1 2018-01-29 Richard Biener <rguenther@suse.de>
3 PR tree-optimization/84086
4 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
5 (flush_ssaname_freelist): When SSA names were released reset
8 2018-01-29 Richard Biener <rguenther@suse.de>
10 PR tree-optimization/84057
11 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
12 removed paths when removing edges.
14 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
16 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
17 -mfunction-return=@var{choice}.
19 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
22 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
24 (layout::print_source_line): Likewise.
25 (test_get_line_width_without_trailing_whitespace): Add test cases.
27 2018-01-27 Jakub Jelinek <jakub@redhat.com>
30 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
33 2018-01-26 Jim Wilson <jimw@sifive.com>
35 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
37 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
40 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
42 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
43 and CMP + SUB-immediate -> SUBS.
45 2018-01-26 Martin Sebor <msebor@redhat.com>
47 PR tree-optimization/83896
48 * tree-ssa-strlen.c (get_string_len): Rename...
49 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
50 Avoid assuming length is constant.
51 (handle_char_store): Use HOST_WIDE_INT for string length.
53 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
56 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
57 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
59 2018-01-26 Richard Biener <rguenther@suse.de>
61 PR rtl-optimization/84003
62 * dse.c (record_store): Only record redundant stores when
63 the earlier store aliases at least all accesses the later one does.
65 2018-01-26 Jakub Jelinek <jakub@redhat.com>
67 PR rtl-optimization/83985
68 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
69 REG_CFA_RESTORE insns.
70 (delete_unmarked_insns): Don't ignore separate shrink wrapping
71 REG_CFA_RESTORE insns here.
74 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
75 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
77 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
79 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
80 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
82 (arc_override_options): Likewise.
83 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
85 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
87 * config/arc/arc.h (TARGET_DBNZ): Define.
88 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
89 properly set the tune attribute.
90 (dbnz): Use TARGET_DBNZ guard.
91 * config/arc/arc.opt (mtune): Add core3 option.
93 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
95 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
96 recognize new pic like addresses.
97 (arc_delegitimize_address): Clean up.
99 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
101 * config/arc/arc-arches.def: Option mrf16 valid for all
103 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
104 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
105 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
106 * config/arc/arc-tables.opt: Regenerate.
107 * config/arc/arc.c (arc_conditional_register_usage): Handle
108 reduced register file case.
109 (arc_file_start): Set must have build attributes.
110 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
112 * config/arc/arc.opt (mrf16): Add new option.
113 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
114 * config/arc/genmultilib.awk: Handle new mrf16 option.
115 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
116 * config/arc/t-multilib: Regenerate.
117 * doc/invoke.texi (ARC Options): Document mrf16 option.
119 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
121 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
122 * config/arc/arc.c (arc_handle_secure_attribute): New function.
123 (arc_attribute_table): Add 'secure_call' attribute.
124 (arc_print_operand): Print secure call operand.
125 (arc_function_ok_for_sibcall): Don't optimize tail calls when
127 (arc_is_secure_call_p): New function. * config/arc/arc.md
128 (call_i): Add support for sjli instruction.
129 (call_value_i): Likewise.
130 * config/arc/constraints.md (Csc): New constraint.
132 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
133 John Eric Martin <John.Martin@emmicro-us.com>
135 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
136 * config/arc/arc.c (_arc_jli_section): New struct.
137 (arc_jli_section): New type.
138 (rc_jli_sections): New static variable.
139 (arc_handle_jli_attribute): New function.
140 (arc_attribute_table): Add jli_always and jli_fixed attribute.
141 (arc_file_end): New function.
142 (TARGET_ASM_FILE_END): Define.
143 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
144 (arc_add_jli_section): New function.
145 (jli_call_scan): Likewise.
146 (arc_reorg): Call jli_call_scan.
147 (arc_output_addsi): Remove 'S' from printing asm operand.
148 (arc_is_jli_call_p): New function.
149 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
151 (movhi_insn): Likewise.
152 (movsi_insn): Likewise.
153 (movsi_set_cc_insn): Likewise.
154 (loadqi_update): Likewise.
155 (load_zeroextendqisi_update): Likewise.
156 (load_signextendqisi_update): Likewise.
157 (loadhi_update): Likewise.
158 (load_zeroextendhisi_update): Likewise.
159 (load_signextendhisi_update): Likewise.
160 (loadsi_update): Likewise.
161 (loadsf_update): Likewise.
162 (movsicc_insn): Likewise.
163 (bset_insn): Likewise.
164 (bxor_insn): Likewise.
165 (bclr_insn): Likewise.
166 (bmsk_insn): Likewise.
167 (bicsi3_insn): Likewise.
168 (cmpsi_cc_c_insn): Likewise.
169 (movsi_ne): Likewise.
170 (movsi_cond_exec): Likewise.
171 (clrsbsi2): Likewise.
182 (call_i): Remove 'S' asm letter, add jli instruction.
183 (call_value_i): Likewise.
184 * config/arc/arc.op (mjli-always): New option.
185 * config/arc/constraints.md (Cji): New constraint.
186 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
188 (subsf3_fpx): Likewise.
189 (mulsf3_fpx): Likewise.
190 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
192 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
194 * doc/invoke.texi (ARC): Document mjli-always option.
196 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
198 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
199 and use incw and decw where possible
200 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
202 2018-01-26 Richard Biener <rguenther@suse.de>
204 PR tree-optimization/81082
205 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
206 association if it requires casting to unsigned.
207 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
208 from fold_plusminus_mult_expr to catch important cases late when
209 range info is available.
211 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
213 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
214 * configure.ac (hidden_linkonce): New test.
215 * configure: Regenerate.
216 * config.in: Regenerate.
218 2018-01-26 Julia Koval <julia.koval@intel.com>
220 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
221 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
222 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
223 _mm_mask_bitshuffle_epi64_mask): Fix type.
224 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
225 USI_FTYPE_V4DI_V4DI_USI): Remove.
226 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
227 __builtin_ia32_vpshufbitqmb256_mask,
228 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
229 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
230 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
232 2018-01-26 Alan Modra <amodra@gmail.com>
235 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
236 UNSPEC_VBPERMQ. Sort other unspecs.
238 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
240 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
242 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
245 * predict.c (drop_profile): Do not push/pop cfun; update also
247 (handle_missing_profiles): Fix logic looking for zero profiles.
249 2018-01-25 Jakub Jelinek <jakub@redhat.com>
252 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
253 on functions with #pragma omp declare simd or functions with simd
255 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
256 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
257 Remove trailing \n from warning_at calls.
259 2018-01-25 Tom de Vries <tom@codesourcery.com>
262 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
263 for neutered workers.
265 2018-01-24 Joseph Myers <joseph@codesourcery.com>
268 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
269 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
271 2017-01-08 Jeff Law <law@redhat.com>
274 * i386.c (get_probe_interval): Move to earlier point.
275 (ix86_compute_frame_layout): If -fstack-clash-protection and
276 the frame is larger than the probe interval, then use pushes
277 to save registers rather than reg->mem moves.
278 (ix86_expand_prologue): Remove conditional for int_registers_saved
281 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
284 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
285 min/max for never referenced object.
287 2018-01-24 Jakub Jelinek <jakub@redhat.com>
290 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
292 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
293 attributes from DECL_ATTRIBUTES (decl) without affecting
294 DECL_ATTRIBUTES (current_function_decl).
295 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
296 functions with non-NULL DECL_ABSTRACT_ORIGIN.
298 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
300 PR tree-optimization/83979
301 * fold-const.c (fold_comparison): Use constant_boolean_node
302 instead of boolean_{true,false}_node.
304 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
306 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
309 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
311 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
312 Simplify the clause that sets the length attribute.
313 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
314 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
315 clause that sets the length attribute.
316 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
318 2018-01-24 Tom de Vries <tom@codesourcery.com>
321 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
322 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
323 Add strict parameter.
324 (prevent_branch_around_nothing): Insert dummy insn between branch to
325 label and label with no ptx insn inbetween.
326 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
328 2018-01-24 Tom de Vries <tom@codesourcery.com>
331 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
332 for neutered threads in warp.
333 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
335 2018-01-24 Richard Biener <rguenther@suse.de>
337 PR tree-optimization/83176
338 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
341 2018-01-24 Richard Biener <rguenther@suse.de>
343 PR tree-optimization/82819
344 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
345 code generating pluses that are no-ops in the target precision.
347 2018-01-24 Richard Biener <rguenther@suse.de>
350 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
352 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
354 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
355 to merge probabilities.
356 * predict.c (probably_never_executed): Also mark as cold functions
357 with global 0 profile and guessed local profile.
358 * profile-count.c (profile_probability::combine_with_count): New
360 * profile-count.h (profile_probability::operator*,
361 profile_probability::operator*=, profile_probability::operator/,
362 profile_probability::operator/=): Reduce precision to adjusted
363 and set value to guessed on contradictory divisions.
364 (profile_probability::combine_with_freq): Remove.
365 (profile_probability::combine_wiht_count): Declare.
366 (profile_count::force_nonzero):: Set to adjusted.
367 (profile_count::probability_in):: Set quality to adjusted.
368 * tree-ssa-tail-merge.c (replace_block_by): Use
371 2018-01-23 Andrew Waterman <andrew@sifive.com>
372 Jim Wilson <jimw@sifive.com>
374 * config/riscv/riscv.c (riscv_stack_boundary): New.
375 (riscv_option_override): Set riscv_stack_boundary. Handle
376 riscv_preferred_stack_boundary_arg.
377 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
378 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
379 (STACK_BOUNDARY): Set to riscv_stack_boundary.
380 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
381 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
382 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
384 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
387 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
388 of struct ix86_frame.
389 (ix86_expand_epilogue): Likewise. Add a local variable for
390 the reg_save_offset field in struct ix86_frame.
392 2018-01-23 Bin Cheng <bin.cheng@arm.com>
394 PR tree-optimization/82604
395 * tree-loop-distribution.c (enum partition_kind): New enum item
396 PKIND_PARTIAL_MEMSET.
397 (partition_builtin_p): Support above new enum item.
398 (generate_code_for_partition): Ditto.
399 (compute_access_range): Differentiate cases that equality can be
400 proven at all loops, the innermost loops or no loops.
401 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
402 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
403 (finalize_partitions, distribute_loop): Don't fuse partition of
404 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
405 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
408 2018-01-23 Martin Liska <mliska@suse.cz>
410 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
411 order to ignore the predictor.
412 (PRED_POLYMORPHIC_CALL): Likewise.
413 (PRED_RECURSIVE_CALL): Likewise.
415 2018-01-23 Martin Liska <mliska@suse.cz>
417 * tree-profile.c (tree_profiling): Print function header to
418 aware reader which function we are working on.
419 * value-prof.c (gimple_find_values_to_profile): Do not print
420 not interesting value histograms.
422 2018-01-23 Martin Liska <mliska@suse.cz>
424 * profile-count.h (enum profile_quality): Add
425 profile_uninitialized as the first value. Do not number values
426 as they are zero based.
427 (profile_count::verify): Update sanity check.
428 (profile_probability::verify): Likewise.
430 2018-01-23 Nathan Sidwell <nathan@acm.org>
432 * doc/invoke.texi (ffor-scope): Deprecate.
434 2018-01-23 David Malcolm <dmalcolm@redhat.com>
436 PR tree-optimization/83510
437 * domwalk.c (set_all_edges_as_executable): New function.
438 (dom_walker::dom_walker): Convert bool param
439 "skip_unreachable_blocks" to enum reachability. Move setup of
440 edge flags to set_all_edges_as_executable and only do it when
441 reachability is REACHABLE_BLOCKS.
442 * domwalk.h (enum dom_walker::reachability): New enum.
443 (dom_walker::dom_walker): Convert bool param
444 "skip_unreachable_blocks" to enum reachability.
445 (set_all_edges_as_executable): New decl.
446 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
447 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
449 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
450 but converting true to REACHABLE_BLOCKS.
451 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
453 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
454 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
455 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
457 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
458 if check_all_array_refs will be called.
460 2018-01-23 David Malcolm <dmalcolm@redhat.com>
462 * tree.c (selftest::test_location_wrappers): Add more test
465 2018-01-23 David Malcolm <dmalcolm@redhat.com>
467 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
468 (selftest::test_bit_in_range): Likewise.
470 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
473 * doc/sourcebuild.texi (vect_float): Say that the selector
474 only describes the situation when -funsafe-math-optimizations is on.
475 (vect_float_strict): Document.
477 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
479 PR tree-optimization/83965
480 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
481 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
482 instead of checking only for a reduction.
483 (vect_recog_widen_sum_pattern): Likewise.
485 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
487 * predict.c (probably_never_executed): Only use precise profile info.
488 (compute_function_frequency): Skip after inlining hack since we now
489 have quality checking.
491 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
493 * profile-count.h (profile_probability::very_unlikely,
494 profile_probability::unlikely, profile_probability::even): Set
495 precision to guessed.
497 2018-01-23 Richard Biener <rguenther@suse.de>
499 PR tree-optimization/83963
500 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
501 Properly terminate dominator walk when crossing the exit edge not
502 when visiting its source block.
504 2018-01-23 Jakub Jelinek <jakub@redhat.com>
507 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
508 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
510 2018-01-22 Jakub Jelinek <jakub@redhat.com>
512 PR tree-optimization/83957
513 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
514 semicolon after for body surrounded by braces.
516 PR tree-optimization/83081
517 * profile-count.h (profile_probability::split): New method.
518 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
519 Use profile_probability::split.
520 (do_compare_rtx_and_jump): Fix adjustment of probabilities
521 when splitting a single conditional jump into 2.
523 2018-01-22 David Malcolm <dmalcolm@redhat.com>
525 PR tree-optimization/69452
526 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
529 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
531 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
532 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
533 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
535 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
537 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
538 * config/rl78/rl78.md: New define_expand "movdi"
539 * config/rl78/rl78.c: New function definition rl78_split_movdi
541 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
544 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
546 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
547 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
548 128-bit to produce an UNSPEC move to get the double word with the
549 signbit and then a shift directly to do signbit.
550 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
551 implementation with a new version that just does either a direct
552 move or a regular move. Move memory interface to separate insns.
553 Move insns so they are next to the expander.
554 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
555 with signbit move. Split big and little endian case.
556 (signbit<mode>2_dm_mem_le): Likewise.
557 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
558 (signbit<mode>2_dm2): Likewise.
560 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
562 * config/rl78/rl78.md: New define_expand "anddi3".
564 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
566 * config/rl78/rl78.md: New define_expand "umindi3".
568 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
570 * config/rl78/rl78.md: New define_expand "smindi3".
572 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
574 * config/rl78/rl78.md: New define_expand "smaxdi3".
576 2018-01-22 Carl Love <cel@us.ibm.com>
578 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
579 LVX_V1TI): Add macro expansion.
580 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
581 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
582 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
583 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
584 Change check to determine if the instruction is a byte reversing
585 entry. Fix typo in comment.
586 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
587 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
588 Add def_builtin calls for new builtins.
589 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
590 Add define_insn expansion.
592 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
594 * config/rl78/rl78.md: New define_expand "umaxdi3".
596 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
598 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
599 for non-QImode registers
601 2018-01-22 Richard Biener <rguenther@suse.de>
603 PR tree-optimization/83963
604 * graphite-scop-detection.c (scop_detection::get_sese): Delay
605 including the loop exit block.
606 (scop_detection::merge_sese): Likewise.
607 (scop_detection::add_scop): Do it here instead.
609 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
611 * doc/sourcebuild.texi (arm_softfloat): Document.
613 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
616 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
617 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
618 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
620 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
621 David Edelsohn <dje.gcc@gmail.com>
624 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
625 Change "crset eq" to "crset 2".
626 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
627 (*call_indirect_aix<mode>_nospec): Likewise.
628 (*call_value_indirect_aix<mode>_nospec): Likewise.
629 (*call_indirect_elfv2<mode>_nospec): Likewise.
630 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
631 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
632 change assembly output from . to $.
633 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
634 (indirect_jump<mode>_nospec): Change assembly output from . to $.
635 (*tablejump<mode>_internal1_nospec): Likewise.
637 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
640 * config/sh/sh_optimize_sett_clrt.cc:
641 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
643 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
645 PR tree-optimization/83940
646 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
647 offset_dt to vect_constant_def rather than vect_unknown_def_type.
648 (vect_check_load_store_mask): Add a mask_dt_out parameter and
649 use it to pass back the definition type.
650 (vect_check_store_rhs): Likewise rhs_dt_out.
651 (vect_build_gather_load_calls): Add a mask_dt argument and use
652 it instead of a call to vect_is_simple_use.
653 (vectorizable_store): Update calls to vect_check_load_store_mask
654 and vect_check_store_rhs. Use the dt returned by the latter instead
655 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
656 instead of calls to vect_is_simple_use. Pass the scalar rather
657 than the vector operand to vect_is_simple_use when handling
658 second and subsequent copies of an rhs value.
659 (vectorizable_load): Update calls to vect_check_load_store_mask
660 and vect_build_gather_load_calls. Use the cached mask_dt and
661 gs_info.offset_dt instead of calls to vect_is_simple_use.
663 2018-01-20 Jakub Jelinek <jakub@redhat.com>
666 * tree-emutls.c: Include gimplify.h.
667 (lower_emutls_2): New function.
668 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
669 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
670 it before further processing.
673 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
674 UINTVAL (trueop1) instead of INTVAL (op1).
676 2018-01-19 Jakub Jelinek <jakub@redhat.com>
680 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
681 INCOMING_FRAME_SP_OFFSET if not defined.
682 (scan_trace): Add ENTRY argument. If true and
683 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
684 emit a note to adjust the CFA offset.
685 (create_cfi_notes): Adjust scan_trace callers.
686 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
687 INCOMING_FRAME_SP_OFFSET in the CIE.
688 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
689 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
691 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
692 * doc/tm.texi: Regenerated.
694 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
696 PR rtl-optimization/83147
697 * lra-constraints.c (remove_inheritance_pseudos): Use
698 lra_substitute_pseudo_within_insn.
700 2018-01-19 Tom de Vries <tom@codesourcery.com>
701 Cesar Philippidis <cesar@codesourcery.com>
704 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
706 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
709 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
710 spaces for function labels.
712 2018-01-19 Martin Liska <mliska@suse.cz>
714 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
715 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
716 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
717 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
718 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
719 (PRED_CONST_RETURN): Change from 69 to 65.
720 (PRED_NULL_RETURN): Change from 91 to 71.
721 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
722 (PRED_LOOP_GUARD): Change from 66 to 73.
724 2018-01-19 Martin Liska <mliska@suse.cz>
726 * predict.c (predict_insn_def): Add new assert.
727 (struct branch_predictor): Change type to signed integer.
728 (test_prediction_value_range): Amend test to cover
730 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
731 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
732 (PRED_LOOP_ITERATIONS_MAX): Likewise.
733 (PRED_LOOP_IV_COMPARE): Likewise.
734 * predict.h (PROB_UNINITIALIZED): Define new constant.
736 2018-01-19 Martin Liska <mliska@suse.cz>
738 * predict.c (dump_prediction): Add new format for
739 analyze_brprob.py script which is enabled with -details
741 * profile-count.h (precise_p): New function.
743 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
745 PR tree-optimization/83922
746 * tree-vect-loop.c (vect_verify_full_masking): Return false if
747 there are no statements that need masking.
748 (vect_active_double_reduction_p): New function.
749 (vect_analyze_loop_operations): Use it when handling phis that
750 are not in the loop header.
752 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
754 PR tree-optimization/83914
755 * tree-vect-loop.c (vectorizable_induction): Don't convert
756 init_expr or apply the peeling adjustment for inductions
757 that are nested within the vectorized loop.
759 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
761 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
764 2018-01-18 Jakub Jelinek <jakub@redhat.com>
768 * function.h (gimplify_parameters): Add gimple_seq * argument.
769 * function.c: Include gimple.h and options.h.
770 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
771 for the added local temporaries if needed.
772 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
773 if there are any parameter cleanups, wrap whole body into a
774 try/finally with the cleanups.
776 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
779 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
780 Use GET_MODE_CLASS for scalar floating point.
782 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
786 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
787 Fix call of call_cgraph_insertion_hooks.
789 2018-01-18 Martin Sebor <msebor@redhat.com>
791 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
793 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
796 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
799 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
802 * common.opt: (-ffile-prefix-map): New option.
803 * opts.c (common_handle_option): Defer it.
804 * opts-global.c (handle_common_deferred_options): Handle it.
805 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
806 * file-prefix-map.h: New file.
807 (remap_debug_filename, add_debug_prefix_map): ...here.
