1 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
3 * profile-count.c (profile_count::combine_with_ipa_count): Handle
6 2018-01-30 Richard Biener <rguenther@suse.de>
8 PR tree-optimization/83008
9 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
10 invariant and constant vector uses in stmts when they need
13 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
16 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
17 * configure: Regenerate.
19 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
21 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
23 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
24 Use gen_rtx_REG rather than gen_lowpart.
26 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
28 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
29 rather than 0 when creating partial subregs.
31 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
33 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
36 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
39 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
40 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
41 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
42 flags. This restores the settings used before the 2017-07-24.
43 Turning off pre increment/decrement/modify allows IVOPTS to
44 optimize DF/SF loops where the index is an int.
46 2018-01-29 Richard Biener <rguenther@suse.de>
47 Kelvin Nilsen <kelvin@gcc.gnu.org>
50 * tree-vect-stmts.c (vectorizable_call): Don't call
51 targetm.vectorize_builtin_md_vectorized_function if callee is
54 2018-01-22 Carl Love <cel@us.ibm.com>
56 * doc/extend.tex: Fix typo in second arg in
57 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
59 2018-01-29 Richard Biener <rguenther@suse.de>
61 PR tree-optimization/84086
62 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
63 (flush_ssaname_freelist): When SSA names were released reset
66 2018-01-29 Richard Biener <rguenther@suse.de>
68 PR tree-optimization/84057
69 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
70 removed paths when removing edges.
72 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
74 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
75 -mfunction-return=@var{choice}.
77 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
80 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
82 (layout::print_source_line): Likewise.
83 (test_get_line_width_without_trailing_whitespace): Add test cases.
85 2018-01-27 Jakub Jelinek <jakub@redhat.com>
88 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
91 2018-01-26 Jim Wilson <jimw@sifive.com>
93 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
95 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
98 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
100 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
101 and CMP + SUB-immediate -> SUBS.
103 2018-01-26 Martin Sebor <msebor@redhat.com>
105 PR tree-optimization/83896
106 * tree-ssa-strlen.c (get_string_len): Rename...
107 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
108 Avoid assuming length is constant.
109 (handle_char_store): Use HOST_WIDE_INT for string length.
111 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
114 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
115 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
117 2018-01-26 Richard Biener <rguenther@suse.de>
119 PR rtl-optimization/84003
120 * dse.c (record_store): Only record redundant stores when
121 the earlier store aliases at least all accesses the later one does.
123 2018-01-26 Jakub Jelinek <jakub@redhat.com>
125 PR rtl-optimization/83985
126 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
127 REG_CFA_RESTORE insns.
128 (delete_unmarked_insns): Don't ignore separate shrink wrapping
129 REG_CFA_RESTORE insns here.
132 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
133 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
135 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
137 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
138 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
139 (arc_init): Likewise.
140 (arc_override_options): Likewise.
141 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
143 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
145 * config/arc/arc.h (TARGET_DBNZ): Define.
146 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
147 properly set the tune attribute.
148 (dbnz): Use TARGET_DBNZ guard.
149 * config/arc/arc.opt (mtune): Add core3 option.
151 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
153 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
154 recognize new pic like addresses.
155 (arc_delegitimize_address): Clean up.
157 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
159 * config/arc/arc-arches.def: Option mrf16 valid for all
161 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
162 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
163 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
164 * config/arc/arc-tables.opt: Regenerate.
165 * config/arc/arc.c (arc_conditional_register_usage): Handle
166 reduced register file case.
167 (arc_file_start): Set must have build attributes.
168 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
170 * config/arc/arc.opt (mrf16): Add new option.
171 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
172 * config/arc/genmultilib.awk: Handle new mrf16 option.
173 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
174 * config/arc/t-multilib: Regenerate.
175 * doc/invoke.texi (ARC Options): Document mrf16 option.
177 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
179 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
180 * config/arc/arc.c (arc_handle_secure_attribute): New function.
181 (arc_attribute_table): Add 'secure_call' attribute.
182 (arc_print_operand): Print secure call operand.
183 (arc_function_ok_for_sibcall): Don't optimize tail calls when
185 (arc_is_secure_call_p): New function. * config/arc/arc.md
186 (call_i): Add support for sjli instruction.
187 (call_value_i): Likewise.
188 * config/arc/constraints.md (Csc): New constraint.
190 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
191 John Eric Martin <John.Martin@emmicro-us.com>
193 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
194 * config/arc/arc.c (_arc_jli_section): New struct.
195 (arc_jli_section): New type.
196 (rc_jli_sections): New static variable.
197 (arc_handle_jli_attribute): New function.
198 (arc_attribute_table): Add jli_always and jli_fixed attribute.
199 (arc_file_end): New function.
200 (TARGET_ASM_FILE_END): Define.
201 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
202 (arc_add_jli_section): New function.
203 (jli_call_scan): Likewise.
204 (arc_reorg): Call jli_call_scan.
205 (arc_output_addsi): Remove 'S' from printing asm operand.
206 (arc_is_jli_call_p): New function.
207 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
209 (movhi_insn): Likewise.
210 (movsi_insn): Likewise.
211 (movsi_set_cc_insn): Likewise.
212 (loadqi_update): Likewise.
213 (load_zeroextendqisi_update): Likewise.
214 (load_signextendqisi_update): Likewise.
215 (loadhi_update): Likewise.
216 (load_zeroextendhisi_update): Likewise.
217 (load_signextendhisi_update): Likewise.
218 (loadsi_update): Likewise.
219 (loadsf_update): Likewise.
220 (movsicc_insn): Likewise.
221 (bset_insn): Likewise.
222 (bxor_insn): Likewise.
223 (bclr_insn): Likewise.
224 (bmsk_insn): Likewise.
225 (bicsi3_insn): Likewise.
226 (cmpsi_cc_c_insn): Likewise.
227 (movsi_ne): Likewise.
228 (movsi_cond_exec): Likewise.
229 (clrsbsi2): Likewise.
240 (call_i): Remove 'S' asm letter, add jli instruction.
241 (call_value_i): Likewise.
242 * config/arc/arc.op (mjli-always): New option.
243 * config/arc/constraints.md (Cji): New constraint.
244 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
246 (subsf3_fpx): Likewise.
247 (mulsf3_fpx): Likewise.
248 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
250 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
252 * doc/invoke.texi (ARC): Document mjli-always option.
254 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
256 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
257 and use incw and decw where possible
258 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
260 2018-01-26 Richard Biener <rguenther@suse.de>
262 PR tree-optimization/81082
263 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
264 association if it requires casting to unsigned.
265 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
266 from fold_plusminus_mult_expr to catch important cases late when
267 range info is available.
269 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
271 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
272 * configure.ac (hidden_linkonce): New test.
273 * configure: Regenerate.
274 * config.in: Regenerate.
276 2018-01-26 Julia Koval <julia.koval@intel.com>
278 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
279 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
280 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
281 _mm_mask_bitshuffle_epi64_mask): Fix type.
282 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
283 USI_FTYPE_V4DI_V4DI_USI): Remove.
284 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
285 __builtin_ia32_vpshufbitqmb256_mask,
286 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
287 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
288 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
290 2018-01-26 Alan Modra <amodra@gmail.com>
293 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
294 UNSPEC_VBPERMQ. Sort other unspecs.
296 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
298 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
300 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
303 * predict.c (drop_profile): Do not push/pop cfun; update also
305 (handle_missing_profiles): Fix logic looking for zero profiles.
307 2018-01-25 Jakub Jelinek <jakub@redhat.com>
310 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
311 on functions with #pragma omp declare simd or functions with simd
313 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
314 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
315 Remove trailing \n from warning_at calls.
317 2018-01-25 Tom de Vries <tom@codesourcery.com>
320 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
321 for neutered workers.
323 2018-01-24 Joseph Myers <joseph@codesourcery.com>
326 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
327 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
329 2017-01-08 Jeff Law <law@redhat.com>
332 * i386.c (get_probe_interval): Move to earlier point.
333 (ix86_compute_frame_layout): If -fstack-clash-protection and
334 the frame is larger than the probe interval, then use pushes
335 to save registers rather than reg->mem moves.
336 (ix86_expand_prologue): Remove conditional for int_registers_saved
339 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
342 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
343 min/max for never referenced object.
345 2018-01-24 Jakub Jelinek <jakub@redhat.com>
348 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
350 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
351 attributes from DECL_ATTRIBUTES (decl) without affecting
352 DECL_ATTRIBUTES (current_function_decl).
353 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
354 functions with non-NULL DECL_ABSTRACT_ORIGIN.
356 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
358 PR tree-optimization/83979
359 * fold-const.c (fold_comparison): Use constant_boolean_node
360 instead of boolean_{true,false}_node.
362 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
364 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
367 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
369 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
370 Simplify the clause that sets the length attribute.
371 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
372 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
373 clause that sets the length attribute.
374 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
376 2018-01-24 Tom de Vries <tom@codesourcery.com>
379 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
380 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
381 Add strict parameter.
382 (prevent_branch_around_nothing): Insert dummy insn between branch to
383 label and label with no ptx insn inbetween.
384 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
386 2018-01-24 Tom de Vries <tom@codesourcery.com>
389 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
390 for neutered threads in warp.
391 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
393 2018-01-24 Richard Biener <rguenther@suse.de>
395 PR tree-optimization/83176
396 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
399 2018-01-24 Richard Biener <rguenther@suse.de>
401 PR tree-optimization/82819
402 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
403 code generating pluses that are no-ops in the target precision.
405 2018-01-24 Richard Biener <rguenther@suse.de>
408 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
410 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
412 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
413 to merge probabilities.
414 * predict.c (probably_never_executed): Also mark as cold functions
415 with global 0 profile and guessed local profile.
416 * profile-count.c (profile_probability::combine_with_count): New
418 * profile-count.h (profile_probability::operator*,
419 profile_probability::operator*=, profile_probability::operator/,
420 profile_probability::operator/=): Reduce precision to adjusted
421 and set value to guessed on contradictory divisions.
422 (profile_probability::combine_with_freq): Remove.
423 (profile_probability::combine_wiht_count): Declare.
424 (profile_count::force_nonzero):: Set to adjusted.
425 (profile_count::probability_in):: Set quality to adjusted.
426 * tree-ssa-tail-merge.c (replace_block_by): Use
429 2018-01-23 Andrew Waterman <andrew@sifive.com>
430 Jim Wilson <jimw@sifive.com>
432 * config/riscv/riscv.c (riscv_stack_boundary): New.
433 (riscv_option_override): Set riscv_stack_boundary. Handle
434 riscv_preferred_stack_boundary_arg.
435 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
436 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
437 (STACK_BOUNDARY): Set to riscv_stack_boundary.
438 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
439 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
440 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
442 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
445 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
446 of struct ix86_frame.
447 (ix86_expand_epilogue): Likewise. Add a local variable for
448 the reg_save_offset field in struct ix86_frame.
450 2018-01-23 Bin Cheng <bin.cheng@arm.com>
452 PR tree-optimization/82604
453 * tree-loop-distribution.c (enum partition_kind): New enum item
454 PKIND_PARTIAL_MEMSET.
455 (partition_builtin_p): Support above new enum item.
456 (generate_code_for_partition): Ditto.
457 (compute_access_range): Differentiate cases that equality can be
458 proven at all loops, the innermost loops or no loops.
459 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
460 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
461 (finalize_partitions, distribute_loop): Don't fuse partition of
462 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
463 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
466 2018-01-23 Martin Liska <mliska@suse.cz>
468 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
469 order to ignore the predictor.
470 (PRED_POLYMORPHIC_CALL): Likewise.
471 (PRED_RECURSIVE_CALL): Likewise.
473 2018-01-23 Martin Liska <mliska@suse.cz>
475 * tree-profile.c (tree_profiling): Print function header to
476 aware reader which function we are working on.
477 * value-prof.c (gimple_find_values_to_profile): Do not print
478 not interesting value histograms.
480 2018-01-23 Martin Liska <mliska@suse.cz>
482 * profile-count.h (enum profile_quality): Add
483 profile_uninitialized as the first value. Do not number values
484 as they are zero based.
485 (profile_count::verify): Update sanity check.
486 (profile_probability::verify): Likewise.
488 2018-01-23 Nathan Sidwell <nathan@acm.org>
490 * doc/invoke.texi (ffor-scope): Deprecate.
492 2018-01-23 David Malcolm <dmalcolm@redhat.com>
494 PR tree-optimization/83510
495 * domwalk.c (set_all_edges_as_executable): New function.
496 (dom_walker::dom_walker): Convert bool param
497 "skip_unreachable_blocks" to enum reachability. Move setup of
498 edge flags to set_all_edges_as_executable and only do it when
499 reachability is REACHABLE_BLOCKS.
500 * domwalk.h (enum dom_walker::reachability): New enum.
501 (dom_walker::dom_walker): Convert bool param
502 "skip_unreachable_blocks" to enum reachability.
503 (set_all_edges_as_executable): New decl.
504 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
505 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
507 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
508 but converting true to REACHABLE_BLOCKS.
509 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
511 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
512 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
513 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
515 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
516 if check_all_array_refs will be called.
518 2018-01-23 David Malcolm <dmalcolm@redhat.com>
520 * tree.c (selftest::test_location_wrappers): Add more test
523 2018-01-23 David Malcolm <dmalcolm@redhat.com>
525 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
526 (selftest::test_bit_in_range): Likewise.
528 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
531 * doc/sourcebuild.texi (vect_float): Say that the selector
532 only describes the situation when -funsafe-math-optimizations is on.
533 (vect_float_strict): Document.
535 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
537 PR tree-optimization/83965
538 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
539 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
540 instead of checking only for a reduction.
541 (vect_recog_widen_sum_pattern): Likewise.
543 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
545 * predict.c (probably_never_executed): Only use precise profile info.
546 (compute_function_frequency): Skip after inlining hack since we now
547 have quality checking.
549 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
551 * profile-count.h (profile_probability::very_unlikely,
552 profile_probability::unlikely, profile_probability::even): Set
553 precision to guessed.
555 2018-01-23 Richard Biener <rguenther@suse.de>
557 PR tree-optimization/83963
558 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
559 Properly terminate dominator walk when crossing the exit edge not
560 when visiting its source block.
562 2018-01-23 Jakub Jelinek <jakub@redhat.com>
565 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
566 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
568 2018-01-22 Jakub Jelinek <jakub@redhat.com>
570 PR tree-optimization/83957
571 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
572 semicolon after for body surrounded by braces.
574 PR tree-optimization/83081
575 * profile-count.h (profile_probability::split): New method.
576 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
577 Use profile_probability::split.
578 (do_compare_rtx_and_jump): Fix adjustment of probabilities
579 when splitting a single conditional jump into 2.
581 2018-01-22 David Malcolm <dmalcolm@redhat.com>
583 PR tree-optimization/69452
584 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
587 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
589 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
590 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
591 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
593 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
595 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
596 * config/rl78/rl78.md: New define_expand "movdi"
597 * config/rl78/rl78.c: New function definition rl78_split_movdi
599 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
602 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
604 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
605 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
606 128-bit to produce an UNSPEC move to get the double word with the
607 signbit and then a shift directly to do signbit.
608 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
609 implementation with a new version that just does either a direct
610 move or a regular move. Move memory interface to separate insns.
611 Move insns so they are next to the expander.
612 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
613 with signbit move. Split big and little endian case.
614 (signbit<mode>2_dm_mem_le): Likewise.
615 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
616 (signbit<mode>2_dm2): Likewise.
618 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
620 * config/rl78/rl78.md: New define_expand "anddi3".
622 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
624 * config/rl78/rl78.md: New define_expand "umindi3".
626 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
628 * config/rl78/rl78.md: New define_expand "smindi3".
630 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
632 * config/rl78/rl78.md: New define_expand "smaxdi3".
634 2018-01-22 Carl Love <cel@us.ibm.com>
636 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
637 LVX_V1TI): Add macro expansion.
638 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
639 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
640 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
641 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
642 Change check to determine if the instruction is a byte reversing
643 entry. Fix typo in comment.
644 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
645 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
646 Add def_builtin calls for new builtins.
647 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
648 Add define_insn expansion.
650 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
652 * config/rl78/rl78.md: New define_expand "umaxdi3".
654 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
656 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
657 for non-QImode registers
659 2018-01-22 Richard Biener <rguenther@suse.de>
661 PR tree-optimization/83963
662 * graphite-scop-detection.c (scop_detection::get_sese): Delay
663 including the loop exit block.
664 (scop_detection::merge_sese): Likewise.
665 (scop_detection::add_scop): Do it here instead.
667 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
669 * doc/sourcebuild.texi (arm_softfloat): Document.
671 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
674 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
675 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
676 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
678 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
679 David Edelsohn <dje.gcc@gmail.com>
682 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
683 Change "crset eq" to "crset 2".
684 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
685 (*call_indirect_aix<mode>_nospec): Likewise.
686 (*call_value_indirect_aix<mode>_nospec): Likewise.
687 (*call_indirect_elfv2<mode>_nospec): Likewise.
688 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
689 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
690 change assembly output from . to $.
691 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
692 (indirect_jump<mode>_nospec): Change assembly output from . to $.
693 (*tablejump<mode>_internal1_nospec): Likewise.
695 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
698 * config/sh/sh_optimize_sett_clrt.cc:
699 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
701 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
703 PR tree-optimization/83940
704 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
705 offset_dt to vect_constant_def rather than vect_unknown_def_type.
706 (vect_check_load_store_mask): Add a mask_dt_out parameter and
707 use it to pass back the definition type.
708 (vect_check_store_rhs): Likewise rhs_dt_out.
709 (vect_build_gather_load_calls): Add a mask_dt argument and use
710 it instead of a call to vect_is_simple_use.
711 (vectorizable_store): Update calls to vect_check_load_store_mask
712 and vect_check_store_rhs. Use the dt returned by the latter instead
713 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
714 instead of calls to vect_is_simple_use. Pass the scalar rather
715 than the vector operand to vect_is_simple_use when handling
716 second and subsequent copies of an rhs value.
717 (vectorizable_load): Update calls to vect_check_load_store_mask
718 and vect_build_gather_load_calls. Use the cached mask_dt and
719 gs_info.offset_dt instead of calls to vect_is_simple_use.
721 2018-01-20 Jakub Jelinek <jakub@redhat.com>
724 * tree-emutls.c: Include gimplify.h.
725 (lower_emutls_2): New function.
726 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
727 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
728 it before further processing.
731 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
732 UINTVAL (trueop1) instead of INTVAL (op1).
734 2018-01-19 Jakub Jelinek <jakub@redhat.com>
738 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
739 INCOMING_FRAME_SP_OFFSET if not defined.
740 (scan_trace): Add ENTRY argument. If true and
741 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
742 emit a note to adjust the CFA offset.
743 (create_cfi_notes): Adjust scan_trace callers.
744 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
745 INCOMING_FRAME_SP_OFFSET in the CIE.
746 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
747 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
749 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
750 * doc/tm.texi: Regenerated.
752 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
754 PR rtl-optimization/83147
755 * lra-constraints.c (remove_inheritance_pseudos): Use
756 lra_substitute_pseudo_within_insn.
758 2018-01-19 Tom de Vries <tom@codesourcery.com>
759 Cesar Philippidis <cesar@codesourcery.com>
762 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
764 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
767 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
768 spaces for function labels.
770 2018-01-19 Martin Liska <mliska@suse.cz>
772 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
773 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
774 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
775 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
776 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
777 (PRED_CONST_RETURN): Change from 69 to 65.
778 (PRED_NULL_RETURN): Change from 91 to 71.
779 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
780 (PRED_LOOP_GUARD): Change from 66 to 73.
782 2018-01-19 Martin Liska <mliska@suse.cz>
784 * predict.c (predict_insn_def): Add new assert.
785 (struct branch_predictor): Change type to signed integer.
786 (test_prediction_value_range): Amend test to cover
788 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
789 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
790 (PRED_LOOP_ITERATIONS_MAX): Likewise.
791 (PRED_LOOP_IV_COMPARE): Likewise.
792 * predict.h (PROB_UNINITIALIZED): Define new constant.
794 2018-01-19 Martin Liska <mliska@suse.cz>
796 * predict.c (dump_prediction): Add new format for
797 analyze_brprob.py script which is enabled with -details
799 * profile-count.h (precise_p): New function.
801 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
803 PR tree-optimization/83922
804 * tree-vect-loop.c (vect_verify_full_masking): Return false if
805 there are no statements that need masking.
806 (vect_active_double_reduction_p): New function.
807 (vect_analyze_loop_operations): Use it when handling phis that
808 are not in the loop header.
810 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
812 PR tree-optimization/83914
813 * tree-vect-loop.c (vectorizable_induction): Don't convert
814 init_expr or apply the peeling adjustment for inductions
815 that are nested within the vectorized loop.
