypes.h (fixed_size_mode): Declare.
4543 (fixed_size_mode_pod): New typedef.
4544 * builtins.h (target_builtins::x_apply_args_mode)
4545 (target_builtins::x_apply_result_mode): Change type to
4546 fixed_size_mode_pod.
4547 * builtins.c (apply_args_size, apply_result_size, result_vector)
4548 (expand_builtin_apply_args_1, expand_builtin_apply)
4549 (expand_builtin_return): Update accordingly.
4551 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4553 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4554 * cselib.c (cselib_hash_rtx): Likewise.
4555 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4556 CONST_VECTOR encoding.
4558 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4559 Jeff Law <law@redhat.com>
4562 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4563 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4564 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4565 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4568 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4569 explicitly probe *sp in a noreturn function if there were any callee
4570 register saves or frame pointer is needed.
4572 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4575 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4576 BLKmode for ternary, binary or unary expressions.
4579 * var-tracking.c (delete_vta_debug_insn): New inline function.
4580 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4581 insns from get_insns () to NULL instead of each bb separately.
4582 Use delete_vta_debug_insn. No longer static.
4583 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4584 delete_vta_debug_insns callers.
4585 * rtl.h (delete_vta_debug_insns): Declare.
4586 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4587 instead of variable_tracking_main.
4589 2018-01-03 Martin Sebor <msebor@redhat.com>
4591 PR tree-optimization/83603
4592 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4593 arguments past the endof the argument list in functions declared
4594 without a prototype.
4595 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4596 Avoid checking when arguments are null.
4598 2018-01-03 Martin Sebor <msebor@redhat.com>
4601 * doc/extend.texi (attribute const): Fix a typo.
4602 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4603 issuing -Wsuggest-attribute for void functions.
4605 2018-01-03 Martin Sebor <msebor@redhat.com>
4607 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4608 offset_int::from instead of wide_int::to_shwi.
4609 (maybe_diag_overlap): Remove assertion.
4610 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4611 * gimple-ssa-sprintf.c (format_directive): Same.
4612 (parse_directive): Same.
4613 (sprintf_dom_walker::compute_format_length): Same.
4614 (try_substitute_return_value): Same.
4616 2017-01-03 Jeff Law <law@redhat.com>
4619 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4620 non-constant residual for zero at runtime and avoid probing in
4621 that case. Reorganize code for trailing problem to mirror handling
4624 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4626 PR tree-optimization/83501
4627 * tree-ssa-strlen.c (get_string_cst): New.
4628 (handle_char_store): Call get_string_cst.
4630 2018-01-03 Martin Liska <mliska@suse.cz>
4632 PR tree-optimization/83593
4633 * tree-ssa-strlen.c: Include tree-cfg.h.
4634 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4635 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4636 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4638 (strlen_dom_walker::before_dom_children): Call
4639 gimple_purge_dead_eh_edges. Dump tranformation with details
4641 (strlen_dom_walker::before_dom_children): Update call by adding
4642 new argument cleanup_eh.
4643 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4645 2018-01-03 Martin Liska <mliska@suse.cz>
4648 * cif-code.def (VARIADIC_THUNK): New enum value.
4649 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4652 2018-01-03 Jan Beulich <jbeulich@suse.com>
4654 * sse.md (mov<mode>_internal): Tighten condition for when to use
4655 vmovdqu<ssescalarsize> for TI and OI modes.
4657 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4659 Update copyright years.
4661 2018-01-03 Martin Liska <mliska@suse.cz>
4664 * ipa-visibility.c (function_and_variable_visibility): Skip
4665 functions with noipa attribure.
4667 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4669 * gcc.c (process_command): Update copyright notice dates.
4670 * gcov-dump.c (print_version): Ditto.
4671 * gcov.c (print_version): Ditto.
4672 * gcov-tool.c (print_version): Ditto.
4673 * gengtype.c (create_file): Ditto.
4674 * doc/cpp.texi: Bump @copying's copyright year.
4675 * doc/cppinternals.texi: Ditto.
4676 * doc/gcc.texi: Ditto.
4677 * doc/gccint.texi: Ditto.
4678 * doc/gcov.texi: Ditto.
4679 * doc/install.texi: Ditto.
4680 * doc/invoke.texi: Ditto.
4682 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4684 * vector-builder.h (vector_builder::m_full_nelts): Change from
4685 unsigned int to poly_uint64.
4686 (vector_builder::full_nelts): Update prototype accordingly.
4687 (vector_builder::new_vector): Likewise.