808 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
809 * final.c (debug_prefix_map, add_debug_prefix_map
810 remap_debug_filename): Move to...
811 * file-prefix-map.c: New file.
812 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
813 generalize, get rid of alloca(), use strrchr() instead of strchr().
814 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
815 Implement in terms of add_prefix_map().
816 (remap_macro_filename, remap_debug_filename): Implement in term of
818 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
819 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
820 * dbxout.c: Include file-prefix-map.h.
821 * varasm.c: Likewise.
822 * vmsdbgout.c: Likewise.
823 * xcoffout.c: Likewise.
824 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
825 * doc/cppopts.texi (-fmacro-prefix-map): Document.
826 * doc/invoke.texi (-ffile-prefix-map): Document.
827 (-fdebug-prefix-map): Update description.
829 2018-01-18 Martin Liska <mliska@suse.cz>
831 * config/i386/i386.c (indirect_thunk_name): Document that also
833 (output_indirect_thunk): Document why both instructions
834 (pause and lfence) are generated.
836 2018-01-18 Richard Biener <rguenther@suse.de>
838 PR tree-optimization/83887
839 * graphite-scop-detection.c
840 (scop_detection::get_nearest_dom_with_single_entry): Remove.
841 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
842 (scop_detection::merge_sese): Re-implement with a flood-fill
843 algorithm that properly finds a SESE region if it exists.
845 2018-01-18 Jakub Jelinek <jakub@redhat.com>
848 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
849 pointer_diff optimizations use view_convert instead of convert.
851 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
853 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
854 Generate different code for -mno-speculate-indirect-jumps.
855 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
856 (*call_indirect_aix<mode>): Disable for
857 -mno-speculate-indirect-jumps.
858 (*call_indirect_aix<mode>_nospec): New define_insn.
859 (*call_value_indirect_aix<mode>): Disable for
860 -mno-speculate-indirect-jumps.
861 (*call_value_indirect_aix<mode>_nospec): New define_insn.
862 (*sibcall_nonlocal_sysv<mode>): Generate different code for
863 -mno-speculate-indirect-jumps.
864 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
866 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
868 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
869 long double type, set the flags for noting the default long double
870 type, even if we don't pass or return a long double type.
872 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
875 * ipa-inline.c (flatten_function): Do not overwrite final inlining
878 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
880 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
881 support for merge[hl].
882 (fold_mergehl_helper): New helper function.
883 (tree-vector-builder.h): New #include for tree_vector_builder usage.
884 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
885 (altivec_vmrglw_direct): Add xxmrglw insn.
887 2018-01-17 Andrew Waterman <andrew@sifive.com>
889 * config/riscv/riscv.c (riscv_conditional_register_usage): If
890 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
892 2018-01-17 David Malcolm <dmalcolm@redhat.com>
895 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
896 call the lto_location_cache before reading the
897 DECL_SOURCE_LOCATION of the types.
899 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
900 Richard Sandiford <richard.sandiford@linaro.org>
902 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
903 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
904 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
905 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
906 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
908 * config/aarch64/constraints.md (aarch64_movti_operand):
910 * config/aarch64/predicates.md (Uti): Add new constraint.
912 2018-01-17 Carl Love <cel@us.ibm.com>
913 * config/rs6000/vsx.md (define_expand xl_len_r,
914 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
915 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
917 (define_expand, define_insn): Move the shift left from the
918 define_insn to the define_expand for lxvl and stxvl instructions.
919 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
920 and XL_LEN_R definitions to PURE.
922 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
924 * config/i386/i386.c (indirect_thunk_name): Declare regno
925 as unsigned int. Compare regno with INVALID_REGNUM.
926 (output_indirect_thunk): Ditto.
927 (output_indirect_thunk_function): Ditto.
928 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
929 in the call to output_indirect_thunk_function.
931 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
934 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
935 rather than the size of inner_type to determine the stack slot size
936 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
938 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
941 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
944 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
946 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
947 endian Linux systems to optionally enable multilibs for selecting
948 the long double type if the user configured an explicit type.
949 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
950 have no long double multilibs if not defined.
951 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
952 warn if the user used -mabi={ieee,ibm}longdouble and we built
953 multilibs for long double.
954 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
955 appropriate multilib option.
956 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
958 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
959 for building long double multilibs.
960 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
962 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
964 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
967 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
969 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
972 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
975 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
978 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
980 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
981 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
983 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
985 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
986 different rtl trees depending on TARGET_64BIT.
987 (rs6000_gen_lvx): Likewise.
989 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
991 * config/visium/visium.md (nop): Tweak comment.
992 (hazard_nop): Likewise.
994 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
996 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
997 -mspeculate-indirect-jumps.
998 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
999 for -mno-speculate-indirect-jumps.
1000 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1001 (*call_value_indirect_elfv2<mode>): Disable for
1002 -mno-speculate-indirect-jumps.
1003 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1004 (indirect_jump): Emit different RTL for
1005 -mno-speculate-indirect-jumps.
1006 (*indirect_jump<mode>): Disable for
1007 -mno-speculate-indirect-jumps.
1008 (*indirect_jump<mode>_nospec): New define_insn.
1009 (tablejump): Emit different RTL for
1010 -mno-speculate-indirect-jumps.
1011 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1012 (tablejumpsi_nospec): New define_expand.
1013 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1014 (tablejumpdi_nospec): New define_expand.
1015 (*tablejump<mode>_internal1): Disable for
1016 -mno-speculate-indirect-jumps.
1017 (*tablejump<mode>_internal1_nospec): New define_insn.
1018 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1021 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1023 * caller-save.c (insert_save): Drop unnecessary parameter. All
1026 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1027 Richard Biener <rguenth@suse.de>
1030 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1031 return early, inline manually is_gimple_sizepos. Make sure if we
1032 call gimplify_expr we don't end up with a gimple constant.
1033 * tree.c (variably_modified_type_p): Don't return true for
1034 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1035 * gimplify.h (is_gimple_sizepos): Remove.
1037 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1039 PR tree-optimization/83857
1040 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1041 vectorizable_live_operation for pure SLP statements.
1042 (vectorizable_live_operation): Handle PHIs.
1044 2018-01-16 Richard Biener <rguenther@suse.de>
1046 PR tree-optimization/83867
1047 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1048 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1050 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1053 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1054 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1055 If off is not INTEGER_CST, issue a may not be aligned warning
1056 rather than isn't aligned. Use isn%'t rather than isn't.
1057 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1059 <case MULT_EXPR>: Improve the case when bottom and one of the
1060 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1061 operand, in that case check if the other operand is multiple of
1062 bottom divided by the INTEGER_CST operand.
1064 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1067 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1068 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1069 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1070 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1071 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1072 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1073 (pa_function_arg_size): New function.
1075 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1077 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1078 in a separate statement.
1080 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1082 PR tree-optimization/83847
1083 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1084 group gathers and scatters.
1086 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1088 PR rtl-optimization/86620
1089 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1091 PR rtl-optimization/83213
1092 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1093 to last if both are JUMP_INSNs.
1095 PR tree-optimization/83843
1096 * gimple-ssa-store-merging.c
1097 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1098 store_immediate_info for bswap/nop orig_stores.
1100 2018-01-15 Andrew Waterman <andrew@sifive.com>
1102 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1104 <UDIV>: Increase cost if !TARGET_DIV.
1106 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1108 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1109 (define_attr "cr_logical_3op"): New.
1110 (cceq_ior_compare): Adjust.
1111 (cceq_ior_compare_complement): Adjust.
1112 (*cceq_rev_compare): Adjust.
1113 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1114 (is_cracked_insn): Adjust.
1115 (insn_must_be_first_in_group): Adjust.
1116 * config/rs6000/40x.md: Adjust.
1117 * config/rs6000/440.md: Adjust.
1118 * config/rs6000/476.md: Adjust.
1119 * config/rs6000/601.md: Adjust.
1120 * config/rs6000/603.md: Adjust.
1121 * config/rs6000/6xx.md: Adjust.
1122 * config/rs6000/7450.md: Adjust.
1123 * config/rs6000/7xx.md: Adjust.
1124 * config/rs6000/8540.md: Adjust.
1125 * config/rs6000/cell.md: Adjust.
1126 * config/rs6000/e300c2c3.md: Adjust.
1127 * config/rs6000/e500mc.md: Adjust.
1128 * config/rs6000/e500mc64.md: Adjust.
1129 * config/rs6000/e5500.md: Adjust.
1130 * config/rs6000/e6500.md: Adjust.
1131 * config/rs6000/mpc.md: Adjust.
1132 * config/rs6000/power4.md: Adjust.
1133 * config/rs6000/power5.md: Adjust.
1134 * config/rs6000/power6.md: Adjust.
1135 * config/rs6000/power7.md: Adjust.
1136 * config/rs6000/power8.md: Adjust.
1137 * config/rs6000/power9.md: Adjust.
1138 * config/rs6000/rs64.md: Adjust.
1139 * config/rs6000/titan.md: Adjust.
1141 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1143 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1144 ix86_indirect_branch_register logic.
1146 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1148 * config/i386/constraints.md (Bs): Update
1149 ix86_indirect_branch_register check. Don't check
1150 ix86_indirect_branch_register with GOT_memory_operand.
1152 * config/i386/predicates.md (GOT_memory_operand): Don't check
1153 ix86_indirect_branch_register here.
1154 (GOT32_symbol_operand): Likewise.
1156 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1158 * config/i386/predicates.md (constant_call_address_operand):
1159 Rewrite ix86_indirect_branch_register logic.
1160 (sibcall_insn_operand): Likewise.
1162 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1164 * config/i386/constraints.md (Bs): Replace
1165 ix86_indirect_branch_thunk_register with
1166 ix86_indirect_branch_register.
1168 * config/i386/i386.md (indirect_jump): Likewise.
1169 (tablejump): Likewise.
1170 (*sibcall_memory): Likewise.
1171 (*sibcall_value_memory): Likewise.
1172 Peepholes of indirect call and jump via memory: Likewise.
1173 * config/i386/i386.opt: Likewise.
1174 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1175 (GOT_memory_operand): Likewise.
1176 (call_insn_operand): Likewise.
1177 (sibcall_insn_operand): Likewise.
1178 (GOT32_symbol_operand): Likewise.
1180 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1183 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1184 type rather than type addr's type points to.
1185 (expand_omp_atomic_mutex): Likewise.
1186 (expand_omp_atomic): Likewise.
1188 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1191 * config/i386/i386.c (output_indirect_thunk_function): Use
1192 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1193 for __x86_return_thunk.
1195 2018-01-15 Richard Biener <rguenther@suse.de>
1198 * expmed.c (extract_bit_field_1): Fix typo.
1200 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1203 * config/arm/iterators.md (VF): New mode iterator.
1204 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1205 Remove integer-related logic from pattern.
1206 (neon_vabd<mode>_3): Likewise.
1208 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1211 * common.opt (fstrict-overflow): No longer an alias.
1212 (fwrapv-pointer): New option.
1213 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1214 also for pointer types based on flag_wrapv_pointer.
1215 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1216 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1217 opts->x_flag_wrapv got set.
1218 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1219 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1220 POINTER_TYPE_OVERFLOW_UNDEFINED.
1221 * match.pd: Likewise in address comparison pattern.
1222 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1224 2018-01-15 Richard Biener <rguenther@suse.de>
1227 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1228 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1229 Reset type names to their identifier if their TYPE_DECL doesn't
1230 have linkage (and thus is used for ODR and devirt).
1231 (save_debug_info_for_decl): Remove.
1232 (save_debug_info_for_type): Likewise.
1233 (add_tree_to_fld_list): Adjust.
1234 * tree-pretty-print.c (dump_generic_node): Make dumping of
1235 type names more robust.
1237 2018-01-15 Richard Biener <rguenther@suse.de>
1239 * BASE-VER: Bump to 8.0.1.
1241 2018-01-14 Martin Sebor <msebor@redhat.com>
1244 * builtins.c (check_access): Avoid warning when the no-warning bit
1247 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1249 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1250 * ira-color (allocno_hard_regs_compare): Likewise.
1252 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1255 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1256 Use .pushsection/.popsection.
1258 2018-01-14 Martin Sebor <msebor@redhat.com>
1261 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1263 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1265 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1266 entry from extra_headers.
1267 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1268 extra_headers, make the list bitwise identical to the i?86-*-* one.
1270 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1272 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1273 -mcmodel=large with -mindirect-branch=thunk,
1274 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1275 -mfunction-return=thunk-extern.
1276 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1277 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1278 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1280 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1282 * config/i386/i386.c (print_reg): Print the name of the full
1283 integer register without '%'.
1284 (ix86_print_operand): Handle 'V'.
1285 * doc/extend.texi: Document 'V' modifier.
1287 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1289 * config/i386/constraints.md (Bs): Disallow memory operand for
1290 -mindirect-branch-register.
1292 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1293 (GOT_memory_operand): Likewise.
1294 (call_insn_operand): Likewise.
1295 (sibcall_insn_operand): Likewise.
1296 (GOT32_symbol_operand): Likewise.
1297 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1298 for -mindirect-branch-register.
1299 (tablejump): Likewise.
1300 (*sibcall_memory): Likewise.
1301 (*sibcall_value_memory): Likewise.
1302 Disallow peepholes of indirect call and jump via memory for
1303 -mindirect-branch-register.
1304 (*call_pop): Replace m with Bw.
1305 (*call_value_pop): Likewise.
1306 (*sibcall_pop_memory): Replace m with Bs.
1307 * config/i386/i386.opt (mindirect-branch-register): New option.
1308 * doc/invoke.texi: Document -mindirect-branch-register option.
1310 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1312 * config/i386/i386-protos.h (ix86_output_function_return): New.
1313 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1314 set function_return_type.
1315 (indirect_thunk_name): Add ret_p to indicate thunk for function
1317 (output_indirect_thunk_function): Pass false to
1318 indirect_thunk_name.
1319 (ix86_output_indirect_branch_via_reg): Likewise.
1320 (ix86_output_indirect_branch_via_push): Likewise.
1321 (output_indirect_thunk_function): Create alias for function
1322 return thunk if regno < 0.
1323 (ix86_output_function_return): New function.
1324 (ix86_handle_fndecl_attribute): Handle function_return.
1325 (ix86_attribute_table): Add function_return.
1326 * config/i386/i386.h (machine_function): Add
1327 function_return_type.
1328 * config/i386/i386.md (simple_return_internal): Use
1329 ix86_output_function_return.
1330 (simple_return_internal_long): Likewise.
1331 * config/i386/i386.opt (mfunction-return=): New option.
1332 (indirect_branch): Mention -mfunction-return=.
1333 * doc/extend.texi: Document function_return function attribute.
1334 * doc/invoke.texi: Document -mfunction-return= option.
1336 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1338 * config/i386/i386-opts.h (indirect_branch): New.
1339 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1340 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1341 with local indirect jump when converting indirect call and jump.
1342 (ix86_set_indirect_branch_type): New.
1343 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1344 (indirectlabelno): New.
1345 (indirect_thunk_needed): Likewise.
1346 (indirect_thunk_bnd_needed): Likewise.
1347 (indirect_thunks_used): Likewise.
1348 (indirect_thunks_bnd_used): Likewise.
1349 (INDIRECT_LABEL): Likewise.
1350 (indirect_thunk_name): Likewise.
1351 (output_indirect_thunk): Likewise.
1352 (output_indirect_thunk_function): Likewise.
1353 (ix86_output_indirect_branch_via_reg): Likewise.
1354 (ix86_output_indirect_branch_via_push): Likewise.
1355 (ix86_output_indirect_branch): Likewise.
1356 (ix86_output_indirect_jmp): Likewise.
1357 (ix86_code_end): Call output_indirect_thunk_function if needed.
1358 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1360 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1361 (ix86_attribute_table): Add indirect_branch.
1362 * config/i386/i386.h (machine_function): Add indirect_branch_type
1363 and has_local_indirect_jump.
1364 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1366 (tablejump): Likewise.
1367 (*indirect_jump): Use ix86_output_indirect_jmp.
1368 (*tablejump_1): Likewise.
1369 (simple_return_indirect_internal): Likewise.
1370 * config/i386/i386.opt (mindirect-branch=): New option.
1371 (indirect_branch): New.
1374 (thunk-inline): Likewise.
1375 (thunk-extern): Likewise.
1376 * doc/extend.texi: Document indirect_branch function attribute.
1377 * doc/invoke.texi: Document -mindirect-branch= option.
1379 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1382 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1384 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1386 * ipa-inline.c (want_inline_small_function_p): Return false if
1387 inlining has already failed with CIF_FINAL_ERROR.
1388 (update_caller_keys): Call want_inline_small_function_p before
1390 (update_callee_keys): Likewise.
1392 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1394 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1396 (rs6000_quadword_masked_address_p): Likewise.
1397 (quad_aligned_load_p): Likewise.
1398 (quad_aligned_store_p): Likewise.
1399 (const_load_sequence_p): Add comment to describe the outer-most loop.
1400 (mimic_memory_attributes_and_flags): New function.
1401 (rs6000_gen_stvx): Likewise.
1402 (replace_swapped_aligned_store): Likewise.
1403 (rs6000_gen_lvx): Likewise.
1404 (replace_swapped_aligned_load): Likewise.
1405 (replace_swapped_load_constant): Capitalize argument name in
1406 comment describing this function.
1407 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1408 and stores that access quad-word aligned addresses and replace
1409 with stvx or lvx instructions when appropriate.
1410 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1411 New function prototype.
1412 (rs6000_quadword_masked_address_p): Likewise.
1413 (rs6000_gen_lvx): Likewise.
1414 (rs6000_gen_stvx): Likewise.
1415 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1416 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1417 when memory address is aligned.
1418 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1419 this split to select lvx instruction when memory address is aligned.
1420 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1421 instruction when memory address is aligned.
1422 (*vsx_le_perm_load_v16qi): Likewise.
1423 (four unnamed splitters): Modify to select the stvx instruction
1424 when memory is aligned.
1426 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1428 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1429 which appears in the queue multiple times.
1431 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1432 Alan Hayward <alan.hayward@arm.com>
1433 David Sherwood <david.sherwood@arm.com>
1435 * tree-vectorizer.h (vec_lower_bound): New structure.
1436 (_loop_vec_info): Add check_nonzero and lower_bounds.
1437 (LOOP_VINFO_CHECK_NONZERO): New macro.
1438 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1439 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1440 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1441 fields. Make seg_len the distance travelled, not including the
1443 (dr_direction_indicator): Declare.
1444 (dr_zero_step_indicator): Likewise.
1445 (dr_known_forward_stride_p): Likewise.
1446 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1448 (runtime_alias_check_p): Allow runtime alias checks with
1450 (operator ==): Compare access_size and align.
1451 (prune_runtime_alias_test_list): Rework for new distinction between
1452 the access_size and seg_len.
1453 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1455 (get_segment_min_max): New function.
1456 (create_intersect_range_checks): Use it.
1457 (dr_step_indicator): New function.
1458 (dr_direction_indicator): Likewise.
1459 (dr_zero_step_indicator): Likewise.
1460 (dr_known_forward_stride_p): Likewise.
1461 * tree-loop-distribution.c (data_ref_segment_size): Return
1462 DR_STEP * (niters - 1).
1463 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1465 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1466 (vect_preserves_scalar_order_p): New function, split out from...
1467 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1468 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1469 (vect_vfa_access_size): New function.
1470 (vect_vfa_align): Likewise.
1471 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1472 (dump_lower_bound): New function.
1473 (vect_check_lower_bound): Likewise.
1474 (vect_small_gap_p): Likewise.
1475 (vectorizable_with_step_bound_p): Likewise.
1476 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1477 depencies if the vectorization factor is 1. Convert the checks
1478 for nonzero steps into checks on the bounds of DR_STEP. Try using
1479 a bunds check for variable steps if the minimum required step is
1480 relatively small. Update calls to the dr_with_seg_len
1481 constructor and to vect_compile_time_alias.
1482 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1484 (vect_loop_versioning): Call it.
1485 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1487 (vect_estimate_min_profitable_iters): Account for any bounds checks.
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 * doc/sourcebuild.texi (vect_scatter_store): Document.
1494 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1496 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1498 * genopinit.c (main): Add supports_vec_scatter_store and
1499 supports_vec_scatter_store_cached to target_optabs.
1500 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1501 IFN_MASK_SCATTER_STORE.
1502 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1504 * internal-fn.h (internal_store_fn_p): Declare.
1505 (internal_fn_stored_value_index): Likewise.
1506 * internal-fn.c (scatter_store_direct): New macro.
1507 (expand_scatter_store_optab_fn): New function.
1508 (direct_scatter_store_optab_supported_p): New macro.