817 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
819 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
822 2018-01-18 Jakub Jelinek <jakub@redhat.com>
826 * function.h (gimplify_parameters): Add gimple_seq * argument.
827 * function.c: Include gimple.h and options.h.
828 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
829 for the added local temporaries if needed.
830 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
831 if there are any parameter cleanups, wrap whole body into a
832 try/finally with the cleanups.
834 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
837 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
838 Use GET_MODE_CLASS for scalar floating point.
840 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
844 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
845 Fix call of call_cgraph_insertion_hooks.
847 2018-01-18 Martin Sebor <msebor@redhat.com>
849 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
851 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
854 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
857 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
860 * common.opt: (-ffile-prefix-map): New option.
861 * opts.c (common_handle_option): Defer it.
862 * opts-global.c (handle_common_deferred_options): Handle it.
863 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
864 * file-prefix-map.h: New file.
865 (remap_debug_filename, add_debug_prefix_map): ...here.
866 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
867 * final.c (debug_prefix_map, add_debug_prefix_map
868 remap_debug_filename): Move to...
869 * file-prefix-map.c: New file.
870 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
871 generalize, get rid of alloca(), use strrchr() instead of strchr().
872 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
873 Implement in terms of add_prefix_map().
874 (remap_macro_filename, remap_debug_filename): Implement in term of
876 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
877 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
878 * dbxout.c: Include file-prefix-map.h.
879 * varasm.c: Likewise.
880 * vmsdbgout.c: Likewise.
881 * xcoffout.c: Likewise.
882 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
883 * doc/cppopts.texi (-fmacro-prefix-map): Document.
884 * doc/invoke.texi (-ffile-prefix-map): Document.
885 (-fdebug-prefix-map): Update description.
887 2018-01-18 Martin Liska <mliska@suse.cz>
889 * config/i386/i386.c (indirect_thunk_name): Document that also
891 (output_indirect_thunk): Document why both instructions
892 (pause and lfence) are generated.
894 2018-01-18 Richard Biener <rguenther@suse.de>
896 PR tree-optimization/83887
897 * graphite-scop-detection.c
898 (scop_detection::get_nearest_dom_with_single_entry): Remove.
899 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
900 (scop_detection::merge_sese): Re-implement with a flood-fill
901 algorithm that properly finds a SESE region if it exists.
903 2018-01-18 Jakub Jelinek <jakub@redhat.com>
906 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
907 pointer_diff optimizations use view_convert instead of convert.
909 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
911 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
912 Generate different code for -mno-speculate-indirect-jumps.
913 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
914 (*call_indirect_aix<mode>): Disable for
915 -mno-speculate-indirect-jumps.
916 (*call_indirect_aix<mode>_nospec): New define_insn.
917 (*call_value_indirect_aix<mode>): Disable for
918 -mno-speculate-indirect-jumps.
919 (*call_value_indirect_aix<mode>_nospec): New define_insn.
920 (*sibcall_nonlocal_sysv<mode>): Generate different code for
921 -mno-speculate-indirect-jumps.
922 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
924 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
926 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
927 long double type, set the flags for noting the default long double
928 type, even if we don't pass or return a long double type.
930 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
933 * ipa-inline.c (flatten_function): Do not overwrite final inlining
936 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
938 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
939 support for merge[hl].
940 (fold_mergehl_helper): New helper function.
941 (tree-vector-builder.h): New #include for tree_vector_builder usage.
942 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
943 (altivec_vmrglw_direct): Add xxmrglw insn.
945 2018-01-17 Andrew Waterman <andrew@sifive.com>
947 * config/riscv/riscv.c (riscv_conditional_register_usage): If
948 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
950 2018-01-17 David Malcolm <dmalcolm@redhat.com>
953 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
954 call the lto_location_cache before reading the
955 DECL_SOURCE_LOCATION of the types.
957 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
958 Richard Sandiford <richard.sandiford@linaro.org>
960 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
961 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
962 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
963 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
964 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
966 * config/aarch64/constraints.md (aarch64_movti_operand):
968 * config/aarch64/predicates.md (Uti): Add new constraint.
970 2018-01-17 Carl Love <cel@us.ibm.com>
971 * config/rs6000/vsx.md (define_expand xl_len_r,
972 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
973 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
975 (define_expand, define_insn): Move the shift left from the
976 define_insn to the define_expand for lxvl and stxvl instructions.
977 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
978 and XL_LEN_R definitions to PURE.
980 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
982 * config/i386/i386.c (indirect_thunk_name): Declare regno
983 as unsigned int. Compare regno with INVALID_REGNUM.
984 (output_indirect_thunk): Ditto.
985 (output_indirect_thunk_function): Ditto.
986 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
987 in the call to output_indirect_thunk_function.
989 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
992 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
993 rather than the size of inner_type to determine the stack slot size
994 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
996 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
999 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1002 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1004 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1005 endian Linux systems to optionally enable multilibs for selecting
1006 the long double type if the user configured an explicit type.
1007 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1008 have no long double multilibs if not defined.
1009 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1010 warn if the user used -mabi={ieee,ibm}longdouble and we built
1011 multilibs for long double.
1012 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1013 appropriate multilib option.
1014 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1016 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1017 for building long double multilibs.
1018 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1020 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1022 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1025 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1027 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1030 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1033 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1036 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1038 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1039 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1041 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1043 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1044 different rtl trees depending on TARGET_64BIT.
1045 (rs6000_gen_lvx): Likewise.
1047 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1049 * config/visium/visium.md (nop): Tweak comment.
1050 (hazard_nop): Likewise.
1052 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1054 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1055 -mspeculate-indirect-jumps.
1056 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1057 for -mno-speculate-indirect-jumps.
1058 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1059 (*call_value_indirect_elfv2<mode>): Disable for
1060 -mno-speculate-indirect-jumps.
1061 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1062 (indirect_jump): Emit different RTL for
1063 -mno-speculate-indirect-jumps.
1064 (*indirect_jump<mode>): Disable for
1065 -mno-speculate-indirect-jumps.
1066 (*indirect_jump<mode>_nospec): New define_insn.
1067 (tablejump): Emit different RTL for
1068 -mno-speculate-indirect-jumps.
1069 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1070 (tablejumpsi_nospec): New define_expand.
1071 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1072 (tablejumpdi_nospec): New define_expand.
1073 (*tablejump<mode>_internal1): Disable for
1074 -mno-speculate-indirect-jumps.
1075 (*tablejump<mode>_internal1_nospec): New define_insn.
1076 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1079 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1081 * caller-save.c (insert_save): Drop unnecessary parameter. All
1084 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1085 Richard Biener <rguenth@suse.de>
1088 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1089 return early, inline manually is_gimple_sizepos. Make sure if we
1090 call gimplify_expr we don't end up with a gimple constant.
1091 * tree.c (variably_modified_type_p): Don't return true for
1092 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1093 * gimplify.h (is_gimple_sizepos): Remove.
1095 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1097 PR tree-optimization/83857
1098 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1099 vectorizable_live_operation for pure SLP statements.
1100 (vectorizable_live_operation): Handle PHIs.
1102 2018-01-16 Richard Biener <rguenther@suse.de>
1104 PR tree-optimization/83867
1105 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1106 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1108 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1111 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1112 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1113 If off is not INTEGER_CST, issue a may not be aligned warning
1114 rather than isn't aligned. Use isn%'t rather than isn't.
1115 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1117 <case MULT_EXPR>: Improve the case when bottom and one of the
1118 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1119 operand, in that case check if the other operand is multiple of
1120 bottom divided by the INTEGER_CST operand.
1122 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1125 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1126 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1127 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1128 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1129 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1130 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1131 (pa_function_arg_size): New function.
1133 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1135 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1136 in a separate statement.
1138 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1140 PR tree-optimization/83847
1141 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1142 group gathers and scatters.
1144 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1146 PR rtl-optimization/86620
1147 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1149 PR rtl-optimization/83213
1150 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1151 to last if both are JUMP_INSNs.
1153 PR tree-optimization/83843
1154 * gimple-ssa-store-merging.c
1155 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1156 store_immediate_info for bswap/nop orig_stores.
1158 2018-01-15 Andrew Waterman <andrew@sifive.com>
1160 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1162 <UDIV>: Increase cost if !TARGET_DIV.
1164 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1166 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1167 (define_attr "cr_logical_3op"): New.
1168 (cceq_ior_compare): Adjust.
1169 (cceq_ior_compare_complement): Adjust.
1170 (*cceq_rev_compare): Adjust.
1171 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1172 (is_cracked_insn): Adjust.
1173 (insn_must_be_first_in_group): Adjust.
1174 * config/rs6000/40x.md: Adjust.
1175 * config/rs6000/440.md: Adjust.
1176 * config/rs6000/476.md: Adjust.
1177 * config/rs6000/601.md: Adjust.
1178 * config/rs6000/603.md: Adjust.
1179 * config/rs6000/6xx.md: Adjust.
1180 * config/rs6000/7450.md: Adjust.
1181 * config/rs6000/7xx.md: Adjust.
1182 * config/rs6000/8540.md: Adjust.
1183 * config/rs6000/cell.md: Adjust.
1184 * config/rs6000/e300c2c3.md: Adjust.
1185 * config/rs6000/e500mc.md: Adjust.
1186 * config/rs6000/e500mc64.md: Adjust.
1187 * config/rs6000/e5500.md: Adjust.
1188 * config/rs6000/e6500.md: Adjust.
1189 * config/rs6000/mpc.md: Adjust.
1190 * config/rs6000/power4.md: Adjust.
1191 * config/rs6000/power5.md: Adjust.
1192 * config/rs6000/power6.md: Adjust.
1193 * config/rs6000/power7.md: Adjust.
1194 * config/rs6000/power8.md: Adjust.
1195 * config/rs6000/power9.md: Adjust.
1196 * config/rs6000/rs64.md: Adjust.
1197 * config/rs6000/titan.md: Adjust.
1199 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1201 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1202 ix86_indirect_branch_register logic.
1204 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1206 * config/i386/constraints.md (Bs): Update
1207 ix86_indirect_branch_register check. Don't check
1208 ix86_indirect_branch_register with GOT_memory_operand.
1210 * config/i386/predicates.md (GOT_memory_operand): Don't check
1211 ix86_indirect_branch_register here.
1212 (GOT32_symbol_operand): Likewise.
1214 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1216 * config/i386/predicates.md (constant_call_address_operand):
1217 Rewrite ix86_indirect_branch_register logic.
1218 (sibcall_insn_operand): Likewise.
1220 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1222 * config/i386/constraints.md (Bs): Replace
1223 ix86_indirect_branch_thunk_register with
1224 ix86_indirect_branch_register.
1226 * config/i386/i386.md (indirect_jump): Likewise.
1227 (tablejump): Likewise.
1228 (*sibcall_memory): Likewise.
1229 (*sibcall_value_memory): Likewise.
1230 Peepholes of indirect call and jump via memory: Likewise.
1231 * config/i386/i386.opt: Likewise.
1232 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1233 (GOT_memory_operand): Likewise.
1234 (call_insn_operand): Likewise.
1235 (sibcall_insn_operand): Likewise.
1236 (GOT32_symbol_operand): Likewise.
1238 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1241 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1242 type rather than type addr's type points to.
1243 (expand_omp_atomic_mutex): Likewise.
1244 (expand_omp_atomic): Likewise.
1246 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1249 * config/i386/i386.c (output_indirect_thunk_function): Use
1250 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1251 for __x86_return_thunk.
1253 2018-01-15 Richard Biener <rguenther@suse.de>
1256 * expmed.c (extract_bit_field_1): Fix typo.
1258 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1261 * config/arm/iterators.md (VF): New mode iterator.
1262 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1263 Remove integer-related logic from pattern.
1264 (neon_vabd<mode>_3): Likewise.
1266 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1269 * common.opt (fstrict-overflow): No longer an alias.
1270 (fwrapv-pointer): New option.
1271 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1272 also for pointer types based on flag_wrapv_pointer.
1273 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1274 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1275 opts->x_flag_wrapv got set.
1276 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1277 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1278 POINTER_TYPE_OVERFLOW_UNDEFINED.
1279 * match.pd: Likewise in address comparison pattern.
1280 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1282 2018-01-15 Richard Biener <rguenther@suse.de>
1285 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1286 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1287 Reset type names to their identifier if their TYPE_DECL doesn't
1288 have linkage (and thus is used for ODR and devirt).
1289 (save_debug_info_for_decl): Remove.
1290 (save_debug_info_for_type): Likewise.
1291 (add_tree_to_fld_list): Adjust.
1292 * tree-pretty-print.c (dump_generic_node): Make dumping of
1293 type names more robust.
1295 2018-01-15 Richard Biener <rguenther@suse.de>
1297 * BASE-VER: Bump to 8.0.1.
1299 2018-01-14 Martin Sebor <msebor@redhat.com>
1302 * builtins.c (check_access): Avoid warning when the no-warning bit
1305 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1307 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1308 * ira-color (allocno_hard_regs_compare): Likewise.
1310 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1313 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1314 Use .pushsection/.popsection.
1316 2018-01-14 Martin Sebor <msebor@redhat.com>
1319 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1321 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1323 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1324 entry from extra_headers.
1325 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1326 extra_headers, make the list bitwise identical to the i?86-*-* one.
1328 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1330 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1331 -mcmodel=large with -mindirect-branch=thunk,
1332 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1333 -mfunction-return=thunk-extern.
1334 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1335 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1336 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1338 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1340 * config/i386/i386.c (print_reg): Print the name of the full
1341 integer register without '%'.
1342 (ix86_print_operand): Handle 'V'.
1343 * doc/extend.texi: Document 'V' modifier.
1345 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1347 * config/i386/constraints.md (Bs): Disallow memory operand for
1348 -mindirect-branch-register.
1350 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1351 (GOT_memory_operand): Likewise.
1352 (call_insn_operand): Likewise.
1353 (sibcall_insn_operand): Likewise.
1354 (GOT32_symbol_operand): Likewise.
1355 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1356 for -mindirect-branch-register.
1357 (tablejump): Likewise.
1358 (*sibcall_memory): Likewise.
1359 (*sibcall_value_memory): Likewise.
1360 Disallow peepholes of indirect call and jump via memory for
1361 -mindirect-branch-register.
1362 (*call_pop): Replace m with Bw.
1363 (*call_value_pop): Likewise.
1364 (*sibcall_pop_memory): Replace m with Bs.
1365 * config/i386/i386.opt (mindirect-branch-register): New option.
1366 * doc/invoke.texi: Document -mindirect-branch-register option.
1368 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1370 * config/i386/i386-protos.h (ix86_output_function_return): New.
1371 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1372 set function_return_type.
1373 (indirect_thunk_name): Add ret_p to indicate thunk for function
1375 (output_indirect_thunk_function): Pass false to
1376 indirect_thunk_name.
1377 (ix86_output_indirect_branch_via_reg): Likewise.
1378 (ix86_output_indirect_branch_via_push): Likewise.
1379 (output_indirect_thunk_function): Create alias for function
1380 return thunk if regno < 0.
1381 (ix86_output_function_return): New function.
1382 (ix86_handle_fndecl_attribute): Handle function_return.
1383 (ix86_attribute_table): Add function_return.
1384 * config/i386/i386.h (machine_function): Add
1385 function_return_type.
1386 * config/i386/i386.md (simple_return_internal): Use
1387 ix86_output_function_return.
1388 (simple_return_internal_long): Likewise.
1389 * config/i386/i386.opt (mfunction-return=): New option.
1390 (indirect_branch): Mention -mfunction-return=.
1391 * doc/extend.texi: Document function_return function attribute.
1392 * doc/invoke.texi: Document -mfunction-return= option.
1394 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1396 * config/i386/i386-opts.h (indirect_branch): New.
1397 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1398 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1399 with local indirect jump when converting indirect call and jump.
1400 (ix86_set_indirect_branch_type): New.
1401 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1402 (indirectlabelno): New.
1403 (indirect_thunk_needed): Likewise.
1404 (indirect_thunk_bnd_needed): Likewise.
1405 (indirect_thunks_used): Likewise.
1406 (indirect_thunks_bnd_used): Likewise.
1407 (INDIRECT_LABEL): Likewise.
1408 (indirect_thunk_name): Likewise.
1409 (output_indirect_thunk): Likewise.
1410 (output_indirect_thunk_function): Likewise.
1411 (ix86_output_indirect_branch_via_reg): Likewise.
1412 (ix86_output_indirect_branch_via_push): Likewise.
1413 (ix86_output_indirect_branch): Likewise.
1414 (ix86_output_indirect_jmp): Likewise.
1415 (ix86_code_end): Call output_indirect_thunk_function if needed.
1416 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1418 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1419 (ix86_attribute_table): Add indirect_branch.
1420 * config/i386/i386.h (machine_function): Add indirect_branch_type
1421 and has_local_indirect_jump.
1422 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1424 (tablejump): Likewise.
1425 (*indirect_jump): Use ix86_output_indirect_jmp.
1426 (*tablejump_1): Likewise.
1427 (simple_return_indirect_internal): Likewise.
1428 * config/i386/i386.opt (mindirect-branch=): New option.
1429 (indirect_branch): New.
1432 (thunk-inline): Likewise.
1433 (thunk-extern): Likewise.
1434 * doc/extend.texi: Document indirect_branch function attribute.
1435 * doc/invoke.texi: Document -mindirect-branch= option.
1437 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1440 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1442 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1444 * ipa-inline.c (want_inline_small_function_p): Return false if
1445 inlining has already failed with CIF_FINAL_ERROR.
1446 (update_caller_keys): Call want_inline_small_function_p before
1448 (update_callee_keys): Likewise.
1450 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1452 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1454 (rs6000_quadword_masked_address_p): Likewise.
1455 (quad_aligned_load_p): Likewise.
1456 (quad_aligned_store_p): Likewise.
1457 (const_load_sequence_p): Add comment to describe the outer-most loop.
1458 (mimic_memory_attributes_and_flags): New function.
1459 (rs6000_gen_stvx): Likewise.
1460 (replace_swapped_aligned_store): Likewise.
1461 (rs6000_gen_lvx): Likewise.
1462 (replace_swapped_aligned_load): Likewise.
1463 (replace_swapped_load_constant): Capitalize argument name in
1464 comment describing this function.
1465 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1466 and stores that access quad-word aligned addresses and replace
1467 with stvx or lvx instructions when appropriate.
1468 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1469 New function prototype.
1470 (rs6000_quadword_masked_address_p): Likewise.
1471 (rs6000_gen_lvx): Likewise.
1472 (rs6000_gen_stvx): Likewise.
1473 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1474 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1475 when memory address is aligned.
1476 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1477 this split to select lvx instruction when memory address is aligned.
1478 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1479 instruction when memory address is aligned.
1480 (*vsx_le_perm_load_v16qi): Likewise.
1481 (four unnamed splitters): Modify to select the stvx instruction
1482 when memory is aligned.
1484 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1486 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1487 which appears in the queue multiple times.
1489 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1490 Alan Hayward <alan.hayward@arm.com>
1491 David Sherwood <david.sherwood@arm.com>
1493 * tree-vectorizer.h (vec_lower_bound): New structure.
1494 (_loop_vec_info): Add check_nonzero and lower_bounds.
1495 (LOOP_VINFO_CHECK_NONZERO): New macro.
1496 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1497 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1498 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1499 fields. Make seg_len the distance travelled, not including the
1501 (dr_direction_indicator): Declare.
1502 (dr_zero_step_indicator): Likewise.
1503 (dr_known_forward_stride_p): Likewise.
1504 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1506 (runtime_alias_check_p): Allow runtime alias checks with
1508 (operator ==): Compare access_size and align.
1509 (prune_runtime_alias_test_list): Rework for new distinction between
1510 the access_size and seg_len.
1511 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1513 (get_segment_min_max): New function.
1514 (create_intersect_range_checks): Use it.
1515 (dr_step_indicator): New function.
1516 (dr_direction_indicator): Likewise.
1517 (dr_zero_step_indicator): Likewise.
1518 (dr_known_forward_stride_p): Likewise.
1519 * tree-loop-distribution.c (data_ref_segment_size): Return
1520 DR_STEP * (niters - 1).
1521 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1523 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1524 (vect_preserves_scalar_order_p): New function, split out from...
1525 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1526 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1527 (vect_vfa_access_size): New function.
1528 (vect_vfa_align): Likewise.
1529 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1530 (dump_lower_bound): New function.
1531 (vect_check_lower_bound): Likewise.
1532 (vect_small_gap_p): Likewise.
1533 (vectorizable_with_step_bound_p): Likewise.
