1 2018-01-29 Richard Biener <rguenther@suse.de>
3 PR tree-optimization/84057
4 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
5 removed paths when removing edges.
7 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
9 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
10 -mfunction-return=@var{choice}.
12 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
15 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
17 (layout::print_source_line): Likewise.
18 (test_get_line_width_without_trailing_whitespace): Add test cases.
20 2018-01-27 Jakub Jelinek <jakub@redhat.com>
23 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
26 2018-01-26 Jim Wilson <jimw@sifive.com>
28 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
30 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
33 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
35 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
36 and CMP + SUB-immediate -> SUBS.
38 2018-01-26 Martin Sebor <msebor@redhat.com>
40 PR tree-optimization/83896
41 * tree-ssa-strlen.c (get_string_len): Rename...
42 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
43 Avoid assuming length is constant.
44 (handle_char_store): Use HOST_WIDE_INT for string length.
46 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
49 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
50 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
52 2018-01-26 Richard Biener <rguenther@suse.de>
54 PR rtl-optimization/84003
55 * dse.c (record_store): Only record redundant stores when
56 the earlier store aliases at least all accesses the later one does.
58 2018-01-26 Jakub Jelinek <jakub@redhat.com>
60 PR rtl-optimization/83985
61 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
62 REG_CFA_RESTORE insns.
63 (delete_unmarked_insns): Don't ignore separate shrink wrapping
64 REG_CFA_RESTORE insns here.
67 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
68 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
70 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
72 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
73 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
75 (arc_override_options): Likewise.
76 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
78 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
80 * config/arc/arc.h (TARGET_DBNZ): Define.
81 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
82 properly set the tune attribute.
83 (dbnz): Use TARGET_DBNZ guard.
84 * config/arc/arc.opt (mtune): Add core3 option.
86 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
88 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
89 recognize new pic like addresses.
90 (arc_delegitimize_address): Clean up.
92 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
94 * config/arc/arc-arches.def: Option mrf16 valid for all
96 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
97 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
98 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
99 * config/arc/arc-tables.opt: Regenerate.
100 * config/arc/arc.c (arc_conditional_register_usage): Handle
101 reduced register file case.
102 (arc_file_start): Set must have build attributes.
103 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
105 * config/arc/arc.opt (mrf16): Add new option.
106 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
107 * config/arc/genmultilib.awk: Handle new mrf16 option.
108 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
109 * config/arc/t-multilib: Regenerate.
110 * doc/invoke.texi (ARC Options): Document mrf16 option.
112 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
114 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
115 * config/arc/arc.c (arc_handle_secure_attribute): New function.
116 (arc_attribute_table): Add 'secure_call' attribute.
117 (arc_print_operand): Print secure call operand.
118 (arc_function_ok_for_sibcall): Don't optimize tail calls when
120 (arc_is_secure_call_p): New function. * config/arc/arc.md
121 (call_i): Add support for sjli instruction.
122 (call_value_i): Likewise.
123 * config/arc/constraints.md (Csc): New constraint.
125 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
126 John Eric Martin <John.Martin@emmicro-us.com>
128 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
129 * config/arc/arc.c (_arc_jli_section): New struct.
130 (arc_jli_section): New type.
131 (rc_jli_sections): New static variable.
132 (arc_handle_jli_attribute): New function.
133 (arc_attribute_table): Add jli_always and jli_fixed attribute.
134 (arc_file_end): New function.
135 (TARGET_ASM_FILE_END): Define.
136 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
137 (arc_add_jli_section): New function.
138 (jli_call_scan): Likewise.
139 (arc_reorg): Call jli_call_scan.
140 (arc_output_addsi): Remove 'S' from printing asm operand.
141 (arc_is_jli_call_p): New function.
142 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
144 (movhi_insn): Likewise.
145 (movsi_insn): Likewise.
146 (movsi_set_cc_insn): Likewise.
147 (loadqi_update): Likewise.
148 (load_zeroextendqisi_update): Likewise.
149 (load_signextendqisi_update): Likewise.
150 (loadhi_update): Likewise.
151 (load_zeroextendhisi_update): Likewise.
152 (load_signextendhisi_update): Likewise.
153 (loadsi_update): Likewise.
154 (loadsf_update): Likewise.
155 (movsicc_insn): Likewise.
156 (bset_insn): Likewise.
157 (bxor_insn): Likewise.
158 (bclr_insn): Likewise.
159 (bmsk_insn): Likewise.
160 (bicsi3_insn): Likewise.
161 (cmpsi_cc_c_insn): Likewise.
162 (movsi_ne): Likewise.
163 (movsi_cond_exec): Likewise.
164 (clrsbsi2): Likewise.
175 (call_i): Remove 'S' asm letter, add jli instruction.
176 (call_value_i): Likewise.
177 * config/arc/arc.op (mjli-always): New option.
178 * config/arc/constraints.md (Cji): New constraint.
179 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
181 (subsf3_fpx): Likewise.
182 (mulsf3_fpx): Likewise.
183 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
185 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
187 * doc/invoke.texi (ARC): Document mjli-always option.
189 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
191 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
192 and use incw and decw where possible
193 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
195 2018-01-26 Richard Biener <rguenther@suse.de>
197 PR tree-optimization/81082
198 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
199 association if it requires casting to unsigned.
200 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
201 from fold_plusminus_mult_expr to catch important cases late when
202 range info is available.
204 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
206 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
207 * configure.ac (hidden_linkonce): New test.
208 * configure: Regenerate.
209 * config.in: Regenerate.
211 2018-01-26 Julia Koval <julia.koval@intel.com>
213 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
214 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
215 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
216 _mm_mask_bitshuffle_epi64_mask): Fix type.
217 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
218 USI_FTYPE_V4DI_V4DI_USI): Remove.
219 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
220 __builtin_ia32_vpshufbitqmb256_mask,
221 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
222 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
223 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
225 2018-01-26 Alan Modra <amodra@gmail.com>
228 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
229 UNSPEC_VBPERMQ. Sort other unspecs.
231 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
233 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
235 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
238 * predict.c (drop_profile): Do not push/pop cfun; update also
240 (handle_missing_profiles): Fix logic looking for zero profiles.
242 2018-01-25 Jakub Jelinek <jakub@redhat.com>
245 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
246 on functions with #pragma omp declare simd or functions with simd
248 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
249 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
250 Remove trailing \n from warning_at calls.
252 2018-01-25 Tom de Vries <tom@codesourcery.com>
255 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
256 for neutered workers.
258 2018-01-24 Joseph Myers <joseph@codesourcery.com>
261 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
262 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
264 2017-01-08 Jeff Law <law@redhat.com>
267 * i386.c (get_probe_interval): Move to earlier point.
268 (ix86_compute_frame_layout): If -fstack-clash-protection and
269 the frame is larger than the probe interval, then use pushes
270 to save registers rather than reg->mem moves.
271 (ix86_expand_prologue): Remove conditional for int_registers_saved
274 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
277 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
278 min/max for never referenced object.
280 2018-01-24 Jakub Jelinek <jakub@redhat.com>
283 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
285 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
286 attributes from DECL_ATTRIBUTES (decl) without affecting
287 DECL_ATTRIBUTES (current_function_decl).
288 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
289 functions with non-NULL DECL_ABSTRACT_ORIGIN.
291 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
293 PR tree-optimization/83979
294 * fold-const.c (fold_comparison): Use constant_boolean_node
295 instead of boolean_{true,false}_node.
297 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
299 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
302 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
304 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
305 Simplify the clause that sets the length attribute.
306 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
307 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
308 clause that sets the length attribute.
309 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
311 2018-01-24 Tom de Vries <tom@codesourcery.com>
314 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
315 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
316 Add strict parameter.
317 (prevent_branch_around_nothing): Insert dummy insn between branch to
318 label and label with no ptx insn inbetween.
319 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
321 2018-01-24 Tom de Vries <tom@codesourcery.com>
324 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
325 for neutered threads in warp.
326 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
328 2018-01-24 Richard Biener <rguenther@suse.de>
330 PR tree-optimization/83176
331 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
334 2018-01-24 Richard Biener <rguenther@suse.de>
336 PR tree-optimization/82819
337 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
338 code generating pluses that are no-ops in the target precision.
340 2018-01-24 Richard Biener <rguenther@suse.de>
343 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
345 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
347 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
348 to merge probabilities.
349 * predict.c (probably_never_executed): Also mark as cold functions
350 with global 0 profile and guessed local profile.
351 * profile-count.c (profile_probability::combine_with_count): New
353 * profile-count.h (profile_probability::operator*,
354 profile_probability::operator*=, profile_probability::operator/,
355 profile_probability::operator/=): Reduce precision to adjusted
356 and set value to guessed on contradictory divisions.
357 (profile_probability::combine_with_freq): Remove.
358 (profile_probability::combine_wiht_count): Declare.
359 (profile_count::force_nonzero):: Set to adjusted.
360 (profile_count::probability_in):: Set quality to adjusted.
361 * tree-ssa-tail-merge.c (replace_block_by): Use
364 2018-01-23 Andrew Waterman <andrew@sifive.com>
365 Jim Wilson <jimw@sifive.com>
367 * config/riscv/riscv.c (riscv_stack_boundary): New.
368 (riscv_option_override): Set riscv_stack_boundary. Handle
369 riscv_preferred_stack_boundary_arg.
370 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
371 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
372 (STACK_BOUNDARY): Set to riscv_stack_boundary.
373 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
374 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
375 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
377 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
380 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
381 of struct ix86_frame.
382 (ix86_expand_epilogue): Likewise. Add a local variable for
383 the reg_save_offset field in struct ix86_frame.
385 2018-01-23 Bin Cheng <bin.cheng@arm.com>
387 PR tree-optimization/82604
388 * tree-loop-distribution.c (enum partition_kind): New enum item
389 PKIND_PARTIAL_MEMSET.
390 (partition_builtin_p): Support above new enum item.
391 (generate_code_for_partition): Ditto.
392 (compute_access_range): Differentiate cases that equality can be
393 proven at all loops, the innermost loops or no loops.
394 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
395 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
396 (finalize_partitions, distribute_loop): Don't fuse partition of
397 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
398 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
401 2018-01-23 Martin Liska <mliska@suse.cz>
403 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
404 order to ignore the predictor.
405 (PRED_POLYMORPHIC_CALL): Likewise.
406 (PRED_RECURSIVE_CALL): Likewise.
408 2018-01-23 Martin Liska <mliska@suse.cz>
410 * tree-profile.c (tree_profiling): Print function header to
411 aware reader which function we are working on.
412 * value-prof.c (gimple_find_values_to_profile): Do not print
413 not interesting value histograms.
415 2018-01-23 Martin Liska <mliska@suse.cz>
417 * profile-count.h (enum profile_quality): Add
418 profile_uninitialized as the first value. Do not number values
419 as they are zero based.
420 (profile_count::verify): Update sanity check.
421 (profile_probability::verify): Likewise.
423 2018-01-23 Nathan Sidwell <nathan@acm.org>
425 * doc/invoke.texi (ffor-scope): Deprecate.
427 2018-01-23 David Malcolm <dmalcolm@redhat.com>
429 PR tree-optimization/83510
430 * domwalk.c (set_all_edges_as_executable): New function.
431 (dom_walker::dom_walker): Convert bool param
432 "skip_unreachable_blocks" to enum reachability. Move setup of
433 edge flags to set_all_edges_as_executable and only do it when
434 reachability is REACHABLE_BLOCKS.
435 * domwalk.h (enum dom_walker::reachability): New enum.
436 (dom_walker::dom_walker): Convert bool param
437 "skip_unreachable_blocks" to enum reachability.
438 (set_all_edges_as_executable): New decl.
439 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
440 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
442 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
443 but converting true to REACHABLE_BLOCKS.
444 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
446 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
447 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
448 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
450 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
451 if check_all_array_refs will be called.
453 2018-01-23 David Malcolm <dmalcolm@redhat.com>
455 * tree.c (selftest::test_location_wrappers): Add more test
458 2018-01-23 David Malcolm <dmalcolm@redhat.com>
460 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
461 (selftest::test_bit_in_range): Likewise.
463 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
466 * doc/sourcebuild.texi (vect_float): Say that the selector
467 only describes the situation when -funsafe-math-optimizations is on.
468 (vect_float_strict): Document.
470 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
472 PR tree-optimization/83965
473 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
474 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
475 instead of checking only for a reduction.
476 (vect_recog_widen_sum_pattern): Likewise.
478 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
480 * predict.c (probably_never_executed): Only use precise profile info.
481 (compute_function_frequency): Skip after inlining hack since we now
482 have quality checking.
484 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
486 * profile-count.h (profile_probability::very_unlikely,
487 profile_probability::unlikely, profile_probability::even): Set
488 precision to guessed.
490 2018-01-23 Richard Biener <rguenther@suse.de>
492 PR tree-optimization/83963
493 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
494 Properly terminate dominator walk when crossing the exit edge not
495 when visiting its source block.
497 2018-01-23 Jakub Jelinek <jakub@redhat.com>
500 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
501 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
503 2018-01-22 Jakub Jelinek <jakub@redhat.com>
505 PR tree-optimization/83957
506 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
507 semicolon after for body surrounded by braces.
509 PR tree-optimization/83081
510 * profile-count.h (profile_probability::split): New method.
511 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
512 Use profile_probability::split.
513 (do_compare_rtx_and_jump): Fix adjustment of probabilities
514 when splitting a single conditional jump into 2.
516 2018-01-22 David Malcolm <dmalcolm@redhat.com>
518 PR tree-optimization/69452
519 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
522 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
524 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
525 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
526 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
528 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
530 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
531 * config/rl78/rl78.md: New define_expand "movdi"
532 * config/rl78/rl78.c: New function definition rl78_split_movdi
534 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
537 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
539 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
540 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
541 128-bit to produce an UNSPEC move to get the double word with the
542 signbit and then a shift directly to do signbit.
543 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
544 implementation with a new version that just does either a direct
545 move or a regular move. Move memory interface to separate insns.
546 Move insns so they are next to the expander.
547 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
548 with signbit move. Split big and little endian case.
549 (signbit<mode>2_dm_mem_le): Likewise.
550 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
551 (signbit<mode>2_dm2): Likewise.
553 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
555 * config/rl78/rl78.md: New define_expand "anddi3".
557 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
559 * config/rl78/rl78.md: New define_expand "umindi3".
561 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
563 * config/rl78/rl78.md: New define_expand "smindi3".
565 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
567 * config/rl78/rl78.md: New define_expand "smaxdi3".
569 2018-01-22 Carl Love <cel@us.ibm.com>
571 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
572 LVX_V1TI): Add macro expansion.
573 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
574 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
575 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
576 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
577 Change check to determine if the instruction is a byte reversing
578 entry. Fix typo in comment.
579 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
580 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
581 Add def_builtin calls for new builtins.
582 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
583 Add define_insn expansion.
585 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
587 * config/rl78/rl78.md: New define_expand "umaxdi3".
589 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
591 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
592 for non-QImode registers
594 2018-01-22 Richard Biener <rguenther@suse.de>
596 PR tree-optimization/83963
597 * graphite-scop-detection.c (scop_detection::get_sese): Delay
598 including the loop exit block.
599 (scop_detection::merge_sese): Likewise.
600 (scop_detection::add_scop): Do it here instead.
602 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
604 * doc/sourcebuild.texi (arm_softfloat): Document.
606 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
609 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
610 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
611 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
613 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
614 David Edelsohn <dje.gcc@gmail.com>
617 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
618 Change "crset eq" to "crset 2".
619 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
620 (*call_indirect_aix<mode>_nospec): Likewise.
621 (*call_value_indirect_aix<mode>_nospec): Likewise.
622 (*call_indirect_elfv2<mode>_nospec): Likewise.
623 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
624 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
625 change assembly output from . to $.
626 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
627 (indirect_jump<mode>_nospec): Change assembly output from . to $.
628 (*tablejump<mode>_internal1_nospec): Likewise.
630 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
633 * config/sh/sh_optimize_sett_clrt.cc:
634 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
636 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
638 PR tree-optimization/83940
639 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
640 offset_dt to vect_constant_def rather than vect_unknown_def_type.
641 (vect_check_load_store_mask): Add a mask_dt_out parameter and
642 use it to pass back the definition type.
643 (vect_check_store_rhs): Likewise rhs_dt_out.
644 (vect_build_gather_load_calls): Add a mask_dt argument and use
645 it instead of a call to vect_is_simple_use.
646 (vectorizable_store): Update calls to vect_check_load_store_mask
647 and vect_check_store_rhs. Use the dt returned by the latter instead
648 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
649 instead of calls to vect_is_simple_use. Pass the scalar rather
650 than the vector operand to vect_is_simple_use when handling
651 second and subsequent copies of an rhs value.
652 (vectorizable_load): Update calls to vect_check_load_store_mask
653 and vect_build_gather_load_calls. Use the cached mask_dt and
654 gs_info.offset_dt instead of calls to vect_is_simple_use.
656 2018-01-20 Jakub Jelinek <jakub@redhat.com>
659 * tree-emutls.c: Include gimplify.h.
660 (lower_emutls_2): New function.
661 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
662 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
663 it before further processing.
666 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
667 UINTVAL (trueop1) instead of INTVAL (op1).
669 2018-01-19 Jakub Jelinek <jakub@redhat.com>
673 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
674 INCOMING_FRAME_SP_OFFSET if not defined.
675 (scan_trace): Add ENTRY argument. If true and
676 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
677 emit a note to adjust the CFA offset.
678 (create_cfi_notes): Adjust scan_trace callers.
679 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
680 INCOMING_FRAME_SP_OFFSET in the CIE.
681 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
682 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
684 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
685 * doc/tm.texi: Regenerated.
687 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
689 PR rtl-optimization/83147
690 * lra-constraints.c (remove_inheritance_pseudos): Use
691 lra_substitute_pseudo_within_insn.
693 2018-01-19 Tom de Vries <tom@codesourcery.com>
694 Cesar Philippidis <cesar@codesourcery.com>
697 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
699 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
702 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
703 spaces for function labels.
705 2018-01-19 Martin Liska <mliska@suse.cz>
707 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
708 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
709 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
710 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
711 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
712 (PRED_CONST_RETURN): Change from 69 to 65.
713 (PRED_NULL_RETURN): Change from 91 to 71.
714 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
715 (PRED_LOOP_GUARD): Change from 66 to 73.
717 2018-01-19 Martin Liska <mliska@suse.cz>
719 * predict.c (predict_insn_def): Add new assert.
720 (struct branch_predictor): Change type to signed integer.
721 (test_prediction_value_range): Amend test to cover
723 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
724 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
725 (PRED_LOOP_ITERATIONS_MAX): Likewise.
726 (PRED_LOOP_IV_COMPARE): Likewise.
727 * predict.h (PROB_UNINITIALIZED): Define new constant.
729 2018-01-19 Martin Liska <mliska@suse.cz>
731 * predict.c (dump_prediction): Add new format for
732 analyze_brprob.py script which is enabled with -details
734 * profile-count.h (precise_p): New function.
736 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
738 PR tree-optimization/83922
739 * tree-vect-loop.c (vect_verify_full_masking): Return false if
740 there are no statements that need masking.
741 (vect_active_double_reduction_p): New function.
742 (vect_analyze_loop_operations): Use it when handling phis that
743 are not in the loop header.
745 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
747 PR tree-optimization/83914
748 * tree-vect-loop.c (vectorizable_induction): Don't convert
749 init_expr or apply the peeling adjustment for inductions
750 that are nested within the vectorized loop.
752 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
754 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
757 2018-01-18 Jakub Jelinek <jakub@redhat.com>
761 * function.h (gimplify_parameters): Add gimple_seq * argument.
762 * function.c: Include gimple.h and options.h.
763 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
764 for the added local temporaries if needed.
765 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
766 if there are any parameter cleanups, wrap whole body into a
767 try/finally with the cleanups.
769 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
772 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
773 Use GET_MODE_CLASS for scalar floating point.
775 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
779 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
780 Fix call of call_cgraph_insertion_hooks.
782 2018-01-18 Martin Sebor <msebor@redhat.com>
784 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
786 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
789 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
792 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
795 * common.opt: (-ffile-prefix-map): New option.
796 * opts.c (common_handle_option): Defer it.
797 * opts-global.c (handle_common_deferred_options): Handle it.
798 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
799 * file-prefix-map.h: New file.
800 (remap_debug_filename, add_debug_prefix_map): ...here.
801 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
802 * final.c (debug_prefix_map, add_debug_prefix_map
803 remap_debug_filename): Move to...