1509 (internal_store_fn_p): New function.
1510 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1511 IFN_MASK_SCATTER_STORE.
1512 (internal_fn_mask_index): Likewise.
1513 (internal_fn_stored_value_index): New function.
1514 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1516 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1517 * optabs-query.c (supports_vec_scatter_store_p): New function.
1518 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1519 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1520 true for scatter stores.
1521 (vect_gather_scatter_fn_p): Handle scatter stores too.
1522 (vect_check_gather_scatter): Consider using scatter stores if
1523 supports_vec_scatter_store_p.
1524 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1526 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1527 internal_fn_stored_value_index.
1528 (check_load_store_masking): Handle scatter stores too.
1529 (vect_get_store_rhs): Make public.
1530 (vectorizable_call): Use internal_store_fn_p.
1531 (vectorizable_store): Handle scatter store internal functions.
1532 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1533 when deciding whether the end of the group has been reached.
1534 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1535 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1536 (mask_scatter_store<mode>): New insns.
1538 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1539 Alan Hayward <alan.hayward@arm.com>
1540 David Sherwood <david.sherwood@arm.com>
1542 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1543 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1544 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1546 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1547 Use vect_truncate_gather_scatter_offset if we can't treat the
1548 operation as a normal gather load or scatter store.
1549 (get_group_load_store_type): Take the gather_scatter_info
1550 as argument. Try using a gather load or scatter store for
1551 single-element groups.
1552 (get_load_store_type): Update calls to get_group_load_store_type
1553 and vect_use_strided_gather_scatters_p.
1555 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1556 Alan Hayward <alan.hayward@arm.com>
1557 David Sherwood <david.sherwood@arm.com>
1559 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1560 optional tree argument.
1561 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1563 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1564 but continue to use the current value as a fallback.
1565 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1566 to compare the updates.
1567 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1568 (get_load_store_type): Use it when handling a strided access.
1569 (vect_get_strided_load_store_ops): New function.
1570 (vect_get_data_ptr_increment): Likewise.
1571 (vectorizable_load): Handle strided gather loads. Always pass
1572 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1574 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1575 Alan Hayward <alan.hayward@arm.com>
1576 David Sherwood <david.sherwood@arm.com>
1578 * doc/md.texi (gather_load@var{m}): Document.
1579 (mask_gather_load@var{m}): Likewise.
1580 * genopinit.c (main): Add supports_vec_gather_load and
1581 supports_vec_gather_load_cached to target_optabs.
1582 * optabs-tree.c (init_tree_optimization_optabs): Use
1583 ggc_cleared_alloc to allocate target_optabs.
1584 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1585 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1587 * internal-fn.h (internal_load_fn_p): Declare.
1588 (internal_gather_scatter_fn_p): Likewise.
1589 (internal_fn_mask_index): Likewise.
1590 (internal_gather_scatter_fn_supported_p): Likewise.
1591 * internal-fn.c (gather_load_direct): New macro.
1592 (expand_gather_load_optab_fn): New function.
1593 (direct_gather_load_optab_supported_p): New macro.
1594 (direct_internal_fn_optab): New function.
1595 (internal_load_fn_p): Likewise.
1596 (internal_gather_scatter_fn_p): Likewise.
1597 (internal_fn_mask_index): Likewise.
1598 (internal_gather_scatter_fn_supported_p): Likewise.
1599 * optabs-query.c (supports_at_least_one_mode_p): New function.
1600 (supports_vec_gather_load_p): Likewise.
1601 * optabs-query.h (supports_vec_gather_load_p): Declare.
1602 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1603 and memory_type field.
1604 (NUM_PATTERNS): Bump to 15.
1605 * tree-vect-data-refs.c: Include internal-fn.h.
1606 (vect_gather_scatter_fn_p): New function.
1607 (vect_describe_gather_scatter_call): Likewise.
1608 (vect_check_gather_scatter): Try using internal functions for
1609 gather loads. Recognize existing calls to a gather load function.
1610 (vect_analyze_data_refs): Consider using gather loads if
1611 supports_vec_gather_load_p.
1612 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1613 (vect_get_gather_scatter_offset_type): Likewise.
1614 (vect_convert_mask_for_vectype): Likewise.
1615 (vect_add_conversion_to_patterm): Likewise.
1616 (vect_try_gather_scatter_pattern): Likewise.
1617 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1618 (vect_vect_recog_func_ptrs): Add it.
1619 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1620 internal_fn_mask_index and internal_gather_scatter_fn_p.
1621 (check_load_store_masking): Take the gather_scatter_info as an
1622 argument and handle gather loads.
1623 (vect_get_gather_scatter_ops): New function.
1624 (vectorizable_call): Check internal_load_fn_p.
1625 (vectorizable_load): Likewise. Handle gather load internal
1627 (vectorizable_store): Update call to check_load_store_masking.
1628 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1629 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1630 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1631 (aarch64_gather_scale_operand_d): New predicates.
1632 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1633 (mask_gather_load<mode>): New insns.
1635 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1636 Alan Hayward <alan.hayward@arm.com>
1637 David Sherwood <david.sherwood@arm.com>
1639 * optabs.def (fold_left_plus_optab): New optab.
1640 * doc/md.texi (fold_left_plus_@var{m}): Document.
1641 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1642 * internal-fn.c (fold_left_direct): Define.
1643 (expand_fold_left_optab_fn): Likewise.
1644 (direct_fold_left_optab_supported_p): Likewise.
1645 * fold-const-call.c (fold_const_fold_left): New function.
1646 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1647 * tree-parloops.c (valid_reduction_p): New function.
1648 (gather_scalar_reductions): Use it.
1649 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1650 (vect_finish_replace_stmt): Declare.
1651 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1652 (needs_fold_left_reduction_p): New function, split out from...
1653 (vect_is_simple_reduction): ...here. Accept reductions that
1654 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1655 (vect_force_simple_reduction): Also store the reduction type in
1656 the assignment's STMT_VINFO_REDUC_TYPE.
1657 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1658 (merge_with_identity): New function.
1659 (vect_expand_fold_left): Likewise.
1660 (vectorize_fold_left_reduction): Likewise.
1661 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1662 scalar phi in place for it. Check for target support and reject
1663 cases that would reassociate the operation. Defer the transform
1664 phase to vectorize_fold_left_reduction.
1665 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1666 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1667 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1669 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1671 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1672 call to ifc_temp_var.
1674 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1675 Alan Hayward <alan.hayward@arm.com>
1676 David Sherwood <david.sherwood@arm.com>
1678 * target.def (legitimize_address_displacement): Take the original
1679 offset as a poly_int.
1680 * targhooks.h (default_legitimize_address_displacement): Update
1682 * targhooks.c (default_legitimize_address_displacement): Likewise.
1683 * doc/tm.texi: Regenerate.
1684 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1685 as an argument, moving assert of ad->disp == ad->disp_term to...
1686 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1687 Try calling targetm.legitimize_address_displacement before expanding
1688 the address rather than afterwards, and adjust for the new interface.
1689 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1690 Match the new hook interface. Handle SVE addresses.
1691 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1694 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1696 * Makefile.in (OBJS): Add early-remat.o.
1697 * target.def (select_early_remat_modes): New hook.
1698 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1699 * doc/tm.texi: Regenerate.
1700 * targhooks.h (default_select_early_remat_modes): Declare.
1701 * targhooks.c (default_select_early_remat_modes): New function.
1702 * timevar.def (TV_EARLY_REMAT): New timevar.
1703 * passes.def (pass_early_remat): New pass.
1704 * tree-pass.h (make_pass_early_remat): Declare.
1705 * early-remat.c: New file.
1706 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1708 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1710 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1711 Alan Hayward <alan.hayward@arm.com>
1712 David Sherwood <david.sherwood@arm.com>
1714 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1715 vfm1 with a bound_epilog parameter.
1716 (vect_do_peeling): Update calls accordingly, and move the prologue
1717 call earlier in the function. Treat the base bound_epilog as 0 for
1718 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1719 this base when peeling for gaps.
1720 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1721 with fully-masked loops.
1722 (vect_estimate_min_profitable_iters): Handle the single peeled
1723 iteration in that case.
1725 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1726 Alan Hayward <alan.hayward@arm.com>
1727 David Sherwood <david.sherwood@arm.com>
1729 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1730 single-element interleaving even if the size is not a power of 2.
1731 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1732 accesses for single-element interleaving if the group size is
1735 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1736 Alan Hayward <alan.hayward@arm.com>
1737 David Sherwood <david.sherwood@arm.com>
1739 * doc/md.texi (fold_extract_last_@var{m}): Document.
1740 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1741 * optabs.def (fold_extract_last_optab): New optab.
1742 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1743 * internal-fn.c (fold_extract_direct): New macro.
1744 (expand_fold_extract_optab_fn): Likewise.
1745 (direct_fold_extract_optab_supported_p): Likewise.
1746 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1747 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1748 EXTRACT_LAST_REDUCTION.
1749 (get_initial_def_for_reduction): Do not create an initial vector
1750 for EXTRACT_LAST_REDUCTION reductions.
1751 (vectorizable_reduction): Leave the scalar phi in place for
1752 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1753 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1754 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1755 transform phase to vectorizable_condition.
1756 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1758 (vect_finish_stmt_generation): ...here.
1759 (vect_finish_replace_stmt): New function.
1760 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1761 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1763 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1765 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1766 Alan Hayward <alan.hayward@arm.com>
1767 David Sherwood <david.sherwood@arm.com>
1769 * doc/md.texi (extract_last_@var{m}): Document.
1770 * optabs.def (extract_last_optab): New optab.
1771 * internal-fn.def (EXTRACT_LAST): New internal function.
1772 * internal-fn.c (cond_unary_direct): New macro.
1773 (expand_cond_unary_optab_fn): Likewise.
1774 (direct_cond_unary_optab_supported_p): Likewise.
1775 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1776 loops using EXTRACT_LAST.
1777 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1778 (extract_last_<mode>): ...this optab.
1779 (vec_extract<mode><Vel>): Update accordingly.
1781 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1782 Alan Hayward <alan.hayward@arm.com>
1783 David Sherwood <david.sherwood@arm.com>
1785 * target.def (empty_mask_is_expensive): New hook.
1786 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
1787 * doc/tm.texi: Regenerate.
1788 * targhooks.h (default_empty_mask_is_expensive): Declare.
1789 * targhooks.c (default_empty_mask_is_expensive): New function.
1790 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
1791 if the target says that empty masks are expensive.
1792 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
1794 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
1796 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1797 Alan Hayward <alan.hayward@arm.com>
1798 David Sherwood <david.sherwood@arm.com>
1800 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
1801 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
1802 (vect_use_loop_mask_for_alignment_p): New function.
1803 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
1804 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
1805 niters_skip argument. Make sure that the first niters_skip elements
1806 of the first iteration are inactive.
1807 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
1808 Update call to vect_set_loop_masks_directly.
1809 (get_misalign_in_elems): New function, split out from...
1810 (vect_gen_prolog_loop_niters): ...here.
1811 (vect_update_init_of_dr): Take a code argument that specifies whether
1812 the adjustment should be added or subtracted.
1813 (vect_update_init_of_drs): Likewise.
1814 (vect_prepare_for_masked_peels): New function.
1815 (vect_do_peeling): Skip prologue peeling if we're using a mask
1816 instead. Update call to vect_update_inits_of_drs.
1817 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1819 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
1820 alignment. Do not include the number of peeled iterations in
1821 the minimum threshold in that case.
1822 (vectorizable_induction): Adjust the start value down by
1823 LOOP_VINFO_MASK_SKIP_NITERS iterations.
1824 (vect_transform_loop): Call vect_prepare_for_masked_peels.
1825 Take the number of skipped iterations into account when calculating
1827 * tree-vect-stmts.c (vect_gen_while_not): New function.
1829 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1830 Alan Hayward <alan.hayward@arm.com>
1831 David Sherwood <david.sherwood@arm.com>
1833 * doc/sourcebuild.texi (vect_fully_masked): Document.
1834 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
1836 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
1838 (vect_analyze_loop_2): ...here. Don't check the vectorization
1839 factor against the number of loop iterations if the loop is
1842 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1843 Alan Hayward <alan.hayward@arm.com>
1844 David Sherwood <david.sherwood@arm.com>
1846 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
1847 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
1848 (dump_groups): Update accordingly.
1849 (iv_use::mem_type): New member variable.
1850 (address_p): New function.
1851 (record_use): Add a mem_type argument and initialize the new
1853 (record_group_use): Add a mem_type argument. Use address_p.
1854 Remove obsolete null checks of base_object. Update call to record_use.
1855 (find_interesting_uses_op): Update call to record_group_use.
1856 (find_interesting_uses_cond): Likewise.
1857 (find_interesting_uses_address): Likewise.
1858 (get_mem_type_for_internal_fn): New function.
1859 (find_address_like_use): Likewise.
1860 (find_interesting_uses_stmt): Try find_address_like_use before
1861 calling find_interesting_uses_op.
1862 (addr_offset_valid_p): Use the iv mem_type field as the type
1863 of the addressed memory.
1864 (add_autoinc_candidates): Likewise.
1865 (get_address_cost): Likewise.
1866 (split_small_address_groups_p): Use address_p.
1867 (split_address_groups): Likewise.
1868 (add_iv_candidate_for_use): Likewise.
1869 (autoinc_possible_for_pair): Likewise.
1870 (rewrite_groups): Likewise.
1871 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
1872 (determine_group_iv_cost): Update after split of USE_ADDRESS.
1873 (get_alias_ptr_type_for_ptr_address): New function.
1874 (rewrite_use_address): Rewrite address uses in calls that were
1875 identified by find_address_like_use.
1877 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1878 Alan Hayward <alan.hayward@arm.com>
1879 David Sherwood <david.sherwood@arm.com>
1881 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
1883 * gimple-expr.h (is_gimple_addressable: Likewise.
1884 * gimple-expr.c (is_gimple_address): Likewise.
1885 * internal-fn.c (expand_call_mem_ref): New function.
1886 (expand_mask_load_optab_fn): Use it.
1887 (expand_mask_store_optab_fn): Likewise.
1889 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1890 Alan Hayward <alan.hayward@arm.com>
1891 David Sherwood <david.sherwood@arm.com>
1893 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
1894 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
1895 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
1896 (cond_umax@var{mode}): Document.
1897 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
1898 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
1899 (cond_umin_optab, cond_umax_optab): New optabs.
1900 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
1901 (COND_IOR, COND_XOR): New internal functions.
1902 * internal-fn.h (get_conditional_internal_fn): Declare.
1903 * internal-fn.c (cond_binary_direct): New macro.
1904 (expand_cond_binary_optab_fn): Likewise.
1905 (direct_cond_binary_optab_supported_p): Likewise.
1906 (get_conditional_internal_fn): New function.
1907 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
1908 Cope with reduction statements that are vectorized as calls rather
1910 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
1911 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
1912 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
1913 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
1914 (UNSPEC_COND_EOR): New unspecs.
1915 (optab): Add mappings for them.
1916 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
1917 (sve_int_op, sve_fp_op): New int attributes.
1919 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1920 Alan Hayward <alan.hayward@arm.com>
1921 David Sherwood <david.sherwood@arm.com>
1923 * optabs.def (while_ult_optab): New optab.
1924 * doc/md.texi (while_ult@var{m}@var{n}): Document.
1925 * internal-fn.def (WHILE_ULT): New internal function.
1926 * internal-fn.h (direct_internal_fn_supported_p): New override
1927 that takes two types as argument.
1928 * internal-fn.c (while_direct): New macro.
1929 (expand_while_optab_fn): New function.
1930 (convert_optab_supported_p): Likewise.
1931 (direct_while_optab_supported_p): New macro.
1932 * wide-int.h (wi::udiv_ceil): New function.
1933 * tree-vectorizer.h (rgroup_masks): New structure.
1934 (vec_loop_masks): New typedef.
1935 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
1937 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
1938 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
1939 (vect_max_vf): New function.
1940 (slpeel_make_loop_iterate_ntimes): Delete.
1941 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
1942 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
1943 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
1944 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
1945 internal-fn.h, stor-layout.h and optabs-query.h.
1946 (vect_set_loop_mask): New function.
1947 (add_preheader_seq): Likewise.
1948 (add_header_seq): Likewise.
1949 (interleave_supported_p): Likewise.
1950 (vect_maybe_permute_loop_masks): Likewise.
1951 (vect_set_loop_masks_directly): Likewise.
1952 (vect_set_loop_condition_masked): Likewise.
1953 (vect_set_loop_condition_unmasked): New function, split out from
1954 slpeel_make_loop_iterate_ntimes.
1955 (slpeel_make_loop_iterate_ntimes): Rename to..
1956 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
1957 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
1958 (vect_do_peeling): Update call accordingly.
1959 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
1961 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1962 mask_compare_type, can_fully_mask_p and fully_masked_p.
1963 (release_vec_loop_masks): New function.
1964 (_loop_vec_info): Use it to free the loop masks.
1965 (can_produce_all_loop_masks_p): New function.
1966 (vect_get_max_nscalars_per_iter): Likewise.
1967 (vect_verify_full_masking): Likewise.
1968 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
1969 retries, and free the mask rgroups before retrying. Check loop-wide
1970 reasons for disallowing fully-masked loops. Make the final decision
1971 about whether use a fully-masked loop or not.
1972 (vect_estimate_min_profitable_iters): Do not assume that peeling
1973 for the number of iterations will be needed for fully-masked loops.
1974 (vectorizable_reduction): Disable fully-masked loops.
1975 (vectorizable_live_operation): Likewise.
1976 (vect_halve_mask_nunits): New function.
1977 (vect_double_mask_nunits): Likewise.
1978 (vect_record_loop_mask): Likewise.
1979 (vect_get_loop_mask): Likewise.
1980 (vect_transform_loop): Handle the case in which the final loop
1981 iteration might handle a partial vector. Call vect_set_loop_condition
1982 instead of slpeel_make_loop_iterate_ntimes.
1983 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
1984 (check_load_store_masking): New function.
1985 (prepare_load_store_mask): Likewise.
1986 (vectorizable_store): Handle fully-masked loops.
1987 (vectorizable_load): Likewise.
1988 (supportable_widening_operation): Use vect_halve_mask_nunits for
1990 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
1991 (vect_gen_while): New function.
1992 * config/aarch64/aarch64.md (umax<mode>3): New expander.
1993 (aarch64_uqdec<mode>): New insn.
1995 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1996 Alan Hayward <alan.hayward@arm.com>
1997 David Sherwood <david.sherwood@arm.com>
1999 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2000 (reduc_xor_scal_optab): New optabs.
2001 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2002 (reduc_xor_scal_@var{m}): Document.
2003 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2004 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2006 * fold-const-call.c (fold_const_call): Handle them.
2007 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2008 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2009 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2010 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2011 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2012 (UNSPEC_XORV): New unspecs.
2013 (optab): Add entries for them.
2014 (BITWISEV): New int iterator.
2015 (bit_reduc_op): New int attributes.
2017 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2018 Alan Hayward <alan.hayward@arm.com>
2019 David Sherwood <david.sherwood@arm.com>
2021 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2022 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2023 * optabs.def (vec_shl_insert_optab): New optab.
2024 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2025 (duplicate_and_interleave): Likewise.
2026 * tree-vect-loop.c: Include internal-fn.h.
2027 (neutral_op_for_slp_reduction): New function, split out from
2028 get_initial_defs_for_reduction.
2029 (get_initial_def_for_reduction): Handle option 2 for variable-length
2030 vectors by loading the neutral value into a vector and then shifting
2031 the initial value into element 0.
2032 (get_initial_defs_for_reduction): Replace the code argument with
2033 the neutral value calculated by neutral_op_for_slp_reduction.
2034 Use gimple_build_vector for constant-length vectors.
2035 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2036 but the first group_size elements have a neutral value.
2037 Use duplicate_and_interleave otherwise.
2038 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2039 Update call to get_initial_defs_for_reduction. Handle SLP
2040 reductions for variable-length vectors by creating one vector
2041 result for each scalar result, with the elements associated
2042 with other scalar results stubbed out with the neutral value.
2043 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2044 Require IFN_VEC_SHL_INSERT for double reductions on
2045 variable-length vectors, or SLP reductions that have
2046 a neutral value. Require can_duplicate_and_interleave_p
2047 support for variable-length unchained SLP reductions if there
2048 is no neutral value, such as for MIN/MAX reductions. Also require
2049 the number of vector elements to be a multiple of the number of
2050 SLP statements when doing variable-length unchained SLP reductions.
2051 Update call to vect_create_epilog_for_reduction.
2052 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2053 and remove initial values.
2054 (duplicate_and_interleave): Make public.