4688 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4689 (vector_builder::operator ==): Likewise.
4690 (vector_builder::finalize): Likewise.
4691 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4692 Take the number of elements as a poly_uint64 rather than an
4694 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4695 from unsigned int to poly_uint64.
4696 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4697 (vec_perm_indices::new_vector): Likewise.
4698 (vec_perm_indices::length): Likewise.
4699 (vec_perm_indices::nelts_per_input): Likewise.
4700 (vec_perm_indices::input_nelts): Likewise.
4701 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4702 number of elements per input as a poly_uint64 rather than an
4703 unsigned int. Use the original encoding for variable-length
4704 vectors, rather than clamping each individual element.
4705 For the second and subsequent elements in each pattern,
4706 clamp the step and base before clamping their sum.
4707 (vec_perm_indices::series_p): Handle polynomial element counts.
4708 (vec_perm_indices::all_in_range_p): Likewise.
4709 (vec_perm_indices_to_tree): Likewise.
4710 (vec_perm_indices_to_rtx): Likewise.
4711 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4712 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4713 (tree_vector_builder::new_binary_operation): Handle polynomial
4714 element counts. Return false if we need to know the number
4715 of elements at compile time.
4716 * fold-const.c (fold_vec_perm): Punt if the number of elements
4717 isn't known at compile time.
4719 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4721 * vec-perm-indices.h (vec_perm_builder): Change element type
4722 from HOST_WIDE_INT to poly_int64.
4723 (vec_perm_indices::element_type): Update accordingly.
4724 (vec_perm_indices::clamp): Handle polynomial element_types.
4725 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4726 (vec_perm_indices::all_in_range_p): Likewise.
4727 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4729 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4730 polynomial vec_perm_indices element types.
4731 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4732 * fold-const.c (fold_vec_perm): Likewise.
4733 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4734 * tree-vect-generic.c (lower_vec_perm): Likewise.
4735 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4736 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4737 element type to HOST_WIDE_INT.
4739 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4740 Alan Hayward <alan.hayward@arm.com>
4741 David Sherwood <david.sherwood@arm.com>
4743 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4744 rather than an int. Use plus_constant.
4745 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4746 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4748 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4749 Alan Hayward <alan.hayward@arm.com>
4750 David Sherwood <david.sherwood@arm.com>
4752 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4753 a HOST_WIDE_INT to a poly_int64.
4755 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4756 Alan Hayward <alan.hayward@arm.com>
4757 David Sherwood <david.sherwood@arm.com>
4759 * calls.c (load_register_parameters): Cope with polynomial
4760 mode sizes. Require a constant size for BLKmode parameters
4761 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4762 forces a parameter to be padded at the lsb end in order to
4763 fill a complete number of words, require the parameter size
4764 to be ordered wrt UNITS_PER_WORD.
4766 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4767 Alan Hayward <alan.hayward@arm.com>
4768 David Sherwood <david.sherwood@arm.com>
4770 * reload1.c (spill_stack_slot_width): Change element type
4771 from unsigned int to poly_uint64_pod.
4772 (alter_reg): Treat mode sizes as polynomial.
4774 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4775 Alan Hayward <alan.hayward@arm.com>
4776 David Sherwood <david.sherwood@arm.com>
4778 * reload.c (complex_word_subreg_p): New function.
4779 (reload_inner_reg_of_subreg, push_reload): Use it.
4781 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4782 Alan Hayward <alan.hayward@arm.com>
4783 David Sherwood <david.sherwood@arm.com>
4785 * lra-constraints.c (process_alt_operands): Reject matched
4786 operands whose sizes aren't ordered.
4787 (match_reload): Refer to this check here.
4789 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4790 Alan Hayward <alan.hayward@arm.com>
4791 David Sherwood <david.sherwood@arm.com>
4793 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4794 that the mode size is in the set {1, 2, 4, 8, 16}.
4796 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4797 Alan Hayward <alan.hayward@arm.com>
4798 David Sherwood <david.sherwood@arm.com>
4800 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4801 Use plus_constant instead of gen_rtx_PLUS.
4803 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4804 Alan Hayward <alan.hayward@arm.com>
4805 David Sherwood <david.sherwood@arm.com>
4807 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4808 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4809 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4810 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4811 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4812 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4813 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4814 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4815 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4816 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4818 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4819 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4820 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4821 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4822 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4823 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4824 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4825 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
4826 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
4827 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
4829 * expr.c (emit_move_resolve_push): Treat the input and result
4830 of PUSH_ROUNDING as a poly_int64.