1534 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1535 depencies if the vectorization factor is 1. Convert the checks
1536 for nonzero steps into checks on the bounds of DR_STEP. Try using
1537 a bunds check for variable steps if the minimum required step is
1538 relatively small. Update calls to the dr_with_seg_len
1539 constructor and to vect_compile_time_alias.
1540 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1542 (vect_loop_versioning): Call it.
1543 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1545 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1547 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1548 Alan Hayward <alan.hayward@arm.com>
1549 David Sherwood <david.sherwood@arm.com>
1551 * doc/sourcebuild.texi (vect_scatter_store): Document.
1552 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1554 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1556 * genopinit.c (main): Add supports_vec_scatter_store and
1557 supports_vec_scatter_store_cached to target_optabs.
1558 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1559 IFN_MASK_SCATTER_STORE.
1560 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1562 * internal-fn.h (internal_store_fn_p): Declare.
1563 (internal_fn_stored_value_index): Likewise.
1564 * internal-fn.c (scatter_store_direct): New macro.
1565 (expand_scatter_store_optab_fn): New function.
1566 (direct_scatter_store_optab_supported_p): New macro.
1567 (internal_store_fn_p): New function.
1568 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1569 IFN_MASK_SCATTER_STORE.
1570 (internal_fn_mask_index): Likewise.
1571 (internal_fn_stored_value_index): New function.
1572 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1574 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1575 * optabs-query.c (supports_vec_scatter_store_p): New function.
1576 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1577 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1578 true for scatter stores.
1579 (vect_gather_scatter_fn_p): Handle scatter stores too.
1580 (vect_check_gather_scatter): Consider using scatter stores if
1581 supports_vec_scatter_store_p.
1582 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1584 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1585 internal_fn_stored_value_index.
1586 (check_load_store_masking): Handle scatter stores too.
1587 (vect_get_store_rhs): Make public.
1588 (vectorizable_call): Use internal_store_fn_p.
1589 (vectorizable_store): Handle scatter store internal functions.
1590 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1591 when deciding whether the end of the group has been reached.
1592 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1593 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1594 (mask_scatter_store<mode>): New insns.
1596 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1597 Alan Hayward <alan.hayward@arm.com>
1598 David Sherwood <david.sherwood@arm.com>
1600 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1601 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1602 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1604 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1605 Use vect_truncate_gather_scatter_offset if we can't treat the
1606 operation as a normal gather load or scatter store.
1607 (get_group_load_store_type): Take the gather_scatter_info
1608 as argument. Try using a gather load or scatter store for
1609 single-element groups.
1610 (get_load_store_type): Update calls to get_group_load_store_type
1611 and vect_use_strided_gather_scatters_p.
1613 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1614 Alan Hayward <alan.hayward@arm.com>
1615 David Sherwood <david.sherwood@arm.com>
1617 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1618 optional tree argument.
1619 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1621 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1622 but continue to use the current value as a fallback.
1623 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1624 to compare the updates.
1625 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1626 (get_load_store_type): Use it when handling a strided access.
1627 (vect_get_strided_load_store_ops): New function.
1628 (vect_get_data_ptr_increment): Likewise.
1629 (vectorizable_load): Handle strided gather loads. Always pass
1630 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1632 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1633 Alan Hayward <alan.hayward@arm.com>
1634 David Sherwood <david.sherwood@arm.com>
1636 * doc/md.texi (gather_load@var{m}): Document.
1637 (mask_gather_load@var{m}): Likewise.
1638 * genopinit.c (main): Add supports_vec_gather_load and
1639 supports_vec_gather_load_cached to target_optabs.
1640 * optabs-tree.c (init_tree_optimization_optabs): Use
1641 ggc_cleared_alloc to allocate target_optabs.
1642 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1643 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1645 * internal-fn.h (internal_load_fn_p): Declare.
1646 (internal_gather_scatter_fn_p): Likewise.
1647 (internal_fn_mask_index): Likewise.
1648 (internal_gather_scatter_fn_supported_p): Likewise.
1649 * internal-fn.c (gather_load_direct): New macro.
1650 (expand_gather_load_optab_fn): New function.
1651 (direct_gather_load_optab_supported_p): New macro.
1652 (direct_internal_fn_optab): New function.
1653 (internal_load_fn_p): Likewise.
1654 (internal_gather_scatter_fn_p): Likewise.
1655 (internal_fn_mask_index): Likewise.
1656 (internal_gather_scatter_fn_supported_p): Likewise.
1657 * optabs-query.c (supports_at_least_one_mode_p): New function.
1658 (supports_vec_gather_load_p): Likewise.
1659 * optabs-query.h (supports_vec_gather_load_p): Declare.
1660 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1661 and memory_type field.
1662 (NUM_PATTERNS): Bump to 15.
1663 * tree-vect-data-refs.c: Include internal-fn.h.
1664 (vect_gather_scatter_fn_p): New function.
1665 (vect_describe_gather_scatter_call): Likewise.
1666 (vect_check_gather_scatter): Try using internal functions for
1667 gather loads. Recognize existing calls to a gather load function.
1668 (vect_analyze_data_refs): Consider using gather loads if
1669 supports_vec_gather_load_p.
1670 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1671 (vect_get_gather_scatter_offset_type): Likewise.
1672 (vect_convert_mask_for_vectype): Likewise.
1673 (vect_add_conversion_to_patterm): Likewise.
1674 (vect_try_gather_scatter_pattern): Likewise.
1675 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1676 (vect_vect_recog_func_ptrs): Add it.
1677 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1678 internal_fn_mask_index and internal_gather_scatter_fn_p.
1679 (check_load_store_masking): Take the gather_scatter_info as an
1680 argument and handle gather loads.
1681 (vect_get_gather_scatter_ops): New function.
1682 (vectorizable_call): Check internal_load_fn_p.
1683 (vectorizable_load): Likewise. Handle gather load internal
1685 (vectorizable_store): Update call to check_load_store_masking.
1686 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1687 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1688 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1689 (aarch64_gather_scale_operand_d): New predicates.
1690 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1691 (mask_gather_load<mode>): New insns.
1693 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1694 Alan Hayward <alan.hayward@arm.com>
1695 David Sherwood <david.sherwood@arm.com>
1697 * optabs.def (fold_left_plus_optab): New optab.
1698 * doc/md.texi (fold_left_plus_@var{m}): Document.
1699 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1700 * internal-fn.c (fold_left_direct): Define.
1701 (expand_fold_left_optab_fn): Likewise.
1702 (direct_fold_left_optab_supported_p): Likewise.
1703 * fold-const-call.c (fold_const_fold_left): New function.
1704 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1705 * tree-parloops.c (valid_reduction_p): New function.
1706 (gather_scalar_reductions): Use it.
1707 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1708 (vect_finish_replace_stmt): Declare.
1709 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1710 (needs_fold_left_reduction_p): New function, split out from...
1711 (vect_is_simple_reduction): ...here. Accept reductions that
1712 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1713 (vect_force_simple_reduction): Also store the reduction type in
1714 the assignment's STMT_VINFO_REDUC_TYPE.
1715 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1716 (merge_with_identity): New function.
1717 (vect_expand_fold_left): Likewise.
1718 (vectorize_fold_left_reduction): Likewise.
1719 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1720 scalar phi in place for it. Check for target support and reject
1721 cases that would reassociate the operation. Defer the transform
1722 phase to vectorize_fold_left_reduction.
1723 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1724 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1725 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1727 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1729 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1730 call to ifc_temp_var.
1732 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1733 Alan Hayward <alan.hayward@arm.com>
1734 David Sherwood <david.sherwood@arm.com>
1736 * target.def (legitimize_address_displacement): Take the original
1737 offset as a poly_int.
1738 * targhooks.h (default_legitimize_address_displacement): Update
1740 * targhooks.c (default_legitimize_address_displacement): Likewise.
1741 * doc/tm.texi: Regenerate.
1742 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1743 as an argument, moving assert of ad->disp == ad->disp_term to...
1744 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1745 Try calling targetm.legitimize_address_displacement before expanding
1746 the address rather than afterwards, and adjust for the new interface.
1747 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1748 Match the new hook interface. Handle SVE addresses.
1749 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1752 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1754 * Makefile.in (OBJS): Add early-remat.o.
1755 * target.def (select_early_remat_modes): New hook.
1756 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1757 * doc/tm.texi: Regenerate.
1758 * targhooks.h (default_select_early_remat_modes): Declare.
1759 * targhooks.c (default_select_early_remat_modes): New function.
1760 * timevar.def (TV_EARLY_REMAT): New timevar.
1761 * passes.def (pass_early_remat): New pass.
1762 * tree-pass.h (make_pass_early_remat): Declare.
1763 * early-remat.c: New file.
1764 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1766 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1768 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1769 Alan Hayward <alan.hayward@arm.com>
1770 David Sherwood <david.sherwood@arm.com>
1772 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1773 vfm1 with a bound_epilog parameter.
1774 (vect_do_peeling): Update calls accordingly, and move the prologue
1775 call earlier in the function. Treat the base bound_epilog as 0 for
1776 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1777 this base when peeling for gaps.
1778 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1779 with fully-masked loops.
1780 (vect_estimate_min_profitable_iters): Handle the single peeled
1781 iteration in that case.
1783 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1784 Alan Hayward <alan.hayward@arm.com>
1785 David Sherwood <david.sherwood@arm.com>
1787 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1788 single-element interleaving even if the size is not a power of 2.
1789 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1790 accesses for single-element interleaving if the group size is
1793 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1794 Alan Hayward <alan.hayward@arm.com>
1795 David Sherwood <david.sherwood@arm.com>
1797 * doc/md.texi (fold_extract_last_@var{m}): Document.
1798 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1799 * optabs.def (fold_extract_last_optab): New optab.
1800 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1801 * internal-fn.c (fold_extract_direct): New macro.
1802 (expand_fold_extract_optab_fn): Likewise.
1803 (direct_fold_extract_optab_supported_p): Likewise.
1804 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1805 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1806 EXTRACT_LAST_REDUCTION.
1807 (get_initial_def_for_reduction): Do not create an initial vector
1808 for EXTRACT_LAST_REDUCTION reductions.
1809 (vectorizable_reduction): Leave the scalar phi in place for
1810 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1811 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1812 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1813 transform phase to vectorizable_condition.
1814 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1816 (vect_finish_stmt_generation): ...here.
1817 (vect_finish_replace_stmt): New function.
1818 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1819 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1821 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1823 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1824 Alan Hayward <alan.hayward@arm.com>
1825 David Sherwood <david.sherwood@arm.com>
1827 * doc/md.texi (extract_last_@var{m}): Document.
1828 * optabs.def (extract_last_optab): New optab.
1829 * internal-fn.def (EXTRACT_LAST): New internal function.
1830 * internal-fn.c (cond_unary_direct): New macro.
1831 (expand_cond_unary_optab_fn): Likewise.
1832 (direct_cond_unary_optab_supported_p): Likewise.
1833 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1834 loops using EXTRACT_LAST.
1835 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1836 (extract_last_<mode>): ...this optab.
1837 (vec_extract<mode><Vel>): Update accordingly.
1839 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1840 Alan Hayward <alan.hayward@arm.com>
1841 David Sherwood <david.sherwood@arm.com>
1843 * target.def (empty_mask_is_expensive): New hook.
1844 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
1845 * doc/tm.texi: Regenerate.
1846 * targhooks.h (default_empty_mask_is_expensive): Declare.
1847 * targhooks.c (default_empty_mask_is_expensive): New function.
1848 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
1849 if the target says that empty masks are expensive.
1850 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
1852 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
1854 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1855 Alan Hayward <alan.hayward@arm.com>
1856 David Sherwood <david.sherwood@arm.com>
1858 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
1859 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
1860 (vect_use_loop_mask_for_alignment_p): New function.
1861 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
1862 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
1863 niters_skip argument. Make sure that the first niters_skip elements
1864 of the first iteration are inactive.
1865 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
1866 Update call to vect_set_loop_masks_directly.
1867 (get_misalign_in_elems): New function, split out from...
1868 (vect_gen_prolog_loop_niters): ...here.
1869 (vect_update_init_of_dr): Take a code argument that specifies whether
1870 the adjustment should be added or subtracted.
1871 (vect_update_init_of_drs): Likewise.
1872 (vect_prepare_for_masked_peels): New function.
1873 (vect_do_peeling): Skip prologue peeling if we're using a mask
1874 instead. Update call to vect_update_inits_of_drs.
1875 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1877 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
1878 alignment. Do not include the number of peeled iterations in
1879 the minimum threshold in that case.
1880 (vectorizable_induction): Adjust the start value down by
1881 LOOP_VINFO_MASK_SKIP_NITERS iterations.
1882 (vect_transform_loop): Call vect_prepare_for_masked_peels.
1883 Take the number of skipped iterations into account when calculating
1885 * tree-vect-stmts.c (vect_gen_while_not): New function.
1887 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1888 Alan Hayward <alan.hayward@arm.com>
1889 David Sherwood <david.sherwood@arm.com>
1891 * doc/sourcebuild.texi (vect_fully_masked): Document.
1892 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
1894 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
1896 (vect_analyze_loop_2): ...here. Don't check the vectorization
1897 factor against the number of loop iterations if the loop is
1900 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1901 Alan Hayward <alan.hayward@arm.com>
1902 David Sherwood <david.sherwood@arm.com>
1904 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
1905 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
1906 (dump_groups): Update accordingly.
1907 (iv_use::mem_type): New member variable.
1908 (address_p): New function.
1909 (record_use): Add a mem_type argument and initialize the new
1911 (record_group_use): Add a mem_type argument. Use address_p.
1912 Remove obsolete null checks of base_object. Update call to record_use.
1913 (find_interesting_uses_op): Update call to record_group_use.
1914 (find_interesting_uses_cond): Likewise.
1915 (find_interesting_uses_address): Likewise.
1916 (get_mem_type_for_internal_fn): New function.
1917 (find_address_like_use): Likewise.
1918 (find_interesting_uses_stmt): Try find_address_like_use before
1919 calling find_interesting_uses_op.
1920 (addr_offset_valid_p): Use the iv mem_type field as the type
1921 of the addressed memory.
1922 (add_autoinc_candidates): Likewise.
1923 (get_address_cost): Likewise.
1924 (split_small_address_groups_p): Use address_p.
1925 (split_address_groups): Likewise.
1926 (add_iv_candidate_for_use): Likewise.
1927 (autoinc_possible_for_pair): Likewise.
1928 (rewrite_groups): Likewise.
1929 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
1930 (determine_group_iv_cost): Update after split of USE_ADDRESS.
1931 (get_alias_ptr_type_for_ptr_address): New function.
1932 (rewrite_use_address): Rewrite address uses in calls that were
1933 identified by find_address_like_use.
1935 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1936 Alan Hayward <alan.hayward@arm.com>
1937 David Sherwood <david.sherwood@arm.com>
1939 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
1941 * gimple-expr.h (is_gimple_addressable: Likewise.
1942 * gimple-expr.c (is_gimple_address): Likewise.
1943 * internal-fn.c (expand_call_mem_ref): New function.
1944 (expand_mask_load_optab_fn): Use it.
1945 (expand_mask_store_optab_fn): Likewise.
1947 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1948 Alan Hayward <alan.hayward@arm.com>
1949 David Sherwood <david.sherwood@arm.com>
1951 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
1952 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
1953 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
1954 (cond_umax@var{mode}): Document.
1955 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
1956 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
1957 (cond_umin_optab, cond_umax_optab): New optabs.
1958 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
1959 (COND_IOR, COND_XOR): New internal functions.
1960 * internal-fn.h (get_conditional_internal_fn): Declare.
1961 * internal-fn.c (cond_binary_direct): New macro.
1962 (expand_cond_binary_optab_fn): Likewise.
1963 (direct_cond_binary_optab_supported_p): Likewise.
1964 (get_conditional_internal_fn): New function.
1965 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
1966 Cope with reduction statements that are vectorized as calls rather
1968 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
1969 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
1970 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
1971 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
1972 (UNSPEC_COND_EOR): New unspecs.
1973 (optab): Add mappings for them.
1974 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
1975 (sve_int_op, sve_fp_op): New int attributes.
1977 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1978 Alan Hayward <alan.hayward@arm.com>
1979 David Sherwood <david.sherwood@arm.com>
1981 * optabs.def (while_ult_optab): New optab.
1982 * doc/md.texi (while_ult@var{m}@var{n}): Document.
1983 * internal-fn.def (WHILE_ULT): New internal function.
1984 * internal-fn.h (direct_internal_fn_supported_p): New override
1985 that takes two types as argument.
1986 * internal-fn.c (while_direct): New macro.
1987 (expand_while_optab_fn): New function.
1988 (convert_optab_supported_p): Likewise.
1989 (direct_while_optab_supported_p): New macro.
1990 * wide-int.h (wi::udiv_ceil): New function.
1991 * tree-vectorizer.h (rgroup_masks): New structure.
1992 (vec_loop_masks): New typedef.
1993 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
1995 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
1996 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
1997 (vect_max_vf): New function.
1998 (slpeel_make_loop_iterate_ntimes): Delete.
1999 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2000 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2001 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2002 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2003 internal-fn.h, stor-layout.h and optabs-query.h.
2004 (vect_set_loop_mask): New function.
2005 (add_preheader_seq): Likewise.
2006 (add_header_seq): Likewise.
2007 (interleave_supported_p): Likewise.
2008 (vect_maybe_permute_loop_masks): Likewise.
2009 (vect_set_loop_masks_directly): Likewise.
2010 (vect_set_loop_condition_masked): Likewise.
2011 (vect_set_loop_condition_unmasked): New function, split out from
2012 slpeel_make_loop_iterate_ntimes.
2013 (slpeel_make_loop_iterate_ntimes): Rename to..
2014 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2015 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2016 (vect_do_peeling): Update call accordingly.
2017 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2019 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2020 mask_compare_type, can_fully_mask_p and fully_masked_p.
2021 (release_vec_loop_masks): New function.
2022 (_loop_vec_info): Use it to free the loop masks.
2023 (can_produce_all_loop_masks_p): New function.
2024 (vect_get_max_nscalars_per_iter): Likewise.
2025 (vect_verify_full_masking): Likewise.
2026 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2027 retries, and free the mask rgroups before retrying. Check loop-wide
2028 reasons for disallowing fully-masked loops. Make the final decision
2029 about whether use a fully-masked loop or not.
2030 (vect_estimate_min_profitable_iters): Do not assume that peeling
2031 for the number of iterations will be needed for fully-masked loops.
2032 (vectorizable_reduction): Disable fully-masked loops.
2033 (vectorizable_live_operation): Likewise.
2034 (vect_halve_mask_nunits): New function.
2035 (vect_double_mask_nunits): Likewise.
2036 (vect_record_loop_mask): Likewise.
2037 (vect_get_loop_mask): Likewise.
2038 (vect_transform_loop): Handle the case in which the final loop
2039 iteration might handle a partial vector. Call vect_set_loop_condition
2040 instead of slpeel_make_loop_iterate_ntimes.
2041 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2042 (check_load_store_masking): New function.
2043 (prepare_load_store_mask): Likewise.
2044 (vectorizable_store): Handle fully-masked loops.
2045 (vectorizable_load): Likewise.
2046 (supportable_widening_operation): Use vect_halve_mask_nunits for
2048 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2049 (vect_gen_while): New function.
2050 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2051 (aarch64_uqdec<mode>): New insn.
2053 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2054 Alan Hayward <alan.hayward@arm.com>
2055 David Sherwood <david.sherwood@arm.com>
2057 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2058 (reduc_xor_scal_optab): New optabs.
2059 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2060 (reduc_xor_scal_@var{m}): Document.
2061 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2062 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2064 * fold-const-call.c (fold_const_call): Handle them.
2065 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2066 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2067 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2068 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2069 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2070 (UNSPEC_XORV): New unspecs.
2071 (optab): Add entries for them.
2072 (BITWISEV): New int iterator.
2073 (bit_reduc_op): New int attributes.
2075 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2076 Alan Hayward <alan.hayward@arm.com>
2077 David Sherwood <david.sherwood@arm.com>
2079 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2080 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2081 * optabs.def (vec_shl_insert_optab): New optab.
2082 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2083 (duplicate_and_interleave): Likewise.
2084 * tree-vect-loop.c: Include internal-fn.h.
2085 (neutral_op_for_slp_reduction): New function, split out from
2086 get_initial_defs_for_reduction.
2087 (get_initial_def_for_reduction): Handle option 2 for variable-length
2088 vectors by loading the neutral value into a vector and then shifting
2089 the initial value into element 0.
2090 (get_initial_defs_for_reduction): Replace the code argument with
2091 the neutral value calculated by neutral_op_for_slp_reduction.
2092 Use gimple_build_vector for constant-length vectors.
2093 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2094 but the first group_size elements have a neutral value.