2055 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2056 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2058 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2059 Alan Hayward <alan.hayward@arm.com>
2060 David Sherwood <david.sherwood@arm.com>
2062 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2063 (can_duplicate_and_interleave_p): New function.
2064 (vect_get_and_check_slp_defs): Take the vector of statements
2065 rather than just the current one. Remove excess parentheses.
2066 Restriction rejectinon of vect_constant_def and vect_external_def
2067 for variable-length vectors to boolean types, or types for which
2068 can_duplicate_and_interleave_p is false.
2069 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2070 (duplicate_and_interleave): New function.
2071 (vect_get_constant_vectors): Use gimple_build_vector for
2072 constant-length vectors and suitable variable-length constant
2073 vectors. Use duplicate_and_interleave for other variable-length
2074 vectors. Don't defer the update when inserting new statements.
2076 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2077 Alan Hayward <alan.hayward@arm.com>
2078 David Sherwood <david.sherwood@arm.com>
2080 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2081 min_profitable_iters doesn't go negative.
2083 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2084 Alan Hayward <alan.hayward@arm.com>
2085 David Sherwood <david.sherwood@arm.com>
2087 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2088 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2089 * optabs.def (vec_mask_load_lanes_optab): New optab.
2090 (vec_mask_store_lanes_optab): Likewise.
2091 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2092 (MASK_STORE_LANES): Likewise.
2093 * internal-fn.c (mask_load_lanes_direct): New macro.
2094 (mask_store_lanes_direct): Likewise.
2095 (expand_mask_load_optab_fn): Handle masked operations.
2096 (expand_mask_load_lanes_optab_fn): New macro.
2097 (expand_mask_store_optab_fn): Handle masked operations.
2098 (expand_mask_store_lanes_optab_fn): New macro.
2099 (direct_mask_load_lanes_optab_supported_p): Likewise.
2100 (direct_mask_store_lanes_optab_supported_p): Likewise.
2101 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2103 (vect_load_lanes_supported): Likewise.
2104 * tree-vect-data-refs.c (strip_conversion): New function.
2105 (can_group_stmts_p): Likewise.
2106 (vect_analyze_data_ref_accesses): Use it instead of checking
2107 for a pair of assignments.
2108 (vect_store_lanes_supported): Take a masked_p parameter.
2109 (vect_load_lanes_supported): Likewise.
2110 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2111 vect_store_lanes_supported and vect_load_lanes_supported.
2112 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2113 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2114 parameter. Don't allow gaps for masked accesses.
2115 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2116 and vect_load_lanes_supported.
2117 (get_load_store_type): Take a masked_p parameter and update
2118 call to get_group_load_store_type.
2119 (vectorizable_store): Update call to get_load_store_type.
2120 Handle IFN_MASK_STORE_LANES.
2121 (vectorizable_load): Update call to get_load_store_type.
2122 Handle IFN_MASK_LOAD_LANES.
2124 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2125 Alan Hayward <alan.hayward@arm.com>
2126 David Sherwood <david.sherwood@arm.com>
2128 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2130 * config/aarch64/aarch64-protos.h
2131 (aarch64_sve_struct_memory_operand_p): Declare.
2132 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2133 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2134 (VPRED, vpred): Handle SVE structure modes.
2135 * config/aarch64/constraints.md (Utx): New constraint.
2136 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2137 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2138 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2139 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2140 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2141 structure modes. Split into pieces after RA.
2142 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2143 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2145 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2146 SVE structure modes.
2147 (aarch64_classify_address): Likewise.
2148 (sizetochar): Move earlier in file.
2149 (aarch64_print_operand): Handle SVE register lists.
2150 (aarch64_array_mode): New function.
2151 (aarch64_sve_struct_memory_operand_p): Likewise.
2152 (TARGET_ARRAY_MODE): Redefine.
2154 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2155 Alan Hayward <alan.hayward@arm.com>
2156 David Sherwood <david.sherwood@arm.com>
2158 * target.def (array_mode): New target hook.
2159 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2160 * doc/tm.texi: Regenerate.
2161 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2162 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2163 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2165 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2168 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2169 Alan Hayward <alan.hayward@arm.com>
2170 David Sherwood <david.sherwood@arm.com>
2172 * fold-const.c (fold_binary_loc): Check the argument types
2173 rather than the result type when testing for a vector operation.
2175 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2177 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2178 * doc/tm.texi: Regenerate.
2180 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2181 Alan Hayward <alan.hayward@arm.com>
2182 David Sherwood <david.sherwood@arm.com>
2184 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2185 (sve): Document new AArch64 extension.
2186 * doc/md.texi (w): Extend the description of the AArch64
2187 constraint to include SVE vectors.
2188 (Upl, Upa): Document new AArch64 predicate constraints.
2189 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2191 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2192 (msve-vector-bits=): New option.
2193 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2194 SVE when these are disabled.
2195 (sve): New extension.
2196 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2197 modes. Adjust their number of units based on aarch64_sve_vg.
2198 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2199 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2200 aarch64_addr_query_type.
2201 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2202 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2203 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2204 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2205 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2206 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2207 (aarch64_simd_imm_zero_p): Delete.
2208 (aarch64_check_zero_based_sve_index_immediate): Declare.
2209 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2210 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2211 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2212 (aarch64_sve_float_mul_immediate_p): Likewise.
2213 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2215 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2216 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2217 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2218 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2219 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2220 (aarch64_regmode_natural_size): Likewise.
2221 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2222 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2224 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2225 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2226 for VG and the SVE predicate registers.
2227 (V_ALIASES): Add a "z"-prefixed alias.
2228 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2229 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2230 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2231 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2232 (REG_CLASS_NAMES): Add entries for them.
2233 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2234 and the predicate registers.
2235 (aarch64_sve_vg): Declare.
2236 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2237 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2238 (REGMODE_NATURAL_SIZE): Define.
2239 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2241 * config/aarch64/aarch64.c: Include cfgrtl.h.
2242 (simd_immediate_info): Add a constructor for series vectors,
2243 and an associated step field.
2244 (aarch64_sve_vg): New variable.
2245 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2246 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2247 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2248 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2249 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2250 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2251 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2252 (aarch64_get_mask_mode): New functions.
2253 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2254 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2255 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2256 predicate modes and predicate registers. Explicitly restrict
2257 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2258 to store a vector mode if it is recognized by
2259 aarch64_classify_vector_mode.
2260 (aarch64_regmode_natural_size): New function.
2261 (aarch64_hard_regno_caller_save_mode): Return the original mode
2263 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2264 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2265 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2266 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2268 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2269 does not overlap dest if the function is frame-related. Handle
2271 (aarch64_split_add_offset): New function.
2272 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2273 them aarch64_add_offset.
2274 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2275 and update call to aarch64_sub_sp.
2276 (aarch64_add_cfa_expression): New function.
2277 (aarch64_expand_prologue): Pass extra temporary registers to the
2278 functions above. Handle the case in which we need to emit new
2279 DW_CFA_expressions for registers that were originally saved
2280 relative to the stack pointer, but now have to be expressed
2281 relative to the frame pointer.
2282 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2284 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2285 IP0 and IP1 values for SVE frames.
2286 (aarch64_expand_vec_series): New function.
2287 (aarch64_expand_sve_widened_duplicate): Likewise.
2288 (aarch64_expand_sve_const_vector): Likewise.
2289 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2290 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2291 into the register, rather than emitting a SET directly.
2292 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2293 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2294 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2295 (offset_9bit_signed_scaled_p): New functions.
2296 (aarch64_replicate_bitmask_imm): New function.
2297 (aarch64_bitmask_imm): Use it.
2298 (aarch64_cannot_force_const_mem): Reject expressions involving
2299 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2300 (aarch64_classify_index): Handle SVE indices, by requiring
2301 a plain register index with a scale that matches the element size.
2302 (aarch64_classify_address): Handle SVE addresses. Assert that
2303 the mode of the address is VOIDmode or an integer mode.
2304 Update call to aarch64_classify_symbol.
2305 (aarch64_classify_symbolic_expression): Update call to
2306 aarch64_classify_symbol.
2307 (aarch64_const_vec_all_in_range_p): New function.
2308 (aarch64_print_vector_float_operand): Likewise.
2309 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2310 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2311 and the FP immediates 1.0 and 0.5.
2312 (aarch64_print_address_internal): Handle SVE addresses.
2313 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2314 (aarch64_regno_regclass): Handle predicate registers.
2315 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2317 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2318 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2319 (aarch64_convert_sve_vector_bits): New function.
2320 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2321 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2323 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2324 Handle SVE vector and predicate modes. Accept VL-based constants
2325 that need only one temporary register, and VL offsets that require
2326 no temporary registers.
2327 (aarch64_conditional_register_usage): Mark the predicate registers
2328 as fixed if SVE isn't available.
2329 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2330 Return true for SVE vector and predicate modes.
2331 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2332 rather than an unsigned int. Handle SVE modes.
2333 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2335 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2337 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2338 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2339 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2340 (aarch64_sve_float_mul_immediate_p): New functions.
2341 (aarch64_sve_valid_immediate): New function.
2342 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2343 Explicitly reject structure modes. Check for INDEX constants.
2344 Handle PTRUE and PFALSE constants.
2345 (aarch64_check_zero_based_sve_index_immediate): New function.
2346 (aarch64_simd_imm_zero_p): Delete.
2347 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2348 vector modes. Accept constants in the range of CNT[BHWD].
2349 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2350 ask for an Advanced SIMD mode.
2351 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2352 (aarch64_simd_vector_alignment): Handle SVE predicates.
2353 (aarch64_vectorize_preferred_vector_alignment): New function.
2354 (aarch64_simd_vector_alignment_reachable): Use it instead of
2356 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2357 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2359 (MAX_VECT_LEN): Delete.
2360 (expand_vec_perm_d): Add a vec_flags field.
2361 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2362 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2363 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2365 (aarch64_evpc_rev): Rename to...
2366 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2367 (aarch64_evpc_rev_global): New function.
2368 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2369 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2371 (aarch64_evpc_sve_tbl): New function.
2372 (aarch64_expand_vec_perm_const_1): Update after rename of
2373 aarch64_evpc_rev. Handle SVE permutes too, trying
2374 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2375 than aarch64_evpc_tbl.
2376 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2377 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2378 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2379 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2380 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2381 (aarch64_expand_sve_vcond): New functions.
2382 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2383 of aarch64_vector_mode_p.
2384 (aarch64_dwarf_poly_indeterminate_value): New function.
2385 (aarch64_compute_pressure_classes): Likewise.
2386 (aarch64_can_change_mode_class): Likewise.
2387 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2388 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2389 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2390 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2391 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2392 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2393 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2394 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2396 (Dn, Dl, Dr): Accept const as well as const_vector.
2397 (Dz): Likewise. Compare against CONST0_RTX.
2398 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2399 of "vector" where appropriate.
2400 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2401 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2402 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2403 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2404 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2405 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2406 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2407 (v_int_equiv): Extend to SVE modes.
2408 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2410 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2411 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2412 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2413 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2414 (SVE_COND_FP_CMP): New int iterators.
2415 (perm_hilo): Handle the new unpack unspecs.
2416 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2418 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2419 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2420 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2421 (aarch64_equality_operator, aarch64_constant_vector_operand)
2422 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2423 (aarch64_sve_nonimmediate_operand): Likewise.
2424 (aarch64_sve_general_operand): Likewise.
2425 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2426 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2427 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2428 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2429 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2430 (aarch64_sve_float_arith_immediate): Likewise.
2431 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2432 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2433 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2434 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2435 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2436 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2437 (aarch64_sve_float_arith_operand): Likewise.
2438 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2439 (aarch64_sve_float_mul_operand): Likewise.
2440 (aarch64_sve_vec_perm_operand): Likewise.
2441 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2442 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2443 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2444 as well as const_vector.
2445 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2446 in file. Use CONST0_RTX and CONSTM1_RTX.
2447 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2448 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2449 Use aarch64_simd_imm_zero.
2450 * config/aarch64/aarch64-sve.md: New file.
2451 * config/aarch64/aarch64.md: Include it.
2452 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2453 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2454 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2455 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2456 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2457 (sve): New attribute.
2458 (enabled): Disable instructions with the sve attribute unless
2460 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2461 aarch64_expand_mov_immediate.
2462 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2463 CNT[BHSD] immediates.
2464 (movti): Split CONST_POLY_INT moves into two halves.
2465 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2466 Split additions that need a temporary here if the destination
2467 is the stack pointer.
2468 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2469 (*add<mode>3_poly_1): New instruction.
2470 (set_clobber_cc): New expander.
2472 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2474 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2475 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2476 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2477 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2478 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2479 Change innermode from fixed_mode_size to machine_mode.
2480 (simplify_subreg): Update call accordingly. Handle a constant-sized
2481 subreg of a variable-length CONST_VECTOR.
2483 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2484 Alan Hayward <alan.hayward@arm.com>
2485 David Sherwood <david.sherwood@arm.com>
2487 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2488 (add_offset_to_base): New function, split out from...
2489 (create_mem_ref): ...here. When handling a scale other than 1,
2490 check first whether the address is valid without the offset.
2491 Add it into the base if so, leaving the index and scale as-is.
2493 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2496 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2497 fold_for_warn before checking if arg2 is INTEGER_CST.
2499 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2501 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2502 (store_multiple_operation): Delete.
2503 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2504 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2505 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2506 guarded by TARGET_STRING.
2507 (rs6000_output_load_multiple): Delete.
2508 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2509 OPTION_MASK_STRING / TARGET_STRING handling.
2510 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2511 (const rs6000_opt_masks) <"string">: Change mask to 0.
2512 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2513 (MASK_STRING): Delete.
2514 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2516 (load_multiple): Delete.
2523 (store_multiple): Delete.
2530 (movmemsi_8reg): Delete.
2531 (corresponding unnamed define_insn): Delete.
2532 (movmemsi_6reg): Delete.
2533 (corresponding unnamed define_insn): Delete.
2534 (movmemsi_4reg): Delete.
2535 (corresponding unnamed define_insn): Delete.
2536 (movmemsi_2reg): Delete.
2537 (corresponding unnamed define_insn): Delete.
2538 (movmemsi_1reg): Delete.
2539 (corresponding unnamed define_insn): Delete.
2540 * config/rs6000/rs6000.opt (mno-string): New.
2541 (mstring): Replace by deprecation warning stub.
2542 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2544 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2546 * regrename.c (regrename_do_replace): If replacing the same
2547 reg multiple times, try to reuse last created gen_raw_REG.
2550 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2551 main to workaround a bug in GDB.
2553 2018-01-12 Tom de Vries <tom@codesourcery.com>
2556 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2558 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2560 PR rtl-optimization/80481
2561 * ira-color.c (get_cap_member): New function.
2562 (allocnos_conflict_by_live_ranges_p): Use it.
2563 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2564 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2566 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2569 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2570 (*saddl_se_1): Ditto.
2572 (*ssubl_se_1): Ditto.
2574 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2576 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2577 rather than wi::to_widest for DR_INITs.
2578 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2579 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2580 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2582 (vect_analyze_group_access_1): Note that here.
2584 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2586 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2587 polynomial type sizes.
2589 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2591 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2592 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2593 (gimple_add_tmp_var): Likewise.
2595 2018-01-12 Martin Liska <mliska@suse.cz>
2597 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2598 (gimple_alloc_sizes): Likewise.
2599 (dump_gimple_statistics): Use PRIu64 in printf format.
2600 * gimple.h: Change uint64_t to int.
2602 2018-01-12 Martin Liska <mliska@suse.cz>
2604 * tree-core.h: Use uint64_t instead of int.
2605 * tree.c (tree_node_counts): Likewise.
2606 (tree_node_sizes): Likewise.
2607 (dump_tree_statistics): Use PRIu64 in printf format.
2609 2018-01-12 Martin Liska <mliska@suse.cz>
2611 * Makefile.in: As qsort_chk is implemented in vec.c, add
2612 vec.o to linkage of gencfn-macros.
2613 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2614 passing the info to record_node_allocation_statistics.
2615 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2617 * ggc-common.c (struct ggc_usage): Add operator== and use
2618 it in operator< and compare function.
2619 * mem-stats.h (struct mem_usage): Likewise.
2620 * vec.c (struct vec_usage): Remove operator< and compare
2621 function. Can be simply inherited.
2623 2018-01-12 Martin Jambor <mjambor@suse.cz>
2626 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2627 * tree-ssa-math-opts.c: Include domwalk.h.
2628 (convert_mult_to_fma_1): New function.
2629 (fma_transformation_info): New type.
2630 (fma_deferring_state): Likewise.
2631 (cancel_fma_deferring): New function.
2632 (result_of_phi): Likewise.
2633 (last_fma_candidate_feeds_initial_phi): Likewise.
2634 (convert_mult_to_fma): Added deferring logic, split actual
2635 transformation to convert_mult_to_fma_1.
2636 (math_opts_dom_walker): New type.
2637 (math_opts_dom_walker::after_dom_children): New method, body moved
2638 here from pass_optimize_widening_mul::execute, added deferring logic
2640 (pass_optimize_widening_mul::execute): Moved most of code to
2641 math_opts_dom_walker::after_dom_children.
2642 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2643 * config/i386/i386.c (ix86_option_override_internal): Added
2644 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2646 2018-01-12 Richard Biener <rguenther@suse.de>
2649 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2650 inline instance vars.
2652 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2655 * config/rx/rx.c (rx_is_restricted_memory_address):
2658 2018-01-12 Richard Biener <rguenther@suse.de>
2660 PR tree-optimization/80846
2661 * target.def (split_reduction): New target hook.
2662 * targhooks.c (default_split_reduction): New function.
2663 * targhooks.h (default_split_reduction): Declare.
2664 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2665 target requests first reduce vectors by combining low and high
2667 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2668 (get_vectype_for_scalar_type_and_size): Export.
2669 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2670 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2671 * doc/tm.texi: Regenerate.
2672 * config/i386/i386.c (ix86_split_reduction): Implement
2673 TARGET_VECTORIZE_SPLIT_REDUCTION.
2675 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2678 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2679 in PIC mode except for TARGET_VXWORKS_RTP.
2680 * config/sparc/sparc.c: Include cfgrtl.h.
2681 (TARGET_INIT_PIC_REG): Define.
2682 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2683 (sparc_pic_register_p): New predicate.
2684 (sparc_legitimate_address_p): Use it.
2685 (sparc_legitimize_pic_address): Likewise.
2686 (sparc_delegitimize_address): Likewise.
2687 (sparc_mode_dependent_address_p): Likewise.
2688 (gen_load_pcrel_sym): Remove 4th parameter.
2689 (load_got_register): Adjust call to above. Remove obsolete stuff.
2690 (sparc_expand_prologue): Do not call load_got_register here.
2691 (sparc_flat_expand_prologue): Likewise.
2692 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2693 (sparc_use_pseudo_pic_reg): New function.
2694 (sparc_init_pic_reg): Likewise.
2695 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2696 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2698 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2700 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2701 Add item for branch_cost.
2703 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2705 PR rtl-optimization/83565
2706 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2707 not extend the result to a larger mode for rotate operations.
2708 (num_sign_bit_copies1): Likewise.
2710 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2713 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2715 Use values-Xc.o for -pedantic.
2716 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2718 2018-01-12 Martin Liska <mliska@suse.cz>
2721 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2723 (possible_polymorphic_call_targets): Use it.
2724 (ipa_devirt): Likewise.
2726 2018-01-12 Martin Liska <mliska@suse.cz>
2728 * profile-count.h (enum profile_quality): Use 0 as invalid
2729 enum value of profile_quality.
2731 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2733 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2734 -mext-string options.
2736 2018-01-12 Richard Biener <rguenther@suse.de>
2738 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2739 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2740 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2742 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2744 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2746 * configure.ac (--with-long-double-format): Add support for the
2747 configuration option to change the default long double format on
2749 * config.gcc (powerpc*-linux*-*): Likewise.
2750 * configure: Regenerate.
2751 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2752 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2753 used without modification.
2755 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2757 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2758 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2759 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2760 MISC_BUILTIN_SPEC_BARRIER.
2761 (rs6000_init_builtins): Likewise.
2762 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2764 (speculation_barrier): New define_insn.
2765 * doc/extend.texi: Document __builtin_speculation_barrier.
2767 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2770 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2771 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2772 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2774 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2775 integral modes instead of "ss" and "sd".
2776 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2777 vectors with 32-bit and 64-bit elements.
2778 (vecdupssescalarmodesuffix): New mode attribute.
2779 (vec_dup<mode>): Use it.
2781 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2784 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2785 frame if argument is passed on stack.