804 * file-prefix-map.c: New file.
805 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
806 generalize, get rid of alloca(), use strrchr() instead of strchr().
807 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
808 Implement in terms of add_prefix_map().
809 (remap_macro_filename, remap_debug_filename): Implement in term of
811 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
812 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
813 * dbxout.c: Include file-prefix-map.h.
814 * varasm.c: Likewise.
815 * vmsdbgout.c: Likewise.
816 * xcoffout.c: Likewise.
817 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
818 * doc/cppopts.texi (-fmacro-prefix-map): Document.
819 * doc/invoke.texi (-ffile-prefix-map): Document.
820 (-fdebug-prefix-map): Update description.
822 2018-01-18 Martin Liska <mliska@suse.cz>
824 * config/i386/i386.c (indirect_thunk_name): Document that also
826 (output_indirect_thunk): Document why both instructions
827 (pause and lfence) are generated.
829 2018-01-18 Richard Biener <rguenther@suse.de>
831 PR tree-optimization/83887
832 * graphite-scop-detection.c
833 (scop_detection::get_nearest_dom_with_single_entry): Remove.
834 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
835 (scop_detection::merge_sese): Re-implement with a flood-fill
836 algorithm that properly finds a SESE region if it exists.
838 2018-01-18 Jakub Jelinek <jakub@redhat.com>
841 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
842 pointer_diff optimizations use view_convert instead of convert.
844 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
846 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
847 Generate different code for -mno-speculate-indirect-jumps.
848 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
849 (*call_indirect_aix<mode>): Disable for
850 -mno-speculate-indirect-jumps.
851 (*call_indirect_aix<mode>_nospec): New define_insn.
852 (*call_value_indirect_aix<mode>): Disable for
853 -mno-speculate-indirect-jumps.
854 (*call_value_indirect_aix<mode>_nospec): New define_insn.
855 (*sibcall_nonlocal_sysv<mode>): Generate different code for
856 -mno-speculate-indirect-jumps.
857 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
859 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
861 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
862 long double type, set the flags for noting the default long double
863 type, even if we don't pass or return a long double type.
865 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
868 * ipa-inline.c (flatten_function): Do not overwrite final inlining
871 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
873 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
874 support for merge[hl].
875 (fold_mergehl_helper): New helper function.
876 (tree-vector-builder.h): New #include for tree_vector_builder usage.
877 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
878 (altivec_vmrglw_direct): Add xxmrglw insn.
880 2018-01-17 Andrew Waterman <andrew@sifive.com>
882 * config/riscv/riscv.c (riscv_conditional_register_usage): If
883 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
885 2018-01-17 David Malcolm <dmalcolm@redhat.com>
888 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
889 call the lto_location_cache before reading the
890 DECL_SOURCE_LOCATION of the types.
892 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
893 Richard Sandiford <richard.sandiford@linaro.org>
895 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
896 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
897 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
898 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
899 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
901 * config/aarch64/constraints.md (aarch64_movti_operand):
903 * config/aarch64/predicates.md (Uti): Add new constraint.
905 2018-01-17 Carl Love <cel@us.ibm.com>
906 * config/rs6000/vsx.md (define_expand xl_len_r,
907 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
908 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
910 (define_expand, define_insn): Move the shift left from the
911 define_insn to the define_expand for lxvl and stxvl instructions.
912 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
913 and XL_LEN_R definitions to PURE.
915 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
917 * config/i386/i386.c (indirect_thunk_name): Declare regno
918 as unsigned int. Compare regno with INVALID_REGNUM.
919 (output_indirect_thunk): Ditto.
920 (output_indirect_thunk_function): Ditto.
921 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
922 in the call to output_indirect_thunk_function.
924 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
927 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
928 rather than the size of inner_type to determine the stack slot size
929 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
931 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
934 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
937 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
939 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
940 endian Linux systems to optionally enable multilibs for selecting
941 the long double type if the user configured an explicit type.
942 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
943 have no long double multilibs if not defined.
944 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
945 warn if the user used -mabi={ieee,ibm}longdouble and we built
946 multilibs for long double.
947 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
948 appropriate multilib option.
949 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
951 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
952 for building long double multilibs.
953 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
955 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
957 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
960 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
962 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
965 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
968 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
971 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
973 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
974 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
976 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
978 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
979 different rtl trees depending on TARGET_64BIT.
980 (rs6000_gen_lvx): Likewise.
982 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
984 * config/visium/visium.md (nop): Tweak comment.
985 (hazard_nop): Likewise.
987 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
989 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
990 -mspeculate-indirect-jumps.
991 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
992 for -mno-speculate-indirect-jumps.
993 (*call_indirect_elfv2<mode>_nospec): New define_insn.
994 (*call_value_indirect_elfv2<mode>): Disable for
995 -mno-speculate-indirect-jumps.
996 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
997 (indirect_jump): Emit different RTL for
998 -mno-speculate-indirect-jumps.
999 (*indirect_jump<mode>): Disable for
1000 -mno-speculate-indirect-jumps.
1001 (*indirect_jump<mode>_nospec): New define_insn.
1002 (tablejump): Emit different RTL for
1003 -mno-speculate-indirect-jumps.
1004 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1005 (tablejumpsi_nospec): New define_expand.
1006 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1007 (tablejumpdi_nospec): New define_expand.
1008 (*tablejump<mode>_internal1): Disable for
1009 -mno-speculate-indirect-jumps.
1010 (*tablejump<mode>_internal1_nospec): New define_insn.
1011 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1014 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1016 * caller-save.c (insert_save): Drop unnecessary parameter. All
1019 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1020 Richard Biener <rguenth@suse.de>
1023 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1024 return early, inline manually is_gimple_sizepos. Make sure if we
1025 call gimplify_expr we don't end up with a gimple constant.
1026 * tree.c (variably_modified_type_p): Don't return true for
1027 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1028 * gimplify.h (is_gimple_sizepos): Remove.
1030 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1032 PR tree-optimization/83857
1033 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1034 vectorizable_live_operation for pure SLP statements.
1035 (vectorizable_live_operation): Handle PHIs.
1037 2018-01-16 Richard Biener <rguenther@suse.de>
1039 PR tree-optimization/83867
1040 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1041 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1043 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1046 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1047 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1048 If off is not INTEGER_CST, issue a may not be aligned warning
1049 rather than isn't aligned. Use isn%'t rather than isn't.
1050 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1052 <case MULT_EXPR>: Improve the case when bottom and one of the
1053 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1054 operand, in that case check if the other operand is multiple of
1055 bottom divided by the INTEGER_CST operand.
1057 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1060 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1061 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1062 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1063 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1064 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1065 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1066 (pa_function_arg_size): New function.
1068 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1070 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1071 in a separate statement.
1073 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1075 PR tree-optimization/83847
1076 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1077 group gathers and scatters.
1079 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1081 PR rtl-optimization/86620
1082 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1084 PR rtl-optimization/83213
1085 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1086 to last if both are JUMP_INSNs.
1088 PR tree-optimization/83843
1089 * gimple-ssa-store-merging.c
1090 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1091 store_immediate_info for bswap/nop orig_stores.
1093 2018-01-15 Andrew Waterman <andrew@sifive.com>
1095 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1097 <UDIV>: Increase cost if !TARGET_DIV.
1099 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1101 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1102 (define_attr "cr_logical_3op"): New.
1103 (cceq_ior_compare): Adjust.
1104 (cceq_ior_compare_complement): Adjust.
1105 (*cceq_rev_compare): Adjust.
1106 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1107 (is_cracked_insn): Adjust.
1108 (insn_must_be_first_in_group): Adjust.
1109 * config/rs6000/40x.md: Adjust.
1110 * config/rs6000/440.md: Adjust.
1111 * config/rs6000/476.md: Adjust.
1112 * config/rs6000/601.md: Adjust.
1113 * config/rs6000/603.md: Adjust.
1114 * config/rs6000/6xx.md: Adjust.
1115 * config/rs6000/7450.md: Adjust.
1116 * config/rs6000/7xx.md: Adjust.
1117 * config/rs6000/8540.md: Adjust.
1118 * config/rs6000/cell.md: Adjust.
1119 * config/rs6000/e300c2c3.md: Adjust.
1120 * config/rs6000/e500mc.md: Adjust.
1121 * config/rs6000/e500mc64.md: Adjust.
1122 * config/rs6000/e5500.md: Adjust.
1123 * config/rs6000/e6500.md: Adjust.
1124 * config/rs6000/mpc.md: Adjust.
1125 * config/rs6000/power4.md: Adjust.
1126 * config/rs6000/power5.md: Adjust.
1127 * config/rs6000/power6.md: Adjust.
1128 * config/rs6000/power7.md: Adjust.
1129 * config/rs6000/power8.md: Adjust.
1130 * config/rs6000/power9.md: Adjust.
1131 * config/rs6000/rs64.md: Adjust.
1132 * config/rs6000/titan.md: Adjust.
1134 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1136 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1137 ix86_indirect_branch_register logic.
1139 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1141 * config/i386/constraints.md (Bs): Update
1142 ix86_indirect_branch_register check. Don't check
1143 ix86_indirect_branch_register with GOT_memory_operand.
1145 * config/i386/predicates.md (GOT_memory_operand): Don't check
1146 ix86_indirect_branch_register here.
1147 (GOT32_symbol_operand): Likewise.
1149 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1151 * config/i386/predicates.md (constant_call_address_operand):
1152 Rewrite ix86_indirect_branch_register logic.
1153 (sibcall_insn_operand): Likewise.
1155 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1157 * config/i386/constraints.md (Bs): Replace
1158 ix86_indirect_branch_thunk_register with
1159 ix86_indirect_branch_register.
1161 * config/i386/i386.md (indirect_jump): Likewise.
1162 (tablejump): Likewise.
1163 (*sibcall_memory): Likewise.
1164 (*sibcall_value_memory): Likewise.
1165 Peepholes of indirect call and jump via memory: Likewise.
1166 * config/i386/i386.opt: Likewise.
1167 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1168 (GOT_memory_operand): Likewise.
1169 (call_insn_operand): Likewise.
1170 (sibcall_insn_operand): Likewise.
1171 (GOT32_symbol_operand): Likewise.
1173 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1176 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1177 type rather than type addr's type points to.
1178 (expand_omp_atomic_mutex): Likewise.
1179 (expand_omp_atomic): Likewise.
1181 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1184 * config/i386/i386.c (output_indirect_thunk_function): Use
1185 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1186 for __x86_return_thunk.
1188 2018-01-15 Richard Biener <rguenther@suse.de>
1191 * expmed.c (extract_bit_field_1): Fix typo.
1193 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1196 * config/arm/iterators.md (VF): New mode iterator.
1197 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1198 Remove integer-related logic from pattern.
1199 (neon_vabd<mode>_3): Likewise.
1201 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1204 * common.opt (fstrict-overflow): No longer an alias.
1205 (fwrapv-pointer): New option.
1206 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1207 also for pointer types based on flag_wrapv_pointer.
1208 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1209 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1210 opts->x_flag_wrapv got set.
1211 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1212 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1213 POINTER_TYPE_OVERFLOW_UNDEFINED.
1214 * match.pd: Likewise in address comparison pattern.
1215 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1217 2018-01-15 Richard Biener <rguenther@suse.de>
1220 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1221 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1222 Reset type names to their identifier if their TYPE_DECL doesn't
1223 have linkage (and thus is used for ODR and devirt).
1224 (save_debug_info_for_decl): Remove.
1225 (save_debug_info_for_type): Likewise.
1226 (add_tree_to_fld_list): Adjust.
1227 * tree-pretty-print.c (dump_generic_node): Make dumping of
1228 type names more robust.
1230 2018-01-15 Richard Biener <rguenther@suse.de>
1232 * BASE-VER: Bump to 8.0.1.
1234 2018-01-14 Martin Sebor <msebor@redhat.com>
1237 * builtins.c (check_access): Avoid warning when the no-warning bit
1240 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1242 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1243 * ira-color (allocno_hard_regs_compare): Likewise.
1245 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1248 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1249 Use .pushsection/.popsection.
1251 2018-01-14 Martin Sebor <msebor@redhat.com>
1254 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1256 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1258 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1259 entry from extra_headers.
1260 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1261 extra_headers, make the list bitwise identical to the i?86-*-* one.
1263 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1265 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1266 -mcmodel=large with -mindirect-branch=thunk,
1267 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1268 -mfunction-return=thunk-extern.
1269 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1270 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1271 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1273 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1275 * config/i386/i386.c (print_reg): Print the name of the full
1276 integer register without '%'.
1277 (ix86_print_operand): Handle 'V'.
1278 * doc/extend.texi: Document 'V' modifier.
1280 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1282 * config/i386/constraints.md (Bs): Disallow memory operand for
1283 -mindirect-branch-register.
1285 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1286 (GOT_memory_operand): Likewise.
1287 (call_insn_operand): Likewise.
1288 (sibcall_insn_operand): Likewise.
1289 (GOT32_symbol_operand): Likewise.
1290 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1291 for -mindirect-branch-register.
1292 (tablejump): Likewise.
1293 (*sibcall_memory): Likewise.
1294 (*sibcall_value_memory): Likewise.
1295 Disallow peepholes of indirect call and jump via memory for
1296 -mindirect-branch-register.
1297 (*call_pop): Replace m with Bw.
1298 (*call_value_pop): Likewise.
1299 (*sibcall_pop_memory): Replace m with Bs.
1300 * config/i386/i386.opt (mindirect-branch-register): New option.
1301 * doc/invoke.texi: Document -mindirect-branch-register option.
1303 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1305 * config/i386/i386-protos.h (ix86_output_function_return): New.
1306 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1307 set function_return_type.
1308 (indirect_thunk_name): Add ret_p to indicate thunk for function
1310 (output_indirect_thunk_function): Pass false to
1311 indirect_thunk_name.
1312 (ix86_output_indirect_branch_via_reg): Likewise.
1313 (ix86_output_indirect_branch_via_push): Likewise.
1314 (output_indirect_thunk_function): Create alias for function
1315 return thunk if regno < 0.
1316 (ix86_output_function_return): New function.
1317 (ix86_handle_fndecl_attribute): Handle function_return.
1318 (ix86_attribute_table): Add function_return.
1319 * config/i386/i386.h (machine_function): Add
1320 function_return_type.
1321 * config/i386/i386.md (simple_return_internal): Use
1322 ix86_output_function_return.
1323 (simple_return_internal_long): Likewise.
1324 * config/i386/i386.opt (mfunction-return=): New option.
1325 (indirect_branch): Mention -mfunction-return=.
1326 * doc/extend.texi: Document function_return function attribute.
1327 * doc/invoke.texi: Document -mfunction-return= option.
1329 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1331 * config/i386/i386-opts.h (indirect_branch): New.
1332 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1333 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1334 with local indirect jump when converting indirect call and jump.
1335 (ix86_set_indirect_branch_type): New.
1336 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1337 (indirectlabelno): New.
1338 (indirect_thunk_needed): Likewise.
1339 (indirect_thunk_bnd_needed): Likewise.
1340 (indirect_thunks_used): Likewise.
1341 (indirect_thunks_bnd_used): Likewise.
1342 (INDIRECT_LABEL): Likewise.
1343 (indirect_thunk_name): Likewise.
1344 (output_indirect_thunk): Likewise.
1345 (output_indirect_thunk_function): Likewise.
1346 (ix86_output_indirect_branch_via_reg): Likewise.
1347 (ix86_output_indirect_branch_via_push): Likewise.
1348 (ix86_output_indirect_branch): Likewise.
1349 (ix86_output_indirect_jmp): Likewise.
1350 (ix86_code_end): Call output_indirect_thunk_function if needed.
1351 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1353 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1354 (ix86_attribute_table): Add indirect_branch.
1355 * config/i386/i386.h (machine_function): Add indirect_branch_type
1356 and has_local_indirect_jump.
1357 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1359 (tablejump): Likewise.
1360 (*indirect_jump): Use ix86_output_indirect_jmp.
1361 (*tablejump_1): Likewise.
1362 (simple_return_indirect_internal): Likewise.
1363 * config/i386/i386.opt (mindirect-branch=): New option.
1364 (indirect_branch): New.
1367 (thunk-inline): Likewise.
1368 (thunk-extern): Likewise.
1369 * doc/extend.texi: Document indirect_branch function attribute.
1370 * doc/invoke.texi: Document -mindirect-branch= option.
1372 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1375 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1377 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1379 * ipa-inline.c (want_inline_small_function_p): Return false if
1380 inlining has already failed with CIF_FINAL_ERROR.
1381 (update_caller_keys): Call want_inline_small_function_p before
1383 (update_callee_keys): Likewise.
1385 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1387 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1389 (rs6000_quadword_masked_address_p): Likewise.
1390 (quad_aligned_load_p): Likewise.
1391 (quad_aligned_store_p): Likewise.
1392 (const_load_sequence_p): Add comment to describe the outer-most loop.
1393 (mimic_memory_attributes_and_flags): New function.
1394 (rs6000_gen_stvx): Likewise.
1395 (replace_swapped_aligned_store): Likewise.
1396 (rs6000_gen_lvx): Likewise.
1397 (replace_swapped_aligned_load): Likewise.
1398 (replace_swapped_load_constant): Capitalize argument name in
1399 comment describing this function.
1400 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1401 and stores that access quad-word aligned addresses and replace
1402 with stvx or lvx instructions when appropriate.
1403 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1404 New function prototype.
1405 (rs6000_quadword_masked_address_p): Likewise.
1406 (rs6000_gen_lvx): Likewise.
1407 (rs6000_gen_stvx): Likewise.
1408 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1409 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1410 when memory address is aligned.
1411 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1412 this split to select lvx instruction when memory address is aligned.
1413 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1414 instruction when memory address is aligned.
1415 (*vsx_le_perm_load_v16qi): Likewise.
1416 (four unnamed splitters): Modify to select the stvx instruction
1417 when memory is aligned.
1419 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1421 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1422 which appears in the queue multiple times.
1424 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1425 Alan Hayward <alan.hayward@arm.com>
1426 David Sherwood <david.sherwood@arm.com>
1428 * tree-vectorizer.h (vec_lower_bound): New structure.
1429 (_loop_vec_info): Add check_nonzero and lower_bounds.
1430 (LOOP_VINFO_CHECK_NONZERO): New macro.
1431 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1432 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1433 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1434 fields. Make seg_len the distance travelled, not including the
1436 (dr_direction_indicator): Declare.
1437 (dr_zero_step_indicator): Likewise.
1438 (dr_known_forward_stride_p): Likewise.
1439 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1441 (runtime_alias_check_p): Allow runtime alias checks with
1443 (operator ==): Compare access_size and align.
1444 (prune_runtime_alias_test_list): Rework for new distinction between
1445 the access_size and seg_len.
1446 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1448 (get_segment_min_max): New function.
1449 (create_intersect_range_checks): Use it.
1450 (dr_step_indicator): New function.
1451 (dr_direction_indicator): Likewise.
1452 (dr_zero_step_indicator): Likewise.
1453 (dr_known_forward_stride_p): Likewise.
1454 * tree-loop-distribution.c (data_ref_segment_size): Return
1455 DR_STEP * (niters - 1).
1456 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1458 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1459 (vect_preserves_scalar_order_p): New function, split out from...
1460 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1461 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1462 (vect_vfa_access_size): New function.
1463 (vect_vfa_align): Likewise.
1464 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1465 (dump_lower_bound): New function.
1466 (vect_check_lower_bound): Likewise.
1467 (vect_small_gap_p): Likewise.
1468 (vectorizable_with_step_bound_p): Likewise.
1469 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1470 depencies if the vectorization factor is 1. Convert the checks
1471 for nonzero steps into checks on the bounds of DR_STEP. Try using
1472 a bunds check for variable steps if the minimum required step is
1473 relatively small. Update calls to the dr_with_seg_len
1474 constructor and to vect_compile_time_alias.
1475 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1477 (vect_loop_versioning): Call it.
1478 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1480 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1482 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1483 Alan Hayward <alan.hayward@arm.com>
1484 David Sherwood <david.sherwood@arm.com>
1486 * doc/sourcebuild.texi (vect_scatter_store): Document.
1487 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1489 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1491 * genopinit.c (main): Add supports_vec_scatter_store and
1492 supports_vec_scatter_store_cached to target_optabs.
1493 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1494 IFN_MASK_SCATTER_STORE.
1495 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1497 * internal-fn.h (internal_store_fn_p): Declare.
1498 (internal_fn_stored_value_index): Likewise.