4831 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
4832 (emit_push_insn): Likewise.
4833 * lra-eliminations.c (mark_not_eliminable): Likewise.
4834 * recog.c (push_operand): Likewise.
4835 * reload1.c (elimination_effects): Likewise.
4836 * rtlanal.c (nonzero_bits1): Likewise.
4837 * calls.c (store_one_arg): Likewise. Require the padding to be
4838 known at compile time.
4840 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4841 Alan Hayward <alan.hayward@arm.com>
4842 David Sherwood <david.sherwood@arm.com>
4844 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
4845 Use plus_constant instead of gen_rtx_PLUS.
4847 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4848 Alan Hayward <alan.hayward@arm.com>
4849 David Sherwood <david.sherwood@arm.com>
4851 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
4854 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4855 Alan Hayward <alan.hayward@arm.com>
4856 David Sherwood <david.sherwood@arm.com>
4858 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
4859 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
4860 via stack temporaries. Treat the mode size as polynomial too.
4862 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4863 Alan Hayward <alan.hayward@arm.com>
4864 David Sherwood <david.sherwood@arm.com>
4866 * expr.c (expand_expr_real_2): When handling conversions involving
4867 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
4868 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
4869 as a poly_uint64 too.
4871 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4872 Alan Hayward <alan.hayward@arm.com>
4873 David Sherwood <david.sherwood@arm.com>
4875 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
4877 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4878 Alan Hayward <alan.hayward@arm.com>
4879 David Sherwood <david.sherwood@arm.com>
4881 * combine.c (can_change_dest_mode): Handle polynomial
4882 REGMODE_NATURAL_SIZE.
4883 * expmed.c (store_bit_field_1): Likewise.
4884 * expr.c (store_constructor): Likewise.
4885 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
4886 and polynomial REGMODE_NATURAL_SIZE.
4887 (gen_lowpart_common): Likewise.
4888 * reginfo.c (record_subregs_of_mode): Likewise.
4889 * rtlanal.c (read_modify_subreg_p): Likewise.
4891 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4892 Alan Hayward <alan.hayward@arm.com>
4893 David Sherwood <david.sherwood@arm.com>
4895 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
4896 numbers of elements.
4898 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4899 Alan Hayward <alan.hayward@arm.com>
4900 David Sherwood <david.sherwood@arm.com>
4902 * match.pd: Cope with polynomial numbers of vector elements.
4904 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4905 Alan Hayward <alan.hayward@arm.com>
4906 David Sherwood <david.sherwood@arm.com>
4908 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
4909 in a POINTER_PLUS_EXPR.
4911 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4912 Alan Hayward <alan.hayward@arm.com>
4913 David Sherwood <david.sherwood@arm.com>
4915 * omp-simd-clone.c (simd_clone_subparts): New function.
4916 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
4917 (ipa_simd_modify_function_body): Likewise.
4919 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4920 Alan Hayward <alan.hayward@arm.com>
4921 David Sherwood <david.sherwood@arm.com>
4923 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
4924 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
4925 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
4926 (expand_vector_condition, vector_element): Likewise.
4927 (subparts_gt): New function.
4928 (get_compute_type): Use subparts_gt.
4929 (count_type_subparts): Delete.
4930 (expand_vector_operations_1): Use subparts_gt instead of
4931 count_type_subparts.
4933 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4934 Alan Hayward <alan.hayward@arm.com>
4935 David Sherwood <david.sherwood@arm.com>
4937 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
4938 (vect_compile_time_alias): ...this new function. Do the calculation
4939 on poly_ints rather than trees.
4940 (vect_prune_runtime_alias_test_list): Update call accordingly.
4942 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4943 Alan Hayward <alan.hayward@arm.com>
4944 David Sherwood <david.sherwood@arm.com>
4946 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
4948 (vect_schedule_slp_instance): Likewise.
4950 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4951 Alan Hayward <alan.hayward@arm.com>
4952 David Sherwood <david.sherwood@arm.com>
4954 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
4955 constant and extern definitions for variable-length vectors.
4956 (vect_get_constant_vectors): Note that the number of units
4957 is known to be constant.
4959 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4960 Alan Hayward <alan.hayward@arm.com>
4961 David Sherwood <david.sherwood@arm.com>
4963 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
4964 of units as polynomial. Choose between WIDE and NARROW based
4967 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4968 Alan Hayward <alan.hayward@arm.com>
4969 David Sherwood <david.sherwood@arm.com>
4971 * tree-vect-stmts.c (simd_clone_subparts): New function.