2787 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2790 * ree.c (combine_reaching_defs): Optimize also
2791 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
2792 reg2=any_extend(exp); reg1=reg2;, formatting fix.
2794 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2797 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
2799 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2802 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
2803 after they are computed.
2805 2018-01-11 Bin Cheng <bin.cheng@arm.com>
2807 PR tree-optimization/83695
2808 * gimple-loop-linterchange.cc
2809 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
2810 reset cached scev information after interchange.
2811 (pass_linterchange::execute): Remove call to scev_reset_htab.
2813 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2815 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
2816 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
2817 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
2818 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
2819 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
2820 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
2821 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
2822 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
2823 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
2824 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
2825 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
2826 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
2827 (V_lane_reg): Likewise.
2828 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
2830 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
2831 (vfmal_lane_low<mode>_intrinsic,
2832 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
2833 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
2834 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
2835 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
2836 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
2837 vfmsl_lane_high<mode>_intrinsic): New define_insns.
2839 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2841 * config/arm/arm-cpus.in (fp16fml): New feature.
2842 (ALL_SIMD): Add fp16fml.
2843 (armv8.2-a): Add fp16fml as an option.
2844 (armv8.3-a): Likewise.
2845 (armv8.4-a): Add fp16fml as part of fp16.
2846 * config/arm/arm.h (TARGET_FP16FML): Define.
2847 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
2849 * config/arm/arm-modes.def (V2HF): Define.
2850 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
2851 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
2852 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
2853 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
2854 vfmsl_low, vfmsl_high): New set of builtins.
2855 * config/arm/iterators.md (PLUSMINUS): New code iterator.
2856 (vfml_op): New code attribute.
2857 (VFMLHALVES): New int iterator.
2858 (VFML, VFMLSEL): New mode attributes.
2859 (V_reg): Define mapping for V2HF.
2860 (V_hi, V_lo): New mode attributes.
2861 (VF_constraint): Likewise.
2862 (vfml_half, vfml_half_selector): New int attributes.
2863 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
2865 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
2866 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
2868 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
2869 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
2870 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
2871 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
2873 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
2874 Document new effective target and option set.
2876 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2878 * config/arm/arm-cpus.in (armv8_4): New feature.
2879 (ARMv8_4a): New fgroup.
2880 (armv8.4-a): New arch.
2881 * config/arm/arm-tables.opt: Regenerate.
2882 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
2883 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
2884 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
2885 Add matching rules for -march=armv8.4-a and extensions.
2886 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
2888 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
2891 * config/rx/rx.md (BW): New mode attribute.
2892 (sync_lock_test_and_setsi): Add mode suffix to insn output.
2894 2018-01-11 Richard Biener <rguenther@suse.de>
2896 PR tree-optimization/83435
2897 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
2898 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
2899 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
2901 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2902 Alan Hayward <alan.hayward@arm.com>
2903 David Sherwood <david.sherwood@arm.com>
2905 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
2907 (aarch64_classify_address): Initialize it. Track polynomial offsets.
2908 (aarch64_print_address_internal): Use it to check for a zero offset.
2910 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2911 Alan Hayward <alan.hayward@arm.com>
2912 David Sherwood <david.sherwood@arm.com>
2914 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
2915 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
2916 Return a poly_int64 rather than a HOST_WIDE_INT.
2917 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
2918 rather than a HOST_WIDE_INT.
2919 * config/aarch64/aarch64.h (aarch64_frame): Protect with
2920 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
2921 hard_fp_offset, frame_size, initial_adjust, callee_offset and
2922 final_offset from HOST_WIDE_INT to poly_int64.
2923 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
2924 to_constant when getting the number of units in an Advanced SIMD
2926 (aarch64_builtin_vectorized_function): Check for a constant number
2928 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
2930 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
2931 attribute instead of GET_MODE_NUNITS.
2932 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
2933 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
2934 GET_MODE_SIZE for fixed-size registers.
2935 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
2936 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
2937 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
2938 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
2939 (aarch64_print_operand, aarch64_print_address_internal)
2940 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
2941 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
2942 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
2943 Handle polynomial GET_MODE_SIZE.
2944 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
2945 wider than SImode without modification.
2946 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
2947 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
2948 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
2949 passing and returning SVE modes.
2950 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
2951 rather than GEN_INT.
2952 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
2953 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
2954 (aarch64_allocate_and_probe_stack_space): Likewise.
2955 (aarch64_layout_frame): Cope with polynomial offsets.
2956 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
2957 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
2959 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
2960 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
2961 poly_int64 rather than a HOST_WIDE_INT.
2962 (aarch64_get_separate_components, aarch64_process_components)
2963 (aarch64_expand_prologue, aarch64_expand_epilogue)
2964 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
2965 (aarch64_anchor_offset): New function, split out from...
2966 (aarch64_legitimize_address): ...here.
2967 (aarch64_builtin_vectorization_cost): Handle polynomial
2968 TYPE_VECTOR_SUBPARTS.
2969 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
2971 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
2972 number of elements from the PARALLEL rather than the mode.
2973 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
2974 rather than GET_MODE_BITSIZE.
2975 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
2976 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
2977 (aarch64_expand_vec_perm_const_1): Handle polynomial
2978 d->perm.length () and d->perm elements.
2979 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
2980 Apply to_constant to d->perm elements.
2981 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
2982 polynomial CONST_VECTOR_NUNITS.
2983 (aarch64_move_pointer): Take amount as a poly_int64 rather
2985 (aarch64_progress_pointer): Avoid temporary variable.
2986 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
2987 the mode attribute instead of GET_MODE.
2989 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2990 Alan Hayward <alan.hayward@arm.com>
2991 David Sherwood <david.sherwood@arm.com>
2993 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
2994 x exists before using it.
2995 (aarch64_add_constant_internal): Rename to...
2996 (aarch64_add_offset_1): ...this. Replace regnum with separate
2997 src and dest rtxes. Handle the case in which they're different,
2998 including when the offset is zero. Replace scratchreg with an rtx.
2999 Use 2 additions if there is no spare register into which we can
3000 move a 16-bit constant.
3001 (aarch64_add_constant): Delete.
3002 (aarch64_add_offset): Replace reg with separate src and dest
3003 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3004 Use aarch64_add_offset_1.
3005 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3006 an rtx rather than an int. Take the delta as a poly_int64
3007 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3008 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3009 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3010 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3011 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3013 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3014 aarch64_add_constant.
3016 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3018 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3019 Use scalar_float_mode.
3021 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3023 * config/aarch64/aarch64-simd.md
3024 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3025 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3026 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3027 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3028 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3029 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3030 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3031 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3032 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3033 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3035 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3038 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3039 targ_options->x_arm_arch_string is non NULL.
3041 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3043 * config/aarch64/aarch64.h
3044 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3046 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3049 * expmed.c (emit_store_flag_force): Swap if const op0
3050 and change VOIDmode to mode of op0.
3052 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3054 PR rtl-optimization/83761
3055 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3056 than bytes to mode_for_size.
3058 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3061 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3062 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3065 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3068 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3069 when in layout mode.
3070 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3071 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3074 2018-01-10 Michael Collison <michael.collison@arm.com>
3076 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3077 * config/aarch64/aarch64-option-extension.def: Add
3078 AARCH64_OPT_EXTENSION of 'fp16fml'.
3079 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3080 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3081 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3082 * config/aarch64/constraints.md (Ui7): New constraint.
3083 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3084 (VFMLA_SEL_W): Ditto.
3087 (VFMLA16_LOW): New int iterator.
3088 (VFMLA16_HIGH): Ditto.
3089 (UNSPEC_FMLAL): New unspec.
3090 (UNSPEC_FMLSL): Ditto.
3091 (UNSPEC_FMLAL2): Ditto.
3092 (UNSPEC_FMLSL2): Ditto.
3093 (f16mac): New code attribute.
3094 * config/aarch64/aarch64-simd-builtins.def
3095 (aarch64_fmlal_lowv2sf): Ditto.
3096 (aarch64_fmlsl_lowv2sf): Ditto.
3097 (aarch64_fmlalq_lowv4sf): Ditto.
3098 (aarch64_fmlslq_lowv4sf): Ditto.
3099 (aarch64_fmlal_highv2sf): Ditto.
3100 (aarch64_fmlsl_highv2sf): Ditto.
3101 (aarch64_fmlalq_highv4sf): Ditto.
3102 (aarch64_fmlslq_highv4sf): Ditto.
3103 (aarch64_fmlal_lane_lowv2sf): Ditto.
3104 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3105 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3106 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3107 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3108 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3109 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3110 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3111 (aarch64_fmlal_lane_highv2sf): Ditto.
3112 (aarch64_fmlsl_lane_highv2sf): Ditto.
3113 (aarch64_fmlal_laneq_highv2sf): Ditto.
3114 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3115 (aarch64_fmlalq_lane_highv4sf): Ditto.
3116 (aarch64_fmlsl_lane_highv4sf): Ditto.
3117 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3118 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3119 * config/aarch64/aarch64-simd.md:
3120 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3121 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3122 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3123 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3124 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3125 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3126 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3127 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3128 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3129 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3130 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3131 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3132 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3133 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3134 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3135 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3136 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3137 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3138 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3139 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3140 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3141 (vfmlsl_low_u32): Ditto.
3142 (vfmlalq_low_u32): Ditto.
3143 (vfmlslq_low_u32): Ditto.
3144 (vfmlal_high_u32): Ditto.
3145 (vfmlsl_high_u32): Ditto.
3146 (vfmlalq_high_u32): Ditto.
3147 (vfmlslq_high_u32): Ditto.
3148 (vfmlal_lane_low_u32): Ditto.
3149 (vfmlsl_lane_low_u32): Ditto.
3150 (vfmlal_laneq_low_u32): Ditto.
3151 (vfmlsl_laneq_low_u32): Ditto.
3152 (vfmlalq_lane_low_u32): Ditto.
3153 (vfmlslq_lane_low_u32): Ditto.
3154 (vfmlalq_laneq_low_u32): Ditto.
3155 (vfmlslq_laneq_low_u32): Ditto.
3156 (vfmlal_lane_high_u32): Ditto.
3157 (vfmlsl_lane_high_u32): Ditto.
3158 (vfmlal_laneq_high_u32): Ditto.
3159 (vfmlsl_laneq_high_u32): Ditto.
3160 (vfmlalq_lane_high_u32): Ditto.
3161 (vfmlslq_lane_high_u32): Ditto.
3162 (vfmlalq_laneq_high_u32): Ditto.
3163 (vfmlslq_laneq_high_u32): Ditto.
3164 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3165 (AARCH64_FL_FOR_ARCH8_4): New.
3166 (AARCH64_ISA_F16FML): New ISA flag.
3167 (TARGET_F16FML): New feature flag for fp16fml.
3168 (doc/invoke.texi): Document new fp16fml option.
3170 2018-01-10 Michael Collison <michael.collison@arm.com>
3172 * config/aarch64/aarch64-builtins.c:
3173 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3174 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3175 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3176 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3177 (AARCH64_ISA_SHA3): New ISA flag.
3178 (TARGET_SHA3): New feature flag for sha3.
3179 * config/aarch64/iterators.md (sha512_op): New int attribute.
3180 (CRYPTO_SHA512): New int iterator.
3181 (UNSPEC_SHA512H): New unspec.
3182 (UNSPEC_SHA512H2): Ditto.
3183 (UNSPEC_SHA512SU0): Ditto.
3184 (UNSPEC_SHA512SU1): Ditto.
3185 * config/aarch64/aarch64-simd-builtins.def
3186 (aarch64_crypto_sha512hqv2di): New builtin.
3187 (aarch64_crypto_sha512h2qv2di): Ditto.
3188 (aarch64_crypto_sha512su0qv2di): Ditto.
3189 (aarch64_crypto_sha512su1qv2di): Ditto.
3190 (aarch64_eor3qv8hi): Ditto.
3191 (aarch64_rax1qv2di): Ditto.
3192 (aarch64_xarqv2di): Ditto.
3193 (aarch64_bcaxqv8hi): Ditto.
3194 * config/aarch64/aarch64-simd.md:
3195 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3196 (aarch64_crypto_sha512su0qv2di): Ditto.
3197 (aarch64_crypto_sha512su1qv2di): Ditto.
3198 (aarch64_eor3qv8hi): Ditto.
3199 (aarch64_rax1qv2di): Ditto.
3200 (aarch64_xarqv2di): Ditto.
3201 (aarch64_bcaxqv8hi): Ditto.
3202 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3203 (vsha512h2q_u64): Ditto.
3204 (vsha512su0q_u64): Ditto.
3205 (vsha512su1q_u64): Ditto.
3206 (veor3q_u16): Ditto.
3207 (vrax1q_u64): Ditto.
3209 (vbcaxq_u16): Ditto.
3210 * config/arm/types.md (crypto_sha512): New type attribute.
3211 (crypto_sha3): Ditto.
3212 (doc/invoke.texi): Document new sha3 option.
3214 2018-01-10 Michael Collison <michael.collison@arm.com>
3216 * config/aarch64/aarch64-builtins.c:
3217 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3218 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3219 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3220 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3221 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3222 (AARCH64_ISA_SM4): New ISA flag.
3223 (TARGET_SM4): New feature flag for sm4.
3224 * config/aarch64/aarch64-simd-builtins.def
3225 (aarch64_sm3ss1qv4si): Ditto.
3226 (aarch64_sm3tt1aq4si): Ditto.
3227 (aarch64_sm3tt1bq4si): Ditto.
3228 (aarch64_sm3tt2aq4si): Ditto.
3229 (aarch64_sm3tt2bq4si): Ditto.
3230 (aarch64_sm3partw1qv4si): Ditto.
3231 (aarch64_sm3partw2qv4si): Ditto.
3232 (aarch64_sm4eqv4si): Ditto.
3233 (aarch64_sm4ekeyqv4si): Ditto.
3234 * config/aarch64/aarch64-simd.md:
3235 (aarch64_sm3ss1qv4si): Ditto.
3236 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3237 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3238 (aarch64_sm4eqv4si): Ditto.
3239 (aarch64_sm4ekeyqv4si): Ditto.
3240 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3241 (sm3part_op): Ditto.
3242 (CRYPTO_SM3TT): Ditto.
3243 (CRYPTO_SM3PART): Ditto.
3244 (UNSPEC_SM3SS1): New unspec.
3245 (UNSPEC_SM3TT1A): Ditto.
3246 (UNSPEC_SM3TT1B): Ditto.
3247 (UNSPEC_SM3TT2A): Ditto.
3248 (UNSPEC_SM3TT2B): Ditto.
3249 (UNSPEC_SM3PARTW1): Ditto.
3250 (UNSPEC_SM3PARTW2): Ditto.
3251 (UNSPEC_SM4E): Ditto.
3252 (UNSPEC_SM4EKEY): Ditto.
3253 * config/aarch64/constraints.md (Ui2): New constraint.
3254 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3255 * config/arm/types.md (crypto_sm3): New type attribute.
3256 (crypto_sm4): Ditto.
3257 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3258 (vsm3tt1aq_u32): Ditto.
3259 (vsm3tt1bq_u32): Ditto.
3260 (vsm3tt2aq_u32): Ditto.
3261 (vsm3tt2bq_u32): Ditto.
3262 (vsm3partw1q_u32): Ditto.
3263 (vsm3partw2q_u32): Ditto.
3264 (vsm4eq_u32): Ditto.
3265 (vsm4ekeyq_u32): Ditto.
3266 (doc/invoke.texi): Document new sm4 option.
3268 2018-01-10 Michael Collison <michael.collison@arm.com>
3270 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3271 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3272 (AARCH64_FL_FOR_ARCH8_4): New.
3273 (AARCH64_FL_V8_4): New flag.
3274 (doc/invoke.texi): Document new armv8.4-a option.
3276 2018-01-10 Michael Collison <michael.collison@arm.com>
3278 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3279 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3280 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3281 * config/aarch64/aarch64-option-extension.def: Add
3282 AARCH64_OPT_EXTENSION of 'sha2'.
3283 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3284 (crypto): Disable sha2 and aes if crypto disabled.
3285 (crypto): Enable aes and sha2 if enabled.
3286 (simd): Disable sha2 and aes if simd disabled.
3287 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3289 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3290 (TARGET_SHA2): New feature flag for sha2.
3291 (TARGET_AES): New feature flag for aes.
3292 * config/aarch64/aarch64-simd.md:
3293 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3294 conditional on TARGET_AES.
3295 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3296 (aarch64_crypto_sha1hsi): Make pattern conditional
3298 (aarch64_crypto_sha1hv4si): Ditto.
3299 (aarch64_be_crypto_sha1hv4si): Ditto.
3300 (aarch64_crypto_sha1su1v4si): Ditto.
3301 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3302 (aarch64_crypto_sha1su0v4si): Ditto.
3303 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3304 (aarch64_crypto_sha256su0v4si): Ditto.
3305 (aarch64_crypto_sha256su1v4si): Ditto.
3306 (doc/invoke.texi): Document new aes and sha2 options.
3308 2018-01-10 Martin Sebor <msebor@redhat.com>
3310 PR tree-optimization/83781
3311 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3314 2018-01-11 Martin Sebor <msebor@gmail.com>
3315 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3317 PR tree-optimization/83501
3318 PR tree-optimization/81703
3320 * tree-ssa-strlen.c (get_string_cst): Rename...
3321 (get_string_len): ...to this. Handle global constants.
3322 (handle_char_store): Adjust.
3324 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3325 Jim Wilson <jimw@sifive.com>
3327 * config/riscv/riscv-protos.h (riscv_output_return): New.
3328 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3329 (riscv_attribute_table, riscv_output_return),
3330 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3331 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3332 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3334 (riscv_expand_prologue): Add early return for naked function.
3335 (riscv_expand_epilogue): Likewise.
3336 (riscv_function_ok_for_sibcall): Return false for naked function.
3337 (riscv_set_current_function): New.
3338 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3339 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3340 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3341 * doc/extend.texi (RISC-V Function Attributes): New.
3343 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3345 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3346 check for 128-bit long double before checking TCmode.
3347 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3348 128-bit long doubles before checking TFmode or TCmode.
3349 (FLOAT128_IBM_P): Likewise.
3351 2018-01-10 Martin Sebor <msebor@redhat.com>
3353 PR tree-optimization/83671
3354 * builtins.c (c_strlen): Unconditionally return zero for the empty
3356 Use -Warray-bounds for warnings.
3357 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3358 for non-constant array indices with COMPONENT_REF, arrays of
3359 arrays, and pointers to arrays.
3360 (gimple_fold_builtin_strlen): Determine and set length range for
3361 non-constant character arrays.
3363 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3366 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3369 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3371 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3373 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3376 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3377 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3378 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3379 indexed_or_indirect_operand predicate.
3380 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3381 (*vsx_le_perm_load_v8hi): Likewise.
3382 (*vsx_le_perm_load_v16qi): Likewise.
3383 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3384 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3385 (*vsx_le_perm_store_v8hi): Likewise.
3386 (*vsx_le_perm_store_v16qi): Likewise.
3387 (eight unnamed splitters): Likewise.
3389 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3391 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3392 * config/rs6000/emmintrin.h: Likewise.
3393 * config/rs6000/mmintrin.h: Likewise.
3394 * config/rs6000/xmmintrin.h: Likewise.
3396 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3399 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3401 * tree.c (tree_nop_conversion): Return true for location wrapper
3403 (maybe_wrap_with_location): New function.
3404 (selftest::check_strip_nops): New function.
3405 (selftest::test_location_wrappers): New function.
3406 (selftest::tree_c_tests): Call it.
3407 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3408 (maybe_wrap_with_location): New decl.
3409 (EXPR_LOCATION_WRAPPER_P): New macro.
3410 (location_wrapper_p): New inline function.
3411 (tree_strip_any_location_wrapper): New inline function.
3413 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3416 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3417 stack_realign_offset for the largest alignment of stack slot
3419 (ix86_find_max_used_stack_alignment): New function.
3420 (ix86_finalize_stack_frame_flags): Use it. Set
3421 max_used_stack_alignment if we don't realign stack.
3422 * config/i386/i386.h (machine_function): Add
3423 max_used_stack_alignment.
3425 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3427 * config/arm/arm.opt (-mbranch-cost): New option.
3428 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3431 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3434 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3435 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3437 2018-01-10 Richard Biener <rguenther@suse.de>
3440 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3441 early out so it also covers the case where we have a non-NULL
3444 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3446 PR tree-optimization/83753
3447 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3448 for non-strided grouped accesses if the number of elements is 1.
3450 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3453 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3454 * i386.h (TARGET_USE_GATHER): Define.