1499 * internal-fn.c (scatter_store_direct): New macro.
1500 (expand_scatter_store_optab_fn): New function.
1501 (direct_scatter_store_optab_supported_p): New macro.
1502 (internal_store_fn_p): New function.
1503 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1504 IFN_MASK_SCATTER_STORE.
1505 (internal_fn_mask_index): Likewise.
1506 (internal_fn_stored_value_index): New function.
1507 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1509 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1510 * optabs-query.c (supports_vec_scatter_store_p): New function.
1511 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1512 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1513 true for scatter stores.
1514 (vect_gather_scatter_fn_p): Handle scatter stores too.
1515 (vect_check_gather_scatter): Consider using scatter stores if
1516 supports_vec_scatter_store_p.
1517 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1519 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1520 internal_fn_stored_value_index.
1521 (check_load_store_masking): Handle scatter stores too.
1522 (vect_get_store_rhs): Make public.
1523 (vectorizable_call): Use internal_store_fn_p.
1524 (vectorizable_store): Handle scatter store internal functions.
1525 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1526 when deciding whether the end of the group has been reached.
1527 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1528 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1529 (mask_scatter_store<mode>): New insns.
1531 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1532 Alan Hayward <alan.hayward@arm.com>
1533 David Sherwood <david.sherwood@arm.com>
1535 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1536 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1537 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1539 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1540 Use vect_truncate_gather_scatter_offset if we can't treat the
1541 operation as a normal gather load or scatter store.
1542 (get_group_load_store_type): Take the gather_scatter_info
1543 as argument. Try using a gather load or scatter store for
1544 single-element groups.
1545 (get_load_store_type): Update calls to get_group_load_store_type
1546 and vect_use_strided_gather_scatters_p.
1548 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1549 Alan Hayward <alan.hayward@arm.com>
1550 David Sherwood <david.sherwood@arm.com>
1552 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1553 optional tree argument.
1554 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1556 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1557 but continue to use the current value as a fallback.
1558 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1559 to compare the updates.
1560 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1561 (get_load_store_type): Use it when handling a strided access.
1562 (vect_get_strided_load_store_ops): New function.
1563 (vect_get_data_ptr_increment): Likewise.
1564 (vectorizable_load): Handle strided gather loads. Always pass
1565 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1567 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1568 Alan Hayward <alan.hayward@arm.com>
1569 David Sherwood <david.sherwood@arm.com>
1571 * doc/md.texi (gather_load@var{m}): Document.
1572 (mask_gather_load@var{m}): Likewise.
1573 * genopinit.c (main): Add supports_vec_gather_load and
1574 supports_vec_gather_load_cached to target_optabs.
1575 * optabs-tree.c (init_tree_optimization_optabs): Use
1576 ggc_cleared_alloc to allocate target_optabs.
1577 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1578 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1580 * internal-fn.h (internal_load_fn_p): Declare.
1581 (internal_gather_scatter_fn_p): Likewise.
1582 (internal_fn_mask_index): Likewise.
1583 (internal_gather_scatter_fn_supported_p): Likewise.
1584 * internal-fn.c (gather_load_direct): New macro.
1585 (expand_gather_load_optab_fn): New function.
1586 (direct_gather_load_optab_supported_p): New macro.
1587 (direct_internal_fn_optab): New function.
1588 (internal_load_fn_p): Likewise.
1589 (internal_gather_scatter_fn_p): Likewise.
1590 (internal_fn_mask_index): Likewise.
1591 (internal_gather_scatter_fn_supported_p): Likewise.
1592 * optabs-query.c (supports_at_least_one_mode_p): New function.
1593 (supports_vec_gather_load_p): Likewise.
1594 * optabs-query.h (supports_vec_gather_load_p): Declare.
1595 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1596 and memory_type field.
1597 (NUM_PATTERNS): Bump to 15.
1598 * tree-vect-data-refs.c: Include internal-fn.h.
1599 (vect_gather_scatter_fn_p): New function.
1600 (vect_describe_gather_scatter_call): Likewise.
1601 (vect_check_gather_scatter): Try using internal functions for
1602 gather loads. Recognize existing calls to a gather load function.
1603 (vect_analyze_data_refs): Consider using gather loads if
1604 supports_vec_gather_load_p.
1605 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1606 (vect_get_gather_scatter_offset_type): Likewise.
1607 (vect_convert_mask_for_vectype): Likewise.
1608 (vect_add_conversion_to_patterm): Likewise.
1609 (vect_try_gather_scatter_pattern): Likewise.
1610 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1611 (vect_vect_recog_func_ptrs): Add it.
1612 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1613 internal_fn_mask_index and internal_gather_scatter_fn_p.
1614 (check_load_store_masking): Take the gather_scatter_info as an
1615 argument and handle gather loads.
1616 (vect_get_gather_scatter_ops): New function.
1617 (vectorizable_call): Check internal_load_fn_p.
1618 (vectorizable_load): Likewise. Handle gather load internal
1620 (vectorizable_store): Update call to check_load_store_masking.
1621 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1622 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1623 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1624 (aarch64_gather_scale_operand_d): New predicates.
1625 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1626 (mask_gather_load<mode>): New insns.
1628 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1629 Alan Hayward <alan.hayward@arm.com>
1630 David Sherwood <david.sherwood@arm.com>
1632 * optabs.def (fold_left_plus_optab): New optab.
1633 * doc/md.texi (fold_left_plus_@var{m}): Document.
1634 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1635 * internal-fn.c (fold_left_direct): Define.
1636 (expand_fold_left_optab_fn): Likewise.
1637 (direct_fold_left_optab_supported_p): Likewise.
1638 * fold-const-call.c (fold_const_fold_left): New function.
1639 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1640 * tree-parloops.c (valid_reduction_p): New function.
1641 (gather_scalar_reductions): Use it.
1642 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1643 (vect_finish_replace_stmt): Declare.
1644 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1645 (needs_fold_left_reduction_p): New function, split out from...
1646 (vect_is_simple_reduction): ...here. Accept reductions that
1647 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1648 (vect_force_simple_reduction): Also store the reduction type in
1649 the assignment's STMT_VINFO_REDUC_TYPE.
1650 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1651 (merge_with_identity): New function.
1652 (vect_expand_fold_left): Likewise.
1653 (vectorize_fold_left_reduction): Likewise.
1654 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1655 scalar phi in place for it. Check for target support and reject
1656 cases that would reassociate the operation. Defer the transform
1657 phase to vectorize_fold_left_reduction.
1658 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1659 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1660 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1662 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1664 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1665 call to ifc_temp_var.
1667 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1668 Alan Hayward <alan.hayward@arm.com>
1669 David Sherwood <david.sherwood@arm.com>
1671 * target.def (legitimize_address_displacement): Take the original
1672 offset as a poly_int.
1673 * targhooks.h (default_legitimize_address_displacement): Update
1675 * targhooks.c (default_legitimize_address_displacement): Likewise.
1676 * doc/tm.texi: Regenerate.
1677 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1678 as an argument, moving assert of ad->disp == ad->disp_term to...
1679 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1680 Try calling targetm.legitimize_address_displacement before expanding
1681 the address rather than afterwards, and adjust for the new interface.
1682 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1683 Match the new hook interface. Handle SVE addresses.
1684 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1687 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1689 * Makefile.in (OBJS): Add early-remat.o.
1690 * target.def (select_early_remat_modes): New hook.
1691 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1692 * doc/tm.texi: Regenerate.
1693 * targhooks.h (default_select_early_remat_modes): Declare.
1694 * targhooks.c (default_select_early_remat_modes): New function.
1695 * timevar.def (TV_EARLY_REMAT): New timevar.
1696 * passes.def (pass_early_remat): New pass.
1697 * tree-pass.h (make_pass_early_remat): Declare.
1698 * early-remat.c: New file.
1699 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1701 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1703 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1704 Alan Hayward <alan.hayward@arm.com>
1705 David Sherwood <david.sherwood@arm.com>
1707 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1708 vfm1 with a bound_epilog parameter.
1709 (vect_do_peeling): Update calls accordingly, and move the prologue
1710 call earlier in the function. Treat the base bound_epilog as 0 for
1711 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1712 this base when peeling for gaps.
1713 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1714 with fully-masked loops.
1715 (vect_estimate_min_profitable_iters): Handle the single peeled
1716 iteration in that case.
1718 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1719 Alan Hayward <alan.hayward@arm.com>
1720 David Sherwood <david.sherwood@arm.com>
1722 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1723 single-element interleaving even if the size is not a power of 2.
1724 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1725 accesses for single-element interleaving if the group size is
1728 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1729 Alan Hayward <alan.hayward@arm.com>
1730 David Sherwood <david.sherwood@arm.com>
1732 * doc/md.texi (fold_extract_last_@var{m}): Document.
1733 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1734 * optabs.def (fold_extract_last_optab): New optab.
1735 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1736 * internal-fn.c (fold_extract_direct): New macro.
1737 (expand_fold_extract_optab_fn): Likewise.
1738 (direct_fold_extract_optab_supported_p): Likewise.
1739 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1740 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1741 EXTRACT_LAST_REDUCTION.
1742 (get_initial_def_for_reduction): Do not create an initial vector
1743 for EXTRACT_LAST_REDUCTION reductions.
1744 (vectorizable_reduction): Leave the scalar phi in place for
1745 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1746 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1747 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1748 transform phase to vectorizable_condition.
1749 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1751 (vect_finish_stmt_generation): ...here.
1752 (vect_finish_replace_stmt): New function.
1753 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1754 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1756 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1758 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1759 Alan Hayward <alan.hayward@arm.com>
1760 David Sherwood <david.sherwood@arm.com>
1762 * doc/md.texi (extract_last_@var{m}): Document.
1763 * optabs.def (extract_last_optab): New optab.
1764 * internal-fn.def (EXTRACT_LAST): New internal function.
1765 * internal-fn.c (cond_unary_direct): New macro.
1766 (expand_cond_unary_optab_fn): Likewise.
1767 (direct_cond_unary_optab_supported_p): Likewise.
1768 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1769 loops using EXTRACT_LAST.
1770 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1771 (extract_last_<mode>): ...this optab.
1772 (vec_extract<mode><Vel>): Update accordingly.
1774 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1775 Alan Hayward <alan.hayward@arm.com>
1776 David Sherwood <david.sherwood@arm.com>
1778 * target.def (empty_mask_is_expensive): New hook.
1779 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
1780 * doc/tm.texi: Regenerate.
1781 * targhooks.h (default_empty_mask_is_expensive): Declare.
1782 * targhooks.c (default_empty_mask_is_expensive): New function.
1783 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
1784 if the target says that empty masks are expensive.
1785 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
1787 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
1789 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1790 Alan Hayward <alan.hayward@arm.com>
1791 David Sherwood <david.sherwood@arm.com>
1793 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
1794 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
1795 (vect_use_loop_mask_for_alignment_p): New function.
1796 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
1797 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
1798 niters_skip argument. Make sure that the first niters_skip elements
1799 of the first iteration are inactive.
1800 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
1801 Update call to vect_set_loop_masks_directly.
1802 (get_misalign_in_elems): New function, split out from...
1803 (vect_gen_prolog_loop_niters): ...here.
1804 (vect_update_init_of_dr): Take a code argument that specifies whether
1805 the adjustment should be added or subtracted.
1806 (vect_update_init_of_drs): Likewise.
1807 (vect_prepare_for_masked_peels): New function.
1808 (vect_do_peeling): Skip prologue peeling if we're using a mask
1809 instead. Update call to vect_update_inits_of_drs.
1810 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1812 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
1813 alignment. Do not include the number of peeled iterations in
1814 the minimum threshold in that case.
1815 (vectorizable_induction): Adjust the start value down by
1816 LOOP_VINFO_MASK_SKIP_NITERS iterations.
1817 (vect_transform_loop): Call vect_prepare_for_masked_peels.
1818 Take the number of skipped iterations into account when calculating
1820 * tree-vect-stmts.c (vect_gen_while_not): New function.
1822 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1823 Alan Hayward <alan.hayward@arm.com>
1824 David Sherwood <david.sherwood@arm.com>
1826 * doc/sourcebuild.texi (vect_fully_masked): Document.
1827 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
1829 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
1831 (vect_analyze_loop_2): ...here. Don't check the vectorization
1832 factor against the number of loop iterations if the loop is
1835 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1836 Alan Hayward <alan.hayward@arm.com>
1837 David Sherwood <david.sherwood@arm.com>
1839 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
1840 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
1841 (dump_groups): Update accordingly.
1842 (iv_use::mem_type): New member variable.
1843 (address_p): New function.
1844 (record_use): Add a mem_type argument and initialize the new
1846 (record_group_use): Add a mem_type argument. Use address_p.
1847 Remove obsolete null checks of base_object. Update call to record_use.
1848 (find_interesting_uses_op): Update call to record_group_use.
1849 (find_interesting_uses_cond): Likewise.
1850 (find_interesting_uses_address): Likewise.
1851 (get_mem_type_for_internal_fn): New function.
1852 (find_address_like_use): Likewise.
1853 (find_interesting_uses_stmt): Try find_address_like_use before
1854 calling find_interesting_uses_op.
1855 (addr_offset_valid_p): Use the iv mem_type field as the type
1856 of the addressed memory.
1857 (add_autoinc_candidates): Likewise.
1858 (get_address_cost): Likewise.
1859 (split_small_address_groups_p): Use address_p.
1860 (split_address_groups): Likewise.
1861 (add_iv_candidate_for_use): Likewise.
1862 (autoinc_possible_for_pair): Likewise.
1863 (rewrite_groups): Likewise.
1864 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
1865 (determine_group_iv_cost): Update after split of USE_ADDRESS.
1866 (get_alias_ptr_type_for_ptr_address): New function.
1867 (rewrite_use_address): Rewrite address uses in calls that were
1868 identified by find_address_like_use.
1870 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1871 Alan Hayward <alan.hayward@arm.com>
1872 David Sherwood <david.sherwood@arm.com>
1874 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
1876 * gimple-expr.h (is_gimple_addressable: Likewise.
1877 * gimple-expr.c (is_gimple_address): Likewise.
1878 * internal-fn.c (expand_call_mem_ref): New function.
1879 (expand_mask_load_optab_fn): Use it.
1880 (expand_mask_store_optab_fn): Likewise.
1882 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1883 Alan Hayward <alan.hayward@arm.com>
1884 David Sherwood <david.sherwood@arm.com>
1886 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
1887 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
1888 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
1889 (cond_umax@var{mode}): Document.
1890 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
1891 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
1892 (cond_umin_optab, cond_umax_optab): New optabs.
1893 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
1894 (COND_IOR, COND_XOR): New internal functions.
1895 * internal-fn.h (get_conditional_internal_fn): Declare.
1896 * internal-fn.c (cond_binary_direct): New macro.
1897 (expand_cond_binary_optab_fn): Likewise.
1898 (direct_cond_binary_optab_supported_p): Likewise.
1899 (get_conditional_internal_fn): New function.
1900 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
1901 Cope with reduction statements that are vectorized as calls rather
1903 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
1904 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
1905 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
1906 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
1907 (UNSPEC_COND_EOR): New unspecs.
1908 (optab): Add mappings for them.
1909 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
1910 (sve_int_op, sve_fp_op): New int attributes.
1912 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1913 Alan Hayward <alan.hayward@arm.com>
1914 David Sherwood <david.sherwood@arm.com>
1916 * optabs.def (while_ult_optab): New optab.
1917 * doc/md.texi (while_ult@var{m}@var{n}): Document.
1918 * internal-fn.def (WHILE_ULT): New internal function.
1919 * internal-fn.h (direct_internal_fn_supported_p): New override
1920 that takes two types as argument.
1921 * internal-fn.c (while_direct): New macro.
1922 (expand_while_optab_fn): New function.
1923 (convert_optab_supported_p): Likewise.
1924 (direct_while_optab_supported_p): New macro.
1925 * wide-int.h (wi::udiv_ceil): New function.
1926 * tree-vectorizer.h (rgroup_masks): New structure.
1927 (vec_loop_masks): New typedef.
1928 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
1930 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
1931 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
1932 (vect_max_vf): New function.
1933 (slpeel_make_loop_iterate_ntimes): Delete.
1934 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
1935 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
1936 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
1937 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
1938 internal-fn.h, stor-layout.h and optabs-query.h.
1939 (vect_set_loop_mask): New function.
1940 (add_preheader_seq): Likewise.
1941 (add_header_seq): Likewise.
1942 (interleave_supported_p): Likewise.
1943 (vect_maybe_permute_loop_masks): Likewise.
1944 (vect_set_loop_masks_directly): Likewise.
1945 (vect_set_loop_condition_masked): Likewise.
1946 (vect_set_loop_condition_unmasked): New function, split out from
1947 slpeel_make_loop_iterate_ntimes.
1948 (slpeel_make_loop_iterate_ntimes): Rename to..
1949 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
1950 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
1951 (vect_do_peeling): Update call accordingly.
1952 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
1954 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1955 mask_compare_type, can_fully_mask_p and fully_masked_p.
1956 (release_vec_loop_masks): New function.
1957 (_loop_vec_info): Use it to free the loop masks.
1958 (can_produce_all_loop_masks_p): New function.
1959 (vect_get_max_nscalars_per_iter): Likewise.
1960 (vect_verify_full_masking): Likewise.
1961 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
1962 retries, and free the mask rgroups before retrying. Check loop-wide
1963 reasons for disallowing fully-masked loops. Make the final decision
1964 about whether use a fully-masked loop or not.
1965 (vect_estimate_min_profitable_iters): Do not assume that peeling
1966 for the number of iterations will be needed for fully-masked loops.
1967 (vectorizable_reduction): Disable fully-masked loops.
1968 (vectorizable_live_operation): Likewise.
1969 (vect_halve_mask_nunits): New function.
1970 (vect_double_mask_nunits): Likewise.
1971 (vect_record_loop_mask): Likewise.
1972 (vect_get_loop_mask): Likewise.
1973 (vect_transform_loop): Handle the case in which the final loop
1974 iteration might handle a partial vector. Call vect_set_loop_condition
1975 instead of slpeel_make_loop_iterate_ntimes.
1976 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
1977 (check_load_store_masking): New function.
1978 (prepare_load_store_mask): Likewise.
1979 (vectorizable_store): Handle fully-masked loops.
1980 (vectorizable_load): Likewise.
1981 (supportable_widening_operation): Use vect_halve_mask_nunits for
1983 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
1984 (vect_gen_while): New function.
1985 * config/aarch64/aarch64.md (umax<mode>3): New expander.
1986 (aarch64_uqdec<mode>): New insn.
1988 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1989 Alan Hayward <alan.hayward@arm.com>
1990 David Sherwood <david.sherwood@arm.com>
1992 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
1993 (reduc_xor_scal_optab): New optabs.
1994 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
1995 (reduc_xor_scal_@var{m}): Document.
1996 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
1997 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
1999 * fold-const-call.c (fold_const_call): Handle them.
2000 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2001 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2002 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2003 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2004 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2005 (UNSPEC_XORV): New unspecs.
2006 (optab): Add entries for them.
2007 (BITWISEV): New int iterator.
2008 (bit_reduc_op): New int attributes.
2010 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2011 Alan Hayward <alan.hayward@arm.com>
2012 David Sherwood <david.sherwood@arm.com>
2014 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2015 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2016 * optabs.def (vec_shl_insert_optab): New optab.
2017 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2018 (duplicate_and_interleave): Likewise.
2019 * tree-vect-loop.c: Include internal-fn.h.
2020 (neutral_op_for_slp_reduction): New function, split out from
2021 get_initial_defs_for_reduction.
2022 (get_initial_def_for_reduction): Handle option 2 for variable-length
2023 vectors by loading the neutral value into a vector and then shifting
2024 the initial value into element 0.
2025 (get_initial_defs_for_reduction): Replace the code argument with
2026 the neutral value calculated by neutral_op_for_slp_reduction.
2027 Use gimple_build_vector for constant-length vectors.
2028 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2029 but the first group_size elements have a neutral value.
2030 Use duplicate_and_interleave otherwise.
2031 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2032 Update call to get_initial_defs_for_reduction. Handle SLP
2033 reductions for variable-length vectors by creating one vector
2034 result for each scalar result, with the elements associated
2035 with other scalar results stubbed out with the neutral value.