2095 Use duplicate_and_interleave otherwise.
2096 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2097 Update call to get_initial_defs_for_reduction. Handle SLP
2098 reductions for variable-length vectors by creating one vector
2099 result for each scalar result, with the elements associated
2100 with other scalar results stubbed out with the neutral value.
2101 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2102 Require IFN_VEC_SHL_INSERT for double reductions on
2103 variable-length vectors, or SLP reductions that have
2104 a neutral value. Require can_duplicate_and_interleave_p
2105 support for variable-length unchained SLP reductions if there
2106 is no neutral value, such as for MIN/MAX reductions. Also require
2107 the number of vector elements to be a multiple of the number of
2108 SLP statements when doing variable-length unchained SLP reductions.
2109 Update call to vect_create_epilog_for_reduction.
2110 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2111 and remove initial values.
2112 (duplicate_and_interleave): Make public.
2113 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2114 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
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-slp.c: Include gimple-fold.h and internal-fn.h
2121 (can_duplicate_and_interleave_p): New function.
2122 (vect_get_and_check_slp_defs): Take the vector of statements
2123 rather than just the current one. Remove excess parentheses.
2124 Restriction rejectinon of vect_constant_def and vect_external_def
2125 for variable-length vectors to boolean types, or types for which
2126 can_duplicate_and_interleave_p is false.
2127 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2128 (duplicate_and_interleave): New function.
2129 (vect_get_constant_vectors): Use gimple_build_vector for
2130 constant-length vectors and suitable variable-length constant
2131 vectors. Use duplicate_and_interleave for other variable-length
2132 vectors. Don't defer the update when inserting new statements.
2134 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2135 Alan Hayward <alan.hayward@arm.com>
2136 David Sherwood <david.sherwood@arm.com>
2138 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2139 min_profitable_iters doesn't go negative.
2141 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2142 Alan Hayward <alan.hayward@arm.com>
2143 David Sherwood <david.sherwood@arm.com>
2145 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2146 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2147 * optabs.def (vec_mask_load_lanes_optab): New optab.
2148 (vec_mask_store_lanes_optab): Likewise.
2149 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2150 (MASK_STORE_LANES): Likewise.
2151 * internal-fn.c (mask_load_lanes_direct): New macro.
2152 (mask_store_lanes_direct): Likewise.
2153 (expand_mask_load_optab_fn): Handle masked operations.
2154 (expand_mask_load_lanes_optab_fn): New macro.
2155 (expand_mask_store_optab_fn): Handle masked operations.
2156 (expand_mask_store_lanes_optab_fn): New macro.
2157 (direct_mask_load_lanes_optab_supported_p): Likewise.
2158 (direct_mask_store_lanes_optab_supported_p): Likewise.
2159 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2161 (vect_load_lanes_supported): Likewise.
2162 * tree-vect-data-refs.c (strip_conversion): New function.
2163 (can_group_stmts_p): Likewise.
2164 (vect_analyze_data_ref_accesses): Use it instead of checking
2165 for a pair of assignments.
2166 (vect_store_lanes_supported): Take a masked_p parameter.
2167 (vect_load_lanes_supported): Likewise.
2168 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2169 vect_store_lanes_supported and vect_load_lanes_supported.
2170 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2171 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2172 parameter. Don't allow gaps for masked accesses.
2173 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2174 and vect_load_lanes_supported.
2175 (get_load_store_type): Take a masked_p parameter and update
2176 call to get_group_load_store_type.
2177 (vectorizable_store): Update call to get_load_store_type.
2178 Handle IFN_MASK_STORE_LANES.
2179 (vectorizable_load): Update call to get_load_store_type.
2180 Handle IFN_MASK_LOAD_LANES.
2182 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2183 Alan Hayward <alan.hayward@arm.com>
2184 David Sherwood <david.sherwood@arm.com>
2186 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2188 * config/aarch64/aarch64-protos.h
2189 (aarch64_sve_struct_memory_operand_p): Declare.
2190 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2191 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2192 (VPRED, vpred): Handle SVE structure modes.
2193 * config/aarch64/constraints.md (Utx): New constraint.
2194 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2195 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2196 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2197 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2198 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2199 structure modes. Split into pieces after RA.
2200 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2201 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2203 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2204 SVE structure modes.
2205 (aarch64_classify_address): Likewise.
2206 (sizetochar): Move earlier in file.
2207 (aarch64_print_operand): Handle SVE register lists.
2208 (aarch64_array_mode): New function.
2209 (aarch64_sve_struct_memory_operand_p): Likewise.
2210 (TARGET_ARRAY_MODE): Redefine.
2212 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2213 Alan Hayward <alan.hayward@arm.com>
2214 David Sherwood <david.sherwood@arm.com>
2216 * target.def (array_mode): New target hook.
2217 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2218 * doc/tm.texi: Regenerate.
2219 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2220 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2221 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2223 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2226 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2227 Alan Hayward <alan.hayward@arm.com>
2228 David Sherwood <david.sherwood@arm.com>
2230 * fold-const.c (fold_binary_loc): Check the argument types
2231 rather than the result type when testing for a vector operation.
2233 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2235 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2236 * doc/tm.texi: Regenerate.
2238 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2239 Alan Hayward <alan.hayward@arm.com>
2240 David Sherwood <david.sherwood@arm.com>
2242 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2243 (sve): Document new AArch64 extension.
2244 * doc/md.texi (w): Extend the description of the AArch64
2245 constraint to include SVE vectors.
2246 (Upl, Upa): Document new AArch64 predicate constraints.
2247 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2249 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2250 (msve-vector-bits=): New option.
2251 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2252 SVE when these are disabled.
2253 (sve): New extension.
2254 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2255 modes. Adjust their number of units based on aarch64_sve_vg.
2256 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2257 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2258 aarch64_addr_query_type.
2259 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2260 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2261 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2262 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2263 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2264 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2265 (aarch64_simd_imm_zero_p): Delete.
2266 (aarch64_check_zero_based_sve_index_immediate): Declare.
2267 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2268 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2269 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2270 (aarch64_sve_float_mul_immediate_p): Likewise.
2271 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2273 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2274 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2275 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2276 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2277 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2278 (aarch64_regmode_natural_size): Likewise.
2279 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2280 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2282 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2283 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2284 for VG and the SVE predicate registers.
2285 (V_ALIASES): Add a "z"-prefixed alias.
2286 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2287 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2288 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2289 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2290 (REG_CLASS_NAMES): Add entries for them.
2291 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2292 and the predicate registers.
2293 (aarch64_sve_vg): Declare.
2294 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2295 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2296 (REGMODE_NATURAL_SIZE): Define.
2297 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2299 * config/aarch64/aarch64.c: Include cfgrtl.h.
2300 (simd_immediate_info): Add a constructor for series vectors,
2301 and an associated step field.
2302 (aarch64_sve_vg): New variable.
2303 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2304 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2305 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2306 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2307 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2308 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2309 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2310 (aarch64_get_mask_mode): New functions.
2311 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2312 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2313 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2314 predicate modes and predicate registers. Explicitly restrict
2315 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2316 to store a vector mode if it is recognized by
2317 aarch64_classify_vector_mode.
2318 (aarch64_regmode_natural_size): New function.
2319 (aarch64_hard_regno_caller_save_mode): Return the original mode
2321 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2322 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2323 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2324 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2326 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2327 does not overlap dest if the function is frame-related. Handle
2329 (aarch64_split_add_offset): New function.
2330 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2331 them aarch64_add_offset.
2332 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2333 and update call to aarch64_sub_sp.
2334 (aarch64_add_cfa_expression): New function.
2335 (aarch64_expand_prologue): Pass extra temporary registers to the
2336 functions above. Handle the case in which we need to emit new
2337 DW_CFA_expressions for registers that were originally saved
2338 relative to the stack pointer, but now have to be expressed
2339 relative to the frame pointer.
2340 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2342 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2343 IP0 and IP1 values for SVE frames.
2344 (aarch64_expand_vec_series): New function.
2345 (aarch64_expand_sve_widened_duplicate): Likewise.
2346 (aarch64_expand_sve_const_vector): Likewise.
2347 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2348 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2349 into the register, rather than emitting a SET directly.
2350 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2351 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2352 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2353 (offset_9bit_signed_scaled_p): New functions.
2354 (aarch64_replicate_bitmask_imm): New function.
2355 (aarch64_bitmask_imm): Use it.
2356 (aarch64_cannot_force_const_mem): Reject expressions involving
2357 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2358 (aarch64_classify_index): Handle SVE indices, by requiring
2359 a plain register index with a scale that matches the element size.
2360 (aarch64_classify_address): Handle SVE addresses. Assert that
2361 the mode of the address is VOIDmode or an integer mode.
2362 Update call to aarch64_classify_symbol.
2363 (aarch64_classify_symbolic_expression): Update call to
2364 aarch64_classify_symbol.
2365 (aarch64_const_vec_all_in_range_p): New function.
2366 (aarch64_print_vector_float_operand): Likewise.
2367 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2368 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2369 and the FP immediates 1.0 and 0.5.
2370 (aarch64_print_address_internal): Handle SVE addresses.
2371 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2372 (aarch64_regno_regclass): Handle predicate registers.
2373 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2375 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2376 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2377 (aarch64_convert_sve_vector_bits): New function.
2378 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2379 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2381 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2382 Handle SVE vector and predicate modes. Accept VL-based constants
2383 that need only one temporary register, and VL offsets that require
2384 no temporary registers.
2385 (aarch64_conditional_register_usage): Mark the predicate registers
2386 as fixed if SVE isn't available.
2387 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2388 Return true for SVE vector and predicate modes.
2389 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2390 rather than an unsigned int. Handle SVE modes.
2391 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2393 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2395 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2396 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2397 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2398 (aarch64_sve_float_mul_immediate_p): New functions.
2399 (aarch64_sve_valid_immediate): New function.
2400 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2401 Explicitly reject structure modes. Check for INDEX constants.
2402 Handle PTRUE and PFALSE constants.
2403 (aarch64_check_zero_based_sve_index_immediate): New function.
2404 (aarch64_simd_imm_zero_p): Delete.
2405 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2406 vector modes. Accept constants in the range of CNT[BHWD].
2407 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2408 ask for an Advanced SIMD mode.
2409 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2410 (aarch64_simd_vector_alignment): Handle SVE predicates.
2411 (aarch64_vectorize_preferred_vector_alignment): New function.
2412 (aarch64_simd_vector_alignment_reachable): Use it instead of
2414 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2415 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2417 (MAX_VECT_LEN): Delete.
2418 (expand_vec_perm_d): Add a vec_flags field.
2419 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2420 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2421 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2423 (aarch64_evpc_rev): Rename to...
2424 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2425 (aarch64_evpc_rev_global): New function.
2426 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2427 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2429 (aarch64_evpc_sve_tbl): New function.
2430 (aarch64_expand_vec_perm_const_1): Update after rename of
2431 aarch64_evpc_rev. Handle SVE permutes too, trying
2432 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2433 than aarch64_evpc_tbl.
2434 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2435 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2436 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2437 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2438 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2439 (aarch64_expand_sve_vcond): New functions.
2440 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2441 of aarch64_vector_mode_p.
2442 (aarch64_dwarf_poly_indeterminate_value): New function.
2443 (aarch64_compute_pressure_classes): Likewise.
2444 (aarch64_can_change_mode_class): Likewise.
2445 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2446 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2447 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2448 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2449 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2450 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2451 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2452 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2454 (Dn, Dl, Dr): Accept const as well as const_vector.
2455 (Dz): Likewise. Compare against CONST0_RTX.
2456 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2457 of "vector" where appropriate.
2458 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2459 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2460 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2461 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2462 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2463 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2464 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2465 (v_int_equiv): Extend to SVE modes.
2466 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2468 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2469 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2470 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2471 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2472 (SVE_COND_FP_CMP): New int iterators.
2473 (perm_hilo): Handle the new unpack unspecs.
2474 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2476 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2477 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2478 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2479 (aarch64_equality_operator, aarch64_constant_vector_operand)
2480 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2481 (aarch64_sve_nonimmediate_operand): Likewise.
2482 (aarch64_sve_general_operand): Likewise.
2483 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2484 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2485 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2486 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2487 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2488 (aarch64_sve_float_arith_immediate): Likewise.
2489 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2490 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2491 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2492 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2493 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2494 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2495 (aarch64_sve_float_arith_operand): Likewise.
2496 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2497 (aarch64_sve_float_mul_operand): Likewise.
2498 (aarch64_sve_vec_perm_operand): Likewise.
2499 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2500 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2501 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2502 as well as const_vector.
2503 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2504 in file. Use CONST0_RTX and CONSTM1_RTX.
2505 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2506 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2507 Use aarch64_simd_imm_zero.
2508 * config/aarch64/aarch64-sve.md: New file.
2509 * config/aarch64/aarch64.md: Include it.
2510 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2511 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2512 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2513 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2514 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2515 (sve): New attribute.
2516 (enabled): Disable instructions with the sve attribute unless
2518 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2519 aarch64_expand_mov_immediate.
2520 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2521 CNT[BHSD] immediates.
2522 (movti): Split CONST_POLY_INT moves into two halves.
2523 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2524 Split additions that need a temporary here if the destination
2525 is the stack pointer.
2526 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2527 (*add<mode>3_poly_1): New instruction.
2528 (set_clobber_cc): New expander.
2530 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2532 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2533 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2534 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2535 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2536 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2537 Change innermode from fixed_mode_size to machine_mode.
2538 (simplify_subreg): Update call accordingly. Handle a constant-sized
2539 subreg of a variable-length CONST_VECTOR.
2541 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2542 Alan Hayward <alan.hayward@arm.com>
2543 David Sherwood <david.sherwood@arm.com>
2545 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2546 (add_offset_to_base): New function, split out from...
2547 (create_mem_ref): ...here. When handling a scale other than 1,
2548 check first whether the address is valid without the offset.
2549 Add it into the base if so, leaving the index and scale as-is.
2551 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2554 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2555 fold_for_warn before checking if arg2 is INTEGER_CST.
2557 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2559 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2560 (store_multiple_operation): Delete.
2561 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2562 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2563 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2564 guarded by TARGET_STRING.
2565 (rs6000_output_load_multiple): Delete.
2566 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2567 OPTION_MASK_STRING / TARGET_STRING handling.
2568 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2569 (const rs6000_opt_masks) <"string">: Change mask to 0.
2570 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2571 (MASK_STRING): Delete.
2572 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2574 (load_multiple): Delete.
2581 (store_multiple): Delete.
2588 (movmemsi_8reg): Delete.
2589 (corresponding unnamed define_insn): Delete.
2590 (movmemsi_6reg): Delete.
2591 (corresponding unnamed define_insn): Delete.
2592 (movmemsi_4reg): Delete.
2593 (corresponding unnamed define_insn): Delete.
2594 (movmemsi_2reg): Delete.
2595 (corresponding unnamed define_insn): Delete.
2596 (movmemsi_1reg): Delete.
2597 (corresponding unnamed define_insn): Delete.
2598 * config/rs6000/rs6000.opt (mno-string): New.
2599 (mstring): Replace by deprecation warning stub.
2600 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2602 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2604 * regrename.c (regrename_do_replace): If replacing the same
2605 reg multiple times, try to reuse last created gen_raw_REG.
2608 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2609 main to workaround a bug in GDB.
2611 2018-01-12 Tom de Vries <tom@codesourcery.com>
2614 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2616 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2618 PR rtl-optimization/80481
2619 * ira-color.c (get_cap_member): New function.
2620 (allocnos_conflict_by_live_ranges_p): Use it.
2621 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2622 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2624 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2627 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2628 (*saddl_se_1): Ditto.
2630 (*ssubl_se_1): Ditto.
2632 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2634 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2635 rather than wi::to_widest for DR_INITs.
2636 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2637 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2638 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2640 (vect_analyze_group_access_1): Note that here.
2642 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2644 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2645 polynomial type sizes.
2647 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2649 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2650 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2651 (gimple_add_tmp_var): Likewise.
2653 2018-01-12 Martin Liska <mliska@suse.cz>
2655 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2656 (gimple_alloc_sizes): Likewise.
2657 (dump_gimple_statistics): Use PRIu64 in printf format.
2658 * gimple.h: Change uint64_t to int.
2660 2018-01-12 Martin Liska <mliska@suse.cz>
2662 * tree-core.h: Use uint64_t instead of int.
2663 * tree.c (tree_node_counts): Likewise.
2664 (tree_node_sizes): Likewise.
2665 (dump_tree_statistics): Use PRIu64 in printf format.
2667 2018-01-12 Martin Liska <mliska@suse.cz>
2669 * Makefile.in: As qsort_chk is implemented in vec.c, add
2670 vec.o to linkage of gencfn-macros.
2671 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2672 passing the info to record_node_allocation_statistics.
2673 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2675 * ggc-common.c (struct ggc_usage): Add operator== and use
2676 it in operator< and compare function.
2677 * mem-stats.h (struct mem_usage): Likewise.
2678 * vec.c (struct vec_usage): Remove operator< and compare
2679 function. Can be simply inherited.
2681 2018-01-12 Martin Jambor <mjambor@suse.cz>
2684 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2685 * tree-ssa-math-opts.c: Include domwalk.h.
2686 (convert_mult_to_fma_1): New function.
2687 (fma_transformation_info): New type.
2688 (fma_deferring_state): Likewise.
2689 (cancel_fma_deferring): New function.
2690 (result_of_phi): Likewise.
2691 (last_fma_candidate_feeds_initial_phi): Likewise.
2692 (convert_mult_to_fma): Added deferring logic, split actual
2693 transformation to convert_mult_to_fma_1.
2694 (math_opts_dom_walker): New type.
2695 (math_opts_dom_walker::after_dom_children): New method, body moved
2696 here from pass_optimize_widening_mul::execute, added deferring logic
2698 (pass_optimize_widening_mul::execute): Moved most of code to
2699 math_opts_dom_walker::after_dom_children.
2700 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2701 * config/i386/i386.c (ix86_option_override_internal): Added
2702 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2704 2018-01-12 Richard Biener <rguenther@suse.de>
2707 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2708 inline instance vars.
2710 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2713 * config/rx/rx.c (rx_is_restricted_memory_address):
2716 2018-01-12 Richard Biener <rguenther@suse.de>
2718 PR tree-optimization/80846
2719 * target.def (split_reduction): New target hook.
2720 * targhooks.c (default_split_reduction): New function.
2721 * targhooks.h (default_split_reduction): Declare.
2722 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2723 target requests first reduce vectors by combining low and high
2725 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2726 (get_vectype_for_scalar_type_and_size): Export.
2727 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2728 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2729 * doc/tm.texi: Regenerate.
2730 * config/i386/i386.c (ix86_split_reduction): Implement
2731 TARGET_VECTORIZE_SPLIT_REDUCTION.
2733 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2736 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2737 in PIC mode except for TARGET_VXWORKS_RTP.
2738 * config/sparc/sparc.c: Include cfgrtl.h.
2739 (TARGET_INIT_PIC_REG): Define.
2740 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2741 (sparc_pic_register_p): New predicate.
2742 (sparc_legitimate_address_p): Use it.
2743 (sparc_legitimize_pic_address): Likewise.
2744 (sparc_delegitimize_address): Likewise.
2745 (sparc_mode_dependent_address_p): Likewise.
2746 (gen_load_pcrel_sym): Remove 4th parameter.
2747 (load_got_register): Adjust call to above. Remove obsolete stuff.
2748 (sparc_expand_prologue): Do not call load_got_register here.
2749 (sparc_flat_expand_prologue): Likewise.
2750 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2751 (sparc_use_pseudo_pic_reg): New function.
2752 (sparc_init_pic_reg): Likewise.
2753 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2754 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2756 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2758 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2759 Add item for branch_cost.
2761 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2763 PR rtl-optimization/83565
2764 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2765 not extend the result to a larger mode for rotate operations.
2766 (num_sign_bit_copies1): Likewise.
2768 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2771 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2773 Use values-Xc.o for -pedantic.
2774 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2776 2018-01-12 Martin Liska <mliska@suse.cz>
2779 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2781 (possible_polymorphic_call_targets): Use it.
2782 (ipa_devirt): Likewise.
2784 2018-01-12 Martin Liska <mliska@suse.cz>
2786 * profile-count.h (enum profile_quality): Use 0 as invalid
2787 enum value of profile_quality.
2789 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2791 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2792 -mext-string options.
2794 2018-01-12 Richard Biener <rguenther@suse.de>
2796 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2797 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2798 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2800 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2802 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2804 * configure.ac (--with-long-double-format): Add support for the
2805 configuration option to change the default long double format on
2807 * config.gcc (powerpc*-linux*-*): Likewise.