3455 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3457 2018-01-10 Martin Liska <mliska@suse.cz>
3460 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3461 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3463 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3464 CLEANUP_NO_PARTITIONING is not set.
3466 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3468 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3469 for vectors, as a partial revert of r254296.
3470 * rtl.h (const_vec_p): Delete.
3471 (const_vec_duplicate_p): Don't test for vector CONSTs.
3472 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3473 * expmed.c (make_tree): Likewise.
3476 * common.md (E, F): Use CONSTANT_P instead of checking for
3478 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3479 checking for CONST_VECTOR.
3481 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3484 * predict.c (force_edge_cold): Handle in more sane way edges
3487 2018-01-09 Carl Love <cel@us.ibm.com>
3489 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3491 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3492 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3493 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3494 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3495 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3496 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3497 * config/rs6000/rs6000-protos.h: Add extern defition for
3498 rs6000_generate_float2_double_code.
3499 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3501 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3502 (float2_v2df): Add define_expand.
3504 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3507 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3508 op_mode in the force_to_mode call.
3510 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3512 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3513 instead of checking each element individually.
3514 (aarch64_evpc_uzp): Likewise.
3515 (aarch64_evpc_zip): Likewise.
3516 (aarch64_evpc_ext): Likewise.
3517 (aarch64_evpc_rev): Likewise.
3518 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3519 instead of checking each element individually. Return true without
3521 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3522 whether all selected elements come from the same input, instead of
3523 checking each element individually. Remove calls to gen_rtx_REG,
3524 start_sequence and end_sequence and instead assert that no rtl is
3527 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3529 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3530 order of HIGH and CONST checks.
3532 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3534 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3535 if the destination isn't an SSA_NAME.
3537 2018-01-09 Richard Biener <rguenther@suse.de>
3539 PR tree-optimization/83668
3540 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3542 (canonicalize_loop_form): ... here, renamed from ...
3543 (canonicalize_loop_closed_ssa_form): ... this and amended to
3544 swap successor edges for loop exit blocks to make us use
3545 the RPO order we need for initial schedule generation.
3547 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3549 PR tree-optimization/64811
3550 * match.pd: When optimizing comparisons with Inf, avoid
3551 introducing or losing exceptions from comparisons with NaN.
3553 2018-01-09 Martin Liska <mliska@suse.cz>
3556 * asan.c (shadow_mem_size): Add gcc_assert.
3558 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3560 Don't save registers in main().
3563 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3564 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3565 * config/avr/avr.c (avr_set_current_function): Don't error if
3566 naked, OS_task or OS_main are specified at the same time.
3567 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3569 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3571 * common/config/avr/avr-common.c (avr_option_optimization_table):
3572 Switch on -mmain-is-OS_task for optimizing compilations.
3574 2018-01-09 Richard Biener <rguenther@suse.de>
3576 PR tree-optimization/83572
3577 * graphite.c: Include cfganal.h.
3578 (graphite_transform_loops): Connect infinite loops to exit
3579 and remove fake edges at the end.
3581 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3583 * ipa-inline.c (edge_badness): Revert accidental checkin.
3585 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3588 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3589 symbols; not inline clones.
3591 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3594 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3595 hard registers. Formatting fixes.
3597 PR preprocessor/83722
3598 * gcc.c (try_generate_repro): Pass
3599 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3600 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3603 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3604 Kito Cheng <kito.cheng@gmail.com>
3606 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3607 (riscv_leaf_function_p): Delete.
3608 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3610 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3612 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3614 (do_ifelse): New function.
3615 (do_isel): New function.
3616 (do_sub3): New function.
3617 (do_add3): New function.
3618 (do_load_mask_compare): New function.
3619 (do_overlap_load_compare): New function.
3620 (expand_compare_loop): New function.
3621 (expand_block_compare): Call expand_compare_loop() when appropriate.
3622 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3624 (-mblock-compare-inline-loop-limit): New option.
3626 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3629 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3630 Reverse order of second and third operands in first alternative.
3631 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3632 of first and second elements in UNSPEC_VPERMR vector.
3633 (altivec_expand_vec_perm_le): Likewise.
3635 2017-01-08 Jeff Law <law@redhat.com>
3637 PR rtl-optimizatin/81308
3638 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3639 (process_switch): If group_case_labels makes a change, then set
3641 (pass_convert_switch::execute): If a switch is converted, then
3642 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3644 PR rtl-optimization/81308
3645 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3648 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3650 PR target/83663 - Revert r255946
3651 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3652 generation for cases where splatting a value is not useful.
3653 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3654 across a vec_duplicate and a paradoxical subreg forming a vector
3655 mode to a vec_concat.
3657 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3659 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3660 -march=armv8.3-a variants.
3661 * config/arm/t-multilib: Likewise.
3662 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3664 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3666 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3668 (cceq_ior_compare_complement): Give it a name so I can use it, and
3669 change boolean_or_operator predicate to boolean_operator so it can
3670 be used to generate a crand.
3671 (eqne): New code iterator.
3672 (bd/bd_neg): New code_attrs.
3673 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3674 a single define_insn.
3675 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3676 decrement (bdnzt/bdnzf/bdzt/bdzf).
3677 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3678 with the new names of the branch decrement patterns, and added the
3679 names of the branch decrement conditional patterns.
3681 2018-01-08 Richard Biener <rguenther@suse.de>
3683 PR tree-optimization/83563
3684 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3687 2018-01-08 Richard Biener <rguenther@suse.de>
3690 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3692 2018-01-08 Richard Biener <rguenther@suse.de>
3694 PR tree-optimization/83685
3695 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3696 references to abnormals.
3698 2018-01-08 Richard Biener <rguenther@suse.de>
3701 * dwarf2out.c (output_indirect_strings): Handle empty
3702 skeleton_debug_str_hash.
3703 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3705 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3707 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3708 (emit_store_direct): Likewise.
3709 (arc_trampoline_adjust_address): Likewise.
3710 (arc_asm_trampoline_template): New function.
3711 (arc_initialize_trampoline): Use asm_trampoline_template.
3712 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3713 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3714 * config/arc/arc.md (flush_icache): Delete pattern.
3716 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3718 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3719 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3722 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3725 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3726 by not USED_FOR_TARGET.
3727 (make_pass_resolve_sw_modes): Likewise.
3729 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3731 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3734 2018-01-08 Richard Biener <rguenther@suse.de>
3737 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3739 2018-01-08 Richard Biener <rguenther@suse.de>
3742 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3744 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3747 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3748 basic blocks with a small number of successors.
3749 (convert_control_dep_chain_into_preds): Improve handling of
3751 (dump_predicates): Split apart into...
3752 (dump_pred_chain): ...here...
3753 (dump_pred_info): ...and here.
3754 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3755 (can_chain_union_be_invalidated_p): Improve check for invalidation
3757 (uninit_uses_cannot_happen): Avoid unnecessary if
3758 convert_control_dep_chain_into_preds yielded nothing.
3760 2018-01-06 Martin Sebor <msebor@redhat.com>
3762 PR tree-optimization/83640
3763 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3764 subtracting negative offset from size.
3765 (builtin_access::overlap): Adjust offset bounds of the access to fall
3766 within the size of the object if possible.
3768 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3770 PR rtl-optimization/83699
3771 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3772 extract_bit_field_as_subreg to cases in which the extracted
3773 value is also a vector.
3775 * lra-constraints.c (process_alt_operands): Test for the equivalence
3776 substitutions when detecting a possible reload cycle.
3778 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3781 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3782 by default if flag_selective_schedling{,2}. Formatting fixes.
3784 PR rtl-optimization/83682
3785 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
3786 if it has non-VECTOR_MODE element mode.
3787 (vec_duplicate_p): Likewise.
3790 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
3791 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
3793 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3796 * config/i386/i386-builtin.def
3797 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
3798 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
3799 Require also OPTION_MASK_ISA_AVX512F in addition to
3800 OPTION_MASK_ISA_GFNI.
3801 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
3802 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
3803 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
3804 to OPTION_MASK_ISA_GFNI.
3805 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
3806 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
3807 OPTION_MASK_ISA_AVX512BW.
3808 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
3809 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
3810 addition to OPTION_MASK_ISA_GFNI.
3811 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
3812 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
3813 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
3814 to OPTION_MASK_ISA_GFNI.
3815 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
3816 a requirement for all ISAs rather than any of them with a few
3818 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
3820 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
3821 bitmasks to be enabled with 3 exceptions, instead of requiring any
3822 enabled ISA with lots of exceptions.
3823 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
3824 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
3825 Change avx512bw in isa attribute to avx512f.
3826 * config/i386/sgxintrin.h: Add license boilerplate.
3827 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
3828 to __AVX512F__ and __AVX512VL to __AVX512VL__.
3829 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
3830 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
3832 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
3833 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
3834 temporarily sse2 rather than sse if not enabled already.
3837 * config/i386/sse.md (VI248_VLBW): Rename to ...
3838 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
3839 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
3840 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
3841 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
3842 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
3843 mode iterator instead of VI248_VLBW.
3845 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
3847 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
3848 (record_modified): Skip clobbers; add debug output.
3849 (param_change_prob): Use sreal frequencies.
3851 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3853 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
3854 punt for user-aligned variables.
3856 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3858 * tree-chrec.c (chrec_contains_symbols): Return true for
3861 2018-01-05 Sudakshina Das <sudi.das@arm.com>
3864 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
3865 of (x|y) == x for BICS pattern.
3867 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3869 PR tree-optimization/83605
3870 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
3871 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
3874 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
3876 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
3877 * config/epiphany/rtems.h: New file.
3879 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3880 Uros Bizjak <ubizjak@gmail.com>
3883 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
3884 QIreg_operand instead of register_operand predicate.
3885 * config/i386/i386.c (ix86_rop_should_change_byte_p,
3886 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
3887 comments instead of -fmitigate[-_]rop.
3889 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3892 * cgraphunit.c (symbol_table::compile): Switch to text_section
3893 before calling assembly_start debug hook.
3894 * run-rtl-passes.c (run_rtl_passes): Likewise.
3897 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3899 * tree-vrp.c (extract_range_from_binary_expr_1): Check
3900 range_int_cst_p rather than !symbolic_range_p before calling
3901 extract_range_from_multiplicative_op_1.
3903 2017-01-04 Jeff Law <law@redhat.com>
3905 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
3906 redundant test in assertion.
3908 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3910 * doc/rtl.texi: Document machine_mode wrapper classes.
3912 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3914 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
3917 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3919 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
3920 the VEC_PERM_EXPR fold to fail.
3922 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3925 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
3926 to switched_sections.
3928 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3931 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
3934 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
3937 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
3938 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
3940 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3943 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
3944 is BLKmode and bitpos not zero or mode change is needed.
3946 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3949 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
3952 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
3955 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
3956 instead of MULT rtx. Update all corresponding splitters.
3958 (*ssub<modesuffix>): Ditto.
3960 (*cmp_sadd_di): Update split patterns.
3961 (*cmp_sadd_si): Ditto.
3962 (*cmp_sadd_sidi): Ditto.
3963 (*cmp_ssub_di): Ditto.
3964 (*cmp_ssub_si): Ditto.
3965 (*cmp_ssub_sidi): Ditto.
3966 * config/alpha/predicates.md (const23_operand): New predicate.
3967 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
3968 Look for ASHIFT, not MULT inner operand.
3969 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
3971 2018-01-04 Martin Liska <mliska@suse.cz>
3973 PR gcov-profile/83669
3974 * gcov.c (output_intermediate_file): Add version to intermediate
3976 * doc/gcov.texi: Document new field 'version' in intermediate
3977 file format. Fix location of '-k' option of gcov command.
3979 2018-01-04 Martin Liska <mliska@suse.cz>
3982 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
3984 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3986 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
3988 2018-01-03 Martin Sebor <msebor@redhat.com>
3990 PR tree-optimization/83655
3991 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
3992 checking calls with invalid arguments.
3994 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3996 * tree-vect-stmts.c (vect_get_store_rhs): New function.
3997 (vectorizable_mask_load_store): Delete.
3998 (vectorizable_call): Return false for masked loads and stores.
3999 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4000 instead of gimple_assign_rhs1.
4001 (vectorizable_load): Handle IFN_MASK_LOAD.
4002 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4004 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4006 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4008 (vectorizable_mask_load_store): ...here.
4009 (vectorizable_load): ...and here.
4011 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4013 * tree-vect-stmts.c (vect_build_all_ones_mask)
4014 (vect_build_zero_merge_argument): New functions, split out from...
4015 (vectorizable_load): ...here.
4017 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4019 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4021 (vectorizable_mask_load_store): ...here.
4022 (vectorizable_store): ...and here.
4024 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4026 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4028 (vectorizable_mask_load_store): ...here.
4030 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4032 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4033 (vect_model_store_cost): Take a vec_load_store_type instead of a
4035 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4036 (vect_model_store_cost): Take a vec_load_store_type instead of a
4038 (vectorizable_mask_load_store): Update accordingly.
4039 (vectorizable_store): Likewise.
4040 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4042 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4044 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4045 IFN_MASK_LOAD calls here rather than...
4046 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4048 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4049 Alan Hayward <alan.hayward@arm.com>
4050 David Sherwood <david.sherwood@arm.com>
4052 * expmed.c (extract_bit_field_1): For vector extracts,
4053 fall back to extract_bit_field_as_subreg if vec_extract
4056 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4057 Alan Hayward <alan.hayward@arm.com>
4058 David Sherwood <david.sherwood@arm.com>
4060 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4061 they are variable or constant sized.
4062 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4063 slots for constant-sized data.
4065 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4066 Alan Hayward <alan.hayward@arm.com>
4067 David Sherwood <david.sherwood@arm.com>
4069 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4070 handling COND_EXPRs with boolean comparisons, try to find a better
4071 basis for the mask type than the boolean itself.
4073 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4075 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4076 is calculated and how it can be overridden.
4077 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4078 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4080 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4083 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4084 Alan Hayward <alan.hayward@arm.com>
4085 David Sherwood <david.sherwood@arm.com>
4087 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4088 Remove the mode argument.
4089 (aarch64_simd_valid_immediate): Remove the mode and inverse
4091 * config/aarch64/iterators.md (bitsize): New iterator.
4092 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4093 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4094 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4095 aarch64_simd_valid_immediate.
4096 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4097 (aarch64_reg_or_bic_imm): Likewise.
4098 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4099 with an insn_type enum and msl with a modifier_type enum.
4100 Replace element_width with a scalar_mode. Change the shift
4101 to unsigned int. Add constructors for scalar_float_mode and
4102 scalar_int_mode elements.
4103 (aarch64_vect_float_const_representable_p): Delete.
4104 (aarch64_can_const_movi_rtx_p)
4105 (aarch64_simd_scalar_immediate_valid_for_move)
4106 (aarch64_simd_make_constant): Update call to
4107 aarch64_simd_valid_immediate.
4108 (aarch64_advsimd_valid_immediate_hs): New function.
4109 (aarch64_advsimd_valid_immediate): Likewise.
4110 (aarch64_simd_valid_immediate): Remove mode and inverse
4111 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4112 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4113 and aarch64_float_const_representable_p on the result.
4114 (aarch64_output_simd_mov_immediate): Remove mode argument.
4115 Update call to aarch64_simd_valid_immediate and use of
4116 simd_immediate_info.
4117 (aarch64_output_scalar_simd_mov_immediate): Update call
4120 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4121 Alan Hayward <alan.hayward@arm.com>
4122 David Sherwood <david.sherwood@arm.com>
4124 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4125 (mode_nunits): Likewise CONST_MODE_NUNITS.
4126 * machmode.def (ADJUST_NUNITS): Document.
4127 * genmodes.c (mode_data::need_nunits_adj): New field.
4128 (blank_mode): Update accordingly.
4129 (adj_nunits): New variable.
4130 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4132 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4133 listed in adj_nunits.
4134 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4135 listed in adj_nunits. Don't emit case statements for such modes.
4136 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4137 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4138 nothing if adj_nunits is nonnull.
4139 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4140 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4141 (emit_mode_fbit): Update use of print_maybe_const_decl.
4142 (emit_move_size): Likewise. Treat the array as non-const
4144 (emit_mode_adjustments): Handle adj_nunits.
4146 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4148 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4149 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4150 (VECTOR_MODES): Use it.
4151 (make_vector_modes): Take the prefix as an argument.
4153 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4154 Alan Hayward <alan.hayward@arm.com>
4155 David Sherwood <david.sherwood@arm.com>
4157 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4158 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4159 for MODE_VECTOR_BOOL.
4160 * machmode.def (VECTOR_BOOL_MODE): Document.
4161 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4162 (make_vector_bool_mode): New function.
4163 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4165 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4166 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4168 * stor-layout.c (int_mode_for_mode): Likewise.
4169 * tree.c (build_vector_type_for_mode): Likewise.
4170 * varasm.c (output_constant_pool_2): Likewise.
4171 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4172 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4173 for MODE_VECTOR_BOOL.
4174 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4175 of mode class checks.
4176 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4177 instead of a list of mode class checks.
4178 (expand_vector_scalar_condition): Likewise.
4179 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4181 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4182 Alan Hayward <alan.hayward@arm.com>
4183 David Sherwood <david.sherwood@arm.com>
4185 * machmode.h (mode_size): Change from unsigned short to
4187 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4188 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4189 or if measurement_type is not polynomial.
4190 (fixed_size_mode::includes_p): Check for constant-sized modes.
4191 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4192 return a poly_uint16 rather than an unsigned short.
4193 (emit_mode_size): Change the type of mode_size from unsigned short
4194 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4195 (emit_mode_adjustments): Cope with polynomial vector sizes.
4196 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4198 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4200 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4201 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4202 * caller-save.c (setup_save_areas): Likewise.
4203 (replace_reg_with_saved_mem): Likewise.
4204 * calls.c (emit_library_call_value_1): Likewise.
4205 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4206 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4207 (gen_lowpart_for_combine): Likewise.
4208 * convert.c (convert_to_integer_1): Likewise.
4209 * cse.c (equiv_constant, cse_insn): Likewise.
4210 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4211 (cselib_subst_to_values): Likewise.
4212 * dce.c (word_dce_process_block): Likewise.
4213 * df-problems.c (df_word_lr_mark_ref): Likewise.
4214 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4215 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4216 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4217 (rtl_for_decl_location): Likewise.
4218 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4219 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4220 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4221 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4222 (expand_expr_real_1): Likewise.
4223 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4224 (pad_below): Likewise.
4225 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4226 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4227 * ira.c (get_subreg_tracking_sizes): Likewise.
4228 * ira-build.c (ira_create_allocno_objects): Likewise.
4229 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4230 (ira_sort_regnos_for_alter_reg): Likewise.
4231 * ira-costs.c (record_operand_costs): Likewise.
4232 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4233 (resolve_simple_move): Likewise.
4234 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4235 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4236 (lra_constraints): Likewise.
4237 (CONST_POOL_OK_P): Reject variable-sized modes.
4238 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4239 (add_pseudo_to_slot, lra_spill): Likewise.
4240 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4241 * optabs-query.c (get_best_extraction_insn): Likewise.
4242 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4243 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4244 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4245 * recog.c (offsettable_address_addr_space_p): Likewise.
4246 * regcprop.c (maybe_mode_change): Likewise.
4247 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4248 * regrename.c (build_def_use): Likewise.
4249 * regstat.c (dump_reg_info): Likewise.
4250 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4251 (find_reloads, find_reloads_subreg_address): Likewise.
4252 * reload1.c (eliminate_regs_1): Likewise.
4253 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4254 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4255 (simplify_binary_operation_1, simplify_subreg): Likewise.
4256 * targhooks.c (default_function_arg_padding): Likewise.
4257 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4258 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4259 (verify_gimple_assign_ternary): Likewise.
4260 * tree-inline.c (estimate_move_cost): Likewise.
4261 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4262 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4263 (get_address_cost_ainc): Likewise.
4264 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4265 (vect_supportable_dr_alignment): Likewise.
4266 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4267 (vectorizable_reduction): Likewise.
4268 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4269 (vectorizable_operation, vectorizable_load): Likewise.
4270 * tree.c (build_same_sized_truth_vector_type): Likewise.
4271 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4272 * var-tracking.c (emit_note_insn_var_location): Likewise.
4273 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4274 (ADDR_VEC_ALIGN): Likewise.
4276 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4277 Alan Hayward <alan.hayward@arm.com>
4278 David Sherwood <david.sherwood@arm.com>
4280 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4282 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4283 or if measurement_type is polynomial.
4284 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4285 * combine.c (make_extraction): Likewise.
4286 * dse.c (find_shift_sequence): Likewise.
4287 * dwarf2out.c (mem_loc_descriptor): Likewise.