2036 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2037 Require IFN_VEC_SHL_INSERT for double reductions on
2038 variable-length vectors, or SLP reductions that have
2039 a neutral value. Require can_duplicate_and_interleave_p
2040 support for variable-length unchained SLP reductions if there
2041 is no neutral value, such as for MIN/MAX reductions. Also require
2042 the number of vector elements to be a multiple of the number of
2043 SLP statements when doing variable-length unchained SLP reductions.
2044 Update call to vect_create_epilog_for_reduction.
2045 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2046 and remove initial values.
2047 (duplicate_and_interleave): Make public.
2048 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2049 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2051 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2052 Alan Hayward <alan.hayward@arm.com>
2053 David Sherwood <david.sherwood@arm.com>
2055 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2056 (can_duplicate_and_interleave_p): New function.
2057 (vect_get_and_check_slp_defs): Take the vector of statements
2058 rather than just the current one. Remove excess parentheses.
2059 Restriction rejectinon of vect_constant_def and vect_external_def
2060 for variable-length vectors to boolean types, or types for which
2061 can_duplicate_and_interleave_p is false.
2062 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2063 (duplicate_and_interleave): New function.
2064 (vect_get_constant_vectors): Use gimple_build_vector for
2065 constant-length vectors and suitable variable-length constant
2066 vectors. Use duplicate_and_interleave for other variable-length
2067 vectors. Don't defer the update when inserting new statements.
2069 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2070 Alan Hayward <alan.hayward@arm.com>
2071 David Sherwood <david.sherwood@arm.com>
2073 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2074 min_profitable_iters doesn't go negative.
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 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2081 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2082 * optabs.def (vec_mask_load_lanes_optab): New optab.
2083 (vec_mask_store_lanes_optab): Likewise.
2084 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2085 (MASK_STORE_LANES): Likewise.
2086 * internal-fn.c (mask_load_lanes_direct): New macro.
2087 (mask_store_lanes_direct): Likewise.
2088 (expand_mask_load_optab_fn): Handle masked operations.
2089 (expand_mask_load_lanes_optab_fn): New macro.
2090 (expand_mask_store_optab_fn): Handle masked operations.
2091 (expand_mask_store_lanes_optab_fn): New macro.
2092 (direct_mask_load_lanes_optab_supported_p): Likewise.
2093 (direct_mask_store_lanes_optab_supported_p): Likewise.
2094 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2096 (vect_load_lanes_supported): Likewise.
2097 * tree-vect-data-refs.c (strip_conversion): New function.
2098 (can_group_stmts_p): Likewise.
2099 (vect_analyze_data_ref_accesses): Use it instead of checking
2100 for a pair of assignments.
2101 (vect_store_lanes_supported): Take a masked_p parameter.
2102 (vect_load_lanes_supported): Likewise.
2103 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2104 vect_store_lanes_supported and vect_load_lanes_supported.
2105 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2106 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2107 parameter. Don't allow gaps for masked accesses.
2108 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2109 and vect_load_lanes_supported.
2110 (get_load_store_type): Take a masked_p parameter and update
2111 call to get_group_load_store_type.
2112 (vectorizable_store): Update call to get_load_store_type.
2113 Handle IFN_MASK_STORE_LANES.
2114 (vectorizable_load): Update call to get_load_store_type.
2115 Handle IFN_MASK_LOAD_LANES.
2117 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2118 Alan Hayward <alan.hayward@arm.com>
2119 David Sherwood <david.sherwood@arm.com>
2121 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2123 * config/aarch64/aarch64-protos.h
2124 (aarch64_sve_struct_memory_operand_p): Declare.
2125 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2126 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2127 (VPRED, vpred): Handle SVE structure modes.
2128 * config/aarch64/constraints.md (Utx): New constraint.
2129 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2130 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2131 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2132 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2133 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2134 structure modes. Split into pieces after RA.
2135 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2136 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2138 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2139 SVE structure modes.
2140 (aarch64_classify_address): Likewise.
2141 (sizetochar): Move earlier in file.
2142 (aarch64_print_operand): Handle SVE register lists.
2143 (aarch64_array_mode): New function.
2144 (aarch64_sve_struct_memory_operand_p): Likewise.
2145 (TARGET_ARRAY_MODE): Redefine.
2147 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2148 Alan Hayward <alan.hayward@arm.com>
2149 David Sherwood <david.sherwood@arm.com>
2151 * target.def (array_mode): New target hook.
2152 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2153 * doc/tm.texi: Regenerate.
2154 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2155 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2156 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2158 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2161 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2162 Alan Hayward <alan.hayward@arm.com>
2163 David Sherwood <david.sherwood@arm.com>
2165 * fold-const.c (fold_binary_loc): Check the argument types
2166 rather than the result type when testing for a vector operation.
2168 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2170 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2171 * doc/tm.texi: Regenerate.
2173 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2174 Alan Hayward <alan.hayward@arm.com>
2175 David Sherwood <david.sherwood@arm.com>
2177 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2178 (sve): Document new AArch64 extension.
2179 * doc/md.texi (w): Extend the description of the AArch64
2180 constraint to include SVE vectors.
2181 (Upl, Upa): Document new AArch64 predicate constraints.
2182 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2184 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2185 (msve-vector-bits=): New option.
2186 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2187 SVE when these are disabled.
2188 (sve): New extension.
2189 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2190 modes. Adjust their number of units based on aarch64_sve_vg.
2191 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2192 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2193 aarch64_addr_query_type.
2194 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2195 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2196 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2197 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2198 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2199 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2200 (aarch64_simd_imm_zero_p): Delete.
2201 (aarch64_check_zero_based_sve_index_immediate): Declare.
2202 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2203 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2204 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2205 (aarch64_sve_float_mul_immediate_p): Likewise.
2206 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2208 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2209 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2210 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2211 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2212 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2213 (aarch64_regmode_natural_size): Likewise.
2214 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2215 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2217 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2218 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2219 for VG and the SVE predicate registers.
2220 (V_ALIASES): Add a "z"-prefixed alias.
2221 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2222 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2223 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2224 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2225 (REG_CLASS_NAMES): Add entries for them.
2226 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2227 and the predicate registers.
2228 (aarch64_sve_vg): Declare.
2229 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2230 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2231 (REGMODE_NATURAL_SIZE): Define.
2232 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2234 * config/aarch64/aarch64.c: Include cfgrtl.h.
2235 (simd_immediate_info): Add a constructor for series vectors,
2236 and an associated step field.
2237 (aarch64_sve_vg): New variable.
2238 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2239 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2240 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2241 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2242 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2243 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2244 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2245 (aarch64_get_mask_mode): New functions.
2246 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2247 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2248 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2249 predicate modes and predicate registers. Explicitly restrict
2250 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2251 to store a vector mode if it is recognized by
2252 aarch64_classify_vector_mode.
2253 (aarch64_regmode_natural_size): New function.
2254 (aarch64_hard_regno_caller_save_mode): Return the original mode
2256 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2257 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2258 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2259 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2261 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2262 does not overlap dest if the function is frame-related. Handle
2264 (aarch64_split_add_offset): New function.
2265 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2266 them aarch64_add_offset.
2267 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2268 and update call to aarch64_sub_sp.
2269 (aarch64_add_cfa_expression): New function.
2270 (aarch64_expand_prologue): Pass extra temporary registers to the
2271 functions above. Handle the case in which we need to emit new
2272 DW_CFA_expressions for registers that were originally saved
2273 relative to the stack pointer, but now have to be expressed
2274 relative to the frame pointer.
2275 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2277 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2278 IP0 and IP1 values for SVE frames.
2279 (aarch64_expand_vec_series): New function.
2280 (aarch64_expand_sve_widened_duplicate): Likewise.
2281 (aarch64_expand_sve_const_vector): Likewise.
2282 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2283 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2284 into the register, rather than emitting a SET directly.
2285 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2286 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2287 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2288 (offset_9bit_signed_scaled_p): New functions.
2289 (aarch64_replicate_bitmask_imm): New function.
2290 (aarch64_bitmask_imm): Use it.
2291 (aarch64_cannot_force_const_mem): Reject expressions involving
2292 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2293 (aarch64_classify_index): Handle SVE indices, by requiring
2294 a plain register index with a scale that matches the element size.
2295 (aarch64_classify_address): Handle SVE addresses. Assert that
2296 the mode of the address is VOIDmode or an integer mode.
2297 Update call to aarch64_classify_symbol.
2298 (aarch64_classify_symbolic_expression): Update call to
2299 aarch64_classify_symbol.
2300 (aarch64_const_vec_all_in_range_p): New function.
2301 (aarch64_print_vector_float_operand): Likewise.
2302 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2303 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2304 and the FP immediates 1.0 and 0.5.
2305 (aarch64_print_address_internal): Handle SVE addresses.
2306 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2307 (aarch64_regno_regclass): Handle predicate registers.
2308 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2310 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2311 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2312 (aarch64_convert_sve_vector_bits): New function.
2313 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2314 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2316 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2317 Handle SVE vector and predicate modes. Accept VL-based constants
2318 that need only one temporary register, and VL offsets that require
2319 no temporary registers.
2320 (aarch64_conditional_register_usage): Mark the predicate registers
2321 as fixed if SVE isn't available.
2322 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2323 Return true for SVE vector and predicate modes.
2324 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2325 rather than an unsigned int. Handle SVE modes.
2326 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2328 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2330 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2331 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2332 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2333 (aarch64_sve_float_mul_immediate_p): New functions.
2334 (aarch64_sve_valid_immediate): New function.
2335 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2336 Explicitly reject structure modes. Check for INDEX constants.
2337 Handle PTRUE and PFALSE constants.
2338 (aarch64_check_zero_based_sve_index_immediate): New function.
2339 (aarch64_simd_imm_zero_p): Delete.
2340 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2341 vector modes. Accept constants in the range of CNT[BHWD].
2342 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2343 ask for an Advanced SIMD mode.
2344 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2345 (aarch64_simd_vector_alignment): Handle SVE predicates.
2346 (aarch64_vectorize_preferred_vector_alignment): New function.
2347 (aarch64_simd_vector_alignment_reachable): Use it instead of
2349 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2350 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2352 (MAX_VECT_LEN): Delete.
2353 (expand_vec_perm_d): Add a vec_flags field.
2354 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2355 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2356 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2358 (aarch64_evpc_rev): Rename to...
2359 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2360 (aarch64_evpc_rev_global): New function.
2361 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2362 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2364 (aarch64_evpc_sve_tbl): New function.
2365 (aarch64_expand_vec_perm_const_1): Update after rename of
2366 aarch64_evpc_rev. Handle SVE permutes too, trying
2367 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2368 than aarch64_evpc_tbl.
2369 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2370 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2371 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2372 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2373 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2374 (aarch64_expand_sve_vcond): New functions.
2375 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2376 of aarch64_vector_mode_p.
2377 (aarch64_dwarf_poly_indeterminate_value): New function.
2378 (aarch64_compute_pressure_classes): Likewise.
2379 (aarch64_can_change_mode_class): Likewise.
2380 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2381 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2382 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2383 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2384 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2385 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2386 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2387 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2389 (Dn, Dl, Dr): Accept const as well as const_vector.
2390 (Dz): Likewise. Compare against CONST0_RTX.
2391 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2392 of "vector" where appropriate.
2393 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2394 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2395 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2396 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2397 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2398 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2399 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2400 (v_int_equiv): Extend to SVE modes.
2401 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2403 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2404 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2405 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2406 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2407 (SVE_COND_FP_CMP): New int iterators.
2408 (perm_hilo): Handle the new unpack unspecs.
2409 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2411 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2412 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2413 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2414 (aarch64_equality_operator, aarch64_constant_vector_operand)
2415 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2416 (aarch64_sve_nonimmediate_operand): Likewise.
2417 (aarch64_sve_general_operand): Likewise.
2418 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2419 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2420 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2421 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2422 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2423 (aarch64_sve_float_arith_immediate): Likewise.
2424 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2425 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2426 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2427 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2428 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2429 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2430 (aarch64_sve_float_arith_operand): Likewise.
2431 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2432 (aarch64_sve_float_mul_operand): Likewise.
2433 (aarch64_sve_vec_perm_operand): Likewise.
2434 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2435 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2436 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2437 as well as const_vector.
2438 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2439 in file. Use CONST0_RTX and CONSTM1_RTX.
2440 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2441 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2442 Use aarch64_simd_imm_zero.
2443 * config/aarch64/aarch64-sve.md: New file.
2444 * config/aarch64/aarch64.md: Include it.
2445 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2446 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2447 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2448 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2449 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2450 (sve): New attribute.
2451 (enabled): Disable instructions with the sve attribute unless
2453 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2454 aarch64_expand_mov_immediate.
2455 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2456 CNT[BHSD] immediates.
2457 (movti): Split CONST_POLY_INT moves into two halves.
2458 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2459 Split additions that need a temporary here if the destination
2460 is the stack pointer.
2461 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2462 (*add<mode>3_poly_1): New instruction.
2463 (set_clobber_cc): New expander.
2465 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2467 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2468 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2469 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2470 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2471 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2472 Change innermode from fixed_mode_size to machine_mode.
2473 (simplify_subreg): Update call accordingly. Handle a constant-sized
2474 subreg of a variable-length CONST_VECTOR.
2476 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2477 Alan Hayward <alan.hayward@arm.com>
2478 David Sherwood <david.sherwood@arm.com>
2480 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2481 (add_offset_to_base): New function, split out from...
2482 (create_mem_ref): ...here. When handling a scale other than 1,
2483 check first whether the address is valid without the offset.
2484 Add it into the base if so, leaving the index and scale as-is.
2486 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2489 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2490 fold_for_warn before checking if arg2 is INTEGER_CST.
2492 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2494 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2495 (store_multiple_operation): Delete.
2496 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2497 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2498 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2499 guarded by TARGET_STRING.
2500 (rs6000_output_load_multiple): Delete.
2501 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2502 OPTION_MASK_STRING / TARGET_STRING handling.
2503 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2504 (const rs6000_opt_masks) <"string">: Change mask to 0.
2505 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2506 (MASK_STRING): Delete.
2507 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2509 (load_multiple): Delete.
2516 (store_multiple): Delete.
2523 (movmemsi_8reg): Delete.
2524 (corresponding unnamed define_insn): Delete.
2525 (movmemsi_6reg): Delete.
2526 (corresponding unnamed define_insn): Delete.
2527 (movmemsi_4reg): Delete.
2528 (corresponding unnamed define_insn): Delete.
2529 (movmemsi_2reg): Delete.
2530 (corresponding unnamed define_insn): Delete.
2531 (movmemsi_1reg): Delete.
2532 (corresponding unnamed define_insn): Delete.
2533 * config/rs6000/rs6000.opt (mno-string): New.
2534 (mstring): Replace by deprecation warning stub.
2535 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2537 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2539 * regrename.c (regrename_do_replace): If replacing the same
2540 reg multiple times, try to reuse last created gen_raw_REG.
2543 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2544 main to workaround a bug in GDB.
2546 2018-01-12 Tom de Vries <tom@codesourcery.com>
2549 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2551 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2553 PR rtl-optimization/80481
2554 * ira-color.c (get_cap_member): New function.
2555 (allocnos_conflict_by_live_ranges_p): Use it.
2556 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2557 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2559 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2562 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2563 (*saddl_se_1): Ditto.
2565 (*ssubl_se_1): Ditto.
2567 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2569 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2570 rather than wi::to_widest for DR_INITs.
2571 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2572 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2573 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2575 (vect_analyze_group_access_1): Note that here.
2577 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2579 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2580 polynomial type sizes.
2582 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2584 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2585 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2586 (gimple_add_tmp_var): Likewise.
2588 2018-01-12 Martin Liska <mliska@suse.cz>
2590 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2591 (gimple_alloc_sizes): Likewise.
2592 (dump_gimple_statistics): Use PRIu64 in printf format.
2593 * gimple.h: Change uint64_t to int.
2595 2018-01-12 Martin Liska <mliska@suse.cz>
2597 * tree-core.h: Use uint64_t instead of int.
2598 * tree.c (tree_node_counts): Likewise.
2599 (tree_node_sizes): Likewise.
2600 (dump_tree_statistics): Use PRIu64 in printf format.
2602 2018-01-12 Martin Liska <mliska@suse.cz>
2604 * Makefile.in: As qsort_chk is implemented in vec.c, add
2605 vec.o to linkage of gencfn-macros.
2606 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2607 passing the info to record_node_allocation_statistics.
2608 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2610 * ggc-common.c (struct ggc_usage): Add operator== and use
2611 it in operator< and compare function.
2612 * mem-stats.h (struct mem_usage): Likewise.
2613 * vec.c (struct vec_usage): Remove operator< and compare
2614 function. Can be simply inherited.
2616 2018-01-12 Martin Jambor <mjambor@suse.cz>
2619 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2620 * tree-ssa-math-opts.c: Include domwalk.h.
2621 (convert_mult_to_fma_1): New function.
2622 (fma_transformation_info): New type.
2623 (fma_deferring_state): Likewise.
2624 (cancel_fma_deferring): New function.
2625 (result_of_phi): Likewise.
2626 (last_fma_candidate_feeds_initial_phi): Likewise.
2627 (convert_mult_to_fma): Added deferring logic, split actual
2628 transformation to convert_mult_to_fma_1.
2629 (math_opts_dom_walker): New type.
2630 (math_opts_dom_walker::after_dom_children): New method, body moved
2631 here from pass_optimize_widening_mul::execute, added deferring logic
2633 (pass_optimize_widening_mul::execute): Moved most of code to
2634 math_opts_dom_walker::after_dom_children.
2635 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2636 * config/i386/i386.c (ix86_option_override_internal): Added
2637 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2639 2018-01-12 Richard Biener <rguenther@suse.de>
2642 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2643 inline instance vars.
2645 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2648 * config/rx/rx.c (rx_is_restricted_memory_address):
2651 2018-01-12 Richard Biener <rguenther@suse.de>
2653 PR tree-optimization/80846
2654 * target.def (split_reduction): New target hook.
2655 * targhooks.c (default_split_reduction): New function.
2656 * targhooks.h (default_split_reduction): Declare.
2657 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2658 target requests first reduce vectors by combining low and high
2660 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2661 (get_vectype_for_scalar_type_and_size): Export.
2662 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2663 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2664 * doc/tm.texi: Regenerate.
2665 * config/i386/i386.c (ix86_split_reduction): Implement
2666 TARGET_VECTORIZE_SPLIT_REDUCTION.
2668 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2671 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2672 in PIC mode except for TARGET_VXWORKS_RTP.
2673 * config/sparc/sparc.c: Include cfgrtl.h.
2674 (TARGET_INIT_PIC_REG): Define.
2675 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2676 (sparc_pic_register_p): New predicate.
2677 (sparc_legitimate_address_p): Use it.
2678 (sparc_legitimize_pic_address): Likewise.
2679 (sparc_delegitimize_address): Likewise.
2680 (sparc_mode_dependent_address_p): Likewise.
2681 (gen_load_pcrel_sym): Remove 4th parameter.
2682 (load_got_register): Adjust call to above. Remove obsolete stuff.
2683 (sparc_expand_prologue): Do not call load_got_register here.
2684 (sparc_flat_expand_prologue): Likewise.
2685 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2686 (sparc_use_pseudo_pic_reg): New function.
2687 (sparc_init_pic_reg): Likewise.
2688 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2689 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2691 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2693 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2694 Add item for branch_cost.
2696 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2698 PR rtl-optimization/83565
2699 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2700 not extend the result to a larger mode for rotate operations.
2701 (num_sign_bit_copies1): Likewise.
2703 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2706 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2708 Use values-Xc.o for -pedantic.
2709 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2711 2018-01-12 Martin Liska <mliska@suse.cz>
2714 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2716 (possible_polymorphic_call_targets): Use it.
2717 (ipa_devirt): Likewise.
2719 2018-01-12 Martin Liska <mliska@suse.cz>
2721 * profile-count.h (enum profile_quality): Use 0 as invalid
2722 enum value of profile_quality.
2724 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2726 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2727 -mext-string options.
2729 2018-01-12 Richard Biener <rguenther@suse.de>
2731 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2732 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2733 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2735 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2737 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2739 * configure.ac (--with-long-double-format): Add support for the
2740 configuration option to change the default long double format on
2742 * config.gcc (powerpc*-linux*-*): Likewise.
2743 * configure: Regenerate.
2744 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2745 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2746 used without modification.
2748 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2750 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2751 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2752 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2753 MISC_BUILTIN_SPEC_BARRIER.
2754 (rs6000_init_builtins): Likewise.
2755 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2757 (speculation_barrier): New define_insn.
2758 * doc/extend.texi: Document __builtin_speculation_barrier.