4972 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
4974 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4975 Alan Hayward <alan.hayward@arm.com>
4976 David Sherwood <david.sherwood@arm.com>
4978 * tree-vect-stmts.c (vectorizable_call): Treat the number of
4979 vectors as polynomial. Use build_index_vector for
4982 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4983 Alan Hayward <alan.hayward@arm.com>
4984 David Sherwood <david.sherwood@arm.com>
4986 * tree-vect-stmts.c (get_load_store_type): Treat the number of
4987 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
4988 for variable-length vectors.
4989 (vectorizable_mask_load_store): Treat the number of units as
4990 polynomial, asserting that it is constant if the condition has
4991 already been enforced.
4992 (vectorizable_store, vectorizable_load): Likewise.
4994 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4995 Alan Hayward <alan.hayward@arm.com>
4996 David Sherwood <david.sherwood@arm.com>
4998 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
4999 of units as polynomial. Punt if we can't tell at compile time
5000 which vector contains the final result.
5002 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5003 Alan Hayward <alan.hayward@arm.com>
5004 David Sherwood <david.sherwood@arm.com>
5006 * tree-vect-loop.c (vectorizable_induction): Treat the number
5007 of units as polynomial. Punt on SLP inductions. Use an integer
5008 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5009 cast of such a series for variable-length floating-point
5012 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5013 Alan Hayward <alan.hayward@arm.com>
5014 David Sherwood <david.sherwood@arm.com>
5016 * tree.h (build_index_vector): Declare.
5017 * tree.c (build_index_vector): New function.
5018 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5019 of units as polynomial, forcibly converting it to a constant if
5020 vectorizable_reduction has already enforced the condition.
5021 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5022 to create a {1,2,3,...} vector.
5023 (vectorizable_reduction): Treat the number of units as polynomial.
5024 Choose vectype_in based on the largest scalar element size rather
5025 than the smallest number of units. Enforce the restrictions
5028 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5029 Alan Hayward <alan.hayward@arm.com>
5030 David Sherwood <david.sherwood@arm.com>
5032 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5033 number of units as polynomial.
5035 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5036 Alan Hayward <alan.hayward@arm.com>
5037 David Sherwood <david.sherwood@arm.com>
5039 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5040 * target.def (autovectorize_vector_sizes): Return the vector sizes
5041 by pointer, using vector_sizes rather than a bitmask.
5042 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5043 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5044 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5046 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5047 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5048 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5049 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5050 * omp-general.c (omp_max_vf): Likewise.
5051 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5052 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5053 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5054 * tree-vect-slp.c (vect_slp_bb): Likewise.
5055 * doc/tm.texi: Regenerate.
5056 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5058 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5059 the vector size as a poly_uint64 rather than an unsigned int.
5060 (current_vector_size): Change from an unsigned int to a poly_uint64.
5061 (get_vectype_for_scalar_type): Update accordingly.
5062 * tree.h (build_truth_vector_type): Take the size and number of
5063 units as a poly_uint64 rather than an unsigned int.
5064 (build_vector_type): Add a temporary overload that takes
5065 the number of units as a poly_uint64 rather than an unsigned int.
5066 * tree.c (make_vector_type): Likewise.
5067 (build_truth_vector_type): Take the number of units as a poly_uint64
5068 rather than an unsigned int.
5070 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5071 Alan Hayward <alan.hayward@arm.com>
5072 David Sherwood <david.sherwood@arm.com>
5074 * target.def (get_mask_mode): Take the number of units and length
5075 as poly_uint64s rather than unsigned ints.
5076 * targhooks.h (default_get_mask_mode): Update accordingly.
5077 * targhooks.c (default_get_mask_mode): Likewise.
5078 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5079 * doc/tm.texi: Regenerate.
5081 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5082 Alan Hayward <alan.hayward@arm.com>
5083 David Sherwood <david.sherwood@arm.com>
5085 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5086 * omp-general.c (omp_max_vf): Likewise.
5087 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5088 (expand_omp_simd): Handle polynomial safelen.
5089 * omp-low.c (omplow_simd_context): Add a default constructor.
5090 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5091 (lower_rec_simd_input_clauses): Update accordingly.
5092 (lower_rec_input_clauses): Likewise.
5094 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5095 Alan Hayward <alan.hayward@arm.com>
5096 David Sherwood <david.sherwood@arm.com>
5098 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5099 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5100 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5101 (vect_analyze_slp_cost): Likewise.
5102 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5103 (vect_model_load_cost): Likewise.