2808 * configure: Regenerate.
2809 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2810 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2811 used without modification.
2813 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2815 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2816 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2817 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2818 MISC_BUILTIN_SPEC_BARRIER.
2819 (rs6000_init_builtins): Likewise.
2820 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2822 (speculation_barrier): New define_insn.
2823 * doc/extend.texi: Document __builtin_speculation_barrier.
2825 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2828 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2829 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2830 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2832 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2833 integral modes instead of "ss" and "sd".
2834 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2835 vectors with 32-bit and 64-bit elements.
2836 (vecdupssescalarmodesuffix): New mode attribute.
2837 (vec_dup<mode>): Use it.
2839 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2842 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2843 frame if argument is passed on stack.
2845 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2848 * ree.c (combine_reaching_defs): Optimize also
2849 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
2850 reg2=any_extend(exp); reg1=reg2;, formatting fix.
2852 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2855 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
2857 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2860 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
2861 after they are computed.
2863 2018-01-11 Bin Cheng <bin.cheng@arm.com>
2865 PR tree-optimization/83695
2866 * gimple-loop-linterchange.cc
2867 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
2868 reset cached scev information after interchange.
2869 (pass_linterchange::execute): Remove call to scev_reset_htab.
2871 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2873 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
2874 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
2875 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
2876 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
2877 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
2878 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
2879 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
2880 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
2881 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
2882 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
2883 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
2884 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
2885 (V_lane_reg): Likewise.
2886 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
2888 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
2889 (vfmal_lane_low<mode>_intrinsic,
2890 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
2891 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
2892 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
2893 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
2894 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
2895 vfmsl_lane_high<mode>_intrinsic): New define_insns.
2897 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2899 * config/arm/arm-cpus.in (fp16fml): New feature.
2900 (ALL_SIMD): Add fp16fml.
2901 (armv8.2-a): Add fp16fml as an option.
2902 (armv8.3-a): Likewise.
2903 (armv8.4-a): Add fp16fml as part of fp16.
2904 * config/arm/arm.h (TARGET_FP16FML): Define.
2905 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
2907 * config/arm/arm-modes.def (V2HF): Define.
2908 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
2909 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
2910 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
2911 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
2912 vfmsl_low, vfmsl_high): New set of builtins.
2913 * config/arm/iterators.md (PLUSMINUS): New code iterator.
2914 (vfml_op): New code attribute.
2915 (VFMLHALVES): New int iterator.
2916 (VFML, VFMLSEL): New mode attributes.
2917 (V_reg): Define mapping for V2HF.
2918 (V_hi, V_lo): New mode attributes.
2919 (VF_constraint): Likewise.
2920 (vfml_half, vfml_half_selector): New int attributes.
2921 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
2923 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
2924 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
2926 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
2927 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
2928 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
2929 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
2931 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
2932 Document new effective target and option set.
2934 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2936 * config/arm/arm-cpus.in (armv8_4): New feature.
2937 (ARMv8_4a): New fgroup.
2938 (armv8.4-a): New arch.
2939 * config/arm/arm-tables.opt: Regenerate.
2940 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
2941 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
2942 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
2943 Add matching rules for -march=armv8.4-a and extensions.
2944 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
2946 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
2949 * config/rx/rx.md (BW): New mode attribute.
2950 (sync_lock_test_and_setsi): Add mode suffix to insn output.
2952 2018-01-11 Richard Biener <rguenther@suse.de>
2954 PR tree-optimization/83435
2955 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
2956 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
2957 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
2959 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2960 Alan Hayward <alan.hayward@arm.com>
2961 David Sherwood <david.sherwood@arm.com>
2963 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
2965 (aarch64_classify_address): Initialize it. Track polynomial offsets.
2966 (aarch64_print_address_internal): Use it to check for a zero offset.
2968 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2969 Alan Hayward <alan.hayward@arm.com>
2970 David Sherwood <david.sherwood@arm.com>
2972 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
2973 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
2974 Return a poly_int64 rather than a HOST_WIDE_INT.
2975 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
2976 rather than a HOST_WIDE_INT.
2977 * config/aarch64/aarch64.h (aarch64_frame): Protect with
2978 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
2979 hard_fp_offset, frame_size, initial_adjust, callee_offset and
2980 final_offset from HOST_WIDE_INT to poly_int64.
2981 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
2982 to_constant when getting the number of units in an Advanced SIMD
2984 (aarch64_builtin_vectorized_function): Check for a constant number
2986 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
2988 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
2989 attribute instead of GET_MODE_NUNITS.
2990 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
2991 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
2992 GET_MODE_SIZE for fixed-size registers.
2993 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
2994 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
2995 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
2996 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
2997 (aarch64_print_operand, aarch64_print_address_internal)
2998 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
2999 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3000 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3001 Handle polynomial GET_MODE_SIZE.
3002 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3003 wider than SImode without modification.
3004 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3005 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3006 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3007 passing and returning SVE modes.
3008 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3009 rather than GEN_INT.
3010 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3011 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3012 (aarch64_allocate_and_probe_stack_space): Likewise.
3013 (aarch64_layout_frame): Cope with polynomial offsets.
3014 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3015 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3017 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3018 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3019 poly_int64 rather than a HOST_WIDE_INT.
3020 (aarch64_get_separate_components, aarch64_process_components)
3021 (aarch64_expand_prologue, aarch64_expand_epilogue)
3022 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3023 (aarch64_anchor_offset): New function, split out from...
3024 (aarch64_legitimize_address): ...here.
3025 (aarch64_builtin_vectorization_cost): Handle polynomial
3026 TYPE_VECTOR_SUBPARTS.
3027 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3029 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3030 number of elements from the PARALLEL rather than the mode.
3031 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3032 rather than GET_MODE_BITSIZE.
3033 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3034 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3035 (aarch64_expand_vec_perm_const_1): Handle polynomial
3036 d->perm.length () and d->perm elements.
3037 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3038 Apply to_constant to d->perm elements.
3039 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3040 polynomial CONST_VECTOR_NUNITS.
3041 (aarch64_move_pointer): Take amount as a poly_int64 rather
3043 (aarch64_progress_pointer): Avoid temporary variable.
3044 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3045 the mode attribute instead of GET_MODE.
3047 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3048 Alan Hayward <alan.hayward@arm.com>
3049 David Sherwood <david.sherwood@arm.com>
3051 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3052 x exists before using it.
3053 (aarch64_add_constant_internal): Rename to...
3054 (aarch64_add_offset_1): ...this. Replace regnum with separate
3055 src and dest rtxes. Handle the case in which they're different,
3056 including when the offset is zero. Replace scratchreg with an rtx.
3057 Use 2 additions if there is no spare register into which we can
3058 move a 16-bit constant.
3059 (aarch64_add_constant): Delete.
3060 (aarch64_add_offset): Replace reg with separate src and dest
3061 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3062 Use aarch64_add_offset_1.
3063 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3064 an rtx rather than an int. Take the delta as a poly_int64
3065 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3066 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3067 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3068 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3069 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3071 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3072 aarch64_add_constant.
3074 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3076 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3077 Use scalar_float_mode.
3079 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3081 * config/aarch64/aarch64-simd.md
3082 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3083 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3084 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3085 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3086 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3087 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3088 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3089 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3090 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3091 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3093 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3096 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3097 targ_options->x_arm_arch_string is non NULL.
3099 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3101 * config/aarch64/aarch64.h
3102 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3104 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3107 * expmed.c (emit_store_flag_force): Swap if const op0
3108 and change VOIDmode to mode of op0.
3110 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3112 PR rtl-optimization/83761
3113 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3114 than bytes to mode_for_size.
3116 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3119 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3120 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3123 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3126 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3127 when in layout mode.
3128 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3129 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3132 2018-01-10 Michael Collison <michael.collison@arm.com>
3134 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3135 * config/aarch64/aarch64-option-extension.def: Add
3136 AARCH64_OPT_EXTENSION of 'fp16fml'.
3137 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3138 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3139 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3140 * config/aarch64/constraints.md (Ui7): New constraint.
3141 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3142 (VFMLA_SEL_W): Ditto.
3145 (VFMLA16_LOW): New int iterator.
3146 (VFMLA16_HIGH): Ditto.
3147 (UNSPEC_FMLAL): New unspec.
3148 (UNSPEC_FMLSL): Ditto.
3149 (UNSPEC_FMLAL2): Ditto.
3150 (UNSPEC_FMLSL2): Ditto.
3151 (f16mac): New code attribute.
3152 * config/aarch64/aarch64-simd-builtins.def
3153 (aarch64_fmlal_lowv2sf): Ditto.
3154 (aarch64_fmlsl_lowv2sf): Ditto.
3155 (aarch64_fmlalq_lowv4sf): Ditto.
3156 (aarch64_fmlslq_lowv4sf): Ditto.
3157 (aarch64_fmlal_highv2sf): Ditto.
3158 (aarch64_fmlsl_highv2sf): Ditto.
3159 (aarch64_fmlalq_highv4sf): Ditto.
3160 (aarch64_fmlslq_highv4sf): Ditto.
3161 (aarch64_fmlal_lane_lowv2sf): Ditto.
3162 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3163 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3164 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3165 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3166 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3167 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3168 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3169 (aarch64_fmlal_lane_highv2sf): Ditto.
3170 (aarch64_fmlsl_lane_highv2sf): Ditto.
3171 (aarch64_fmlal_laneq_highv2sf): Ditto.
3172 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3173 (aarch64_fmlalq_lane_highv4sf): Ditto.
3174 (aarch64_fmlsl_lane_highv4sf): Ditto.
3175 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3176 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3177 * config/aarch64/aarch64-simd.md:
3178 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3179 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3180 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3181 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3182 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3183 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3184 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3185 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3186 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3187 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3188 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3189 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3190 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3191 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3192 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3193 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3194 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3195 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3196 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3197 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3198 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3199 (vfmlsl_low_u32): Ditto.
3200 (vfmlalq_low_u32): Ditto.
3201 (vfmlslq_low_u32): Ditto.
3202 (vfmlal_high_u32): Ditto.
3203 (vfmlsl_high_u32): Ditto.
3204 (vfmlalq_high_u32): Ditto.
3205 (vfmlslq_high_u32): Ditto.
3206 (vfmlal_lane_low_u32): Ditto.
3207 (vfmlsl_lane_low_u32): Ditto.
3208 (vfmlal_laneq_low_u32): Ditto.
3209 (vfmlsl_laneq_low_u32): Ditto.
3210 (vfmlalq_lane_low_u32): Ditto.
3211 (vfmlslq_lane_low_u32): Ditto.
3212 (vfmlalq_laneq_low_u32): Ditto.
3213 (vfmlslq_laneq_low_u32): Ditto.
3214 (vfmlal_lane_high_u32): Ditto.
3215 (vfmlsl_lane_high_u32): Ditto.
3216 (vfmlal_laneq_high_u32): Ditto.
3217 (vfmlsl_laneq_high_u32): Ditto.
3218 (vfmlalq_lane_high_u32): Ditto.
3219 (vfmlslq_lane_high_u32): Ditto.
3220 (vfmlalq_laneq_high_u32): Ditto.
3221 (vfmlslq_laneq_high_u32): Ditto.
3222 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3223 (AARCH64_FL_FOR_ARCH8_4): New.
3224 (AARCH64_ISA_F16FML): New ISA flag.
3225 (TARGET_F16FML): New feature flag for fp16fml.
3226 (doc/invoke.texi): Document new fp16fml option.
3228 2018-01-10 Michael Collison <michael.collison@arm.com>
3230 * config/aarch64/aarch64-builtins.c:
3231 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3232 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3233 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3234 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3235 (AARCH64_ISA_SHA3): New ISA flag.
3236 (TARGET_SHA3): New feature flag for sha3.
3237 * config/aarch64/iterators.md (sha512_op): New int attribute.
3238 (CRYPTO_SHA512): New int iterator.
3239 (UNSPEC_SHA512H): New unspec.
3240 (UNSPEC_SHA512H2): Ditto.
3241 (UNSPEC_SHA512SU0): Ditto.
3242 (UNSPEC_SHA512SU1): Ditto.
3243 * config/aarch64/aarch64-simd-builtins.def
3244 (aarch64_crypto_sha512hqv2di): New builtin.
3245 (aarch64_crypto_sha512h2qv2di): Ditto.
3246 (aarch64_crypto_sha512su0qv2di): Ditto.
3247 (aarch64_crypto_sha512su1qv2di): Ditto.
3248 (aarch64_eor3qv8hi): Ditto.
3249 (aarch64_rax1qv2di): Ditto.
3250 (aarch64_xarqv2di): Ditto.
3251 (aarch64_bcaxqv8hi): Ditto.
3252 * config/aarch64/aarch64-simd.md:
3253 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3254 (aarch64_crypto_sha512su0qv2di): Ditto.
3255 (aarch64_crypto_sha512su1qv2di): Ditto.
3256 (aarch64_eor3qv8hi): Ditto.
3257 (aarch64_rax1qv2di): Ditto.
3258 (aarch64_xarqv2di): Ditto.
3259 (aarch64_bcaxqv8hi): Ditto.
3260 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3261 (vsha512h2q_u64): Ditto.
3262 (vsha512su0q_u64): Ditto.
3263 (vsha512su1q_u64): Ditto.
3264 (veor3q_u16): Ditto.
3265 (vrax1q_u64): Ditto.
3267 (vbcaxq_u16): Ditto.
3268 * config/arm/types.md (crypto_sha512): New type attribute.
3269 (crypto_sha3): Ditto.
3270 (doc/invoke.texi): Document new sha3 option.
3272 2018-01-10 Michael Collison <michael.collison@arm.com>
3274 * config/aarch64/aarch64-builtins.c:
3275 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3276 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3277 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3278 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3279 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3280 (AARCH64_ISA_SM4): New ISA flag.
3281 (TARGET_SM4): New feature flag for sm4.
3282 * config/aarch64/aarch64-simd-builtins.def
3283 (aarch64_sm3ss1qv4si): Ditto.
3284 (aarch64_sm3tt1aq4si): Ditto.
3285 (aarch64_sm3tt1bq4si): Ditto.
3286 (aarch64_sm3tt2aq4si): Ditto.
3287 (aarch64_sm3tt2bq4si): Ditto.
3288 (aarch64_sm3partw1qv4si): Ditto.
3289 (aarch64_sm3partw2qv4si): Ditto.
3290 (aarch64_sm4eqv4si): Ditto.
3291 (aarch64_sm4ekeyqv4si): Ditto.
3292 * config/aarch64/aarch64-simd.md:
3293 (aarch64_sm3ss1qv4si): Ditto.
3294 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3295 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3296 (aarch64_sm4eqv4si): Ditto.
3297 (aarch64_sm4ekeyqv4si): Ditto.
3298 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3299 (sm3part_op): Ditto.
3300 (CRYPTO_SM3TT): Ditto.
3301 (CRYPTO_SM3PART): Ditto.
3302 (UNSPEC_SM3SS1): New unspec.
3303 (UNSPEC_SM3TT1A): Ditto.
3304 (UNSPEC_SM3TT1B): Ditto.
3305 (UNSPEC_SM3TT2A): Ditto.
3306 (UNSPEC_SM3TT2B): Ditto.
3307 (UNSPEC_SM3PARTW1): Ditto.
3308 (UNSPEC_SM3PARTW2): Ditto.
3309 (UNSPEC_SM4E): Ditto.
3310 (UNSPEC_SM4EKEY): Ditto.
3311 * config/aarch64/constraints.md (Ui2): New constraint.
3312 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3313 * config/arm/types.md (crypto_sm3): New type attribute.
3314 (crypto_sm4): Ditto.
3315 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3316 (vsm3tt1aq_u32): Ditto.
3317 (vsm3tt1bq_u32): Ditto.
3318 (vsm3tt2aq_u32): Ditto.
3319 (vsm3tt2bq_u32): Ditto.
3320 (vsm3partw1q_u32): Ditto.
3321 (vsm3partw2q_u32): Ditto.
3322 (vsm4eq_u32): Ditto.
3323 (vsm4ekeyq_u32): Ditto.
3324 (doc/invoke.texi): Document new sm4 option.
3326 2018-01-10 Michael Collison <michael.collison@arm.com>
3328 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3329 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3330 (AARCH64_FL_FOR_ARCH8_4): New.
3331 (AARCH64_FL_V8_4): New flag.
3332 (doc/invoke.texi): Document new armv8.4-a option.
3334 2018-01-10 Michael Collison <michael.collison@arm.com>
3336 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3337 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3338 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3339 * config/aarch64/aarch64-option-extension.def: Add
3340 AARCH64_OPT_EXTENSION of 'sha2'.
3341 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3342 (crypto): Disable sha2 and aes if crypto disabled.
3343 (crypto): Enable aes and sha2 if enabled.
3344 (simd): Disable sha2 and aes if simd disabled.
3345 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3347 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3348 (TARGET_SHA2): New feature flag for sha2.
3349 (TARGET_AES): New feature flag for aes.
3350 * config/aarch64/aarch64-simd.md:
3351 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3352 conditional on TARGET_AES.
3353 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3354 (aarch64_crypto_sha1hsi): Make pattern conditional
3356 (aarch64_crypto_sha1hv4si): Ditto.
3357 (aarch64_be_crypto_sha1hv4si): Ditto.
3358 (aarch64_crypto_sha1su1v4si): Ditto.
3359 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3360 (aarch64_crypto_sha1su0v4si): Ditto.
3361 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3362 (aarch64_crypto_sha256su0v4si): Ditto.
3363 (aarch64_crypto_sha256su1v4si): Ditto.
3364 (doc/invoke.texi): Document new aes and sha2 options.
3366 2018-01-10 Martin Sebor <msebor@redhat.com>
3368 PR tree-optimization/83781
3369 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3372 2018-01-11 Martin Sebor <msebor@gmail.com>
3373 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3375 PR tree-optimization/83501
3376 PR tree-optimization/81703
3378 * tree-ssa-strlen.c (get_string_cst): Rename...
3379 (get_string_len): ...to this. Handle global constants.
3380 (handle_char_store): Adjust.
3382 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3383 Jim Wilson <jimw@sifive.com>
3385 * config/riscv/riscv-protos.h (riscv_output_return): New.
3386 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3387 (riscv_attribute_table, riscv_output_return),
3388 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3389 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3390 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3392 (riscv_expand_prologue): Add early return for naked function.
3393 (riscv_expand_epilogue): Likewise.
3394 (riscv_function_ok_for_sibcall): Return false for naked function.
3395 (riscv_set_current_function): New.
3396 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3397 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3398 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3399 * doc/extend.texi (RISC-V Function Attributes): New.
3401 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3403 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3404 check for 128-bit long double before checking TCmode.
3405 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3406 128-bit long doubles before checking TFmode or TCmode.
3407 (FLOAT128_IBM_P): Likewise.
3409 2018-01-10 Martin Sebor <msebor@redhat.com>
3411 PR tree-optimization/83671
3412 * builtins.c (c_strlen): Unconditionally return zero for the empty
3414 Use -Warray-bounds for warnings.
3415 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3416 for non-constant array indices with COMPONENT_REF, arrays of
3417 arrays, and pointers to arrays.
3418 (gimple_fold_builtin_strlen): Determine and set length range for
3419 non-constant character arrays.
3421 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3424 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3427 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3429 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3431 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3434 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3435 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3436 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3437 indexed_or_indirect_operand predicate.
3438 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3439 (*vsx_le_perm_load_v8hi): Likewise.
3440 (*vsx_le_perm_load_v16qi): Likewise.
3441 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3442 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3443 (*vsx_le_perm_store_v8hi): Likewise.
3444 (*vsx_le_perm_store_v16qi): Likewise.
3445 (eight unnamed splitters): Likewise.
3447 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3449 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3450 * config/rs6000/emmintrin.h: Likewise.
3451 * config/rs6000/mmintrin.h: Likewise.
3452 * config/rs6000/xmmintrin.h: Likewise.
3454 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3457 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3459 * tree.c (tree_nop_conversion): Return true for location wrapper
3461 (maybe_wrap_with_location): New function.
3462 (selftest::check_strip_nops): New function.
3463 (selftest::test_location_wrappers): New function.
3464 (selftest::tree_c_tests): Call it.
3465 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3466 (maybe_wrap_with_location): New decl.
3467 (EXPR_LOCATION_WRAPPER_P): New macro.
3468 (location_wrapper_p): New inline function.
3469 (tree_strip_any_location_wrapper): New inline function.
3471 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3474 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3475 stack_realign_offset for the largest alignment of stack slot
3477 (ix86_find_max_used_stack_alignment): New function.
3478 (ix86_finalize_stack_frame_flags): Use it. Set
3479 max_used_stack_alignment if we don't realign stack.