2760 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2763 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2764 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2765 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2767 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2768 integral modes instead of "ss" and "sd".
2769 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2770 vectors with 32-bit and 64-bit elements.
2771 (vecdupssescalarmodesuffix): New mode attribute.
2772 (vec_dup<mode>): Use it.
2774 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2777 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2778 frame if argument is passed on stack.
2780 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2783 * ree.c (combine_reaching_defs): Optimize also
2784 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
2785 reg2=any_extend(exp); reg1=reg2;, formatting fix.
2787 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2790 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
2792 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2795 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
2796 after they are computed.
2798 2018-01-11 Bin Cheng <bin.cheng@arm.com>
2800 PR tree-optimization/83695
2801 * gimple-loop-linterchange.cc
2802 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
2803 reset cached scev information after interchange.
2804 (pass_linterchange::execute): Remove call to scev_reset_htab.
2806 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2808 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
2809 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
2810 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
2811 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
2812 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
2813 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
2814 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
2815 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
2816 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
2817 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
2818 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
2819 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
2820 (V_lane_reg): Likewise.
2821 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
2823 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
2824 (vfmal_lane_low<mode>_intrinsic,
2825 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
2826 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
2827 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
2828 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
2829 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
2830 vfmsl_lane_high<mode>_intrinsic): New define_insns.
2832 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2834 * config/arm/arm-cpus.in (fp16fml): New feature.
2835 (ALL_SIMD): Add fp16fml.
2836 (armv8.2-a): Add fp16fml as an option.
2837 (armv8.3-a): Likewise.
2838 (armv8.4-a): Add fp16fml as part of fp16.
2839 * config/arm/arm.h (TARGET_FP16FML): Define.
2840 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
2842 * config/arm/arm-modes.def (V2HF): Define.
2843 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
2844 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
2845 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
2846 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
2847 vfmsl_low, vfmsl_high): New set of builtins.
2848 * config/arm/iterators.md (PLUSMINUS): New code iterator.
2849 (vfml_op): New code attribute.
2850 (VFMLHALVES): New int iterator.
2851 (VFML, VFMLSEL): New mode attributes.
2852 (V_reg): Define mapping for V2HF.
2853 (V_hi, V_lo): New mode attributes.
2854 (VF_constraint): Likewise.
2855 (vfml_half, vfml_half_selector): New int attributes.
2856 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
2858 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
2859 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
2861 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
2862 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
2863 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
2864 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
2866 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
2867 Document new effective target and option set.
2869 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2871 * config/arm/arm-cpus.in (armv8_4): New feature.
2872 (ARMv8_4a): New fgroup.
2873 (armv8.4-a): New arch.
2874 * config/arm/arm-tables.opt: Regenerate.
2875 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
2876 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
2877 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
2878 Add matching rules for -march=armv8.4-a and extensions.
2879 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
2881 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
2884 * config/rx/rx.md (BW): New mode attribute.
2885 (sync_lock_test_and_setsi): Add mode suffix to insn output.
2887 2018-01-11 Richard Biener <rguenther@suse.de>
2889 PR tree-optimization/83435
2890 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
2891 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
2892 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
2894 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2895 Alan Hayward <alan.hayward@arm.com>
2896 David Sherwood <david.sherwood@arm.com>
2898 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
2900 (aarch64_classify_address): Initialize it. Track polynomial offsets.
2901 (aarch64_print_address_internal): Use it to check for a zero offset.
2903 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2904 Alan Hayward <alan.hayward@arm.com>
2905 David Sherwood <david.sherwood@arm.com>
2907 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
2908 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
2909 Return a poly_int64 rather than a HOST_WIDE_INT.
2910 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
2911 rather than a HOST_WIDE_INT.
2912 * config/aarch64/aarch64.h (aarch64_frame): Protect with
2913 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
2914 hard_fp_offset, frame_size, initial_adjust, callee_offset and
2915 final_offset from HOST_WIDE_INT to poly_int64.
2916 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
2917 to_constant when getting the number of units in an Advanced SIMD
2919 (aarch64_builtin_vectorized_function): Check for a constant number
2921 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
2923 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
2924 attribute instead of GET_MODE_NUNITS.
2925 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
2926 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
2927 GET_MODE_SIZE for fixed-size registers.
2928 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
2929 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
2930 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
2931 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
2932 (aarch64_print_operand, aarch64_print_address_internal)
2933 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
2934 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
2935 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
2936 Handle polynomial GET_MODE_SIZE.
2937 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
2938 wider than SImode without modification.
2939 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
2940 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
2941 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
2942 passing and returning SVE modes.
2943 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
2944 rather than GEN_INT.
2945 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
2946 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
2947 (aarch64_allocate_and_probe_stack_space): Likewise.
2948 (aarch64_layout_frame): Cope with polynomial offsets.
2949 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
2950 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
2952 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
2953 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
2954 poly_int64 rather than a HOST_WIDE_INT.
2955 (aarch64_get_separate_components, aarch64_process_components)
2956 (aarch64_expand_prologue, aarch64_expand_epilogue)
2957 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
2958 (aarch64_anchor_offset): New function, split out from...
2959 (aarch64_legitimize_address): ...here.
2960 (aarch64_builtin_vectorization_cost): Handle polynomial
2961 TYPE_VECTOR_SUBPARTS.
2962 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
2964 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
2965 number of elements from the PARALLEL rather than the mode.
2966 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
2967 rather than GET_MODE_BITSIZE.
2968 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
2969 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
2970 (aarch64_expand_vec_perm_const_1): Handle polynomial
2971 d->perm.length () and d->perm elements.
2972 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
2973 Apply to_constant to d->perm elements.
2974 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
2975 polynomial CONST_VECTOR_NUNITS.
2976 (aarch64_move_pointer): Take amount as a poly_int64 rather
2978 (aarch64_progress_pointer): Avoid temporary variable.
2979 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
2980 the mode attribute instead of GET_MODE.
2982 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2983 Alan Hayward <alan.hayward@arm.com>
2984 David Sherwood <david.sherwood@arm.com>
2986 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
2987 x exists before using it.
2988 (aarch64_add_constant_internal): Rename to...
2989 (aarch64_add_offset_1): ...this. Replace regnum with separate
2990 src and dest rtxes. Handle the case in which they're different,
2991 including when the offset is zero. Replace scratchreg with an rtx.
2992 Use 2 additions if there is no spare register into which we can
2993 move a 16-bit constant.
2994 (aarch64_add_constant): Delete.
2995 (aarch64_add_offset): Replace reg with separate src and dest
2996 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
2997 Use aarch64_add_offset_1.
2998 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
2999 an rtx rather than an int. Take the delta as a poly_int64
3000 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3001 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3002 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3003 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3004 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3006 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3007 aarch64_add_constant.
3009 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3011 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3012 Use scalar_float_mode.
3014 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3016 * config/aarch64/aarch64-simd.md
3017 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3018 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3019 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3020 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3021 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3022 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3023 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3024 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3025 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3026 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3028 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3031 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3032 targ_options->x_arm_arch_string is non NULL.
3034 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3036 * config/aarch64/aarch64.h
3037 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3039 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3042 * expmed.c (emit_store_flag_force): Swap if const op0
3043 and change VOIDmode to mode of op0.
3045 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3047 PR rtl-optimization/83761
3048 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3049 than bytes to mode_for_size.
3051 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3054 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3055 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3058 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3061 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3062 when in layout mode.
3063 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3064 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3067 2018-01-10 Michael Collison <michael.collison@arm.com>
3069 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3070 * config/aarch64/aarch64-option-extension.def: Add
3071 AARCH64_OPT_EXTENSION of 'fp16fml'.
3072 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3073 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3074 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3075 * config/aarch64/constraints.md (Ui7): New constraint.
3076 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3077 (VFMLA_SEL_W): Ditto.
3080 (VFMLA16_LOW): New int iterator.
3081 (VFMLA16_HIGH): Ditto.
3082 (UNSPEC_FMLAL): New unspec.
3083 (UNSPEC_FMLSL): Ditto.
3084 (UNSPEC_FMLAL2): Ditto.
3085 (UNSPEC_FMLSL2): Ditto.
3086 (f16mac): New code attribute.
3087 * config/aarch64/aarch64-simd-builtins.def
3088 (aarch64_fmlal_lowv2sf): Ditto.
3089 (aarch64_fmlsl_lowv2sf): Ditto.
3090 (aarch64_fmlalq_lowv4sf): Ditto.
3091 (aarch64_fmlslq_lowv4sf): Ditto.
3092 (aarch64_fmlal_highv2sf): Ditto.
3093 (aarch64_fmlsl_highv2sf): Ditto.
3094 (aarch64_fmlalq_highv4sf): Ditto.
3095 (aarch64_fmlslq_highv4sf): Ditto.
3096 (aarch64_fmlal_lane_lowv2sf): Ditto.
3097 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3098 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3099 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3100 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3101 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3102 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3103 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3104 (aarch64_fmlal_lane_highv2sf): Ditto.
3105 (aarch64_fmlsl_lane_highv2sf): Ditto.
3106 (aarch64_fmlal_laneq_highv2sf): Ditto.
3107 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3108 (aarch64_fmlalq_lane_highv4sf): Ditto.
3109 (aarch64_fmlsl_lane_highv4sf): Ditto.
3110 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3111 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3112 * config/aarch64/aarch64-simd.md:
3113 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3114 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3115 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3116 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3117 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3118 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3119 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3120 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3121 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3122 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3123 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3124 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3125 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3126 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3127 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3128 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3129 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3130 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3131 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3132 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3133 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3134 (vfmlsl_low_u32): Ditto.
3135 (vfmlalq_low_u32): Ditto.
3136 (vfmlslq_low_u32): Ditto.
3137 (vfmlal_high_u32): Ditto.
3138 (vfmlsl_high_u32): Ditto.
3139 (vfmlalq_high_u32): Ditto.
3140 (vfmlslq_high_u32): Ditto.
3141 (vfmlal_lane_low_u32): Ditto.
3142 (vfmlsl_lane_low_u32): Ditto.
3143 (vfmlal_laneq_low_u32): Ditto.
3144 (vfmlsl_laneq_low_u32): Ditto.
3145 (vfmlalq_lane_low_u32): Ditto.
3146 (vfmlslq_lane_low_u32): Ditto.
3147 (vfmlalq_laneq_low_u32): Ditto.
3148 (vfmlslq_laneq_low_u32): Ditto.
3149 (vfmlal_lane_high_u32): Ditto.
3150 (vfmlsl_lane_high_u32): Ditto.
3151 (vfmlal_laneq_high_u32): Ditto.
3152 (vfmlsl_laneq_high_u32): Ditto.
3153 (vfmlalq_lane_high_u32): Ditto.
3154 (vfmlslq_lane_high_u32): Ditto.
3155 (vfmlalq_laneq_high_u32): Ditto.
3156 (vfmlslq_laneq_high_u32): Ditto.
3157 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3158 (AARCH64_FL_FOR_ARCH8_4): New.
3159 (AARCH64_ISA_F16FML): New ISA flag.
3160 (TARGET_F16FML): New feature flag for fp16fml.
3161 (doc/invoke.texi): Document new fp16fml option.
3163 2018-01-10 Michael Collison <michael.collison@arm.com>
3165 * config/aarch64/aarch64-builtins.c:
3166 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3167 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3168 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3169 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3170 (AARCH64_ISA_SHA3): New ISA flag.
3171 (TARGET_SHA3): New feature flag for sha3.
3172 * config/aarch64/iterators.md (sha512_op): New int attribute.
3173 (CRYPTO_SHA512): New int iterator.
3174 (UNSPEC_SHA512H): New unspec.
3175 (UNSPEC_SHA512H2): Ditto.
3176 (UNSPEC_SHA512SU0): Ditto.
3177 (UNSPEC_SHA512SU1): Ditto.
3178 * config/aarch64/aarch64-simd-builtins.def
3179 (aarch64_crypto_sha512hqv2di): New builtin.
3180 (aarch64_crypto_sha512h2qv2di): Ditto.
3181 (aarch64_crypto_sha512su0qv2di): Ditto.
3182 (aarch64_crypto_sha512su1qv2di): Ditto.
3183 (aarch64_eor3qv8hi): Ditto.
3184 (aarch64_rax1qv2di): Ditto.
3185 (aarch64_xarqv2di): Ditto.
3186 (aarch64_bcaxqv8hi): Ditto.
3187 * config/aarch64/aarch64-simd.md:
3188 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3189 (aarch64_crypto_sha512su0qv2di): Ditto.
3190 (aarch64_crypto_sha512su1qv2di): Ditto.
3191 (aarch64_eor3qv8hi): Ditto.
3192 (aarch64_rax1qv2di): Ditto.
3193 (aarch64_xarqv2di): Ditto.
3194 (aarch64_bcaxqv8hi): Ditto.
3195 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3196 (vsha512h2q_u64): Ditto.
3197 (vsha512su0q_u64): Ditto.
3198 (vsha512su1q_u64): Ditto.
3199 (veor3q_u16): Ditto.
3200 (vrax1q_u64): Ditto.
3202 (vbcaxq_u16): Ditto.
3203 * config/arm/types.md (crypto_sha512): New type attribute.
3204 (crypto_sha3): Ditto.
3205 (doc/invoke.texi): Document new sha3 option.
3207 2018-01-10 Michael Collison <michael.collison@arm.com>
3209 * config/aarch64/aarch64-builtins.c:
3210 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3211 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3212 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3213 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3214 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3215 (AARCH64_ISA_SM4): New ISA flag.
3216 (TARGET_SM4): New feature flag for sm4.
3217 * config/aarch64/aarch64-simd-builtins.def
3218 (aarch64_sm3ss1qv4si): Ditto.
3219 (aarch64_sm3tt1aq4si): Ditto.
3220 (aarch64_sm3tt1bq4si): Ditto.
3221 (aarch64_sm3tt2aq4si): Ditto.
3222 (aarch64_sm3tt2bq4si): Ditto.
3223 (aarch64_sm3partw1qv4si): Ditto.
3224 (aarch64_sm3partw2qv4si): Ditto.
3225 (aarch64_sm4eqv4si): Ditto.
3226 (aarch64_sm4ekeyqv4si): Ditto.
3227 * config/aarch64/aarch64-simd.md:
3228 (aarch64_sm3ss1qv4si): Ditto.
3229 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3230 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3231 (aarch64_sm4eqv4si): Ditto.
3232 (aarch64_sm4ekeyqv4si): Ditto.
3233 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3234 (sm3part_op): Ditto.
3235 (CRYPTO_SM3TT): Ditto.
3236 (CRYPTO_SM3PART): Ditto.
3237 (UNSPEC_SM3SS1): New unspec.
3238 (UNSPEC_SM3TT1A): Ditto.
3239 (UNSPEC_SM3TT1B): Ditto.
3240 (UNSPEC_SM3TT2A): Ditto.
3241 (UNSPEC_SM3TT2B): Ditto.
3242 (UNSPEC_SM3PARTW1): Ditto.
3243 (UNSPEC_SM3PARTW2): Ditto.
3244 (UNSPEC_SM4E): Ditto.
3245 (UNSPEC_SM4EKEY): Ditto.
3246 * config/aarch64/constraints.md (Ui2): New constraint.
3247 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3248 * config/arm/types.md (crypto_sm3): New type attribute.
3249 (crypto_sm4): Ditto.
3250 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3251 (vsm3tt1aq_u32): Ditto.
3252 (vsm3tt1bq_u32): Ditto.
3253 (vsm3tt2aq_u32): Ditto.
3254 (vsm3tt2bq_u32): Ditto.
3255 (vsm3partw1q_u32): Ditto.
3256 (vsm3partw2q_u32): Ditto.
3257 (vsm4eq_u32): Ditto.
3258 (vsm4ekeyq_u32): Ditto.
3259 (doc/invoke.texi): Document new sm4 option.
3261 2018-01-10 Michael Collison <michael.collison@arm.com>
3263 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3264 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3265 (AARCH64_FL_FOR_ARCH8_4): New.
3266 (AARCH64_FL_V8_4): New flag.
3267 (doc/invoke.texi): Document new armv8.4-a option.
3269 2018-01-10 Michael Collison <michael.collison@arm.com>
3271 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3272 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3273 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3274 * config/aarch64/aarch64-option-extension.def: Add
3275 AARCH64_OPT_EXTENSION of 'sha2'.
3276 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3277 (crypto): Disable sha2 and aes if crypto disabled.
3278 (crypto): Enable aes and sha2 if enabled.
3279 (simd): Disable sha2 and aes if simd disabled.
3280 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3282 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3283 (TARGET_SHA2): New feature flag for sha2.
3284 (TARGET_AES): New feature flag for aes.
3285 * config/aarch64/aarch64-simd.md:
3286 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3287 conditional on TARGET_AES.
3288 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3289 (aarch64_crypto_sha1hsi): Make pattern conditional
3291 (aarch64_crypto_sha1hv4si): Ditto.
3292 (aarch64_be_crypto_sha1hv4si): Ditto.
3293 (aarch64_crypto_sha1su1v4si): Ditto.
3294 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3295 (aarch64_crypto_sha1su0v4si): Ditto.
3296 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3297 (aarch64_crypto_sha256su0v4si): Ditto.
3298 (aarch64_crypto_sha256su1v4si): Ditto.
3299 (doc/invoke.texi): Document new aes and sha2 options.
3301 2018-01-10 Martin Sebor <msebor@redhat.com>
3303 PR tree-optimization/83781
3304 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3307 2018-01-11 Martin Sebor <msebor@gmail.com>
3308 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3310 PR tree-optimization/83501
3311 PR tree-optimization/81703
3313 * tree-ssa-strlen.c (get_string_cst): Rename...
3314 (get_string_len): ...to this. Handle global constants.
3315 (handle_char_store): Adjust.
3317 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3318 Jim Wilson <jimw@sifive.com>
3320 * config/riscv/riscv-protos.h (riscv_output_return): New.
3321 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3322 (riscv_attribute_table, riscv_output_return),
3323 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3324 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3325 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3327 (riscv_expand_prologue): Add early return for naked function.
3328 (riscv_expand_epilogue): Likewise.
3329 (riscv_function_ok_for_sibcall): Return false for naked function.
3330 (riscv_set_current_function): New.
3331 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3332 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3333 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3334 * doc/extend.texi (RISC-V Function Attributes): New.
3336 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3338 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3339 check for 128-bit long double before checking TCmode.
3340 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3341 128-bit long doubles before checking TFmode or TCmode.
3342 (FLOAT128_IBM_P): Likewise.
3344 2018-01-10 Martin Sebor <msebor@redhat.com>
3346 PR tree-optimization/83671
3347 * builtins.c (c_strlen): Unconditionally return zero for the empty
3349 Use -Warray-bounds for warnings.
3350 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3351 for non-constant array indices with COMPONENT_REF, arrays of
3352 arrays, and pointers to arrays.
3353 (gimple_fold_builtin_strlen): Determine and set length range for
3354 non-constant character arrays.
3356 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3359 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3362 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3364 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3366 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3369 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3370 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3371 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3372 indexed_or_indirect_operand predicate.
3373 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3374 (*vsx_le_perm_load_v8hi): Likewise.
3375 (*vsx_le_perm_load_v16qi): Likewise.
3376 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3377 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3378 (*vsx_le_perm_store_v8hi): Likewise.
3379 (*vsx_le_perm_store_v16qi): Likewise.
3380 (eight unnamed splitters): Likewise.
3382 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3384 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3385 * config/rs6000/emmintrin.h: Likewise.
3386 * config/rs6000/mmintrin.h: Likewise.
3387 * config/rs6000/xmmintrin.h: Likewise.
3389 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3392 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3394 * tree.c (tree_nop_conversion): Return true for location wrapper
3396 (maybe_wrap_with_location): New function.
3397 (selftest::check_strip_nops): New function.
3398 (selftest::test_location_wrappers): New function.
3399 (selftest::tree_c_tests): Call it.
3400 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3401 (maybe_wrap_with_location): New decl.
3402 (EXPR_LOCATION_WRAPPER_P): New macro.
3403 (location_wrapper_p): New inline function.
3404 (tree_strip_any_location_wrapper): New inline function.
3406 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3409 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3410 stack_realign_offset for the largest alignment of stack slot
3412 (ix86_find_max_used_stack_alignment): New function.
3413 (ix86_finalize_stack_frame_flags): Use it. Set
3414 max_used_stack_alignment if we don't realign stack.
3415 * config/i386/i386.h (machine_function): Add
3416 max_used_stack_alignment.
3418 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3420 * config/arm/arm.opt (-mbranch-cost): New option.