5105 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5106 Alan Hayward <alan.hayward@arm.com>
5107 David Sherwood <david.sherwood@arm.com>
5109 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5110 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5111 from an unsigned int * to a poly_uint64_pod *.
5112 (calculate_unrolling_factor): New function.
5113 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5115 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5116 Alan Hayward <alan.hayward@arm.com>
5117 David Sherwood <david.sherwood@arm.com>
5119 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5120 from an unsigned int to a poly_uint64.
5121 (_loop_vec_info::slp_unrolling_factor): Likewise.
5122 (_loop_vec_info::vectorization_factor): Change from an int
5124 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5125 (vect_get_num_vectors): New function.
5126 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5127 (vect_get_num_copies): Use vect_get_num_vectors.
5128 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5129 to an unsigned int *.
5130 (vect_analyze_data_refs): Change min_vf from an int * to a
5132 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5133 than an unsigned HOST_WIDE_INT.
5134 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5135 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5136 to an unsigned int *.
5137 (vect_analyze_data_ref_dependences): Likewise.
5138 (vect_compute_data_ref_alignment): Handle polynomial vf.
5139 (vect_enhance_data_refs_alignment): Likewise.
5140 (vect_prune_runtime_alias_test_list): Likewise.
5141 (vect_shift_permute_load_chain): Likewise.
5142 (vect_supportable_dr_alignment): Likewise.
5143 (dependence_distance_ge_vf): Take the vectorization factor as a
5144 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5145 (vect_analyze_data_refs): Change min_vf from an int * to a
5147 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5148 vfm1 as a poly_uint64 rather than an int. Make the same change
5149 for the returned bound_scalar.
5150 (vect_gen_vector_loop_niters): Handle polynomial vf.
5151 (vect_do_peeling): Likewise. Update call to
5152 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5153 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5155 * tree-vect-loop.c (vect_determine_vectorization_factor)
5156 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5157 (vect_get_known_peeling_cost): Likewise.
5158 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5159 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5160 (vect_transform_loop): Likewise. Use the lowest possible VF when
5161 updating the upper bounds of the loop.
5162 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5164 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5165 polynomial unroll factors.
5166 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5167 (vect_make_slp_decision): Likewise.
5168 (vect_supported_load_permutation_p): Likewise, and polynomial
5170 (vect_analyze_slp_cost): Handle polynomial vf.
5171 (vect_slp_analyze_node_operations): Likewise.
5172 (vect_slp_analyze_bb_1): Likewise.
5173 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5174 than an unsigned HOST_WIDE_INT.
5175 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5176 (vectorizable_load): Handle polynomial vf.
5177 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5179 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5181 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5182 Alan Hayward <alan.hayward@arm.com>
5183 David Sherwood <david.sherwood@arm.com>
5185 * match.pd: Handle bit operations involving three constants
5186 and try to fold one pair.
5188 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5190 * tree-vect-loop-manip.c: Include gimple-fold.h.
5191 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5192 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5193 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5194 Add a path that uses a step of VF instead of 1, but disable it
5196 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5197 and niters_no_overflow parameters. Update calls to
5198 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5199 Create a new SSA name if the latter choses to use a ste other
5200 than zero, and return it via niters_vector_mult_vf_var.
5201 * tree-vect-loop.c (vect_transform_loop): Update calls to
5202 vect_do_peeling, vect_gen_vector_loop_niters and
5203 slpeel_make_loop_iterate_ntimes.
5204 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5205 (vect_gen_vector_loop_niters): Update declarations after above changes.
5207 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5209 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5210 128-bit round to integer instructions.
5211 (ceil<mode>2): Likewise.
5212 (btrunc<mode>2): Likewise.
5213 (round<mode>2): Likewise.
5215 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5217 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5218 unaligned VSX load/store on P8/P9.
5219 (expand_block_clear): Allow the use of unaligned VSX
5220 load/store on P8/P9.
5222 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5224 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5226 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5227 swap associated with both a load and a store.
5229 2018-01-02 Andrew Waterman <andrew@sifive.com>
5231 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5232 * config/riscv/riscv.md (clear_cache): Use it.
5234 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5236 * web.c: Remove out-of-date comment.
5238 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5240 * expr.c (fixup_args_size_notes): Check that any existing
5241 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5242 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5243 (emit_single_push_insn): ...here.
5245 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5247 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5248 (const_vector_encoded_nelts): New function.
5249 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5250 (const_vector_int_elt, const_vector_elt): Declare.
5251 * emit-rtl.c (const_vector_int_elt_1): New function.