4288 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4289 (extract_bit_field, extract_low_bits): Likewise.
4290 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4291 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4292 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4293 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4294 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4295 * reload.c (find_reloads): Likewise.
4296 * reload1.c (alter_reg): Likewise.
4297 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4298 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4299 * tree-if-conv.c (predicate_mem_writes): Likewise.
4300 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4301 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4302 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4303 * valtrack.c (dead_debug_insert_temp): Likewise.
4304 * varasm.c (mergeable_constant_section): Likewise.
4305 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4307 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4308 Alan Hayward <alan.hayward@arm.com>
4309 David Sherwood <david.sherwood@arm.com>
4311 * expr.c (expand_assignment): Cope with polynomial mode sizes
4312 when assigning to a CONCAT.
4314 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4315 Alan Hayward <alan.hayward@arm.com>
4316 David Sherwood <david.sherwood@arm.com>
4318 * machmode.h (mode_precision): Change from unsigned short to
4320 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4322 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4323 or if measurement_type is not polynomial.
4324 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4325 in which the mode is already known to be a scalar_int_mode.
4326 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4327 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4329 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4330 for GET_MODE_PRECISION.
4331 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4332 for GET_MODE_PRECISION.
4333 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4335 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4336 (expand_field_assignment, make_extraction): Likewise.
4337 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4338 (get_last_value): Likewise.
4339 * convert.c (convert_to_integer_1): Likewise.
4340 * cse.c (cse_insn): Likewise.
4341 * expr.c (expand_expr_real_1): Likewise.
4342 * lra-constraints.c (simplify_operand_subreg): Likewise.
4343 * optabs-query.c (can_atomic_load_p): Likewise.
4344 * optabs.c (expand_atomic_load): Likewise.
4345 (expand_atomic_store): Likewise.
4346 * ree.c (combine_reaching_defs): Likewise.
4347 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4348 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4349 * tree.h (type_has_mode_precision_p): Likewise.
4350 * ubsan.c (instrument_si_overflow): Likewise.
4352 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4353 Alan Hayward <alan.hayward@arm.com>
4354 David Sherwood <david.sherwood@arm.com>
4356 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4357 polynomial numbers of units.
4358 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4359 (valid_vector_subparts_p): New function.
4360 (build_vector_type): Remove temporary shim and take the number
4361 of units as a poly_uint64 rather than an int.
4362 (build_opaque_vector_type): Take the number of units as a
4363 poly_uint64 rather than an int.
4364 * tree.c (build_vector_from_ctor): Handle polynomial
4365 TYPE_VECTOR_SUBPARTS.
4366 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4367 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4368 (build_vector_from_val): If the number of units is variable,
4369 use build_vec_duplicate_cst for constant operands and
4370 VEC_DUPLICATE_EXPR otherwise.
4371 (make_vector_type): Remove temporary is_constant ().
4372 (build_vector_type, build_opaque_vector_type): Take the number of
4373 units as a poly_uint64 rather than an int.
4374 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4376 * cfgexpand.c (expand_debug_expr): Likewise.
4377 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4378 (store_constructor, expand_expr_real_1): Likewise.
4379 (const_scalar_mask_from_tree): Likewise.
4380 * fold-const-call.c (fold_const_reduction): Likewise.
4381 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4382 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4383 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4384 (fold_relational_const): Likewise.
4385 (native_interpret_vector): Likewise. Change the size from an
4386 int to an unsigned int.
4387 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4388 TYPE_VECTOR_SUBPARTS.
4389 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4390 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4391 duplicating a non-constant operand into a variable-length vector.
4392 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4393 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4394 * ipa-icf.c (sem_variable::equals): Likewise.
4395 * match.pd: Likewise.
4396 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4397 * print-tree.c (print_node): Likewise.
4398 * stor-layout.c (layout_type): Likewise.
4399 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4400 * tree-cfg.c (verify_gimple_comparison): Likewise.
4401 (verify_gimple_assign_binary): Likewise.
4402 (verify_gimple_assign_ternary): Likewise.
4403 (verify_gimple_assign_single): Likewise.
4404 * tree-pretty-print.c (dump_generic_node): Likewise.
4405 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4406 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4407 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4408 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4409 (vect_shift_permute_load_chain): Likewise.
4410 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4411 (expand_vector_condition, optimize_vector_constructor): Likewise.
4412 (lower_vec_perm, get_compute_type): Likewise.
4413 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4414 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4415 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4416 (vect_recog_mask_conversion_pattern): Likewise.
4417 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4418 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4419 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4420 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4421 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4422 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4423 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4424 (supportable_widening_operation): Likewise.
4425 (supportable_narrowing_operation): Likewise.
4426 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4428 * varasm.c (output_constant): Likewise.
4430 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4431 Alan Hayward <alan.hayward@arm.com>
4432 David Sherwood <david.sherwood@arm.com>
4434 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4435 so that both the length == 3 and length != 3 cases set up their
4436 own permute vectors. Add comments explaining why we know the
4437 number of elements is constant.
4438 (vect_permute_load_chain): Likewise.
4440 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4441 Alan Hayward <alan.hayward@arm.com>
4442 David Sherwood <david.sherwood@arm.com>
4444 * machmode.h (mode_nunits): Change from unsigned char to
4446 (ONLY_FIXED_SIZE_MODES): New macro.
4447 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4448 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4449 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4451 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4452 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4453 or if measurement_type is not polynomial.
4454 * genmodes.c (ZERO_COEFFS): New macro.
4455 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4457 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4458 Use ZERO_COEFFS when emitting initializers.
4459 * data-streamer.h (bp_pack_poly_value): New function.
4460 (bp_unpack_poly_value): Likewise.
4461 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4462 for GET_MODE_NUNITS.
4463 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4464 for GET_MODE_NUNITS.
4465 * tree.c (make_vector_type): Remove temporary shim and make
4466 the real function take the number of units as a poly_uint64
4468 (build_vector_type_for_mode): Handle polynomial nunits.
4469 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4470 * emit-rtl.c (const_vec_series_p_1): Likewise.
4471 (gen_rtx_CONST_VECTOR): Likewise.
4472 * fold-const.c (test_vec_duplicate_folding): Likewise.
4473 * genrecog.c (validate_pattern): Likewise.
4474 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4475 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4476 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4477 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4478 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4479 * rtlanal.c (subreg_get_info): Likewise.
4480 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4481 (vect_grouped_load_supported): Likewise.
4482 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4483 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4484 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4485 (simplify_const_unary_operation, simplify_binary_operation_1)
4486 (simplify_const_binary_operation, simplify_ternary_operation)
4487 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4488 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4489 instead of CONST_VECTOR_NUNITS.
4490 * varasm.c (output_constant_pool_2): Likewise.
4491 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4492 explicit-encoded elements in the XVEC for variable-length vectors.
4494 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4496 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
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 * coretypes.h (fixed_size_mode): Declare.
4503 (fixed_size_mode_pod): New typedef.
4504 * builtins.h (target_builtins::x_apply_args_mode)
4505 (target_builtins::x_apply_result_mode): Change type to
4506 fixed_size_mode_pod.
4507 * builtins.c (apply_args_size, apply_result_size, result_vector)
4508 (expand_builtin_apply_args_1, expand_builtin_apply)
4509 (expand_builtin_return): Update accordingly.
4511 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4513 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4514 * cselib.c (cselib_hash_rtx): Likewise.
4515 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4516 CONST_VECTOR encoding.
4518 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4519 Jeff Law <law@redhat.com>
4522 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4523 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4524 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4525 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4528 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4529 explicitly probe *sp in a noreturn function if there were any callee
4530 register saves or frame pointer is needed.
4532 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4535 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4536 BLKmode for ternary, binary or unary expressions.
4539 * var-tracking.c (delete_vta_debug_insn): New inline function.
4540 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4541 insns from get_insns () to NULL instead of each bb separately.
4542 Use delete_vta_debug_insn. No longer static.
4543 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4544 delete_vta_debug_insns callers.
4545 * rtl.h (delete_vta_debug_insns): Declare.
4546 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4547 instead of variable_tracking_main.
4549 2018-01-03 Martin Sebor <msebor@redhat.com>
4551 PR tree-optimization/83603
4552 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4553 arguments past the endof the argument list in functions declared
4554 without a prototype.
4555 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4556 Avoid checking when arguments are null.
4558 2018-01-03 Martin Sebor <msebor@redhat.com>
4561 * doc/extend.texi (attribute const): Fix a typo.
4562 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4563 issuing -Wsuggest-attribute for void functions.
4565 2018-01-03 Martin Sebor <msebor@redhat.com>
4567 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4568 offset_int::from instead of wide_int::to_shwi.
4569 (maybe_diag_overlap): Remove assertion.
4570 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4571 * gimple-ssa-sprintf.c (format_directive): Same.
4572 (parse_directive): Same.
4573 (sprintf_dom_walker::compute_format_length): Same.
4574 (try_substitute_return_value): Same.
4576 2017-01-03 Jeff Law <law@redhat.com>
4579 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4580 non-constant residual for zero at runtime and avoid probing in
4581 that case. Reorganize code for trailing problem to mirror handling
4584 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4586 PR tree-optimization/83501
4587 * tree-ssa-strlen.c (get_string_cst): New.
4588 (handle_char_store): Call get_string_cst.
4590 2018-01-03 Martin Liska <mliska@suse.cz>
4592 PR tree-optimization/83593
4593 * tree-ssa-strlen.c: Include tree-cfg.h.
4594 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4595 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4596 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4598 (strlen_dom_walker::before_dom_children): Call
4599 gimple_purge_dead_eh_edges. Dump tranformation with details
4601 (strlen_dom_walker::before_dom_children): Update call by adding
4602 new argument cleanup_eh.
4603 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4605 2018-01-03 Martin Liska <mliska@suse.cz>
4608 * cif-code.def (VARIADIC_THUNK): New enum value.
4609 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4612 2018-01-03 Jan Beulich <jbeulich@suse.com>
4614 * sse.md (mov<mode>_internal): Tighten condition for when to use
4615 vmovdqu<ssescalarsize> for TI and OI modes.
4617 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4619 Update copyright years.
4621 2018-01-03 Martin Liska <mliska@suse.cz>
4624 * ipa-visibility.c (function_and_variable_visibility): Skip
4625 functions with noipa attribure.
4627 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4629 * gcc.c (process_command): Update copyright notice dates.
4630 * gcov-dump.c (print_version): Ditto.
4631 * gcov.c (print_version): Ditto.
4632 * gcov-tool.c (print_version): Ditto.
4633 * gengtype.c (create_file): Ditto.
4634 * doc/cpp.texi: Bump @copying's copyright year.
4635 * doc/cppinternals.texi: Ditto.
4636 * doc/gcc.texi: Ditto.
4637 * doc/gccint.texi: Ditto.
4638 * doc/gcov.texi: Ditto.
4639 * doc/install.texi: Ditto.
4640 * doc/invoke.texi: Ditto.
4642 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4644 * vector-builder.h (vector_builder::m_full_nelts): Change from
4645 unsigned int to poly_uint64.
4646 (vector_builder::full_nelts): Update prototype accordingly.
4647 (vector_builder::new_vector): Likewise.
4648 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4649 (vector_builder::operator ==): Likewise.
4650 (vector_builder::finalize): Likewise.
4651 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4652 Take the number of elements as a poly_uint64 rather than an
4654 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4655 from unsigned int to poly_uint64.
4656 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4657 (vec_perm_indices::new_vector): Likewise.
4658 (vec_perm_indices::length): Likewise.
4659 (vec_perm_indices::nelts_per_input): Likewise.
4660 (vec_perm_indices::input_nelts): Likewise.
4661 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4662 number of elements per input as a poly_uint64 rather than an
4663 unsigned int. Use the original encoding for variable-length
4664 vectors, rather than clamping each individual element.
4665 For the second and subsequent elements in each pattern,
4666 clamp the step and base before clamping their sum.
4667 (vec_perm_indices::series_p): Handle polynomial element counts.
4668 (vec_perm_indices::all_in_range_p): Likewise.
4669 (vec_perm_indices_to_tree): Likewise.
4670 (vec_perm_indices_to_rtx): Likewise.
4671 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4672 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4673 (tree_vector_builder::new_binary_operation): Handle polynomial
4674 element counts. Return false if we need to know the number
4675 of elements at compile time.
4676 * fold-const.c (fold_vec_perm): Punt if the number of elements
4677 isn't known at compile time.
4679 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4681 * vec-perm-indices.h (vec_perm_builder): Change element type
4682 from HOST_WIDE_INT to poly_int64.
4683 (vec_perm_indices::element_type): Update accordingly.
4684 (vec_perm_indices::clamp): Handle polynomial element_types.
4685 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4686 (vec_perm_indices::all_in_range_p): Likewise.
4687 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4689 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4690 polynomial vec_perm_indices element types.
4691 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4692 * fold-const.c (fold_vec_perm): Likewise.
4693 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4694 * tree-vect-generic.c (lower_vec_perm): Likewise.
4695 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4696 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4697 element type to HOST_WIDE_INT.
4699 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4700 Alan Hayward <alan.hayward@arm.com>
4701 David Sherwood <david.sherwood@arm.com>
4703 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4704 rather than an int. Use plus_constant.
4705 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4706 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4708 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4709 Alan Hayward <alan.hayward@arm.com>
4710 David Sherwood <david.sherwood@arm.com>
4712 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4713 a HOST_WIDE_INT to a poly_int64.
4715 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4716 Alan Hayward <alan.hayward@arm.com>
4717 David Sherwood <david.sherwood@arm.com>
4719 * calls.c (load_register_parameters): Cope with polynomial
4720 mode sizes. Require a constant size for BLKmode parameters
4721 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4722 forces a parameter to be padded at the lsb end in order to
4723 fill a complete number of words, require the parameter size
4724 to be ordered wrt UNITS_PER_WORD.
4726 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4727 Alan Hayward <alan.hayward@arm.com>
4728 David Sherwood <david.sherwood@arm.com>
4730 * reload1.c (spill_stack_slot_width): Change element type
4731 from unsigned int to poly_uint64_pod.
4732 (alter_reg): Treat mode sizes as polynomial.
4734 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4735 Alan Hayward <alan.hayward@arm.com>
4736 David Sherwood <david.sherwood@arm.com>
4738 * reload.c (complex_word_subreg_p): New function.
4739 (reload_inner_reg_of_subreg, push_reload): Use it.
4741 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4742 Alan Hayward <alan.hayward@arm.com>
4743 David Sherwood <david.sherwood@arm.com>
4745 * lra-constraints.c (process_alt_operands): Reject matched
4746 operands whose sizes aren't ordered.
4747 (match_reload): Refer to this check here.
4749 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4750 Alan Hayward <alan.hayward@arm.com>
4751 David Sherwood <david.sherwood@arm.com>
4753 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4754 that the mode size is in the set {1, 2, 4, 8, 16}.
4756 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4757 Alan Hayward <alan.hayward@arm.com>
4758 David Sherwood <david.sherwood@arm.com>
4760 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4761 Use plus_constant instead of gen_rtx_PLUS.
4763 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4764 Alan Hayward <alan.hayward@arm.com>
4765 David Sherwood <david.sherwood@arm.com>
4767 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4768 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4769 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4770 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4771 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4772 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4773 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4774 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4775 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4776 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4778 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4779 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4780 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4781 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4782 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4783 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4784 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4785 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
4786 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
4787 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
4789 * expr.c (emit_move_resolve_push): Treat the input and result
4790 of PUSH_ROUNDING as a poly_int64.
4791 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
4792 (emit_push_insn): Likewise.
4793 * lra-eliminations.c (mark_not_eliminable): Likewise.
4794 * recog.c (push_operand): Likewise.
4795 * reload1.c (elimination_effects): Likewise.
4796 * rtlanal.c (nonzero_bits1): Likewise.
4797 * calls.c (store_one_arg): Likewise. Require the padding to be
4798 known at compile time.
4800 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4801 Alan Hayward <alan.hayward@arm.com>
4802 David Sherwood <david.sherwood@arm.com>
4804 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
4805 Use plus_constant instead of gen_rtx_PLUS.
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 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
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 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
4819 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
4820 via stack temporaries. Treat the mode size as polynomial too.
4822 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4823 Alan Hayward <alan.hayward@arm.com>
4824 David Sherwood <david.sherwood@arm.com>
4826 * expr.c (expand_expr_real_2): When handling conversions involving
4827 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
4828 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
4829 as a poly_uint64 too.
4831 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4832 Alan Hayward <alan.hayward@arm.com>
4833 David Sherwood <david.sherwood@arm.com>
4835 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
4837 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4838 Alan Hayward <alan.hayward@arm.com>
4839 David Sherwood <david.sherwood@arm.com>
4841 * combine.c (can_change_dest_mode): Handle polynomial
4842 REGMODE_NATURAL_SIZE.
4843 * expmed.c (store_bit_field_1): Likewise.
4844 * expr.c (store_constructor): Likewise.
4845 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
4846 and polynomial REGMODE_NATURAL_SIZE.
4847 (gen_lowpart_common): Likewise.
4848 * reginfo.c (record_subregs_of_mode): Likewise.
4849 * rtlanal.c (read_modify_subreg_p): Likewise.
4851 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4852 Alan Hayward <alan.hayward@arm.com>
4853 David Sherwood <david.sherwood@arm.com>
4855 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
4856 numbers of elements.
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 * match.pd: Cope with polynomial numbers of vector elements.
4864 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4865 Alan Hayward <alan.hayward@arm.com>
4866 David Sherwood <david.sherwood@arm.com>
4868 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
4869 in a POINTER_PLUS_EXPR.
4871 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4872 Alan Hayward <alan.hayward@arm.com>
4873 David Sherwood <david.sherwood@arm.com>
4875 * omp-simd-clone.c (simd_clone_subparts): New function.
4876 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
4877 (ipa_simd_modify_function_body): Likewise.
4879 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4880 Alan Hayward <alan.hayward@arm.com>
4881 David Sherwood <david.sherwood@arm.com>
4883 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
4884 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
4885 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
4886 (expand_vector_condition, vector_element): Likewise.
4887 (subparts_gt): New function.
4888 (get_compute_type): Use subparts_gt.
4889 (count_type_subparts): Delete.
4890 (expand_vector_operations_1): Use subparts_gt instead of
4891 count_type_subparts.
4893 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4894 Alan Hayward <alan.hayward@arm.com>
4895 David Sherwood <david.sherwood@arm.com>
4897 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
4898 (vect_compile_time_alias): ...this new function. Do the calculation
4899 on poly_ints rather than trees.
4900 (vect_prune_runtime_alias_test_list): Update call accordingly.
4902 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4903 Alan Hayward <alan.hayward@arm.com>
4904 David Sherwood <david.sherwood@arm.com>
4906 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
4908 (vect_schedule_slp_instance): Likewise.
4910 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4911 Alan Hayward <alan.hayward@arm.com>
4912 David Sherwood <david.sherwood@arm.com>
4914 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
4915 constant and extern definitions for variable-length vectors.
4916 (vect_get_constant_vectors): Note that the number of units
4917 is known to be constant.
4919 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4920 Alan Hayward <alan.hayward@arm.com>
4921 David Sherwood <david.sherwood@arm.com>
4923 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
4924 of units as polynomial. Choose between WIDE and NARROW based
4927 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4928 Alan Hayward <alan.hayward@arm.com>
4929 David Sherwood <david.sherwood@arm.com>
4931 * tree-vect-stmts.c (simd_clone_subparts): New function.
4932 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
4934 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4935 Alan Hayward <alan.hayward@arm.com>
4936 David Sherwood <david.sherwood@arm.com>
4938 * tree-vect-stmts.c (vectorizable_call): Treat the number of
4939 vectors as polynomial. Use build_index_vector for
4942 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4943 Alan Hayward <alan.hayward@arm.com>
4944 David Sherwood <david.sherwood@arm.com>
4946 * tree-vect-stmts.c (get_load_store_type): Treat the number of
4947 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
4948 for variable-length vectors.
4949 (vectorizable_mask_load_store): Treat the number of units as
4950 polynomial, asserting that it is constant if the condition has
4951 already been enforced.
4952 (vectorizable_store, vectorizable_load): Likewise.
4954 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4955 Alan Hayward <alan.hayward@arm.com>
4956 David Sherwood <david.sherwood@arm.com>
4958 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
4959 of units as polynomial. Punt if we can't tell at compile time
4960 which vector contains the final result.