3421 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3424 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3427 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3428 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3430 2018-01-10 Richard Biener <rguenther@suse.de>
3433 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3434 early out so it also covers the case where we have a non-NULL
3437 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3439 PR tree-optimization/83753
3440 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3441 for non-strided grouped accesses if the number of elements is 1.
3443 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3446 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3447 * i386.h (TARGET_USE_GATHER): Define.
3448 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3450 2018-01-10 Martin Liska <mliska@suse.cz>
3453 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3454 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3456 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3457 CLEANUP_NO_PARTITIONING is not set.
3459 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3461 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3462 for vectors, as a partial revert of r254296.
3463 * rtl.h (const_vec_p): Delete.
3464 (const_vec_duplicate_p): Don't test for vector CONSTs.
3465 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3466 * expmed.c (make_tree): Likewise.
3469 * common.md (E, F): Use CONSTANT_P instead of checking for
3471 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3472 checking for CONST_VECTOR.
3474 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3477 * predict.c (force_edge_cold): Handle in more sane way edges
3480 2018-01-09 Carl Love <cel@us.ibm.com>
3482 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3484 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3485 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3486 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3487 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3488 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3489 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3490 * config/rs6000/rs6000-protos.h: Add extern defition for
3491 rs6000_generate_float2_double_code.
3492 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3494 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3495 (float2_v2df): Add define_expand.
3497 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3500 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3501 op_mode in the force_to_mode call.
3503 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3505 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3506 instead of checking each element individually.
3507 (aarch64_evpc_uzp): Likewise.
3508 (aarch64_evpc_zip): Likewise.
3509 (aarch64_evpc_ext): Likewise.
3510 (aarch64_evpc_rev): Likewise.
3511 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3512 instead of checking each element individually. Return true without
3514 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3515 whether all selected elements come from the same input, instead of
3516 checking each element individually. Remove calls to gen_rtx_REG,
3517 start_sequence and end_sequence and instead assert that no rtl is
3520 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3522 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3523 order of HIGH and CONST checks.
3525 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3527 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3528 if the destination isn't an SSA_NAME.
3530 2018-01-09 Richard Biener <rguenther@suse.de>
3532 PR tree-optimization/83668
3533 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3535 (canonicalize_loop_form): ... here, renamed from ...
3536 (canonicalize_loop_closed_ssa_form): ... this and amended to
3537 swap successor edges for loop exit blocks to make us use
3538 the RPO order we need for initial schedule generation.
3540 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3542 PR tree-optimization/64811
3543 * match.pd: When optimizing comparisons with Inf, avoid
3544 introducing or losing exceptions from comparisons with NaN.
3546 2018-01-09 Martin Liska <mliska@suse.cz>
3549 * asan.c (shadow_mem_size): Add gcc_assert.
3551 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3553 Don't save registers in main().
3556 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3557 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3558 * config/avr/avr.c (avr_set_current_function): Don't error if
3559 naked, OS_task or OS_main are specified at the same time.
3560 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3562 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3564 * common/config/avr/avr-common.c (avr_option_optimization_table):
3565 Switch on -mmain-is-OS_task for optimizing compilations.
3567 2018-01-09 Richard Biener <rguenther@suse.de>
3569 PR tree-optimization/83572
3570 * graphite.c: Include cfganal.h.
3571 (graphite_transform_loops): Connect infinite loops to exit
3572 and remove fake edges at the end.
3574 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3576 * ipa-inline.c (edge_badness): Revert accidental checkin.
3578 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3581 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3582 symbols; not inline clones.
3584 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3587 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3588 hard registers. Formatting fixes.
3590 PR preprocessor/83722
3591 * gcc.c (try_generate_repro): Pass
3592 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3593 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3596 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3597 Kito Cheng <kito.cheng@gmail.com>
3599 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3600 (riscv_leaf_function_p): Delete.
3601 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3603 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3605 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3607 (do_ifelse): New function.
3608 (do_isel): New function.
3609 (do_sub3): New function.
3610 (do_add3): New function.
3611 (do_load_mask_compare): New function.
3612 (do_overlap_load_compare): New function.
3613 (expand_compare_loop): New function.
3614 (expand_block_compare): Call expand_compare_loop() when appropriate.
3615 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3617 (-mblock-compare-inline-loop-limit): New option.
3619 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3622 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3623 Reverse order of second and third operands in first alternative.
3624 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3625 of first and second elements in UNSPEC_VPERMR vector.
3626 (altivec_expand_vec_perm_le): Likewise.
3628 2017-01-08 Jeff Law <law@redhat.com>
3630 PR rtl-optimizatin/81308
3631 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3632 (process_switch): If group_case_labels makes a change, then set
3634 (pass_convert_switch::execute): If a switch is converted, then
3635 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3637 PR rtl-optimization/81308
3638 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3641 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3643 PR target/83663 - Revert r255946
3644 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3645 generation for cases where splatting a value is not useful.
3646 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3647 across a vec_duplicate and a paradoxical subreg forming a vector
3648 mode to a vec_concat.
3650 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3652 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3653 -march=armv8.3-a variants.
3654 * config/arm/t-multilib: Likewise.
3655 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3657 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3659 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3661 (cceq_ior_compare_complement): Give it a name so I can use it, and
3662 change boolean_or_operator predicate to boolean_operator so it can
3663 be used to generate a crand.
3664 (eqne): New code iterator.
3665 (bd/bd_neg): New code_attrs.
3666 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3667 a single define_insn.
3668 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3669 decrement (bdnzt/bdnzf/bdzt/bdzf).
3670 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3671 with the new names of the branch decrement patterns, and added the
3672 names of the branch decrement conditional patterns.
3674 2018-01-08 Richard Biener <rguenther@suse.de>
3676 PR tree-optimization/83563
3677 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3680 2018-01-08 Richard Biener <rguenther@suse.de>
3683 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3685 2018-01-08 Richard Biener <rguenther@suse.de>
3687 PR tree-optimization/83685
3688 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3689 references to abnormals.
3691 2018-01-08 Richard Biener <rguenther@suse.de>
3694 * dwarf2out.c (output_indirect_strings): Handle empty
3695 skeleton_debug_str_hash.
3696 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3698 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3700 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3701 (emit_store_direct): Likewise.
3702 (arc_trampoline_adjust_address): Likewise.
3703 (arc_asm_trampoline_template): New function.
3704 (arc_initialize_trampoline): Use asm_trampoline_template.
3705 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3706 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3707 * config/arc/arc.md (flush_icache): Delete pattern.
3709 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3711 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3712 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3715 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3718 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3719 by not USED_FOR_TARGET.
3720 (make_pass_resolve_sw_modes): Likewise.
3722 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3724 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3727 2018-01-08 Richard Biener <rguenther@suse.de>
3730 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3732 2018-01-08 Richard Biener <rguenther@suse.de>
3735 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3737 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3740 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3741 basic blocks with a small number of successors.
3742 (convert_control_dep_chain_into_preds): Improve handling of
3744 (dump_predicates): Split apart into...
3745 (dump_pred_chain): ...here...
3746 (dump_pred_info): ...and here.
3747 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3748 (can_chain_union_be_invalidated_p): Improve check for invalidation
3750 (uninit_uses_cannot_happen): Avoid unnecessary if
3751 convert_control_dep_chain_into_preds yielded nothing.
3753 2018-01-06 Martin Sebor <msebor@redhat.com>
3755 PR tree-optimization/83640
3756 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3757 subtracting negative offset from size.
3758 (builtin_access::overlap): Adjust offset bounds of the access to fall
3759 within the size of the object if possible.
3761 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3763 PR rtl-optimization/83699
3764 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3765 extract_bit_field_as_subreg to cases in which the extracted
3766 value is also a vector.
3768 * lra-constraints.c (process_alt_operands): Test for the equivalence
3769 substitutions when detecting a possible reload cycle.
3771 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3774 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3775 by default if flag_selective_schedling{,2}. Formatting fixes.
3777 PR rtl-optimization/83682
3778 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
3779 if it has non-VECTOR_MODE element mode.
3780 (vec_duplicate_p): Likewise.
3783 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
3784 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
3786 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3789 * config/i386/i386-builtin.def
3790 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
3791 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
3792 Require also OPTION_MASK_ISA_AVX512F in addition to
3793 OPTION_MASK_ISA_GFNI.
3794 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
3795 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
3796 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
3797 to OPTION_MASK_ISA_GFNI.
3798 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
3799 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
3800 OPTION_MASK_ISA_AVX512BW.
3801 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
3802 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
3803 addition to OPTION_MASK_ISA_GFNI.
3804 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
3805 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
3806 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
3807 to OPTION_MASK_ISA_GFNI.
3808 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
3809 a requirement for all ISAs rather than any of them with a few
3811 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
3813 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
3814 bitmasks to be enabled with 3 exceptions, instead of requiring any
3815 enabled ISA with lots of exceptions.
3816 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
3817 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
3818 Change avx512bw in isa attribute to avx512f.
3819 * config/i386/sgxintrin.h: Add license boilerplate.
3820 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
3821 to __AVX512F__ and __AVX512VL to __AVX512VL__.
3822 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
3823 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
3825 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
3826 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
3827 temporarily sse2 rather than sse if not enabled already.
3830 * config/i386/sse.md (VI248_VLBW): Rename to ...
3831 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
3832 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
3833 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
3834 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
3835 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
3836 mode iterator instead of VI248_VLBW.
3838 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
3840 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
3841 (record_modified): Skip clobbers; add debug output.
3842 (param_change_prob): Use sreal frequencies.
3844 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3846 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
3847 punt for user-aligned variables.
3849 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3851 * tree-chrec.c (chrec_contains_symbols): Return true for
3854 2018-01-05 Sudakshina Das <sudi.das@arm.com>
3857 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
3858 of (x|y) == x for BICS pattern.
3860 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3862 PR tree-optimization/83605
3863 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
3864 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
3867 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
3869 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
3870 * config/epiphany/rtems.h: New file.
3872 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3873 Uros Bizjak <ubizjak@gmail.com>
3876 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
3877 QIreg_operand instead of register_operand predicate.
3878 * config/i386/i386.c (ix86_rop_should_change_byte_p,
3879 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
3880 comments instead of -fmitigate[-_]rop.
3882 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3885 * cgraphunit.c (symbol_table::compile): Switch to text_section
3886 before calling assembly_start debug hook.
3887 * run-rtl-passes.c (run_rtl_passes): Likewise.
3890 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3892 * tree-vrp.c (extract_range_from_binary_expr_1): Check
3893 range_int_cst_p rather than !symbolic_range_p before calling
3894 extract_range_from_multiplicative_op_1.
3896 2017-01-04 Jeff Law <law@redhat.com>
3898 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
3899 redundant test in assertion.
3901 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3903 * doc/rtl.texi: Document machine_mode wrapper classes.
3905 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3907 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
3910 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3912 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
3913 the VEC_PERM_EXPR fold to fail.
3915 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3918 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
3919 to switched_sections.
3921 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3924 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
3927 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
3930 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
3931 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
3933 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3936 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
3937 is BLKmode and bitpos not zero or mode change is needed.
3939 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3942 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
3945 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
3948 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
3949 instead of MULT rtx. Update all corresponding splitters.
3951 (*ssub<modesuffix>): Ditto.
3953 (*cmp_sadd_di): Update split patterns.
3954 (*cmp_sadd_si): Ditto.
3955 (*cmp_sadd_sidi): Ditto.
3956 (*cmp_ssub_di): Ditto.
3957 (*cmp_ssub_si): Ditto.
3958 (*cmp_ssub_sidi): Ditto.
3959 * config/alpha/predicates.md (const23_operand): New predicate.
3960 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
3961 Look for ASHIFT, not MULT inner operand.
3962 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
3964 2018-01-04 Martin Liska <mliska@suse.cz>
3966 PR gcov-profile/83669
3967 * gcov.c (output_intermediate_file): Add version to intermediate
3969 * doc/gcov.texi: Document new field 'version' in intermediate
3970 file format. Fix location of '-k' option of gcov command.
3972 2018-01-04 Martin Liska <mliska@suse.cz>
3975 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
3977 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3979 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
3981 2018-01-03 Martin Sebor <msebor@redhat.com>
3983 PR tree-optimization/83655
3984 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
3985 checking calls with invalid arguments.
3987 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3989 * tree-vect-stmts.c (vect_get_store_rhs): New function.
3990 (vectorizable_mask_load_store): Delete.
3991 (vectorizable_call): Return false for masked loads and stores.
3992 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
3993 instead of gimple_assign_rhs1.
3994 (vectorizable_load): Handle IFN_MASK_LOAD.
3995 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
3997 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3999 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4001 (vectorizable_mask_load_store): ...here.
4002 (vectorizable_load): ...and here.
4004 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4006 * tree-vect-stmts.c (vect_build_all_ones_mask)
4007 (vect_build_zero_merge_argument): New functions, split out from...
4008 (vectorizable_load): ...here.
4010 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4012 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4014 (vectorizable_mask_load_store): ...here.
4015 (vectorizable_store): ...and here.
4017 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4019 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4021 (vectorizable_mask_load_store): ...here.
4023 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4025 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4026 (vect_model_store_cost): Take a vec_load_store_type instead of a
4028 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4029 (vect_model_store_cost): Take a vec_load_store_type instead of a
4031 (vectorizable_mask_load_store): Update accordingly.
4032 (vectorizable_store): Likewise.
4033 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4035 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4037 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4038 IFN_MASK_LOAD calls here rather than...
4039 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4041 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4042 Alan Hayward <alan.hayward@arm.com>
4043 David Sherwood <david.sherwood@arm.com>
4045 * expmed.c (extract_bit_field_1): For vector extracts,
4046 fall back to extract_bit_field_as_subreg if vec_extract
4049 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4050 Alan Hayward <alan.hayward@arm.com>
4051 David Sherwood <david.sherwood@arm.com>
4053 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4054 they are variable or constant sized.
4055 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4056 slots for constant-sized data.
4058 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4059 Alan Hayward <alan.hayward@arm.com>
4060 David Sherwood <david.sherwood@arm.com>
4062 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4063 handling COND_EXPRs with boolean comparisons, try to find a better
4064 basis for the mask type than the boolean itself.
4066 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4068 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4069 is calculated and how it can be overridden.
4070 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4071 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4073 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4076 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4077 Alan Hayward <alan.hayward@arm.com>
4078 David Sherwood <david.sherwood@arm.com>
4080 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4081 Remove the mode argument.
4082 (aarch64_simd_valid_immediate): Remove the mode and inverse
4084 * config/aarch64/iterators.md (bitsize): New iterator.
4085 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4086 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4087 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4088 aarch64_simd_valid_immediate.
4089 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4090 (aarch64_reg_or_bic_imm): Likewise.
4091 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4092 with an insn_type enum and msl with a modifier_type enum.
4093 Replace element_width with a scalar_mode. Change the shift
4094 to unsigned int. Add constructors for scalar_float_mode and
4095 scalar_int_mode elements.
4096 (aarch64_vect_float_const_representable_p): Delete.
4097 (aarch64_can_const_movi_rtx_p)
4098 (aarch64_simd_scalar_immediate_valid_for_move)
4099 (aarch64_simd_make_constant): Update call to
4100 aarch64_simd_valid_immediate.
4101 (aarch64_advsimd_valid_immediate_hs): New function.
4102 (aarch64_advsimd_valid_immediate): Likewise.
4103 (aarch64_simd_valid_immediate): Remove mode and inverse
4104 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4105 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4106 and aarch64_float_const_representable_p on the result.
4107 (aarch64_output_simd_mov_immediate): Remove mode argument.
4108 Update call to aarch64_simd_valid_immediate and use of
4109 simd_immediate_info.
4110 (aarch64_output_scalar_simd_mov_immediate): Update call
4113 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4114 Alan Hayward <alan.hayward@arm.com>
4115 David Sherwood <david.sherwood@arm.com>
4117 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4118 (mode_nunits): Likewise CONST_MODE_NUNITS.
4119 * machmode.def (ADJUST_NUNITS): Document.
4120 * genmodes.c (mode_data::need_nunits_adj): New field.
4121 (blank_mode): Update accordingly.
4122 (adj_nunits): New variable.
4123 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4125 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4126 listed in adj_nunits.
4127 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4128 listed in adj_nunits. Don't emit case statements for such modes.
4129 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4130 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4131 nothing if adj_nunits is nonnull.
4132 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4133 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4134 (emit_mode_fbit): Update use of print_maybe_const_decl.
4135 (emit_move_size): Likewise. Treat the array as non-const
4137 (emit_mode_adjustments): Handle adj_nunits.
4139 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4141 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4142 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4143 (VECTOR_MODES): Use it.
4144 (make_vector_modes): Take the prefix as an argument.
4146 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4147 Alan Hayward <alan.hayward@arm.com>
4148 David Sherwood <david.sherwood@arm.com>
4150 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4151 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4152 for MODE_VECTOR_BOOL.
4153 * machmode.def (VECTOR_BOOL_MODE): Document.
4154 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4155 (make_vector_bool_mode): New function.
4156 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4158 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4159 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4161 * stor-layout.c (int_mode_for_mode): Likewise.
4162 * tree.c (build_vector_type_for_mode): Likewise.
4163 * varasm.c (output_constant_pool_2): Likewise.
4164 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4165 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4166 for MODE_VECTOR_BOOL.
4167 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4168 of mode class checks.
4169 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4170 instead of a list of mode class checks.
4171 (expand_vector_scalar_condition): Likewise.
4172 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4174 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4175 Alan Hayward <alan.hayward@arm.com>
4176 David Sherwood <david.sherwood@arm.com>
4178 * machmode.h (mode_size): Change from unsigned short to
4180 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4181 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4182 or if measurement_type is not polynomial.
4183 (fixed_size_mode::includes_p): Check for constant-sized modes.
4184 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4185 return a poly_uint16 rather than an unsigned short.
4186 (emit_mode_size): Change the type of mode_size from unsigned short
4187 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4188 (emit_mode_adjustments): Cope with polynomial vector sizes.
4189 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4191 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4193 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4194 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4195 * caller-save.c (setup_save_areas): Likewise.
4196 (replace_reg_with_saved_mem): Likewise.
4197 * calls.c (emit_library_call_value_1): Likewise.
4198 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4199 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4200 (gen_lowpart_for_combine): Likewise.
4201 * convert.c (convert_to_integer_1): Likewise.
4202 * cse.c (equiv_constant, cse_insn): Likewise.
4203 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4204 (cselib_subst_to_values): Likewise.
4205 * dce.c (word_dce_process_block): Likewise.
4206 * df-problems.c (df_word_lr_mark_ref): Likewise.
4207 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4208 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4209 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4210 (rtl_for_decl_location): Likewise.
4211 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4212 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4213 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4214 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4215 (expand_expr_real_1): Likewise.
4216 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4217 (pad_below): Likewise.
4218 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4219 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4220 * ira.c (get_subreg_tracking_sizes): Likewise.
4221 * ira-build.c (ira_create_allocno_objects): Likewise.
4222 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4223 (ira_sort_regnos_for_alter_reg): Likewise.
4224 * ira-costs.c (record_operand_costs): Likewise.
4225 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4226 (resolve_simple_move): Likewise.
4227 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4228 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4229 (lra_constraints): Likewise.
4230 (CONST_POOL_OK_P): Reject variable-sized modes.
4231 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4232 (add_pseudo_to_slot, lra_spill): Likewise.
4233 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4234 * optabs-query.c (get_best_extraction_insn): Likewise.
4235 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4236 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4237 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4238 * recog.c (offsettable_address_addr_space_p): Likewise.
4239 * regcprop.c (maybe_mode_change): Likewise.
4240 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4241 * regrename.c (build_def_use): Likewise.
4242 * regstat.c (dump_reg_info): Likewise.
4243 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4244 (find_reloads, find_reloads_subreg_address): Likewise.
4245 * reload1.c (eliminate_regs_1): Likewise.
4246 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4247 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4248 (simplify_binary_operation_1, simplify_subreg): Likewise.
4249 * targhooks.c (default_function_arg_padding): Likewise.
4250 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4251 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4252 (verify_gimple_assign_ternary): Likewise.
4253 * tree-inline.c (estimate_move_cost): Likewise.
4254 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4255 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4256 (get_address_cost_ainc): Likewise.
4257 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4258 (vect_supportable_dr_alignment): Likewise.
4259 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4260 (vectorizable_reduction): Likewise.
4261 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4262 (vectorizable_operation, vectorizable_load): Likewise.