3480 * config/i386/i386.h (machine_function): Add
3481 max_used_stack_alignment.
3483 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3485 * config/arm/arm.opt (-mbranch-cost): New option.
3486 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3489 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3492 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3493 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3495 2018-01-10 Richard Biener <rguenther@suse.de>
3498 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3499 early out so it also covers the case where we have a non-NULL
3502 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3504 PR tree-optimization/83753
3505 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3506 for non-strided grouped accesses if the number of elements is 1.
3508 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3511 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3512 * i386.h (TARGET_USE_GATHER): Define.
3513 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3515 2018-01-10 Martin Liska <mliska@suse.cz>
3518 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3519 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3521 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3522 CLEANUP_NO_PARTITIONING is not set.
3524 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3526 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3527 for vectors, as a partial revert of r254296.
3528 * rtl.h (const_vec_p): Delete.
3529 (const_vec_duplicate_p): Don't test for vector CONSTs.
3530 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3531 * expmed.c (make_tree): Likewise.
3534 * common.md (E, F): Use CONSTANT_P instead of checking for
3536 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3537 checking for CONST_VECTOR.
3539 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3542 * predict.c (force_edge_cold): Handle in more sane way edges
3545 2018-01-09 Carl Love <cel@us.ibm.com>
3547 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3549 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3550 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3551 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3552 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3553 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3554 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3555 * config/rs6000/rs6000-protos.h: Add extern defition for
3556 rs6000_generate_float2_double_code.
3557 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3559 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3560 (float2_v2df): Add define_expand.
3562 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3565 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3566 op_mode in the force_to_mode call.
3568 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3570 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3571 instead of checking each element individually.
3572 (aarch64_evpc_uzp): Likewise.
3573 (aarch64_evpc_zip): Likewise.
3574 (aarch64_evpc_ext): Likewise.
3575 (aarch64_evpc_rev): Likewise.
3576 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3577 instead of checking each element individually. Return true without
3579 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3580 whether all selected elements come from the same input, instead of
3581 checking each element individually. Remove calls to gen_rtx_REG,
3582 start_sequence and end_sequence and instead assert that no rtl is
3585 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3587 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3588 order of HIGH and CONST checks.
3590 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3592 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3593 if the destination isn't an SSA_NAME.
3595 2018-01-09 Richard Biener <rguenther@suse.de>
3597 PR tree-optimization/83668
3598 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3600 (canonicalize_loop_form): ... here, renamed from ...
3601 (canonicalize_loop_closed_ssa_form): ... this and amended to
3602 swap successor edges for loop exit blocks to make us use
3603 the RPO order we need for initial schedule generation.
3605 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3607 PR tree-optimization/64811
3608 * match.pd: When optimizing comparisons with Inf, avoid
3609 introducing or losing exceptions from comparisons with NaN.
3611 2018-01-09 Martin Liska <mliska@suse.cz>
3614 * asan.c (shadow_mem_size): Add gcc_assert.
3616 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3618 Don't save registers in main().
3621 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3622 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3623 * config/avr/avr.c (avr_set_current_function): Don't error if
3624 naked, OS_task or OS_main are specified at the same time.
3625 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3627 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3629 * common/config/avr/avr-common.c (avr_option_optimization_table):
3630 Switch on -mmain-is-OS_task for optimizing compilations.
3632 2018-01-09 Richard Biener <rguenther@suse.de>
3634 PR tree-optimization/83572
3635 * graphite.c: Include cfganal.h.
3636 (graphite_transform_loops): Connect infinite loops to exit
3637 and remove fake edges at the end.
3639 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3641 * ipa-inline.c (edge_badness): Revert accidental checkin.
3643 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3646 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3647 symbols; not inline clones.
3649 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3652 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3653 hard registers. Formatting fixes.
3655 PR preprocessor/83722
3656 * gcc.c (try_generate_repro): Pass
3657 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3658 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3661 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3662 Kito Cheng <kito.cheng@gmail.com>
3664 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3665 (riscv_leaf_function_p): Delete.
3666 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3668 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3670 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3672 (do_ifelse): New function.
3673 (do_isel): New function.
3674 (do_sub3): New function.
3675 (do_add3): New function.
3676 (do_load_mask_compare): New function.
3677 (do_overlap_load_compare): New function.
3678 (expand_compare_loop): New function.
3679 (expand_block_compare): Call expand_compare_loop() when appropriate.
3680 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3682 (-mblock-compare-inline-loop-limit): New option.
3684 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3687 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3688 Reverse order of second and third operands in first alternative.
3689 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3690 of first and second elements in UNSPEC_VPERMR vector.
3691 (altivec_expand_vec_perm_le): Likewise.
3693 2017-01-08 Jeff Law <law@redhat.com>
3695 PR rtl-optimizatin/81308
3696 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3697 (process_switch): If group_case_labels makes a change, then set
3699 (pass_convert_switch::execute): If a switch is converted, then
3700 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3702 PR rtl-optimization/81308
3703 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3706 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3708 PR target/83663 - Revert r255946
3709 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3710 generation for cases where splatting a value is not useful.
3711 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3712 across a vec_duplicate and a paradoxical subreg forming a vector
3713 mode to a vec_concat.
3715 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3717 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3718 -march=armv8.3-a variants.
3719 * config/arm/t-multilib: Likewise.
3720 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3722 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3724 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3726 (cceq_ior_compare_complement): Give it a name so I can use it, and
3727 change boolean_or_operator predicate to boolean_operator so it can
3728 be used to generate a crand.
3729 (eqne): New code iterator.
3730 (bd/bd_neg): New code_attrs.
3731 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3732 a single define_insn.
3733 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3734 decrement (bdnzt/bdnzf/bdzt/bdzf).
3735 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3736 with the new names of the branch decrement patterns, and added the
3737 names of the branch decrement conditional patterns.
3739 2018-01-08 Richard Biener <rguenther@suse.de>
3741 PR tree-optimization/83563
3742 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3745 2018-01-08 Richard Biener <rguenther@suse.de>
3748 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3750 2018-01-08 Richard Biener <rguenther@suse.de>
3752 PR tree-optimization/83685
3753 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3754 references to abnormals.
3756 2018-01-08 Richard Biener <rguenther@suse.de>
3759 * dwarf2out.c (output_indirect_strings): Handle empty
3760 skeleton_debug_str_hash.
3761 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3763 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3765 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3766 (emit_store_direct): Likewise.
3767 (arc_trampoline_adjust_address): Likewise.
3768 (arc_asm_trampoline_template): New function.
3769 (arc_initialize_trampoline): Use asm_trampoline_template.
3770 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3771 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3772 * config/arc/arc.md (flush_icache): Delete pattern.
3774 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3776 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3777 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3780 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3783 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3784 by not USED_FOR_TARGET.
3785 (make_pass_resolve_sw_modes): Likewise.
3787 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3789 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3792 2018-01-08 Richard Biener <rguenther@suse.de>
3795 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3797 2018-01-08 Richard Biener <rguenther@suse.de>
3800 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3802 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3805 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3806 basic blocks with a small number of successors.
3807 (convert_control_dep_chain_into_preds): Improve handling of
3809 (dump_predicates): Split apart into...
3810 (dump_pred_chain): ...here...
3811 (dump_pred_info): ...and here.
3812 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3813 (can_chain_union_be_invalidated_p): Improve check for invalidation
3815 (uninit_uses_cannot_happen): Avoid unnecessary if
3816 convert_control_dep_chain_into_preds yielded nothing.
3818 2018-01-06 Martin Sebor <msebor@redhat.com>
3820 PR tree-optimization/83640
3821 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3822 subtracting negative offset from size.
3823 (builtin_access::overlap): Adjust offset bounds of the access to fall
3824 within the size of the object if possible.
3826 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3828 PR rtl-optimization/83699
3829 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3830 extract_bit_field_as_subreg to cases in which the extracted
3831 value is also a vector.
3833 * lra-constraints.c (process_alt_operands): Test for the equivalence
3834 substitutions when detecting a possible reload cycle.
3836 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3839 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3840 by default if flag_selective_schedling{,2}. Formatting fixes.
3842 PR rtl-optimization/83682
3843 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
3844 if it has non-VECTOR_MODE element mode.
3845 (vec_duplicate_p): Likewise.
3848 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
3849 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
3851 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3854 * config/i386/i386-builtin.def
3855 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
3856 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
3857 Require also OPTION_MASK_ISA_AVX512F in addition to
3858 OPTION_MASK_ISA_GFNI.
3859 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
3860 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
3861 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
3862 to OPTION_MASK_ISA_GFNI.
3863 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
3864 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
3865 OPTION_MASK_ISA_AVX512BW.
3866 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
3867 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
3868 addition to OPTION_MASK_ISA_GFNI.
3869 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
3870 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
3871 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
3872 to OPTION_MASK_ISA_GFNI.
3873 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
3874 a requirement for all ISAs rather than any of them with a few
3876 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
3878 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
3879 bitmasks to be enabled with 3 exceptions, instead of requiring any
3880 enabled ISA with lots of exceptions.
3881 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
3882 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
3883 Change avx512bw in isa attribute to avx512f.
3884 * config/i386/sgxintrin.h: Add license boilerplate.
3885 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
3886 to __AVX512F__ and __AVX512VL to __AVX512VL__.
3887 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
3888 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
3890 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
3891 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
3892 temporarily sse2 rather than sse if not enabled already.
3895 * config/i386/sse.md (VI248_VLBW): Rename to ...
3896 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
3897 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
3898 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
3899 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
3900 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
3901 mode iterator instead of VI248_VLBW.
3903 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
3905 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
3906 (record_modified): Skip clobbers; add debug output.
3907 (param_change_prob): Use sreal frequencies.
3909 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3911 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
3912 punt for user-aligned variables.
3914 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3916 * tree-chrec.c (chrec_contains_symbols): Return true for
3919 2018-01-05 Sudakshina Das <sudi.das@arm.com>
3922 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
3923 of (x|y) == x for BICS pattern.
3925 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3927 PR tree-optimization/83605
3928 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
3929 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
3932 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
3934 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
3935 * config/epiphany/rtems.h: New file.
3937 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3938 Uros Bizjak <ubizjak@gmail.com>
3941 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
3942 QIreg_operand instead of register_operand predicate.
3943 * config/i386/i386.c (ix86_rop_should_change_byte_p,
3944 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
3945 comments instead of -fmitigate[-_]rop.
3947 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3950 * cgraphunit.c (symbol_table::compile): Switch to text_section
3951 before calling assembly_start debug hook.
3952 * run-rtl-passes.c (run_rtl_passes): Likewise.
3955 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3957 * tree-vrp.c (extract_range_from_binary_expr_1): Check
3958 range_int_cst_p rather than !symbolic_range_p before calling
3959 extract_range_from_multiplicative_op_1.
3961 2017-01-04 Jeff Law <law@redhat.com>
3963 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
3964 redundant test in assertion.
3966 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3968 * doc/rtl.texi: Document machine_mode wrapper classes.
3970 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3972 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
3975 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3977 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
3978 the VEC_PERM_EXPR fold to fail.
3980 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3983 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
3984 to switched_sections.
3986 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3989 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
3992 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
3995 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
3996 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
3998 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4001 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4002 is BLKmode and bitpos not zero or mode change is needed.
4004 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4007 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4010 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4013 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4014 instead of MULT rtx. Update all corresponding splitters.
4016 (*ssub<modesuffix>): Ditto.
4018 (*cmp_sadd_di): Update split patterns.
4019 (*cmp_sadd_si): Ditto.
4020 (*cmp_sadd_sidi): Ditto.
4021 (*cmp_ssub_di): Ditto.
4022 (*cmp_ssub_si): Ditto.
4023 (*cmp_ssub_sidi): Ditto.
4024 * config/alpha/predicates.md (const23_operand): New predicate.
4025 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4026 Look for ASHIFT, not MULT inner operand.
4027 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4029 2018-01-04 Martin Liska <mliska@suse.cz>
4031 PR gcov-profile/83669
4032 * gcov.c (output_intermediate_file): Add version to intermediate
4034 * doc/gcov.texi: Document new field 'version' in intermediate
4035 file format. Fix location of '-k' option of gcov command.
4037 2018-01-04 Martin Liska <mliska@suse.cz>
4040 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4042 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4044 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4046 2018-01-03 Martin Sebor <msebor@redhat.com>
4048 PR tree-optimization/83655
4049 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4050 checking calls with invalid arguments.
4052 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4054 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4055 (vectorizable_mask_load_store): Delete.
4056 (vectorizable_call): Return false for masked loads and stores.
4057 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4058 instead of gimple_assign_rhs1.
4059 (vectorizable_load): Handle IFN_MASK_LOAD.
4060 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4062 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4064 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4066 (vectorizable_mask_load_store): ...here.
4067 (vectorizable_load): ...and here.
4069 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4071 * tree-vect-stmts.c (vect_build_all_ones_mask)
4072 (vect_build_zero_merge_argument): New functions, split out from...
4073 (vectorizable_load): ...here.
4075 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4077 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4079 (vectorizable_mask_load_store): ...here.
4080 (vectorizable_store): ...and here.
4082 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4084 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4086 (vectorizable_mask_load_store): ...here.
4088 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4090 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4091 (vect_model_store_cost): Take a vec_load_store_type instead of a
4093 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4094 (vect_model_store_cost): Take a vec_load_store_type instead of a
4096 (vectorizable_mask_load_store): Update accordingly.
4097 (vectorizable_store): Likewise.
4098 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4100 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4102 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4103 IFN_MASK_LOAD calls here rather than...
4104 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4106 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4107 Alan Hayward <alan.hayward@arm.com>
4108 David Sherwood <david.sherwood@arm.com>
4110 * expmed.c (extract_bit_field_1): For vector extracts,
4111 fall back to extract_bit_field_as_subreg if vec_extract
4114 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4115 Alan Hayward <alan.hayward@arm.com>
4116 David Sherwood <david.sherwood@arm.com>
4118 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4119 they are variable or constant sized.
4120 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4121 slots for constant-sized data.
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 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4128 handling COND_EXPRs with boolean comparisons, try to find a better
4129 basis for the mask type than the boolean itself.
4131 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4133 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4134 is calculated and how it can be overridden.
4135 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4136 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4138 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4141 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4142 Alan Hayward <alan.hayward@arm.com>
4143 David Sherwood <david.sherwood@arm.com>
4145 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4146 Remove the mode argument.
4147 (aarch64_simd_valid_immediate): Remove the mode and inverse
4149 * config/aarch64/iterators.md (bitsize): New iterator.
4150 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4151 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4152 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4153 aarch64_simd_valid_immediate.
4154 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4155 (aarch64_reg_or_bic_imm): Likewise.
4156 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4157 with an insn_type enum and msl with a modifier_type enum.
4158 Replace element_width with a scalar_mode. Change the shift
4159 to unsigned int. Add constructors for scalar_float_mode and
4160 scalar_int_mode elements.
4161 (aarch64_vect_float_const_representable_p): Delete.
4162 (aarch64_can_const_movi_rtx_p)
4163 (aarch64_simd_scalar_immediate_valid_for_move)
4164 (aarch64_simd_make_constant): Update call to
4165 aarch64_simd_valid_immediate.
4166 (aarch64_advsimd_valid_immediate_hs): New function.
4167 (aarch64_advsimd_valid_immediate): Likewise.
4168 (aarch64_simd_valid_immediate): Remove mode and inverse
4169 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4170 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4171 and aarch64_float_const_representable_p on the result.
4172 (aarch64_output_simd_mov_immediate): Remove mode argument.
4173 Update call to aarch64_simd_valid_immediate and use of
4174 simd_immediate_info.
4175 (aarch64_output_scalar_simd_mov_immediate): Update call
4178 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4179 Alan Hayward <alan.hayward@arm.com>
4180 David Sherwood <david.sherwood@arm.com>
4182 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4183 (mode_nunits): Likewise CONST_MODE_NUNITS.
4184 * machmode.def (ADJUST_NUNITS): Document.
4185 * genmodes.c (mode_data::need_nunits_adj): New field.
4186 (blank_mode): Update accordingly.
4187 (adj_nunits): New variable.
4188 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4190 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4191 listed in adj_nunits.
4192 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4193 listed in adj_nunits. Don't emit case statements for such modes.
4194 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4195 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4196 nothing if adj_nunits is nonnull.
4197 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4198 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4199 (emit_mode_fbit): Update use of print_maybe_const_decl.
4200 (emit_move_size): Likewise. Treat the array as non-const
4202 (emit_mode_adjustments): Handle adj_nunits.
4204 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4206 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4207 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4208 (VECTOR_MODES): Use it.
4209 (make_vector_modes): Take the prefix as an argument.
4211 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4212 Alan Hayward <alan.hayward@arm.com>
4213 David Sherwood <david.sherwood@arm.com>
4215 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4216 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4217 for MODE_VECTOR_BOOL.
4218 * machmode.def (VECTOR_BOOL_MODE): Document.
4219 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4220 (make_vector_bool_mode): New function.
4221 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4223 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4224 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4226 * stor-layout.c (int_mode_for_mode): Likewise.
4227 * tree.c (build_vector_type_for_mode): Likewise.
4228 * varasm.c (output_constant_pool_2): Likewise.
4229 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4230 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4231 for MODE_VECTOR_BOOL.
4232 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4233 of mode class checks.
4234 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4235 instead of a list of mode class checks.
4236 (expand_vector_scalar_condition): Likewise.
4237 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4239 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4240 Alan Hayward <alan.hayward@arm.com>
4241 David Sherwood <david.sherwood@arm.com>
4243 * machmode.h (mode_size): Change from unsigned short to
4245 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4246 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4247 or if measurement_type is not polynomial.
4248 (fixed_size_mode::includes_p): Check for constant-sized modes.
4249 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4250 return a poly_uint16 rather than an unsigned short.
4251 (emit_mode_size): Change the type of mode_size from unsigned short
4252 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4253 (emit_mode_adjustments): Cope with polynomial vector sizes.
4254 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4256 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4258 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4259 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4260 * caller-save.c (setup_save_areas): Likewise.
4261 (replace_reg_with_saved_mem): Likewise.
4262 * calls.c (emit_library_call_value_1): Likewise.
4263 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4264 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4265 (gen_lowpart_for_combine): Likewise.
4266 * convert.c (convert_to_integer_1): Likewise.
4267 * cse.c (equiv_constant, cse_insn): Likewise.
4268 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4269 (cselib_subst_to_values): Likewise.
4270 * dce.c (word_dce_process_block): Likewise.
4271 * df-problems.c (df_word_lr_mark_ref): Likewise.
4272 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4273 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4274 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4275 (rtl_for_decl_location): Likewise.
4276 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4277 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4278 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4279 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4280 (expand_expr_real_1): Likewise.
4281 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4282 (pad_below): Likewise.
4283 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4284 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4285 * ira.c (get_subreg_tracking_sizes): Likewise.
4286 * ira-build.c (ira_create_allocno_objects): Likewise.
4287 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4288 (ira_sort_regnos_for_alter_reg): Likewise.
4289 * ira-costs.c (record_operand_costs): Likewise.
4290 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4291 (resolve_simple_move): Likewise.
4292 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4293 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4294 (lra_constraints): Likewise.
4295 (CONST_POOL_OK_P): Reject variable-sized modes.
4296 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4297 (add_pseudo_to_slot, lra_spill): Likewise.
4298 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4299 * optabs-query.c (get_best_extraction_insn): Likewise.
4300 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4301 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4302 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4303 * recog.c (offsettable_address_addr_space_p): Likewise.
4304 * regcprop.c (maybe_mode_change): Likewise.
4305 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4306 * regrename.c (build_def_use): Likewise.
4307 * regstat.c (dump_reg_info): Likewise.
4308 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4309 (find_reloads, find_reloads_subreg_address): Likewise.
4310 * reload1.c (eliminate_regs_1): Likewise.
4311 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4312 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4313 (simplify_binary_operation_1, simplify_subreg): Likewise.
4314 * targhooks.c (default_function_arg_padding): Likewise.
4315 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4316 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4317 (verify_gimple_assign_ternary): Likewise.
4318 * tree-inline.c (estimate_move_cost): Likewise.
4319 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4320 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4321 (get_address_cost_ainc): Likewise.
4322 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4323 (vect_supportable_dr_alignment): Likewise.
4324 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4325 (vectorizable_reduction): Likewise.
4326 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4327 (vectorizable_operation, vectorizable_load): Likewise.
4328 * tree.c (build_same_sized_truth_vector_type): Likewise.
4329 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4330 * var-tracking.c (emit_note_insn_var_location): Likewise.
4331 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4332 (ADDR_VEC_ALIGN): Likewise.