4962 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4963 Alan Hayward <alan.hayward@arm.com>
4964 David Sherwood <david.sherwood@arm.com>
4966 * tree-vect-loop.c (vectorizable_induction): Treat the number
4967 of units as polynomial. Punt on SLP inductions. Use an integer
4968 VEC_SERIES_EXPR for variable-length integer reductions. Use a
4969 cast of such a series for variable-length floating-point
4972 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4973 Alan Hayward <alan.hayward@arm.com>
4974 David Sherwood <david.sherwood@arm.com>
4976 * tree.h (build_index_vector): Declare.
4977 * tree.c (build_index_vector): New function.
4978 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
4979 of units as polynomial, forcibly converting it to a constant if
4980 vectorizable_reduction has already enforced the condition.
4981 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
4982 to create a {1,2,3,...} vector.
4983 (vectorizable_reduction): Treat the number of units as polynomial.
4984 Choose vectype_in based on the largest scalar element size rather
4985 than the smallest number of units. Enforce the restrictions
4988 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4989 Alan Hayward <alan.hayward@arm.com>
4990 David Sherwood <david.sherwood@arm.com>
4992 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
4993 number of units as polynomial.
4995 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4996 Alan Hayward <alan.hayward@arm.com>
4997 David Sherwood <david.sherwood@arm.com>
4999 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5000 * target.def (autovectorize_vector_sizes): Return the vector sizes
5001 by pointer, using vector_sizes rather than a bitmask.
5002 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5003 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5004 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5006 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5007 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5008 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5009 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5010 * omp-general.c (omp_max_vf): Likewise.
5011 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5012 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5013 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5014 * tree-vect-slp.c (vect_slp_bb): Likewise.
5015 * doc/tm.texi: Regenerate.
5016 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5018 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5019 the vector size as a poly_uint64 rather than an unsigned int.
5020 (current_vector_size): Change from an unsigned int to a poly_uint64.
5021 (get_vectype_for_scalar_type): Update accordingly.
5022 * tree.h (build_truth_vector_type): Take the size and number of
5023 units as a poly_uint64 rather than an unsigned int.
5024 (build_vector_type): Add a temporary overload that takes
5025 the number of units as a poly_uint64 rather than an unsigned int.
5026 * tree.c (make_vector_type): Likewise.
5027 (build_truth_vector_type): Take the number of units as a poly_uint64
5028 rather than an unsigned int.
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 * target.def (get_mask_mode): Take the number of units and length
5035 as poly_uint64s rather than unsigned ints.
5036 * targhooks.h (default_get_mask_mode): Update accordingly.
5037 * targhooks.c (default_get_mask_mode): Likewise.
5038 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5039 * doc/tm.texi: Regenerate.
5041 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5042 Alan Hayward <alan.hayward@arm.com>
5043 David Sherwood <david.sherwood@arm.com>
5045 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5046 * omp-general.c (omp_max_vf): Likewise.
5047 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5048 (expand_omp_simd): Handle polynomial safelen.
5049 * omp-low.c (omplow_simd_context): Add a default constructor.
5050 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5051 (lower_rec_simd_input_clauses): Update accordingly.
5052 (lower_rec_input_clauses): Likewise.
5054 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5055 Alan Hayward <alan.hayward@arm.com>
5056 David Sherwood <david.sherwood@arm.com>
5058 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5059 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5060 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5061 (vect_analyze_slp_cost): Likewise.
5062 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5063 (vect_model_load_cost): Likewise.
5065 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5066 Alan Hayward <alan.hayward@arm.com>
5067 David Sherwood <david.sherwood@arm.com>
5069 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5070 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5071 from an unsigned int * to a poly_uint64_pod *.
5072 (calculate_unrolling_factor): New function.
5073 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5075 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5076 Alan Hayward <alan.hayward@arm.com>
5077 David Sherwood <david.sherwood@arm.com>
5079 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5080 from an unsigned int to a poly_uint64.
5081 (_loop_vec_info::slp_unrolling_factor): Likewise.
5082 (_loop_vec_info::vectorization_factor): Change from an int
5084 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5085 (vect_get_num_vectors): New function.
5086 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5087 (vect_get_num_copies): Use vect_get_num_vectors.
5088 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5089 to an unsigned int *.
5090 (vect_analyze_data_refs): Change min_vf from an int * to a
5092 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5093 than an unsigned HOST_WIDE_INT.
5094 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5095 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5096 to an unsigned int *.
5097 (vect_analyze_data_ref_dependences): Likewise.
5098 (vect_compute_data_ref_alignment): Handle polynomial vf.
5099 (vect_enhance_data_refs_alignment): Likewise.
5100 (vect_prune_runtime_alias_test_list): Likewise.
5101 (vect_shift_permute_load_chain): Likewise.
5102 (vect_supportable_dr_alignment): Likewise.
5103 (dependence_distance_ge_vf): Take the vectorization factor as a
5104 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5105 (vect_analyze_data_refs): Change min_vf from an int * to a
5107 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5108 vfm1 as a poly_uint64 rather than an int. Make the same change
5109 for the returned bound_scalar.
5110 (vect_gen_vector_loop_niters): Handle polynomial vf.
5111 (vect_do_peeling): Likewise. Update call to
5112 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5113 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5115 * tree-vect-loop.c (vect_determine_vectorization_factor)
5116 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5117 (vect_get_known_peeling_cost): Likewise.
5118 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5119 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5120 (vect_transform_loop): Likewise. Use the lowest possible VF when
5121 updating the upper bounds of the loop.
5122 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5124 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5125 polynomial unroll factors.
5126 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5127 (vect_make_slp_decision): Likewise.
5128 (vect_supported_load_permutation_p): Likewise, and polynomial
5130 (vect_analyze_slp_cost): Handle polynomial vf.
5131 (vect_slp_analyze_node_operations): Likewise.
5132 (vect_slp_analyze_bb_1): Likewise.
5133 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5134 than an unsigned HOST_WIDE_INT.
5135 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5136 (vectorizable_load): Handle polynomial vf.
5137 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5139 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5141 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5142 Alan Hayward <alan.hayward@arm.com>
5143 David Sherwood <david.sherwood@arm.com>
5145 * match.pd: Handle bit operations involving three constants
5146 and try to fold one pair.
5148 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5150 * tree-vect-loop-manip.c: Include gimple-fold.h.
5151 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5152 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5153 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5154 Add a path that uses a step of VF instead of 1, but disable it
5156 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5157 and niters_no_overflow parameters. Update calls to
5158 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5159 Create a new SSA name if the latter choses to use a ste other
5160 than zero, and return it via niters_vector_mult_vf_var.
5161 * tree-vect-loop.c (vect_transform_loop): Update calls to
5162 vect_do_peeling, vect_gen_vector_loop_niters and
5163 slpeel_make_loop_iterate_ntimes.
5164 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5165 (vect_gen_vector_loop_niters): Update declarations after above changes.
5167 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5169 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5170 128-bit round to integer instructions.
5171 (ceil<mode>2): Likewise.
5172 (btrunc<mode>2): Likewise.
5173 (round<mode>2): Likewise.
5175 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5177 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5178 unaligned VSX load/store on P8/P9.
5179 (expand_block_clear): Allow the use of unaligned VSX
5180 load/store on P8/P9.
5182 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5184 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5186 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5187 swap associated with both a load and a store.
5189 2018-01-02 Andrew Waterman <andrew@sifive.com>
5191 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5192 * config/riscv/riscv.md (clear_cache): Use it.
5194 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5196 * web.c: Remove out-of-date comment.
5198 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5200 * expr.c (fixup_args_size_notes): Check that any existing
5201 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5202 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5203 (emit_single_push_insn): ...here.
5205 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5207 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5208 (const_vector_encoded_nelts): New function.
5209 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5210 (const_vector_int_elt, const_vector_elt): Declare.
5211 * emit-rtl.c (const_vector_int_elt_1): New function.
5212 (const_vector_elt): Likewise.
5213 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5214 of CONST_VECTOR_ELT.
5216 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5218 * expr.c: Include rtx-vector-builder.h.
5219 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5220 directly on the tree encoding.
5221 (const_vector_from_tree): Likewise.
5222 * optabs.c: Include rtx-vector-builder.h.
5223 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5224 sequence of "u" values.
5225 * vec-perm-indices.c: Include rtx-vector-builder.h.
5226 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5227 directly on the vec_perm_indices encoding.
5229 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5231 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5232 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5233 * rtx-vector-builder.h: New file.
5234 * rtx-vector-builder.c: Likewise.
5235 * rtl.h (rtx_def::u2): Add a const_vector field.
5236 (CONST_VECTOR_NPATTERNS): New macro.
5237 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5238 (CONST_VECTOR_DUPLICATE_P): Likewise.
5239 (CONST_VECTOR_STEPPED_P): Likewise.
5240 (CONST_VECTOR_ENCODED_ELT): Likewise.
5241 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5242 (unwrap_const_vec_duplicate): Likewise.
5243 (const_vec_series_p): Check for a non-duplicated vector encoding.
5244 Say that the function only returns true for integer vectors.
5245 * emit-rtl.c: Include rtx-vector-builder.h.
5246 (gen_const_vec_duplicate_1): Delete.
5247 (gen_const_vector): Call gen_const_vec_duplicate instead of
5248 gen_const_vec_duplicate_1.
5249 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5250 (gen_const_vec_duplicate): Use rtx_vector_builder.
5251 (gen_const_vec_series): Likewise.
5252 (gen_rtx_CONST_VECTOR): Likewise.
5253 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5254 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5255 Build a new vector rather than modifying a CONST_VECTOR in-place.
5256 (handle_special_swappables): Update call accordingly.
5257 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5258 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5259 Build a new vector rather than modifying a CONST_VECTOR in-place.
5260 (handle_special_swappables): Update call accordingly.
5262 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5264 * simplify-rtx.c (simplify_const_binary_operation): Use
5265 CONST_VECTOR_ELT instead of XVECEXP.
5267 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5269 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5270 the selector elements to be different from the data elements
5271 if the selector is a VECTOR_CST.
5272 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5273 ssizetype for the selector.
5275 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5277 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5278 before testing each element individually.
5279 * tree-vect-generic.c (lower_vec_perm): Likewise.
5281 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5283 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5284 * selftest-run-tests.c (selftest::run_tests): Call it.
5285 * vector-builder.h (vector_builder::operator ==): New function.
5286 (vector_builder::operator !=): Likewise.
5287 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5288 (vec_perm_indices::all_from_input_p): New function.
5289 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5290 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5291 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5292 instead of reading the VECTOR_CST directly. Detect whether both
5293 vector inputs are the same before constructing the vec_perm_indices,
5294 and update the number of inputs argument accordingly. Use the
5295 utility functions added above. Only construct sel2 if we need to.
5297 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5299 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5300 the broadcast of the low byte.
5301 (expand_mult_highpart): Use an explicit encoding for the permutes.
5302 * optabs-query.c (can_mult_highpart_p): Likewise.
5303 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5304 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5305 (vectorizable_bswap): Likewise.
5306 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5307 explicit encoding for the power-of-2 permutes.
5308 (vect_permute_store_chain): Likewise.
5309 (vect_grouped_load_supported): Likewise.
5310 (vect_permute_load_chain): Likewise.
5312 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5314 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5315 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5316 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5317 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5318 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5319 (vect_gen_perm_mask_any): Likewise.
5321 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5323 * int-vector-builder.h: New file.
5324 * vec-perm-indices.h: Include int-vector-builder.h.
5325 (vec_perm_indices): Redefine as an int_vector_builder.
5326 (auto_vec_perm_indices): Delete.
5327 (vec_perm_builder): Redefine as a stand-alone class.
5328 (vec_perm_indices::vec_perm_indices): New function.
5329 (vec_perm_indices::clamp): Likewise.
5330 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5331 (vec_perm_indices::new_vector): New function.
5332 (vec_perm_indices::new_expanded_vector): Update for new
5333 vec_perm_indices class.
5334 (vec_perm_indices::rotate_inputs): New function.
5335 (vec_perm_indices::all_in_range_p): Operate directly on the
5336 encoded form, without computing elided elements.
5337 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5338 encoding. Update for new vec_perm_indices class.
5339 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5340 the given vec_perm_builder.
5341 (expand_vec_perm_var): Update vec_perm_builder constructor.
5342 (expand_mult_highpart): Use vec_perm_builder instead of
5343 auto_vec_perm_indices.
5344 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5345 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5346 or double series encoding as appropriate.
5347 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5348 vec_perm_indices instead of auto_vec_perm_indices.
5349 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5350 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5351 (vect_permute_store_chain): Likewise.
5352 (vect_grouped_load_supported): Likewise.
5353 (vect_permute_load_chain): Likewise.
5354 (vect_shift_permute_load_chain): Likewise.
5355 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5356 (vect_transform_slp_perm_load): Likewise.
5357 (vect_schedule_slp_instance): Likewise.
5358 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5359 (vectorizable_mask_load_store): Likewise.
5360 (vectorizable_bswap): Likewise.
5361 (vectorizable_store): Likewise.
5362 (vectorizable_load): Likewise.
5363 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5364 vec_perm_indices instead of auto_vec_perm_indices. Use
5365 tree_to_vec_perm_builder to read the vector from a tree.
5366 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5367 vec_perm_builder instead of a vec_perm_indices.
5368 (have_whole_vector_shift): Use vec_perm_builder and
5369 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5370 truncation to calc_vec_perm_mask_for_shift.
5371 (vect_create_epilog_for_reduction): Likewise.
5372 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5373 from auto_vec_perm_indices to vec_perm_indices.
5374 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5375 instead of changing individual elements.
5376 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5377 the vector in d.perm.
5378 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5379 from auto_vec_perm_indices to vec_perm_indices.
5380 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5381 instead of changing individual elements.
5382 (arm_vectorize_vec_perm_const): Use new_vector to install
5383 the vector in d.perm.
5384 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5385 Update vec_perm_builder constructor.
5386 (rs6000_expand_interleave): Likewise.
5387 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5388 (rs6000_expand_interleave): Likewise.
5390 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5392 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5393 to qimode could truncate the indices.
5394 * optabs.c (expand_vec_perm_var): Likewise.
5396 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5398 * Makefile.in (OBJS): Add vec-perm-indices.o.
5399 * vec-perm-indices.h: New file.
5400 * vec-perm-indices.c: Likewise.
5401 * target.h (vec_perm_indices): Replace with a forward class
5403 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5404 * optabs.h: Include vec-perm-indices.h.
5405 (expand_vec_perm): Delete.
5406 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5407 (expand_vec_perm_const): Declare.
5408 * target.def (vec_perm_const_ok): Replace with...
5409 (vec_perm_const): ...this new hook.
5410 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5411 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5412 * doc/tm.texi: Regenerate.
5413 * optabs.def (vec_perm_const): Delete.
5414 * doc/md.texi (vec_perm_const): Likewise.
5415 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5416 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5417 expand_vec_perm for constant permutation vectors. Assert that
5418 the mode of variable permutation vectors is the integer equivalent
5419 of the mode that is being permuted.
5420 * optabs-query.h (selector_fits_mode_p): Declare.
5421 * optabs-query.c: Include vec-perm-indices.h.
5422 (selector_fits_mode_p): New function.
5423 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5424 is defined, instead of checking whether the vec_perm_const_optab
5425 exists. Use targetm.vectorize.vec_perm_const instead of
5426 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5427 fit in the vector mode before using a variable permute.
5428 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5429 vec_perm_indices instead of an rtx.
5430 (expand_vec_perm): Replace with...
5431 (expand_vec_perm_const): ...this new function. Take the selector
5432 as a vec_perm_indices rather than an rtx. Also take the mode of
5433 the selector. Update call to shift_amt_for_vec_perm_mask.
5434 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5435 Use vec_perm_indices::new_expanded_vector to expand the original
5436 selector into bytes. Check whether the indices fit in the vector
5437 mode before using a variable permute.
5438 (expand_vec_perm_var): Make global.
5439 (expand_mult_highpart): Use expand_vec_perm_const.
5440 * fold-const.c: Includes vec-perm-indices.h.
5441 * tree-ssa-forwprop.c: Likewise.
5442 * tree-vect-data-refs.c: Likewise.
5443 * tree-vect-generic.c: Likewise.
5444 * tree-vect-loop.c: Likewise.
5445 * tree-vect-slp.c: Likewise.
5446 * tree-vect-stmts.c: Likewise.
5447 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5449 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5450 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5451 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5452 (aarch64_vectorize_vec_perm_const): ...this new function.
5453 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5454 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5455 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5456 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5457 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5458 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5459 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5461 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5462 check for NEON modes.
5463 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5464 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5465 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5466 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5468 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5469 the old VEC_PERM_CONST conditions.
5470 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5471 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5472 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5473 (ia64_vectorize_vec_perm_const_ok): Merge into...
5474 (ia64_vectorize_vec_perm_const): ...this new function.
5475 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5476 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5477 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5478 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5479 * config/mips/mips.c (mips_expand_vec_perm_const)
5480 (mips_vectorize_vec_perm_const_ok): Merge into...
5481 (mips_vectorize_vec_perm_const): ...this new function.
5482 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5483 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5484 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5485 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5486 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5487 (rs6000_expand_vec_perm_const): Delete.
5488 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5490 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5491 (altivec_expand_vec_perm_const_le): Take each operand individually.
5492 Operate on constant selectors rather than rtxes.
5493 (altivec_expand_vec_perm_const): Likewise. Update call to
5494 altivec_expand_vec_perm_const_le.
5495 (rs6000_expand_vec_perm_const): Delete.
5496 (rs6000_vectorize_vec_perm_const_ok): Delete.
5497 (rs6000_vectorize_vec_perm_const): New function.
5498 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5499 an element count and rtx array.
5500 (rs6000_expand_extract_even): Update call accordingly.
5501 (rs6000_expand_interleave): Likewise.
5502 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5503 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5504 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5505 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5506 (rs6000_expand_vec_perm_const): Delete.
5507 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5508 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5509 (altivec_expand_vec_perm_const_le): Take each operand individually.
5510 Operate on constant selectors rather than rtxes.
5511 (altivec_expand_vec_perm_const): Likewise. Update call to
5512 altivec_expand_vec_perm_const_le.
5513 (rs6000_expand_vec_perm_const): Delete.
5514 (rs6000_vectorize_vec_perm_const_ok): Delete.
5515 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5516 reference to the SPE evmerge intructions.
5517 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5518 an element count and rtx array.
5519 (rs6000_expand_extract_even): Update call accordingly.
5520 (rs6000_expand_interleave): Likewise.
5521 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5522 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5524 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5526 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5528 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5529 vector mode and that that mode matches the mode of the data
5531 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5532 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5533 directly using expand_vec_perm_1 when forcing selectors into
5535 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5537 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5539 * optabs-query.h (can_vec_perm_p): Delete.
5540 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5541 * optabs-query.c (can_vec_perm_p): Split into...
5542 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5543 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5544 particular selector is valid.
5545 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5546 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5547 (vect_grouped_load_supported): Likewise.
5548 (vect_shift_permute_load_chain): Likewise.
5549 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5550 (vect_transform_slp_perm_load): Likewise.
5551 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5552 (vectorizable_bswap): Likewise.
5553 (vect_gen_perm_mask_checked): Likewise.
5554 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5555 implementations of variable permutation vectors into account
5556 when deciding which selector to use.
5557 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5558 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5559 with a false third argument.
5560 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5561 to test whether the constant selector is valid and can_vec_perm_var_p
5562 to test whether a variable selector is valid.
5564 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5566 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5567 * optabs-query.c (can_vec_perm_p): Likewise.
5568 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5569 instead of vec_perm_indices.
5570 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5571 (vect_gen_perm_mask_checked): Likewise,
5572 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5573 (vect_gen_perm_mask_checked): Likewise,
5575 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5577 * optabs-query.h (qimode_for_vec_perm): Declare.
5578 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5579 (qimode_for_vec_perm): ...this new function.
5580 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5582 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5584 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5585 does not have a conditional at the top.
5587 2018-01-02 Richard Biener <rguenther@suse.de>
5589 * ipa-inline.c (big_speedup_p): Fix expression.
5591 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5594 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5597 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5601 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5602 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5603 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5604 cond_taken_branch_cost 3->4.
5606 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5608 PR tree-optimization/83581
5609 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5610 TODO_cleanup_cfg if any changes have been made.
5613 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5614 convert_modes if target mode has the right side, but different mode
5618 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5619 last argument when extracting from CONCAT. If either from_real or
5620 from_imag is NULL, use expansion through memory. If result is not
5621 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5622 the parts directly to inner mode, if even that fails, use expansion
5626 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5627 check for bswap in mode rather than HImode and use that in expand_unop
5630 Copyright (C) 2018 Free Software Foundation, Inc.
5632 Copying and distribution of this file, with or without modification,
5633 are permitted in any medium without royalty provided the copyright
5634 notice and this notice are preserved.