4263 * tree.c (build_same_sized_truth_vector_type): Likewise.
4264 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4265 * var-tracking.c (emit_note_insn_var_location): Likewise.
4266 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4267 (ADDR_VEC_ALIGN): Likewise.
4269 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4270 Alan Hayward <alan.hayward@arm.com>
4271 David Sherwood <david.sherwood@arm.com>
4273 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4275 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4276 or if measurement_type is polynomial.
4277 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4278 * combine.c (make_extraction): Likewise.
4279 * dse.c (find_shift_sequence): Likewise.
4280 * dwarf2out.c (mem_loc_descriptor): Likewise.
4281 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4282 (extract_bit_field, extract_low_bits): Likewise.
4283 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4284 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4285 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4286 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4287 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4288 * reload.c (find_reloads): Likewise.
4289 * reload1.c (alter_reg): Likewise.
4290 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4291 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4292 * tree-if-conv.c (predicate_mem_writes): Likewise.
4293 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4294 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4295 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4296 * valtrack.c (dead_debug_insert_temp): Likewise.
4297 * varasm.c (mergeable_constant_section): Likewise.
4298 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4300 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4301 Alan Hayward <alan.hayward@arm.com>
4302 David Sherwood <david.sherwood@arm.com>
4304 * expr.c (expand_assignment): Cope with polynomial mode sizes
4305 when assigning to a CONCAT.
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 * machmode.h (mode_precision): Change from unsigned short to
4313 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4315 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4316 or if measurement_type is not polynomial.
4317 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4318 in which the mode is already known to be a scalar_int_mode.
4319 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4320 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4322 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4323 for GET_MODE_PRECISION.
4324 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4325 for GET_MODE_PRECISION.
4326 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4328 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4329 (expand_field_assignment, make_extraction): Likewise.
4330 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4331 (get_last_value): Likewise.
4332 * convert.c (convert_to_integer_1): Likewise.
4333 * cse.c (cse_insn): Likewise.
4334 * expr.c (expand_expr_real_1): Likewise.
4335 * lra-constraints.c (simplify_operand_subreg): Likewise.
4336 * optabs-query.c (can_atomic_load_p): Likewise.
4337 * optabs.c (expand_atomic_load): Likewise.
4338 (expand_atomic_store): Likewise.
4339 * ree.c (combine_reaching_defs): Likewise.
4340 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4341 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4342 * tree.h (type_has_mode_precision_p): Likewise.
4343 * ubsan.c (instrument_si_overflow): Likewise.
4345 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4346 Alan Hayward <alan.hayward@arm.com>
4347 David Sherwood <david.sherwood@arm.com>
4349 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4350 polynomial numbers of units.
4351 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4352 (valid_vector_subparts_p): New function.
4353 (build_vector_type): Remove temporary shim and take the number
4354 of units as a poly_uint64 rather than an int.
4355 (build_opaque_vector_type): Take the number of units as a
4356 poly_uint64 rather than an int.
4357 * tree.c (build_vector_from_ctor): Handle polynomial
4358 TYPE_VECTOR_SUBPARTS.
4359 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4360 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4361 (build_vector_from_val): If the number of units is variable,
4362 use build_vec_duplicate_cst for constant operands and
4363 VEC_DUPLICATE_EXPR otherwise.
4364 (make_vector_type): Remove temporary is_constant ().
4365 (build_vector_type, build_opaque_vector_type): Take the number of
4366 units as a poly_uint64 rather than an int.
4367 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4369 * cfgexpand.c (expand_debug_expr): Likewise.
4370 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4371 (store_constructor, expand_expr_real_1): Likewise.
4372 (const_scalar_mask_from_tree): Likewise.
4373 * fold-const-call.c (fold_const_reduction): Likewise.
4374 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4375 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4376 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4377 (fold_relational_const): Likewise.
4378 (native_interpret_vector): Likewise. Change the size from an
4379 int to an unsigned int.
4380 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4381 TYPE_VECTOR_SUBPARTS.
4382 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4383 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4384 duplicating a non-constant operand into a variable-length vector.
4385 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4386 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4387 * ipa-icf.c (sem_variable::equals): Likewise.
4388 * match.pd: Likewise.
4389 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4390 * print-tree.c (print_node): Likewise.
4391 * stor-layout.c (layout_type): Likewise.
4392 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4393 * tree-cfg.c (verify_gimple_comparison): Likewise.
4394 (verify_gimple_assign_binary): Likewise.
4395 (verify_gimple_assign_ternary): Likewise.
4396 (verify_gimple_assign_single): Likewise.
4397 * tree-pretty-print.c (dump_generic_node): Likewise.
4398 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4399 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4400 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4401 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4402 (vect_shift_permute_load_chain): Likewise.
4403 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4404 (expand_vector_condition, optimize_vector_constructor): Likewise.
4405 (lower_vec_perm, get_compute_type): Likewise.
4406 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4407 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4408 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4409 (vect_recog_mask_conversion_pattern): Likewise.
4410 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4411 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4412 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4413 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4414 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4415 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4416 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4417 (supportable_widening_operation): Likewise.
4418 (supportable_narrowing_operation): Likewise.
4419 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4421 * varasm.c (output_constant): Likewise.
4423 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4424 Alan Hayward <alan.hayward@arm.com>
4425 David Sherwood <david.sherwood@arm.com>
4427 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4428 so that both the length == 3 and length != 3 cases set up their
4429 own permute vectors. Add comments explaining why we know the
4430 number of elements is constant.
4431 (vect_permute_load_chain): Likewise.
4433 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4434 Alan Hayward <alan.hayward@arm.com>
4435 David Sherwood <david.sherwood@arm.com>
4437 * machmode.h (mode_nunits): Change from unsigned char to
4439 (ONLY_FIXED_SIZE_MODES): New macro.
4440 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4441 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4442 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4444 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4445 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4446 or if measurement_type is not polynomial.
4447 * genmodes.c (ZERO_COEFFS): New macro.
4448 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4450 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4451 Use ZERO_COEFFS when emitting initializers.
4452 * data-streamer.h (bp_pack_poly_value): New function.
4453 (bp_unpack_poly_value): Likewise.
4454 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4455 for GET_MODE_NUNITS.
4456 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4457 for GET_MODE_NUNITS.
4458 * tree.c (make_vector_type): Remove temporary shim and make
4459 the real function take the number of units as a poly_uint64
4461 (build_vector_type_for_mode): Handle polynomial nunits.
4462 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4463 * emit-rtl.c (const_vec_series_p_1): Likewise.
4464 (gen_rtx_CONST_VECTOR): Likewise.
4465 * fold-const.c (test_vec_duplicate_folding): Likewise.
4466 * genrecog.c (validate_pattern): Likewise.
4467 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4468 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4469 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4470 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4471 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4472 * rtlanal.c (subreg_get_info): Likewise.
4473 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4474 (vect_grouped_load_supported): Likewise.
4475 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4476 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4477 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4478 (simplify_const_unary_operation, simplify_binary_operation_1)
4479 (simplify_const_binary_operation, simplify_ternary_operation)
4480 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4481 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4482 instead of CONST_VECTOR_NUNITS.
4483 * varasm.c (output_constant_pool_2): Likewise.
4484 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4485 explicit-encoded elements in the XVEC for variable-length vectors.
4487 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4489 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4491 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4492 Alan Hayward <alan.hayward@arm.com>
4493 David Sherwood <david.sherwood@arm.com>
4495 * coretypes.h (fixed_size_mode): Declare.
4496 (fixed_size_mode_pod): New typedef.
4497 * builtins.h (target_builtins::x_apply_args_mode)
4498 (target_builtins::x_apply_result_mode): Change type to
4499 fixed_size_mode_pod.
4500 * builtins.c (apply_args_size, apply_result_size, result_vector)
4501 (expand_builtin_apply_args_1, expand_builtin_apply)
4502 (expand_builtin_return): Update accordingly.
4504 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4506 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4507 * cselib.c (cselib_hash_rtx): Likewise.
4508 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4509 CONST_VECTOR encoding.
4511 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4512 Jeff Law <law@redhat.com>
4515 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4516 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4517 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4518 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4521 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4522 explicitly probe *sp in a noreturn function if there were any callee
4523 register saves or frame pointer is needed.
4525 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4528 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4529 BLKmode for ternary, binary or unary expressions.
4532 * var-tracking.c (delete_vta_debug_insn): New inline function.
4533 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4534 insns from get_insns () to NULL instead of each bb separately.
4535 Use delete_vta_debug_insn. No longer static.
4536 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4537 delete_vta_debug_insns callers.
4538 * rtl.h (delete_vta_debug_insns): Declare.
4539 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4540 instead of variable_tracking_main.
4542 2018-01-03 Martin Sebor <msebor@redhat.com>
4544 PR tree-optimization/83603
4545 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4546 arguments past the endof the argument list in functions declared
4547 without a prototype.
4548 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4549 Avoid checking when arguments are null.
4551 2018-01-03 Martin Sebor <msebor@redhat.com>
4554 * doc/extend.texi (attribute const): Fix a typo.
4555 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4556 issuing -Wsuggest-attribute for void functions.
4558 2018-01-03 Martin Sebor <msebor@redhat.com>
4560 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4561 offset_int::from instead of wide_int::to_shwi.
4562 (maybe_diag_overlap): Remove assertion.
4563 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4564 * gimple-ssa-sprintf.c (format_directive): Same.
4565 (parse_directive): Same.
4566 (sprintf_dom_walker::compute_format_length): Same.
4567 (try_substitute_return_value): Same.
4569 2017-01-03 Jeff Law <law@redhat.com>
4572 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4573 non-constant residual for zero at runtime and avoid probing in
4574 that case. Reorganize code for trailing problem to mirror handling
4577 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4579 PR tree-optimization/83501
4580 * tree-ssa-strlen.c (get_string_cst): New.
4581 (handle_char_store): Call get_string_cst.
4583 2018-01-03 Martin Liska <mliska@suse.cz>
4585 PR tree-optimization/83593
4586 * tree-ssa-strlen.c: Include tree-cfg.h.
4587 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4588 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4589 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4591 (strlen_dom_walker::before_dom_children): Call
4592 gimple_purge_dead_eh_edges. Dump tranformation with details
4594 (strlen_dom_walker::before_dom_children): Update call by adding
4595 new argument cleanup_eh.
4596 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4598 2018-01-03 Martin Liska <mliska@suse.cz>
4601 * cif-code.def (VARIADIC_THUNK): New enum value.
4602 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4605 2018-01-03 Jan Beulich <jbeulich@suse.com>
4607 * sse.md (mov<mode>_internal): Tighten condition for when to use
4608 vmovdqu<ssescalarsize> for TI and OI modes.
4610 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4612 Update copyright years.
4614 2018-01-03 Martin Liska <mliska@suse.cz>
4617 * ipa-visibility.c (function_and_variable_visibility): Skip
4618 functions with noipa attribure.
4620 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4622 * gcc.c (process_command): Update copyright notice dates.
4623 * gcov-dump.c (print_version): Ditto.
4624 * gcov.c (print_version): Ditto.
4625 * gcov-tool.c (print_version): Ditto.
4626 * gengtype.c (create_file): Ditto.
4627 * doc/cpp.texi: Bump @copying's copyright year.
4628 * doc/cppinternals.texi: Ditto.
4629 * doc/gcc.texi: Ditto.
4630 * doc/gccint.texi: Ditto.
4631 * doc/gcov.texi: Ditto.
4632 * doc/install.texi: Ditto.
4633 * doc/invoke.texi: Ditto.
4635 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4637 * vector-builder.h (vector_builder::m_full_nelts): Change from
4638 unsigned int to poly_uint64.
4639 (vector_builder::full_nelts): Update prototype accordingly.
4640 (vector_builder::new_vector): Likewise.
4641 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4642 (vector_builder::operator ==): Likewise.
4643 (vector_builder::finalize): Likewise.
4644 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4645 Take the number of elements as a poly_uint64 rather than an
4647 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4648 from unsigned int to poly_uint64.
4649 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4650 (vec_perm_indices::new_vector): Likewise.
4651 (vec_perm_indices::length): Likewise.
4652 (vec_perm_indices::nelts_per_input): Likewise.
4653 (vec_perm_indices::input_nelts): Likewise.
4654 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4655 number of elements per input as a poly_uint64 rather than an
4656 unsigned int. Use the original encoding for variable-length
4657 vectors, rather than clamping each individual element.
4658 For the second and subsequent elements in each pattern,
4659 clamp the step and base before clamping their sum.
4660 (vec_perm_indices::series_p): Handle polynomial element counts.
4661 (vec_perm_indices::all_in_range_p): Likewise.
4662 (vec_perm_indices_to_tree): Likewise.
4663 (vec_perm_indices_to_rtx): Likewise.
4664 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4665 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4666 (tree_vector_builder::new_binary_operation): Handle polynomial
4667 element counts. Return false if we need to know the number
4668 of elements at compile time.
4669 * fold-const.c (fold_vec_perm): Punt if the number of elements
4670 isn't known at compile time.
4672 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4674 * vec-perm-indices.h (vec_perm_builder): Change element type
4675 from HOST_WIDE_INT to poly_int64.
4676 (vec_perm_indices::element_type): Update accordingly.
4677 (vec_perm_indices::clamp): Handle polynomial element_types.
4678 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4679 (vec_perm_indices::all_in_range_p): Likewise.
4680 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4682 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4683 polynomial vec_perm_indices element types.
4684 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4685 * fold-const.c (fold_vec_perm): Likewise.
4686 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4687 * tree-vect-generic.c (lower_vec_perm): Likewise.
4688 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4689 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4690 element type to HOST_WIDE_INT.
4692 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4693 Alan Hayward <alan.hayward@arm.com>
4694 David Sherwood <david.sherwood@arm.com>
4696 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4697 rather than an int. Use plus_constant.
4698 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4699 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4701 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4702 Alan Hayward <alan.hayward@arm.com>
4703 David Sherwood <david.sherwood@arm.com>
4705 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4706 a HOST_WIDE_INT to a poly_int64.
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 (load_register_parameters): Cope with polynomial
4713 mode sizes. Require a constant size for BLKmode parameters
4714 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4715 forces a parameter to be padded at the lsb end in order to
4716 fill a complete number of words, require the parameter size
4717 to be ordered wrt UNITS_PER_WORD.
4719 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4720 Alan Hayward <alan.hayward@arm.com>
4721 David Sherwood <david.sherwood@arm.com>
4723 * reload1.c (spill_stack_slot_width): Change element type
4724 from unsigned int to poly_uint64_pod.
4725 (alter_reg): Treat mode sizes as polynomial.
4727 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4728 Alan Hayward <alan.hayward@arm.com>
4729 David Sherwood <david.sherwood@arm.com>
4731 * reload.c (complex_word_subreg_p): New function.
4732 (reload_inner_reg_of_subreg, push_reload): Use it.
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 * lra-constraints.c (process_alt_operands): Reject matched
4739 operands whose sizes aren't ordered.
4740 (match_reload): Refer to this check here.
4742 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4743 Alan Hayward <alan.hayward@arm.com>
4744 David Sherwood <david.sherwood@arm.com>
4746 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4747 that the mode size is in the set {1, 2, 4, 8, 16}.
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 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4754 Use plus_constant instead of gen_rtx_PLUS.
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 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4761 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4762 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4763 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4764 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4765 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4766 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4767 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4768 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4769 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4771 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4772 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4773 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4774 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4775 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4776 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4777 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4778 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
4779 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
4780 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
4782 * expr.c (emit_move_resolve_push): Treat the input and result
4783 of PUSH_ROUNDING as a poly_int64.
4784 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
4785 (emit_push_insn): Likewise.
4786 * lra-eliminations.c (mark_not_eliminable): Likewise.
4787 * recog.c (push_operand): Likewise.
4788 * reload1.c (elimination_effects): Likewise.
4789 * rtlanal.c (nonzero_bits1): Likewise.
4790 * calls.c (store_one_arg): Likewise. Require the padding to be
4791 known at compile time.
4793 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4794 Alan Hayward <alan.hayward@arm.com>
4795 David Sherwood <david.sherwood@arm.com>
4797 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
4798 Use plus_constant instead of gen_rtx_PLUS.
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 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
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 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
4812 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
4813 via stack temporaries. Treat the mode size as polynomial too.
4815 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4816 Alan Hayward <alan.hayward@arm.com>
4817 David Sherwood <david.sherwood@arm.com>
4819 * expr.c (expand_expr_real_2): When handling conversions involving
4820 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
4821 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
4822 as a poly_uint64 too.
4824 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4825 Alan Hayward <alan.hayward@arm.com>
4826 David Sherwood <david.sherwood@arm.com>
4828 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
4830 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4831 Alan Hayward <alan.hayward@arm.com>
4832 David Sherwood <david.sherwood@arm.com>
4834 * combine.c (can_change_dest_mode): Handle polynomial
4835 REGMODE_NATURAL_SIZE.
4836 * expmed.c (store_bit_field_1): Likewise.
4837 * expr.c (store_constructor): Likewise.
4838 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
4839 and polynomial REGMODE_NATURAL_SIZE.
4840 (gen_lowpart_common): Likewise.
4841 * reginfo.c (record_subregs_of_mode): Likewise.
4842 * rtlanal.c (read_modify_subreg_p): Likewise.
4844 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4845 Alan Hayward <alan.hayward@arm.com>
4846 David Sherwood <david.sherwood@arm.com>
4848 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
4849 numbers of elements.
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 * match.pd: Cope with polynomial numbers of vector elements.
4857 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4858 Alan Hayward <alan.hayward@arm.com>
4859 David Sherwood <david.sherwood@arm.com>
4861 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
4862 in a POINTER_PLUS_EXPR.
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 * omp-simd-clone.c (simd_clone_subparts): New function.
4869 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
4870 (ipa_simd_modify_function_body): Likewise.
4872 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4873 Alan Hayward <alan.hayward@arm.com>
4874 David Sherwood <david.sherwood@arm.com>
4876 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
4877 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
4878 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
4879 (expand_vector_condition, vector_element): Likewise.
4880 (subparts_gt): New function.
4881 (get_compute_type): Use subparts_gt.
4882 (count_type_subparts): Delete.
4883 (expand_vector_operations_1): Use subparts_gt instead of
4884 count_type_subparts.
4886 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4887 Alan Hayward <alan.hayward@arm.com>
4888 David Sherwood <david.sherwood@arm.com>
4890 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
4891 (vect_compile_time_alias): ...this new function. Do the calculation
4892 on poly_ints rather than trees.
4893 (vect_prune_runtime_alias_test_list): Update call accordingly.
4895 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4896 Alan Hayward <alan.hayward@arm.com>
4897 David Sherwood <david.sherwood@arm.com>
4899 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
4901 (vect_schedule_slp_instance): Likewise.
4903 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4904 Alan Hayward <alan.hayward@arm.com>
4905 David Sherwood <david.sherwood@arm.com>
4907 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
4908 constant and extern definitions for variable-length vectors.
4909 (vect_get_constant_vectors): Note that the number of units
4910 is known to be constant.
4912 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4913 Alan Hayward <alan.hayward@arm.com>
4914 David Sherwood <david.sherwood@arm.com>
4916 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
4917 of units as polynomial. Choose between WIDE and NARROW based
4920 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4921 Alan Hayward <alan.hayward@arm.com>
4922 David Sherwood <david.sherwood@arm.com>
4924 * tree-vect-stmts.c (simd_clone_subparts): New function.
4925 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
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 (vectorizable_call): Treat the number of
4932 vectors as polynomial. Use build_index_vector for
4935 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4936 Alan Hayward <alan.hayward@arm.com>
4937 David Sherwood <david.sherwood@arm.com>
4939 * tree-vect-stmts.c (get_load_store_type): Treat the number of
4940 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
4941 for variable-length vectors.
4942 (vectorizable_mask_load_store): Treat the number of units as
4943 polynomial, asserting that it is constant if the condition has
4944 already been enforced.
4945 (vectorizable_store, vectorizable_load): Likewise.
4947 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4948 Alan Hayward <alan.hayward@arm.com>
4949 David Sherwood <david.sherwood@arm.com>
4951 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
4952 of units as polynomial. Punt if we can't tell at compile time
4953 which vector contains the final result.
4955 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4956 Alan Hayward <alan.hayward@arm.com>
4957 David Sherwood <david.sherwood@arm.com>
4959 * tree-vect-loop.c (vectorizable_induction): Treat the number
4960 of units as polynomial. Punt on SLP inductions. Use an integer
4961 VEC_SERIES_EXPR for variable-length integer reductions. Use a
4962 cast of such a series for variable-length floating-point
4965 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4966 Alan Hayward <alan.hayward@arm.com>
4967 David Sherwood <david.sherwood@arm.com>
4969 * tree.h (build_index_vector): Declare.