5252 (const_vector_elt): Likewise.
5253 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5254 of CONST_VECTOR_ELT.
5256 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5258 * expr.c: Include rtx-vector-builder.h.
5259 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5260 directly on the tree encoding.
5261 (const_vector_from_tree): Likewise.
5262 * optabs.c: Include rtx-vector-builder.h.
5263 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5264 sequence of "u" values.
5265 * vec-perm-indices.c: Include rtx-vector-builder.h.
5266 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5267 directly on the vec_perm_indices encoding.
5269 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5271 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5272 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5273 * rtx-vector-builder.h: New file.
5274 * rtx-vector-builder.c: Likewise.
5275 * rtl.h (rtx_def::u2): Add a const_vector field.
5276 (CONST_VECTOR_NPATTERNS): New macro.
5277 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5278 (CONST_VECTOR_DUPLICATE_P): Likewise.
5279 (CONST_VECTOR_STEPPED_P): Likewise.
5280 (CONST_VECTOR_ENCODED_ELT): Likewise.
5281 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5282 (unwrap_const_vec_duplicate): Likewise.
5283 (const_vec_series_p): Check for a non-duplicated vector encoding.
5284 Say that the function only returns true for integer vectors.
5285 * emit-rtl.c: Include rtx-vector-builder.h.
5286 (gen_const_vec_duplicate_1): Delete.
5287 (gen_const_vector): Call gen_const_vec_duplicate instead of
5288 gen_const_vec_duplicate_1.
5289 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5290 (gen_const_vec_duplicate): Use rtx_vector_builder.
5291 (gen_const_vec_series): Likewise.
5292 (gen_rtx_CONST_VECTOR): Likewise.
5293 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5294 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5295 Build a new vector rather than modifying a CONST_VECTOR in-place.
5296 (handle_special_swappables): Update call accordingly.
5297 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5298 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5299 Build a new vector rather than modifying a CONST_VECTOR in-place.
5300 (handle_special_swappables): Update call accordingly.
5302 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5304 * simplify-rtx.c (simplify_const_binary_operation): Use
5305 CONST_VECTOR_ELT instead of XVECEXP.
5307 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5309 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5310 the selector elements to be different from the data elements
5311 if the selector is a VECTOR_CST.
5312 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5313 ssizetype for the selector.
5315 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5317 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5318 before testing each element individually.
5319 * tree-vect-generic.c (lower_vec_perm): Likewise.
5321 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5323 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5324 * selftest-run-tests.c (selftest::run_tests): Call it.
5325 * vector-builder.h (vector_builder::operator ==): New function.
5326 (vector_builder::operator !=): Likewise.
5327 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5328 (vec_perm_indices::all_from_input_p): New function.
5329 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5330 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5331 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5332 instead of reading the VECTOR_CST directly. Detect whether both
5333 vector inputs are the same before constructing the vec_perm_indices,
5334 and update the number of inputs argument accordingly. Use the
5335 utility functions added above. Only construct sel2 if we need to.
5337 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5339 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5340 the broadcast of the low byte.
5341 (expand_mult_highpart): Use an explicit encoding for the permutes.
5342 * optabs-query.c (can_mult_highpart_p): Likewise.
5343 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5344 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5345 (vectorizable_bswap): Likewise.
5346 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5347 explicit encoding for the power-of-2 permutes.
5348 (vect_permute_store_chain): Likewise.
5349 (vect_grouped_load_supported): Likewise.
5350 (vect_permute_load_chain): Likewise.
5352 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5354 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5355 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5356 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5357 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5358 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5359 (vect_gen_perm_mask_any): Likewise.
5361 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5363 * int-vector-builder.h: New file.
5364 * vec-perm-indices.h: Include int-vector-builder.h.
5365 (vec_perm_indices): Redefine as an int_vector_builder.
5366 (auto_vec_perm_indices): Delete.
5367 (vec_perm_builder): Redefine as a stand-alone class.
5368 (vec_perm_indices::vec_perm_indices): New function.
5369 (vec_perm_indices::clamp): Likewise.
5370 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5371 (vec_perm_indices::new_vector): New function.
5372 (vec_perm_indices::new_expanded_vector): Update for new
5373 vec_perm_indices class.
5374 (vec_perm_indices::rotate_inputs): New function.
5375 (vec_perm_indices::all_in_range_p): Operate directly on the
5376 encoded form, without computing elided elements.
5377 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5378 encoding. Update for new vec_perm_indices class.
5379 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5380 the given vec_perm_builder.