4334 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4335 Alan Hayward <alan.hayward@arm.com>
4336 David Sherwood <david.sherwood@arm.com>
4338 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4340 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4341 or if measurement_type is polynomial.
4342 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4343 * combine.c (make_extraction): Likewise.
4344 * dse.c (find_shift_sequence): Likewise.
4345 * dwarf2out.c (mem_loc_descriptor): Likewise.
4346 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4347 (extract_bit_field, extract_low_bits): Likewise.
4348 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4349 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4350 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4351 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4352 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4353 * reload.c (find_reloads): Likewise.
4354 * reload1.c (alter_reg): Likewise.
4355 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4356 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4357 * tree-if-conv.c (predicate_mem_writes): Likewise.
4358 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4359 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4360 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4361 * valtrack.c (dead_debug_insert_temp): Likewise.
4362 * varasm.c (mergeable_constant_section): Likewise.
4363 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4365 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4366 Alan Hayward <alan.hayward@arm.com>
4367 David Sherwood <david.sherwood@arm.com>
4369 * expr.c (expand_assignment): Cope with polynomial mode sizes
4370 when assigning to a CONCAT.
4372 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4373 Alan Hayward <alan.hayward@arm.com>
4374 David Sherwood <david.sherwood@arm.com>
4376 * machmode.h (mode_precision): Change from unsigned short to
4378 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4380 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4381 or if measurement_type is not polynomial.
4382 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4383 in which the mode is already known to be a scalar_int_mode.
4384 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4385 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4387 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4388 for GET_MODE_PRECISION.
4389 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4390 for GET_MODE_PRECISION.
4391 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4393 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4394 (expand_field_assignment, make_extraction): Likewise.
4395 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4396 (get_last_value): Likewise.
4397 * convert.c (convert_to_integer_1): Likewise.
4398 * cse.c (cse_insn): Likewise.
4399 * expr.c (expand_expr_real_1): Likewise.
4400 * lra-constraints.c (simplify_operand_subreg): Likewise.
4401 * optabs-query.c (can_atomic_load_p): Likewise.
4402 * optabs.c (expand_atomic_load): Likewise.
4403 (expand_atomic_store): Likewise.
4404 * ree.c (combine_reaching_defs): Likewise.
4405 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4406 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4407 * tree.h (type_has_mode_precision_p): Likewise.
4408 * ubsan.c (instrument_si_overflow): Likewise.
4410 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4411 Alan Hayward <alan.hayward@arm.com>
4412 David Sherwood <david.sherwood@arm.com>
4414 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4415 polynomial numbers of units.
4416 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4417 (valid_vector_subparts_p): New function.
4418 (build_vector_type): Remove temporary shim and take the number
4419 of units as a poly_uint64 rather than an int.
4420 (build_opaque_vector_type): Take the number of units as a
4421 poly_uint64 rather than an int.
4422 * tree.c (build_vector_from_ctor): Handle polynomial
4423 TYPE_VECTOR_SUBPARTS.
4424 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4425 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4426 (build_vector_from_val): If the number of units is variable,
4427 use build_vec_duplicate_cst for constant operands and
4428 VEC_DUPLICATE_EXPR otherwise.
4429 (make_vector_type): Remove temporary is_constant ().
4430 (build_vector_type, build_opaque_vector_type): Take the number of
4431 units as a poly_uint64 rather than an int.
4432 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4434 * cfgexpand.c (expand_debug_expr): Likewise.
4435 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4436 (store_constructor, expand_expr_real_1): Likewise.
4437 (const_scalar_mask_from_tree): Likewise.
4438 * fold-const-call.c (fold_const_reduction): Likewise.
4439 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4440 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4441 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4442 (fold_relational_const): Likewise.
4443 (native_interpret_vector): Likewise. Change the size from an
4444 int to an unsigned int.
4445 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4446 TYPE_VECTOR_SUBPARTS.
4447 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4448 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4449 duplicating a non-constant operand into a variable-length vector.
4450 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4451 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4452 * ipa-icf.c (sem_variable::equals): Likewise.
4453 * match.pd: Likewise.
4454 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4455 * print-tree.c (print_node): Likewise.
4456 * stor-layout.c (layout_type): Likewise.
4457 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4458 * tree-cfg.c (verify_gimple_comparison): Likewise.
4459 (verify_gimple_assign_binary): Likewise.
4460 (verify_gimple_assign_ternary): Likewise.
4461 (verify_gimple_assign_single): Likewise.
4462 * tree-pretty-print.c (dump_generic_node): Likewise.
4463 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4464 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4465 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4466 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4467 (vect_shift_permute_load_chain): Likewise.
4468 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4469 (expand_vector_condition, optimize_vector_constructor): Likewise.
4470 (lower_vec_perm, get_compute_type): Likewise.
4471 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4472 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4473 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4474 (vect_recog_mask_conversion_pattern): Likewise.
4475 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4476 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4477 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4478 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4479 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4480 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4481 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4482 (supportable_widening_operation): Likewise.
4483 (supportable_narrowing_operation): Likewise.
4484 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4486 * varasm.c (output_constant): Likewise.
4488 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4489 Alan Hayward <alan.hayward@arm.com>
4490 David Sherwood <david.sherwood@arm.com>
4492 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4493 so that both the length == 3 and length != 3 cases set up their
4494 own permute vectors. Add comments explaining why we know the
4495 number of elements is constant.
4496 (vect_permute_load_chain): Likewise.
4498 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4499 Alan Hayward <alan.hayward@arm.com>
4500 David Sherwood <david.sherwood@arm.com>
4502 * machmode.h (mode_nunits): Change from unsigned char to
4504 (ONLY_FIXED_SIZE_MODES): New macro.
4505 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4506 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4507 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4509 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4510 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4511 or if measurement_type is not polynomial.
4512 * genmodes.c (ZERO_COEFFS): New macro.
4513 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4515 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4516 Use ZERO_COEFFS when emitting initializers.
4517 * data-streamer.h (bp_pack_poly_value): New function.
4518 (bp_unpack_poly_value): Likewise.
4519 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4520 for GET_MODE_NUNITS.
4521 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4522 for GET_MODE_NUNITS.
4523 * tree.c (make_vector_type): Remove temporary shim and make
4524 the real function take the number of units as a poly_uint64
4526 (build_vector_type_for_mode): Handle polynomial nunits.
4527 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4528 * emit-rtl.c (const_vec_series_p_1): Likewise.
4529 (gen_rtx_CONST_VECTOR): Likewise.
4530 * fold-const.c (test_vec_duplicate_folding): Likewise.
4531 * genrecog.c (validate_pattern): Likewise.
4532 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4533 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4534 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4535 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4536 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4537 * rtlanal.c (subreg_get_info): Likewise.
4538 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4539 (vect_grouped_load_supported): Likewise.
4540 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4541 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4542 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4543 (simplify_const_unary_operation, simplify_binary_operation_1)
4544 (simplify_const_binary_operation, simplify_ternary_operation)
4545 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4546 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4547 instead of CONST_VECTOR_NUNITS.
4548 * varasm.c (output_constant_pool_2): Likewise.
4549 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4550 explicit-encoded elements in the XVEC for variable-length vectors.
4552 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4554 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4556 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4557 Alan Hayward <alan.hayward@arm.com>
4558 David Sherwood <david.sherwood@arm.com>
4560 * coretypes.h (fixed_size_mode): Declare.
4561 (fixed_size_mode_pod): New typedef.
4562 * builtins.h (target_builtins::x_apply_args_mode)
4563 (target_builtins::x_apply_result_mode): Change type to
4564 fixed_size_mode_pod.
4565 * builtins.c (apply_args_size, apply_result_size, result_vector)
4566 (expand_builtin_apply_args_1, expand_builtin_apply)
4567 (expand_builtin_return): Update accordingly.
4569 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4571 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4572 * cselib.c (cselib_hash_rtx): Likewise.
4573 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4574 CONST_VECTOR encoding.
4576 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4577 Jeff Law <law@redhat.com>
4580 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4581 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4582 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4583 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4586 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4587 explicitly probe *sp in a noreturn function if there were any callee
4588 register saves or frame pointer is needed.
4590 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4593 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4594 BLKmode for ternary, binary or unary expressions.
4597 * var-tracking.c (delete_vta_debug_insn): New inline function.
4598 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4599 insns from get_insns () to NULL instead of each bb separately.
4600 Use delete_vta_debug_insn. No longer static.
4601 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4602 delete_vta_debug_insns callers.
4603 * rtl.h (delete_vta_debug_insns): Declare.
4604 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4605 instead of variable_tracking_main.
4607 2018-01-03 Martin Sebor <msebor@redhat.com>
4609 PR tree-optimization/83603
4610 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4611 arguments past the endof the argument list in functions declared
4612 without a prototype.
4613 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4614 Avoid checking when arguments are null.
4616 2018-01-03 Martin Sebor <msebor@redhat.com>
4619 * doc/extend.texi (attribute const): Fix a typo.
4620 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4621 issuing -Wsuggest-attribute for void functions.
4623 2018-01-03 Martin Sebor <msebor@redhat.com>
4625 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4626 offset_int::from instead of wide_int::to_shwi.
4627 (maybe_diag_overlap): Remove assertion.
4628 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4629 * gimple-ssa-sprintf.c (format_directive): Same.
4630 (parse_directive): Same.
4631 (sprintf_dom_walker::compute_format_length): Same.
4632 (try_substitute_return_value): Same.
4634 2017-01-03 Jeff Law <law@redhat.com>
4637 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4638 non-constant residual for zero at runtime and avoid probing in
4639 that case. Reorganize code for trailing problem to mirror handling
4642 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4644 PR tree-optimization/83501
4645 * tree-ssa-strlen.c (get_string_cst): New.
4646 (handle_char_store): Call get_string_cst.
4648 2018-01-03 Martin Liska <mliska@suse.cz>
4650 PR tree-optimization/83593
4651 * tree-ssa-strlen.c: Include tree-cfg.h.
4652 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4653 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4654 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4656 (strlen_dom_walker::before_dom_children): Call
4657 gimple_purge_dead_eh_edges. Dump tranformation with details
4659 (strlen_dom_walker::before_dom_children): Update call by adding
4660 new argument cleanup_eh.
4661 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4663 2018-01-03 Martin Liska <mliska@suse.cz>
4666 * cif-code.def (VARIADIC_THUNK): New enum value.
4667 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4670 2018-01-03 Jan Beulich <jbeulich@suse.com>
4672 * sse.md (mov<mode>_internal): Tighten condition for when to use
4673 vmovdqu<ssescalarsize> for TI and OI modes.
4675 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4677 Update copyright years.
4679 2018-01-03 Martin Liska <mliska@suse.cz>
4682 * ipa-visibility.c (function_and_variable_visibility): Skip
4683 functions with noipa attribure.
4685 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4687 * gcc.c (process_command): Update copyright notice dates.
4688 * gcov-dump.c (print_version): Ditto.
4689 * gcov.c (print_version): Ditto.
4690 * gcov-tool.c (print_version): Ditto.
4691 * gengtype.c (create_file): Ditto.
4692 * doc/cpp.texi: Bump @copying's copyright year.
4693 * doc/cppinternals.texi: Ditto.
4694 * doc/gcc.texi: Ditto.
4695 * doc/gccint.texi: Ditto.
4696 * doc/gcov.texi: Ditto.
4697 * doc/install.texi: Ditto.
4698 * doc/invoke.texi: Ditto.
4700 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4702 * vector-builder.h (vector_builder::m_full_nelts): Change from
4703 unsigned int to poly_uint64.
4704 (vector_builder::full_nelts): Update prototype accordingly.
4705 (vector_builder::new_vector): Likewise.
4706 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4707 (vector_builder::operator ==): Likewise.
4708 (vector_builder::finalize): Likewise.
4709 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4710 Take the number of elements as a poly_uint64 rather than an
4712 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4713 from unsigned int to poly_uint64.
4714 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4715 (vec_perm_indices::new_vector): Likewise.
4716 (vec_perm_indices::length): Likewise.
4717 (vec_perm_indices::nelts_per_input): Likewise.
4718 (vec_perm_indices::input_nelts): Likewise.
4719 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4720 number of elements per input as a poly_uint64 rather than an
4721 unsigned int. Use the original encoding for variable-length
4722 vectors, rather than clamping each individual element.
4723 For the second and subsequent elements in each pattern,
4724 clamp the step and base before clamping their sum.
4725 (vec_perm_indices::series_p): Handle polynomial element counts.
4726 (vec_perm_indices::all_in_range_p): Likewise.
4727 (vec_perm_indices_to_tree): Likewise.
4728 (vec_perm_indices_to_rtx): Likewise.
4729 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4730 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4731 (tree_vector_builder::new_binary_operation): Handle polynomial
4732 element counts. Return false if we need to know the number
4733 of elements at compile time.
4734 * fold-const.c (fold_vec_perm): Punt if the number of elements
4735 isn't known at compile time.
4737 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4739 * vec-perm-indices.h (vec_perm_builder): Change element type
4740 from HOST_WIDE_INT to poly_int64.
4741 (vec_perm_indices::element_type): Update accordingly.
4742 (vec_perm_indices::clamp): Handle polynomial element_types.
4743 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4744 (vec_perm_indices::all_in_range_p): Likewise.
4745 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4747 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4748 polynomial vec_perm_indices element types.
4749 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4750 * fold-const.c (fold_vec_perm): Likewise.
4751 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4752 * tree-vect-generic.c (lower_vec_perm): Likewise.
4753 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4754 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4755 element type to HOST_WIDE_INT.
4757 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4758 Alan Hayward <alan.hayward@arm.com>
4759 David Sherwood <david.sherwood@arm.com>
4761 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4762 rather than an int. Use plus_constant.
4763 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4764 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
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 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4771 a HOST_WIDE_INT to a poly_int64.
4773 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4774 Alan Hayward <alan.hayward@arm.com>
4775 David Sherwood <david.sherwood@arm.com>
4777 * calls.c (load_register_parameters): Cope with polynomial
4778 mode sizes. Require a constant size for BLKmode parameters
4779 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4780 forces a parameter to be padded at the lsb end in order to
4781 fill a complete number of words, require the parameter size
4782 to be ordered wrt UNITS_PER_WORD.
4784 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4785 Alan Hayward <alan.hayward@arm.com>
4786 David Sherwood <david.sherwood@arm.com>
4788 * reload1.c (spill_stack_slot_width): Change element type
4789 from unsigned int to poly_uint64_pod.
4790 (alter_reg): Treat mode sizes as polynomial.
4792 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4793 Alan Hayward <alan.hayward@arm.com>
4794 David Sherwood <david.sherwood@arm.com>
4796 * reload.c (complex_word_subreg_p): New function.
4797 (reload_inner_reg_of_subreg, push_reload): Use it.
4799 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4800 Alan Hayward <alan.hayward@arm.com>
4801 David Sherwood <david.sherwood@arm.com>
4803 * lra-constraints.c (process_alt_operands): Reject matched
4804 operands whose sizes aren't ordered.
4805 (match_reload): Refer to this check here.
4807 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4808 Alan Hayward <alan.hayward@arm.com>
4809 David Sherwood <david.sherwood@arm.com>
4811 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4812 that the mode size is in the set {1, 2, 4, 8, 16}.
4814 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4815 Alan Hayward <alan.hayward@arm.com>
4816 David Sherwood <david.sherwood@arm.com>
4818 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4819 Use plus_constant instead of gen_rtx_PLUS.
4821 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4822 Alan Hayward <alan.hayward@arm.com>
4823 David Sherwood <david.sherwood@arm.com>
4825 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4826 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4827 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4828 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4829 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4830 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4831 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4832 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4833 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4834 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4836 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4837 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4838 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4839 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4840 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4841 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4842 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4843 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
4844 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
4845 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
4847 * expr.c (emit_move_resolve_push): Treat the input and result
4848 of PUSH_ROUNDING as a poly_int64.
4849 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
4850 (emit_push_insn): Likewise.
4851 * lra-eliminations.c (mark_not_eliminable): Likewise.
4852 * recog.c (push_operand): Likewise.
4853 * reload1.c (elimination_effects): Likewise.
4854 * rtlanal.c (nonzero_bits1): Likewise.
4855 * calls.c (store_one_arg): Likewise. Require the padding to be
4856 known at compile time.
4858 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4859 Alan Hayward <alan.hayward@arm.com>
4860 David Sherwood <david.sherwood@arm.com>
4862 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
4863 Use plus_constant instead of gen_rtx_PLUS.
4865 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4866 Alan Hayward <alan.hayward@arm.com>
4867 David Sherwood <david.sherwood@arm.com>
4869 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
4872 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4873 Alan Hayward <alan.hayward@arm.com>
4874 David Sherwood <david.sherwood@arm.com>
4876 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
4877 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
4878 via stack temporaries. Treat the mode size as polynomial too.
4880 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4881 Alan Hayward <alan.hayward@arm.com>
4882 David Sherwood <david.sherwood@arm.com>
4884 * expr.c (expand_expr_real_2): When handling conversions involving
4885 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
4886 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
4887 as a poly_uint64 too.
4889 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4890 Alan Hayward <alan.hayward@arm.com>
4891 David Sherwood <david.sherwood@arm.com>
4893 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
4895 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4896 Alan Hayward <alan.hayward@arm.com>
4897 David Sherwood <david.sherwood@arm.com>
4899 * combine.c (can_change_dest_mode): Handle polynomial
4900 REGMODE_NATURAL_SIZE.
4901 * expmed.c (store_bit_field_1): Likewise.
4902 * expr.c (store_constructor): Likewise.
4903 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
4904 and polynomial REGMODE_NATURAL_SIZE.
4905 (gen_lowpart_common): Likewise.
4906 * reginfo.c (record_subregs_of_mode): Likewise.
4907 * rtlanal.c (read_modify_subreg_p): Likewise.
4909 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4910 Alan Hayward <alan.hayward@arm.com>
4911 David Sherwood <david.sherwood@arm.com>
4913 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
4914 numbers of elements.
4916 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4917 Alan Hayward <alan.hayward@arm.com>
4918 David Sherwood <david.sherwood@arm.com>
4920 * match.pd: Cope with polynomial numbers of vector elements.
4922 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4923 Alan Hayward <alan.hayward@arm.com>
4924 David Sherwood <david.sherwood@arm.com>
4926 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
4927 in a POINTER_PLUS_EXPR.
4929 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4930 Alan Hayward <alan.hayward@arm.com>
4931 David Sherwood <david.sherwood@arm.com>
4933 * omp-simd-clone.c (simd_clone_subparts): New function.
4934 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
4935 (ipa_simd_modify_function_body): Likewise.
4937 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4938 Alan Hayward <alan.hayward@arm.com>
4939 David Sherwood <david.sherwood@arm.com>
4941 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
4942 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
4943 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
4944 (expand_vector_condition, vector_element): Likewise.
4945 (subparts_gt): New function.
4946 (get_compute_type): Use subparts_gt.
4947 (count_type_subparts): Delete.
4948 (expand_vector_operations_1): Use subparts_gt instead of
4949 count_type_subparts.
4951 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4952 Alan Hayward <alan.hayward@arm.com>
4953 David Sherwood <david.sherwood@arm.com>
4955 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
4956 (vect_compile_time_alias): ...this new function. Do the calculation
4957 on poly_ints rather than trees.
4958 (vect_prune_runtime_alias_test_list): Update call accordingly.
4960 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4961 Alan Hayward <alan.hayward@arm.com>
4962 David Sherwood <david.sherwood@arm.com>
4964 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
4966 (vect_schedule_slp_instance): Likewise.
4968 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4969 Alan Hayward <alan.hayward@arm.com>
4970 David Sherwood <david.sherwood@arm.com>
4972 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
4973 constant and extern definitions for variable-length vectors.
4974 (vect_get_constant_vectors): Note that the number of units
4975 is known to be constant.
4977 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4978 Alan Hayward <alan.hayward@arm.com>
4979 David Sherwood <david.sherwood@arm.com>
4981 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
4982 of units as polynomial. Choose between WIDE and NARROW based
4985 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4986 Alan Hayward <alan.hayward@arm.com>
4987 David Sherwood <david.sherwood@arm.com>
4989 * tree-vect-stmts.c (simd_clone_subparts): New function.
4990 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
4992 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4993 Alan Hayward <alan.hayward@arm.com>
4994 David Sherwood <david.sherwood@arm.com>
4996 * tree-vect-stmts.c (vectorizable_call): Treat the number of
4997 vectors as polynomial. Use build_index_vector for
5000 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5001 Alan Hayward <alan.hayward@arm.com>
5002 David Sherwood <david.sherwood@arm.com>
5004 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5005 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5006 for variable-length vectors.
5007 (vectorizable_mask_load_store): Treat the number of units as
5008 polynomial, asserting that it is constant if the condition has
5009 already been enforced.