5381 (expand_vec_perm_var): Update vec_perm_builder constructor.
5382 (expand_mult_highpart): Use vec_perm_builder instead of
5383 auto_vec_perm_indices.
5384 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5385 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5386 or double series encoding as appropriate.
5387 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5388 vec_perm_indices instead of auto_vec_perm_indices.
5389 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5390 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5391 (vect_permute_store_chain): Likewise.
5392 (vect_grouped_load_supported): Likewise.
5393 (vect_permute_load_chain): Likewise.
5394 (vect_shift_permute_load_chain): Likewise.
5395 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5396 (vect_transform_slp_perm_load): Likewise.
5397 (vect_schedule_slp_instance): Likewise.
5398 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5399 (vectorizable_mask_load_store): Likewise.
5400 (vectorizable_bswap): Likewise.
5401 (vectorizable_store): Likewise.
5402 (vectorizable_load): Likewise.
5403 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5404 vec_perm_indices instead of auto_vec_perm_indices. Use
5405 tree_to_vec_perm_builder to read the vector from a tree.
5406 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5407 vec_perm_builder instead of a vec_perm_indices.
5408 (have_whole_vector_shift): Use vec_perm_builder and
5409 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5410 truncation to calc_vec_perm_mask_for_shift.
5411 (vect_create_epilog_for_reduction): Likewise.
5412 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5413 from auto_vec_perm_indices to vec_perm_indices.
5414 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5415 instead of changing individual elements.
5416 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5417 the vector in d.perm.
5418 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5419 from auto_vec_perm_indices to vec_perm_indices.
5420 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5421 instead of changing individual elements.
5422 (arm_vectorize_vec_perm_const): Use new_vector to install
5423 the vector in d.perm.
5424 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5425 Update vec_perm_builder constructor.
5426 (rs6000_expand_interleave): Likewise.
5427 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5428 (rs6000_expand_interleave): Likewise.
5430 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5432 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5433 to qimode could truncate the indices.
5434 * optabs.c (expand_vec_perm_var): Likewise.
5436 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5438 * Makefile.in (OBJS): Add vec-perm-indices.o.
5439 * vec-perm-indices.h: New file.
5440 * vec-perm-indices.c: Likewise.
5441 * target.h (vec_perm_indices): Replace with a forward class
5443 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5444 * optabs.h: Include vec-perm-indices.h.
5445 (expand_vec_perm): Delete.
5446 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5447 (expand_vec_perm_const): Declare.
5448 * target.def (vec_perm_const_ok): Replace with...
5449 (vec_perm_const): ...this new hook.
5450 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5451 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5452 * doc/tm.texi: Regenerate.
5453 * optabs.def (vec_perm_const): Delete.
5454 * doc/md.texi (vec_perm_const): Likewise.
5455 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5456 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5457 expand_vec_perm for constant permutation vectors. Assert that
5458 the mode of variable permutation vectors is the integer equivalent
5459 of the mode that is being permuted.
5460 * optabs-query.h (selector_fits_mode_p): Declare.
5461 * optabs-query.c: Include vec-perm-indices.h.
5462 (selector_fits_mode_p): New function.
5463 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5464 is defined, instead of checking whether the vec_perm_const_optab
5465 exists. Use targetm.vectorize.vec_perm_const instead of
5466 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5467 fit in the vector mode before using a variable permute.
5468 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5469 vec_perm_indices instead of an rtx.
5470 (expand_vec_perm): Replace with...
5471 (expand_vec_perm_const): ...this new function. Take the selector
5472 as a vec_perm_indices rather than an rtx. Also take the mode of
5473 the selector. Update call to shift_amt_for_vec_perm_mask.
5474 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5475 Use vec_perm_indices::new_expanded_vector to expand the original
5476 selector into bytes. Check whether the indices fit in the vector
5477 mode before using a variable permute.
5478 (expand_vec_perm_var): Make global.
5479 (expand_mult_highpart): Use expand_vec_perm_const.
5480 * fold-const.c: Includes vec-perm-indices.h.
5481 * tree-ssa-forwprop.c: Likewise.
5482 * tree-vect-data-refs.c: Likewise.
5483 * tree-vect-generic.c: Likewise.
5484 * tree-vect-loop.c: Likewise.
5485 * tree-vect-slp.c: Likewise.
5486 * tree-vect-stmts.c: Likewise.
5487 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5489 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5490 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5491 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5492 (aarch64_vectorize_vec_perm_const): ...this new function.
5493 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5494 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5495 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5496 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5497 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5498 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5499 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5501 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5502 check for NEON modes.
5503 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5504 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5505 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5506 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5508 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5509 the old VEC_PERM_CONST conditions.
5510 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5511 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5512 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5513 (ia64_vectorize_vec_perm_const_ok): Merge into...
5514 (ia64_vectorize_vec_perm_const): ...this new function.
5515 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5516 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5517 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5518 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5519 * config/mips/mips.c (mips_expand_vec_perm_const)
5520 (mips_vectorize_vec_perm_const_ok): Merge into...
5521 (mips_vectorize_vec_perm_const): ...this new function.
5522 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5523 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5524 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5525 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5526 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5527 (rs6000_expand_vec_perm_const): Delete.
5528 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5530 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5531 (altivec_expand_vec_perm_const_le): Take each operand individually.
5532 Operate on constant selectors rather than rtxes.
5533 (altivec_expand_vec_perm_const): Likewise. Update call to
5534 altivec_expand_vec_perm_const_le.
5535 (rs6000_expand_vec_perm_const): Delete.
5536 (rs6000_vectorize_vec_perm_const_ok): Delete.
5537 (rs6000_vectorize_vec_perm_const): New function.
5538 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5539 an element count and rtx array.
5540 (rs6000_expand_extract_even): Update call accordingly.
5541 (rs6000_expand_interleave): Likewise.
5542 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5543 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5544 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5545 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5546 (rs6000_expand_vec_perm_const): Delete.
5547 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5548 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5549 (altivec_expand_vec_perm_const_le): Take each operand individually.
5550 Operate on constant selectors rather than rtxes.
5551 (altivec_expand_vec_perm_const): Likewise. Update call to
5552 altivec_expand_vec_perm_const_le.
5553 (rs6000_expand_vec_perm_const): Delete.
5554 (rs6000_vectorize_vec_perm_const_ok): Delete.
5555 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5556 reference to the SPE evmerge intructions.
5557 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5558 an element count and rtx array.
5559 (rs6000_expand_extract_even): Update call accordingly.
5560 (rs6000_expand_interleave): Likewise.
5561 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5562 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5564 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5566 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5568 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5569 vector mode and that that mode matches the mode of the data
5571 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5572 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5573 directly using expand_vec_perm_1 when forcing selectors into
5575 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5577 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5579 * optabs-query.h (can_vec_perm_p): Delete.
5580 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5581 * optabs-query.c (can_vec_perm_p): Split into...
5582 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5583 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5584 particular selector is valid.
5585 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5586 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5587 (vect_grouped_load_supported): Likewise.
5588 (vect_shift_permute_load_chain): Likewise.
5589 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5590 (vect_transform_slp_perm_load): Likewise.
5591 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5592 (vectorizable_bswap): Likewise.
5593 (vect_gen_perm_mask_checked): Likewise.
5594 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5595 implementations of variable permutation vectors into account
5596 when deciding which selector to use.
5597 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5598 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5599 with a false third argument.
5600 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5601 to test whether the constant selector is valid and can_vec_perm_var_p
5602 to test whether a variable selector is valid.
5604 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5606 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5607 * optabs-query.c (can_vec_perm_p): Likewise.
5608 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5609 instead of vec_perm_indices.
5610 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5611 (vect_gen_perm_mask_checked): Likewise,
5612 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5613 (vect_gen_perm_mask_checked): Likewise,
5615 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5617 * optabs-query.h (qimode_for_vec_perm): Declare.
5618 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5619 (qimode_for_vec_perm): ...this new function.
5620 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5622 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5624 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5625 does not have a conditional at the top.
5627 2018-01-02 Richard Biener <rguenther@suse.de>
5629 * ipa-inline.c (big_speedup_p): Fix expression.
5631 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5634 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5637 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5641 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5642 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5643 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5644 cond_taken_branch_cost 3->4.
5646 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5648 PR tree-optimization/83581
5649 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5650 TODO_cleanup_cfg if any changes have been made.
5653 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5654 convert_modes if target mode has the right side, but different mode
5658 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5659 last argument when extracting from CONCAT. If either from_real or
5660 from_imag is NULL, use expansion through memory. If result is not
5661 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5662 the parts directly to inner mode, if even that fails, use expansion
5666 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5667 check for bswap in mode rather than HImode and use that in expand_unop
5670 Copyright (C) 2018 Free Software Foundation, Inc.
5672 Copying and distribution of this file, with or without modification,
5673 are permitted in any medium without royalty provided the copyright
5674 notice and this notice are preserved.