5010 (vectorizable_store, vectorizable_load): Likewise.
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-vect-loop.c (vectorizable_live_operation): Treat the number
5017 of units as polynomial. Punt if we can't tell at compile time
5018 which vector contains the final result.
5020 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5021 Alan Hayward <alan.hayward@arm.com>
5022 David Sherwood <david.sherwood@arm.com>
5024 * tree-vect-loop.c (vectorizable_induction): Treat the number
5025 of units as polynomial. Punt on SLP inductions. Use an integer
5026 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5027 cast of such a series for variable-length floating-point
5030 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5031 Alan Hayward <alan.hayward@arm.com>
5032 David Sherwood <david.sherwood@arm.com>
5034 * tree.h (build_index_vector): Declare.
5035 * tree.c (build_index_vector): New function.
5036 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5037 of units as polynomial, forcibly converting it to a constant if
5038 vectorizable_reduction has already enforced the condition.
5039 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5040 to create a {1,2,3,...} vector.
5041 (vectorizable_reduction): Treat the number of units as polynomial.
5042 Choose vectype_in based on the largest scalar element size rather
5043 than the smallest number of units. Enforce the restrictions
5046 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5047 Alan Hayward <alan.hayward@arm.com>
5048 David Sherwood <david.sherwood@arm.com>
5050 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5051 number of units as polynomial.
5053 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5054 Alan Hayward <alan.hayward@arm.com>
5055 David Sherwood <david.sherwood@arm.com>
5057 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5058 * target.def (autovectorize_vector_sizes): Return the vector sizes
5059 by pointer, using vector_sizes rather than a bitmask.
5060 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5061 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5062 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5064 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5065 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5066 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5067 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5068 * omp-general.c (omp_max_vf): Likewise.
5069 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5070 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5071 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5072 * tree-vect-slp.c (vect_slp_bb): Likewise.
5073 * doc/tm.texi: Regenerate.
5074 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5076 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5077 the vector size as a poly_uint64 rather than an unsigned int.
5078 (current_vector_size): Change from an unsigned int to a poly_uint64.
5079 (get_vectype_for_scalar_type): Update accordingly.
5080 * tree.h (build_truth_vector_type): Take the size and number of
5081 units as a poly_uint64 rather than an unsigned int.
5082 (build_vector_type): Add a temporary overload that takes
5083 the number of units as a poly_uint64 rather than an unsigned int.
5084 * tree.c (make_vector_type): Likewise.
5085 (build_truth_vector_type): Take the number of units as a poly_uint64
5086 rather than an unsigned int.
5088 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5089 Alan Hayward <alan.hayward@arm.com>
5090 David Sherwood <david.sherwood@arm.com>
5092 * target.def (get_mask_mode): Take the number of units and length
5093 as poly_uint64s rather than unsigned ints.
5094 * targhooks.h (default_get_mask_mode): Update accordingly.
5095 * targhooks.c (default_get_mask_mode): Likewise.
5096 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5097 * doc/tm.texi: Regenerate.
5099 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5100 Alan Hayward <alan.hayward@arm.com>
5101 David Sherwood <david.sherwood@arm.com>
5103 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5104 * omp-general.c (omp_max_vf): Likewise.
5105 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5106 (expand_omp_simd): Handle polynomial safelen.
5107 * omp-low.c (omplow_simd_context): Add a default constructor.
5108 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5109 (lower_rec_simd_input_clauses): Update accordingly.
5110 (lower_rec_input_clauses): Likewise.
5112 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5113 Alan Hayward <alan.hayward@arm.com>
5114 David Sherwood <david.sherwood@arm.com>
5116 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5117 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5118 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5119 (vect_analyze_slp_cost): Likewise.
5120 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5121 (vect_model_load_cost): Likewise.
5123 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5124 Alan Hayward <alan.hayward@arm.com>
5125 David Sherwood <david.sherwood@arm.com>
5127 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5128 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5129 from an unsigned int * to a poly_uint64_pod *.
5130 (calculate_unrolling_factor): New function.
5131 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5133 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5134 Alan Hayward <alan.hayward@arm.com>
5135 David Sherwood <david.sherwood@arm.com>
5137 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5138 from an unsigned int to a poly_uint64.
5139 (_loop_vec_info::slp_unrolling_factor): Likewise.
5140 (_loop_vec_info::vectorization_factor): Change from an int
5142 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5143 (vect_get_num_vectors): New function.
5144 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5145 (vect_get_num_copies): Use vect_get_num_vectors.
5146 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5147 to an unsigned int *.
5148 (vect_analyze_data_refs): Change min_vf from an int * to a
5150 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5151 than an unsigned HOST_WIDE_INT.
5152 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5153 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5154 to an unsigned int *.
5155 (vect_analyze_data_ref_dependences): Likewise.
5156 (vect_compute_data_ref_alignment): Handle polynomial vf.
5157 (vect_enhance_data_refs_alignment): Likewise.
5158 (vect_prune_runtime_alias_test_list): Likewise.
5159 (vect_shift_permute_load_chain): Likewise.
5160 (vect_supportable_dr_alignment): Likewise.
5161 (dependence_distance_ge_vf): Take the vectorization factor as a
5162 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5163 (vect_analyze_data_refs): Change min_vf from an int * to a
5165 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5166 vfm1 as a poly_uint64 rather than an int. Make the same change
5167 for the returned bound_scalar.
5168 (vect_gen_vector_loop_niters): Handle polynomial vf.
5169 (vect_do_peeling): Likewise. Update call to
5170 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5171 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5173 * tree-vect-loop.c (vect_determine_vectorization_factor)
5174 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5175 (vect_get_known_peeling_cost): Likewise.
5176 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5177 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5178 (vect_transform_loop): Likewise. Use the lowest possible VF when
5179 updating the upper bounds of the loop.
5180 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5182 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5183 polynomial unroll factors.
5184 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5185 (vect_make_slp_decision): Likewise.
5186 (vect_supported_load_permutation_p): Likewise, and polynomial
5188 (vect_analyze_slp_cost): Handle polynomial vf.
5189 (vect_slp_analyze_node_operations): Likewise.
5190 (vect_slp_analyze_bb_1): Likewise.
5191 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5192 than an unsigned HOST_WIDE_INT.
5193 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5194 (vectorizable_load): Handle polynomial vf.
5195 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5197 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5199 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5200 Alan Hayward <alan.hayward@arm.com>
5201 David Sherwood <david.sherwood@arm.com>
5203 * match.pd: Handle bit operations involving three constants
5204 and try to fold one pair.
5206 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5208 * tree-vect-loop-manip.c: Include gimple-fold.h.
5209 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5210 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5211 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5212 Add a path that uses a step of VF instead of 1, but disable it
5214 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5215 and niters_no_overflow parameters. Update calls to
5216 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5217 Create a new SSA name if the latter choses to use a ste other
5218 than zero, and return it via niters_vector_mult_vf_var.
5219 * tree-vect-loop.c (vect_transform_loop): Update calls to
5220 vect_do_peeling, vect_gen_vector_loop_niters and
5221 slpeel_make_loop_iterate_ntimes.
5222 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5223 (vect_gen_vector_loop_niters): Update declarations after above changes.
5225 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5227 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5228 128-bit round to integer instructions.
5229 (ceil<mode>2): Likewise.
5230 (btrunc<mode>2): Likewise.
5231 (round<mode>2): Likewise.
5233 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5235 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5236 unaligned VSX load/store on P8/P9.
5237 (expand_block_clear): Allow the use of unaligned VSX
5238 load/store on P8/P9.
5240 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5242 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5244 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5245 swap associated with both a load and a store.
5247 2018-01-02 Andrew Waterman <andrew@sifive.com>
5249 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5250 * config/riscv/riscv.md (clear_cache): Use it.
5252 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5254 * web.c: Remove out-of-date comment.
5256 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5258 * expr.c (fixup_args_size_notes): Check that any existing
5259 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5260 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5261 (emit_single_push_insn): ...here.
5263 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5265 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5266 (const_vector_encoded_nelts): New function.
5267 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5268 (const_vector_int_elt, const_vector_elt): Declare.
5269 * emit-rtl.c (const_vector_int_elt_1): New function.
5270 (const_vector_elt): Likewise.
5271 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5272 of CONST_VECTOR_ELT.
5274 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5276 * expr.c: Include rtx-vector-builder.h.
5277 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5278 directly on the tree encoding.
5279 (const_vector_from_tree): Likewise.
5280 * optabs.c: Include rtx-vector-builder.h.
5281 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5282 sequence of "u" values.
5283 * vec-perm-indices.c: Include rtx-vector-builder.h.
5284 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5285 directly on the vec_perm_indices encoding.
5287 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5289 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5290 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5291 * rtx-vector-builder.h: New file.
5292 * rtx-vector-builder.c: Likewise.
5293 * rtl.h (rtx_def::u2): Add a const_vector field.
5294 (CONST_VECTOR_NPATTERNS): New macro.
5295 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5296 (CONST_VECTOR_DUPLICATE_P): Likewise.
5297 (CONST_VECTOR_STEPPED_P): Likewise.
5298 (CONST_VECTOR_ENCODED_ELT): Likewise.
5299 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5300 (unwrap_const_vec_duplicate): Likewise.
5301 (const_vec_series_p): Check for a non-duplicated vector encoding.
5302 Say that the function only returns true for integer vectors.
5303 * emit-rtl.c: Include rtx-vector-builder.h.
5304 (gen_const_vec_duplicate_1): Delete.
5305 (gen_const_vector): Call gen_const_vec_duplicate instead of
5306 gen_const_vec_duplicate_1.
5307 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5308 (gen_const_vec_duplicate): Use rtx_vector_builder.
5309 (gen_const_vec_series): Likewise.
5310 (gen_rtx_CONST_VECTOR): Likewise.
5311 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5312 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5313 Build a new vector rather than modifying a CONST_VECTOR in-place.
5314 (handle_special_swappables): Update call accordingly.
5315 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5316 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5317 Build a new vector rather than modifying a CONST_VECTOR in-place.
5318 (handle_special_swappables): Update call accordingly.
5320 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5322 * simplify-rtx.c (simplify_const_binary_operation): Use
5323 CONST_VECTOR_ELT instead of XVECEXP.
5325 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5327 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5328 the selector elements to be different from the data elements
5329 if the selector is a VECTOR_CST.
5330 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5331 ssizetype for the selector.
5333 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5335 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5336 before testing each element individually.
5337 * tree-vect-generic.c (lower_vec_perm): Likewise.
5339 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5341 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5342 * selftest-run-tests.c (selftest::run_tests): Call it.
5343 * vector-builder.h (vector_builder::operator ==): New function.
5344 (vector_builder::operator !=): Likewise.
5345 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5346 (vec_perm_indices::all_from_input_p): New function.
5347 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5348 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5349 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5350 instead of reading the VECTOR_CST directly. Detect whether both
5351 vector inputs are the same before constructing the vec_perm_indices,
5352 and update the number of inputs argument accordingly. Use the
5353 utility functions added above. Only construct sel2 if we need to.
5355 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5357 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5358 the broadcast of the low byte.
5359 (expand_mult_highpart): Use an explicit encoding for the permutes.
5360 * optabs-query.c (can_mult_highpart_p): Likewise.
5361 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5362 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5363 (vectorizable_bswap): Likewise.
5364 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5365 explicit encoding for the power-of-2 permutes.
5366 (vect_permute_store_chain): Likewise.
5367 (vect_grouped_load_supported): Likewise.
5368 (vect_permute_load_chain): Likewise.
5370 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5372 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5373 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5374 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5375 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5376 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5377 (vect_gen_perm_mask_any): Likewise.
5379 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5381 * int-vector-builder.h: New file.
5382 * vec-perm-indices.h: Include int-vector-builder.h.
5383 (vec_perm_indices): Redefine as an int_vector_builder.
5384 (auto_vec_perm_indices): Delete.
5385 (vec_perm_builder): Redefine as a stand-alone class.
5386 (vec_perm_indices::vec_perm_indices): New function.
5387 (vec_perm_indices::clamp): Likewise.
5388 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5389 (vec_perm_indices::new_vector): New function.
5390 (vec_perm_indices::new_expanded_vector): Update for new
5391 vec_perm_indices class.
5392 (vec_perm_indices::rotate_inputs): New function.
5393 (vec_perm_indices::all_in_range_p): Operate directly on the
5394 encoded form, without computing elided elements.
5395 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5396 encoding. Update for new vec_perm_indices class.
5397 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5398 the given vec_perm_builder.
5399 (expand_vec_perm_var): Update vec_perm_builder constructor.
5400 (expand_mult_highpart): Use vec_perm_builder instead of
5401 auto_vec_perm_indices.
5402 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5403 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5404 or double series encoding as appropriate.
5405 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5406 vec_perm_indices instead of auto_vec_perm_indices.
5407 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5408 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5409 (vect_permute_store_chain): Likewise.
5410 (vect_grouped_load_supported): Likewise.
5411 (vect_permute_load_chain): Likewise.
5412 (vect_shift_permute_load_chain): Likewise.
5413 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5414 (vect_transform_slp_perm_load): Likewise.
5415 (vect_schedule_slp_instance): Likewise.
5416 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5417 (vectorizable_mask_load_store): Likewise.
5418 (vectorizable_bswap): Likewise.
5419 (vectorizable_store): Likewise.
5420 (vectorizable_load): Likewise.
5421 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5422 vec_perm_indices instead of auto_vec_perm_indices. Use
5423 tree_to_vec_perm_builder to read the vector from a tree.
5424 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5425 vec_perm_builder instead of a vec_perm_indices.
5426 (have_whole_vector_shift): Use vec_perm_builder and
5427 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5428 truncation to calc_vec_perm_mask_for_shift.
5429 (vect_create_epilog_for_reduction): Likewise.
5430 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5431 from auto_vec_perm_indices to vec_perm_indices.
5432 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5433 instead of changing individual elements.
5434 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5435 the vector in d.perm.
5436 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5437 from auto_vec_perm_indices to vec_perm_indices.
5438 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5439 instead of changing individual elements.
5440 (arm_vectorize_vec_perm_const): Use new_vector to install
5441 the vector in d.perm.
5442 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5443 Update vec_perm_builder constructor.
5444 (rs6000_expand_interleave): Likewise.
5445 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5446 (rs6000_expand_interleave): Likewise.
5448 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5450 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5451 to qimode could truncate the indices.
5452 * optabs.c (expand_vec_perm_var): Likewise.
5454 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5456 * Makefile.in (OBJS): Add vec-perm-indices.o.
5457 * vec-perm-indices.h: New file.
5458 * vec-perm-indices.c: Likewise.
5459 * target.h (vec_perm_indices): Replace with a forward class
5461 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5462 * optabs.h: Include vec-perm-indices.h.
5463 (expand_vec_perm): Delete.
5464 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5465 (expand_vec_perm_const): Declare.
5466 * target.def (vec_perm_const_ok): Replace with...
5467 (vec_perm_const): ...this new hook.
5468 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5469 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5470 * doc/tm.texi: Regenerate.
5471 * optabs.def (vec_perm_const): Delete.
5472 * doc/md.texi (vec_perm_const): Likewise.
5473 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5474 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5475 expand_vec_perm for constant permutation vectors. Assert that
5476 the mode of variable permutation vectors is the integer equivalent
5477 of the mode that is being permuted.
5478 * optabs-query.h (selector_fits_mode_p): Declare.
5479 * optabs-query.c: Include vec-perm-indices.h.
5480 (selector_fits_mode_p): New function.
5481 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5482 is defined, instead of checking whether the vec_perm_const_optab
5483 exists. Use targetm.vectorize.vec_perm_const instead of
5484 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5485 fit in the vector mode before using a variable permute.
5486 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5487 vec_perm_indices instead of an rtx.
5488 (expand_vec_perm): Replace with...
5489 (expand_vec_perm_const): ...this new function. Take the selector
5490 as a vec_perm_indices rather than an rtx. Also take the mode of
5491 the selector. Update call to shift_amt_for_vec_perm_mask.
5492 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5493 Use vec_perm_indices::new_expanded_vector to expand the original
5494 selector into bytes. Check whether the indices fit in the vector
5495 mode before using a variable permute.
5496 (expand_vec_perm_var): Make global.
5497 (expand_mult_highpart): Use expand_vec_perm_const.
5498 * fold-const.c: Includes vec-perm-indices.h.
5499 * tree-ssa-forwprop.c: Likewise.
5500 * tree-vect-data-refs.c: Likewise.
5501 * tree-vect-generic.c: Likewise.
5502 * tree-vect-loop.c: Likewise.
5503 * tree-vect-slp.c: Likewise.
5504 * tree-vect-stmts.c: Likewise.
5505 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5507 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5508 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5509 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5510 (aarch64_vectorize_vec_perm_const): ...this new function.
5511 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5512 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5513 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5514 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5515 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5516 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5517 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5519 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5520 check for NEON modes.
5521 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5522 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5523 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5524 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5526 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5527 the old VEC_PERM_CONST conditions.
5528 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5529 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5530 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5531 (ia64_vectorize_vec_perm_const_ok): Merge into...
5532 (ia64_vectorize_vec_perm_const): ...this new function.
5533 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5534 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5535 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5536 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5537 * config/mips/mips.c (mips_expand_vec_perm_const)
5538 (mips_vectorize_vec_perm_const_ok): Merge into...
5539 (mips_vectorize_vec_perm_const): ...this new function.
5540 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5541 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5542 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5543 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5544 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5545 (rs6000_expand_vec_perm_const): Delete.
5546 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
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.
5556 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5557 an element count and rtx array.
5558 (rs6000_expand_extract_even): Update call accordingly.
5559 (rs6000_expand_interleave): Likewise.
5560 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5561 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5562 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5563 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5564 (rs6000_expand_vec_perm_const): Delete.
5565 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5566 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5567 (altivec_expand_vec_perm_const_le): Take each operand individually.
5568 Operate on constant selectors rather than rtxes.
5569 (altivec_expand_vec_perm_const): Likewise. Update call to
5570 altivec_expand_vec_perm_const_le.
5571 (rs6000_expand_vec_perm_const): Delete.
5572 (rs6000_vectorize_vec_perm_const_ok): Delete.
5573 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5574 reference to the SPE evmerge intructions.
5575 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5576 an element count and rtx array.
5577 (rs6000_expand_extract_even): Update call accordingly.
5578 (rs6000_expand_interleave): Likewise.
5579 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5580 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5582 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5584 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5586 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5587 vector mode and that that mode matches the mode of the data
5589 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5590 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5591 directly using expand_vec_perm_1 when forcing selectors into
5593 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5595 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5597 * optabs-query.h (can_vec_perm_p): Delete.
5598 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5599 * optabs-query.c (can_vec_perm_p): Split into...
5600 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5601 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5602 particular selector is valid.
5603 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5604 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5605 (vect_grouped_load_supported): Likewise.
5606 (vect_shift_permute_load_chain): Likewise.
5607 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5608 (vect_transform_slp_perm_load): Likewise.
5609 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5610 (vectorizable_bswap): Likewise.
5611 (vect_gen_perm_mask_checked): Likewise.
5612 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5613 implementations of variable permutation vectors into account
5614 when deciding which selector to use.
5615 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5616 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5617 with a false third argument.
5618 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5619 to test whether the constant selector is valid and can_vec_perm_var_p
5620 to test whether a variable selector is valid.
5622 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5624 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5625 * optabs-query.c (can_vec_perm_p): Likewise.
5626 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5627 instead of vec_perm_indices.
5628 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5629 (vect_gen_perm_mask_checked): Likewise,
5630 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5631 (vect_gen_perm_mask_checked): Likewise,
5633 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5635 * optabs-query.h (qimode_for_vec_perm): Declare.
5636 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5637 (qimode_for_vec_perm): ...this new function.
5638 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5640 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5642 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5643 does not have a conditional at the top.
5645 2018-01-02 Richard Biener <rguenther@suse.de>
5647 * ipa-inline.c (big_speedup_p): Fix expression.
5649 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5652 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5655 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5659 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5660 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5661 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5662 cond_taken_branch_cost 3->4.
5664 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5666 PR tree-optimization/83581
5667 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5668 TODO_cleanup_cfg if any changes have been made.
5671 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5672 convert_modes if target mode has the right side, but different mode
5676 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5677 last argument when extracting from CONCAT. If either from_real or
5678 from_imag is NULL, use expansion through memory. If result is not
5679 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5680 the parts directly to inner mode, if even that fails, use expansion
5684 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5685 check for bswap in mode rather than HImode and use that in expand_unop
5688 Copyright (C) 2018 Free Software Foundation, Inc.
5690 Copying and distribution of this file, with or without modification,
5691 are permitted in any medium without royalty provided the copyright
5692 notice and this notice are preserved.