4970 * tree.c (build_index_vector): New function.
4971 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
4972 of units as polynomial, forcibly converting it to a constant if
4973 vectorizable_reduction has already enforced the condition.
4974 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
4975 to create a {1,2,3,...} vector.
4976 (vectorizable_reduction): Treat the number of units as polynomial.
4977 Choose vectype_in based on the largest scalar element size rather
4978 than the smallest number of units. Enforce the restrictions
4981 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4982 Alan Hayward <alan.hayward@arm.com>
4983 David Sherwood <david.sherwood@arm.com>
4985 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
4986 number of units as polynomial.
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 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
4993 * target.def (autovectorize_vector_sizes): Return the vector sizes
4994 by pointer, using vector_sizes rather than a bitmask.
4995 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
4996 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
4997 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
4999 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5000 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5001 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5002 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5003 * omp-general.c (omp_max_vf): Likewise.
5004 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5005 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5006 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5007 * tree-vect-slp.c (vect_slp_bb): Likewise.
5008 * doc/tm.texi: Regenerate.
5009 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5011 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5012 the vector size as a poly_uint64 rather than an unsigned int.
5013 (current_vector_size): Change from an unsigned int to a poly_uint64.
5014 (get_vectype_for_scalar_type): Update accordingly.
5015 * tree.h (build_truth_vector_type): Take the size and number of
5016 units as a poly_uint64 rather than an unsigned int.
5017 (build_vector_type): Add a temporary overload that takes
5018 the number of units as a poly_uint64 rather than an unsigned int.
5019 * tree.c (make_vector_type): Likewise.
5020 (build_truth_vector_type): Take the number of units as a poly_uint64
5021 rather than an unsigned int.
5023 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5024 Alan Hayward <alan.hayward@arm.com>
5025 David Sherwood <david.sherwood@arm.com>
5027 * target.def (get_mask_mode): Take the number of units and length
5028 as poly_uint64s rather than unsigned ints.
5029 * targhooks.h (default_get_mask_mode): Update accordingly.
5030 * targhooks.c (default_get_mask_mode): Likewise.
5031 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5032 * doc/tm.texi: Regenerate.
5034 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5035 Alan Hayward <alan.hayward@arm.com>
5036 David Sherwood <david.sherwood@arm.com>
5038 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5039 * omp-general.c (omp_max_vf): Likewise.
5040 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5041 (expand_omp_simd): Handle polynomial safelen.
5042 * omp-low.c (omplow_simd_context): Add a default constructor.
5043 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5044 (lower_rec_simd_input_clauses): Update accordingly.
5045 (lower_rec_input_clauses): Likewise.
5047 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5048 Alan Hayward <alan.hayward@arm.com>
5049 David Sherwood <david.sherwood@arm.com>
5051 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5052 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5053 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5054 (vect_analyze_slp_cost): Likewise.
5055 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5056 (vect_model_load_cost): Likewise.
5058 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5059 Alan Hayward <alan.hayward@arm.com>
5060 David Sherwood <david.sherwood@arm.com>
5062 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5063 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5064 from an unsigned int * to a poly_uint64_pod *.
5065 (calculate_unrolling_factor): New function.
5066 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5068 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5069 Alan Hayward <alan.hayward@arm.com>
5070 David Sherwood <david.sherwood@arm.com>
5072 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5073 from an unsigned int to a poly_uint64.
5074 (_loop_vec_info::slp_unrolling_factor): Likewise.
5075 (_loop_vec_info::vectorization_factor): Change from an int
5077 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5078 (vect_get_num_vectors): New function.
5079 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5080 (vect_get_num_copies): Use vect_get_num_vectors.
5081 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5082 to an unsigned int *.
5083 (vect_analyze_data_refs): Change min_vf from an int * to a
5085 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5086 than an unsigned HOST_WIDE_INT.
5087 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5088 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5089 to an unsigned int *.
5090 (vect_analyze_data_ref_dependences): Likewise.
5091 (vect_compute_data_ref_alignment): Handle polynomial vf.
5092 (vect_enhance_data_refs_alignment): Likewise.
5093 (vect_prune_runtime_alias_test_list): Likewise.
5094 (vect_shift_permute_load_chain): Likewise.
5095 (vect_supportable_dr_alignment): Likewise.
5096 (dependence_distance_ge_vf): Take the vectorization factor as a
5097 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5098 (vect_analyze_data_refs): Change min_vf from an int * to a
5100 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5101 vfm1 as a poly_uint64 rather than an int. Make the same change
5102 for the returned bound_scalar.
5103 (vect_gen_vector_loop_niters): Handle polynomial vf.
5104 (vect_do_peeling): Likewise. Update call to
5105 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5106 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5108 * tree-vect-loop.c (vect_determine_vectorization_factor)
5109 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5110 (vect_get_known_peeling_cost): Likewise.
5111 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5112 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5113 (vect_transform_loop): Likewise. Use the lowest possible VF when
5114 updating the upper bounds of the loop.
5115 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5117 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5118 polynomial unroll factors.
5119 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5120 (vect_make_slp_decision): Likewise.
5121 (vect_supported_load_permutation_p): Likewise, and polynomial
5123 (vect_analyze_slp_cost): Handle polynomial vf.
5124 (vect_slp_analyze_node_operations): Likewise.
5125 (vect_slp_analyze_bb_1): Likewise.
5126 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5127 than an unsigned HOST_WIDE_INT.
5128 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5129 (vectorizable_load): Handle polynomial vf.
5130 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5132 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5134 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5135 Alan Hayward <alan.hayward@arm.com>
5136 David Sherwood <david.sherwood@arm.com>
5138 * match.pd: Handle bit operations involving three constants
5139 and try to fold one pair.
5141 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5143 * tree-vect-loop-manip.c: Include gimple-fold.h.
5144 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5145 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5146 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5147 Add a path that uses a step of VF instead of 1, but disable it
5149 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5150 and niters_no_overflow parameters. Update calls to
5151 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5152 Create a new SSA name if the latter choses to use a ste other
5153 than zero, and return it via niters_vector_mult_vf_var.
5154 * tree-vect-loop.c (vect_transform_loop): Update calls to
5155 vect_do_peeling, vect_gen_vector_loop_niters and
5156 slpeel_make_loop_iterate_ntimes.
5157 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5158 (vect_gen_vector_loop_niters): Update declarations after above changes.
5160 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5162 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5163 128-bit round to integer instructions.
5164 (ceil<mode>2): Likewise.
5165 (btrunc<mode>2): Likewise.
5166 (round<mode>2): Likewise.
5168 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5170 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5171 unaligned VSX load/store on P8/P9.
5172 (expand_block_clear): Allow the use of unaligned VSX
5173 load/store on P8/P9.
5175 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5177 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5179 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5180 swap associated with both a load and a store.
5182 2018-01-02 Andrew Waterman <andrew@sifive.com>
5184 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5185 * config/riscv/riscv.md (clear_cache): Use it.
5187 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5189 * web.c: Remove out-of-date comment.
5191 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5193 * expr.c (fixup_args_size_notes): Check that any existing
5194 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5195 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5196 (emit_single_push_insn): ...here.
5198 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5200 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5201 (const_vector_encoded_nelts): New function.
5202 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5203 (const_vector_int_elt, const_vector_elt): Declare.
5204 * emit-rtl.c (const_vector_int_elt_1): New function.
5205 (const_vector_elt): Likewise.
5206 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5207 of CONST_VECTOR_ELT.
5209 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5211 * expr.c: Include rtx-vector-builder.h.
5212 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5213 directly on the tree encoding.
5214 (const_vector_from_tree): Likewise.
5215 * optabs.c: Include rtx-vector-builder.h.
5216 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5217 sequence of "u" values.
5218 * vec-perm-indices.c: Include rtx-vector-builder.h.
5219 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5220 directly on the vec_perm_indices encoding.
5222 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5224 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5225 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5226 * rtx-vector-builder.h: New file.
5227 * rtx-vector-builder.c: Likewise.
5228 * rtl.h (rtx_def::u2): Add a const_vector field.
5229 (CONST_VECTOR_NPATTERNS): New macro.
5230 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5231 (CONST_VECTOR_DUPLICATE_P): Likewise.
5232 (CONST_VECTOR_STEPPED_P): Likewise.
5233 (CONST_VECTOR_ENCODED_ELT): Likewise.
5234 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5235 (unwrap_const_vec_duplicate): Likewise.
5236 (const_vec_series_p): Check for a non-duplicated vector encoding.
5237 Say that the function only returns true for integer vectors.
5238 * emit-rtl.c: Include rtx-vector-builder.h.
5239 (gen_const_vec_duplicate_1): Delete.
5240 (gen_const_vector): Call gen_const_vec_duplicate instead of
5241 gen_const_vec_duplicate_1.
5242 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5243 (gen_const_vec_duplicate): Use rtx_vector_builder.
5244 (gen_const_vec_series): Likewise.
5245 (gen_rtx_CONST_VECTOR): Likewise.
5246 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5247 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5248 Build a new vector rather than modifying a CONST_VECTOR in-place.
5249 (handle_special_swappables): Update call accordingly.
5250 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5251 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5252 Build a new vector rather than modifying a CONST_VECTOR in-place.
5253 (handle_special_swappables): Update call accordingly.
5255 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5257 * simplify-rtx.c (simplify_const_binary_operation): Use
5258 CONST_VECTOR_ELT instead of XVECEXP.
5260 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5262 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5263 the selector elements to be different from the data elements
5264 if the selector is a VECTOR_CST.
5265 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5266 ssizetype for the selector.
5268 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5270 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5271 before testing each element individually.
5272 * tree-vect-generic.c (lower_vec_perm): Likewise.
5274 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5276 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5277 * selftest-run-tests.c (selftest::run_tests): Call it.
5278 * vector-builder.h (vector_builder::operator ==): New function.
5279 (vector_builder::operator !=): Likewise.
5280 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5281 (vec_perm_indices::all_from_input_p): New function.
5282 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5283 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5284 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5285 instead of reading the VECTOR_CST directly. Detect whether both
5286 vector inputs are the same before constructing the vec_perm_indices,
5287 and update the number of inputs argument accordingly. Use the
5288 utility functions added above. Only construct sel2 if we need to.
5290 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5292 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5293 the broadcast of the low byte.
5294 (expand_mult_highpart): Use an explicit encoding for the permutes.
5295 * optabs-query.c (can_mult_highpart_p): Likewise.
5296 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5297 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5298 (vectorizable_bswap): Likewise.
5299 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5300 explicit encoding for the power-of-2 permutes.
5301 (vect_permute_store_chain): Likewise.
5302 (vect_grouped_load_supported): Likewise.
5303 (vect_permute_load_chain): Likewise.
5305 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5307 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5308 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5309 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5310 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5311 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5312 (vect_gen_perm_mask_any): Likewise.
5314 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5316 * int-vector-builder.h: New file.
5317 * vec-perm-indices.h: Include int-vector-builder.h.
5318 (vec_perm_indices): Redefine as an int_vector_builder.
5319 (auto_vec_perm_indices): Delete.
5320 (vec_perm_builder): Redefine as a stand-alone class.
5321 (vec_perm_indices::vec_perm_indices): New function.
5322 (vec_perm_indices::clamp): Likewise.
5323 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5324 (vec_perm_indices::new_vector): New function.
5325 (vec_perm_indices::new_expanded_vector): Update for new
5326 vec_perm_indices class.
5327 (vec_perm_indices::rotate_inputs): New function.
5328 (vec_perm_indices::all_in_range_p): Operate directly on the
5329 encoded form, without computing elided elements.
5330 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5331 encoding. Update for new vec_perm_indices class.
5332 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5333 the given vec_perm_builder.
5334 (expand_vec_perm_var): Update vec_perm_builder constructor.
5335 (expand_mult_highpart): Use vec_perm_builder instead of
5336 auto_vec_perm_indices.
5337 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5338 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5339 or double series encoding as appropriate.
5340 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5341 vec_perm_indices instead of auto_vec_perm_indices.
5342 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5343 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5344 (vect_permute_store_chain): Likewise.
5345 (vect_grouped_load_supported): Likewise.
5346 (vect_permute_load_chain): Likewise.
5347 (vect_shift_permute_load_chain): Likewise.
5348 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5349 (vect_transform_slp_perm_load): Likewise.
5350 (vect_schedule_slp_instance): Likewise.
5351 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5352 (vectorizable_mask_load_store): Likewise.
5353 (vectorizable_bswap): Likewise.
5354 (vectorizable_store): Likewise.
5355 (vectorizable_load): Likewise.
5356 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5357 vec_perm_indices instead of auto_vec_perm_indices. Use
5358 tree_to_vec_perm_builder to read the vector from a tree.
5359 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5360 vec_perm_builder instead of a vec_perm_indices.
5361 (have_whole_vector_shift): Use vec_perm_builder and
5362 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5363 truncation to calc_vec_perm_mask_for_shift.
5364 (vect_create_epilog_for_reduction): Likewise.
5365 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5366 from auto_vec_perm_indices to vec_perm_indices.
5367 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5368 instead of changing individual elements.
5369 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5370 the vector in d.perm.
5371 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5372 from auto_vec_perm_indices to vec_perm_indices.
5373 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5374 instead of changing individual elements.
5375 (arm_vectorize_vec_perm_const): Use new_vector to install
5376 the vector in d.perm.
5377 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5378 Update vec_perm_builder constructor.
5379 (rs6000_expand_interleave): Likewise.
5380 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5381 (rs6000_expand_interleave): Likewise.
5383 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5385 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5386 to qimode could truncate the indices.
5387 * optabs.c (expand_vec_perm_var): Likewise.
5389 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5391 * Makefile.in (OBJS): Add vec-perm-indices.o.
5392 * vec-perm-indices.h: New file.
5393 * vec-perm-indices.c: Likewise.
5394 * target.h (vec_perm_indices): Replace with a forward class
5396 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5397 * optabs.h: Include vec-perm-indices.h.
5398 (expand_vec_perm): Delete.
5399 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5400 (expand_vec_perm_const): Declare.
5401 * target.def (vec_perm_const_ok): Replace with...
5402 (vec_perm_const): ...this new hook.
5403 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5404 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5405 * doc/tm.texi: Regenerate.
5406 * optabs.def (vec_perm_const): Delete.
5407 * doc/md.texi (vec_perm_const): Likewise.
5408 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5409 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5410 expand_vec_perm for constant permutation vectors. Assert that
5411 the mode of variable permutation vectors is the integer equivalent
5412 of the mode that is being permuted.
5413 * optabs-query.h (selector_fits_mode_p): Declare.
5414 * optabs-query.c: Include vec-perm-indices.h.
5415 (selector_fits_mode_p): New function.
5416 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5417 is defined, instead of checking whether the vec_perm_const_optab
5418 exists. Use targetm.vectorize.vec_perm_const instead of
5419 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5420 fit in the vector mode before using a variable permute.
5421 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5422 vec_perm_indices instead of an rtx.
5423 (expand_vec_perm): Replace with...
5424 (expand_vec_perm_const): ...this new function. Take the selector
5425 as a vec_perm_indices rather than an rtx. Also take the mode of
5426 the selector. Update call to shift_amt_for_vec_perm_mask.
5427 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5428 Use vec_perm_indices::new_expanded_vector to expand the original
5429 selector into bytes. Check whether the indices fit in the vector
5430 mode before using a variable permute.
5431 (expand_vec_perm_var): Make global.
5432 (expand_mult_highpart): Use expand_vec_perm_const.
5433 * fold-const.c: Includes vec-perm-indices.h.
5434 * tree-ssa-forwprop.c: Likewise.
5435 * tree-vect-data-refs.c: Likewise.
5436 * tree-vect-generic.c: Likewise.
5437 * tree-vect-loop.c: Likewise.
5438 * tree-vect-slp.c: Likewise.
5439 * tree-vect-stmts.c: Likewise.
5440 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5442 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5443 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5444 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5445 (aarch64_vectorize_vec_perm_const): ...this new function.
5446 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5447 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5448 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5449 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5450 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5451 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5452 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5454 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5455 check for NEON modes.
5456 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5457 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5458 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5459 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5461 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5462 the old VEC_PERM_CONST conditions.
5463 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5464 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5465 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5466 (ia64_vectorize_vec_perm_const_ok): Merge into...
5467 (ia64_vectorize_vec_perm_const): ...this new function.
5468 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5469 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5470 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5471 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5472 * config/mips/mips.c (mips_expand_vec_perm_const)
5473 (mips_vectorize_vec_perm_const_ok): Merge into...
5474 (mips_vectorize_vec_perm_const): ...this new function.
5475 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5476 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5477 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5478 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5479 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5480 (rs6000_expand_vec_perm_const): Delete.
5481 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5483 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5484 (altivec_expand_vec_perm_const_le): Take each operand individually.
5485 Operate on constant selectors rather than rtxes.
5486 (altivec_expand_vec_perm_const): Likewise. Update call to
5487 altivec_expand_vec_perm_const_le.
5488 (rs6000_expand_vec_perm_const): Delete.
5489 (rs6000_vectorize_vec_perm_const_ok): Delete.
5490 (rs6000_vectorize_vec_perm_const): New function.
5491 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5492 an element count and rtx array.
5493 (rs6000_expand_extract_even): Update call accordingly.
5494 (rs6000_expand_interleave): Likewise.
5495 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5496 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5497 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5498 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5499 (rs6000_expand_vec_perm_const): Delete.
5500 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5501 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5502 (altivec_expand_vec_perm_const_le): Take each operand individually.
5503 Operate on constant selectors rather than rtxes.
5504 (altivec_expand_vec_perm_const): Likewise. Update call to
5505 altivec_expand_vec_perm_const_le.
5506 (rs6000_expand_vec_perm_const): Delete.
5507 (rs6000_vectorize_vec_perm_const_ok): Delete.
5508 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5509 reference to the SPE evmerge intructions.
5510 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5511 an element count and rtx array.
5512 (rs6000_expand_extract_even): Update call accordingly.
5513 (rs6000_expand_interleave): Likewise.
5514 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5515 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5517 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5519 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5521 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5522 vector mode and that that mode matches the mode of the data
5524 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5525 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5526 directly using expand_vec_perm_1 when forcing selectors into
5528 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5530 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5532 * optabs-query.h (can_vec_perm_p): Delete.
5533 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5534 * optabs-query.c (can_vec_perm_p): Split into...
5535 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5536 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5537 particular selector is valid.
5538 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5539 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5540 (vect_grouped_load_supported): Likewise.
5541 (vect_shift_permute_load_chain): Likewise.
5542 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5543 (vect_transform_slp_perm_load): Likewise.
5544 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5545 (vectorizable_bswap): Likewise.
5546 (vect_gen_perm_mask_checked): Likewise.
5547 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5548 implementations of variable permutation vectors into account
5549 when deciding which selector to use.
5550 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5551 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5552 with a false third argument.
5553 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5554 to test whether the constant selector is valid and can_vec_perm_var_p
5555 to test whether a variable selector is valid.
5557 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5559 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5560 * optabs-query.c (can_vec_perm_p): Likewise.
5561 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5562 instead of vec_perm_indices.
5563 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5564 (vect_gen_perm_mask_checked): Likewise,
5565 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5566 (vect_gen_perm_mask_checked): Likewise,
5568 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5570 * optabs-query.h (qimode_for_vec_perm): Declare.
5571 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5572 (qimode_for_vec_perm): ...this new function.
5573 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5575 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5577 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5578 does not have a conditional at the top.
5580 2018-01-02 Richard Biener <rguenther@suse.de>
5582 * ipa-inline.c (big_speedup_p): Fix expression.
5584 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5587 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5590 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5594 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5595 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5596 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5597 cond_taken_branch_cost 3->4.
5599 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5601 PR tree-optimization/83581
5602 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5603 TODO_cleanup_cfg if any changes have been made.
5606 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5607 convert_modes if target mode has the right side, but different mode
5611 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5612 last argument when extracting from CONCAT. If either from_real or
5613 from_imag is NULL, use expansion through memory. If result is not
5614 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5615 the parts directly to inner mode, if even that fails, use expansion
5619 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5620 check for bswap in mode rather than HImode and use that in expand_unop
5623 Copyright (C) 2018 Free Software Foundation, Inc.
5625 Copying and distribution of this file, with or without modification,
5626 are permitted in any medium without royalty provided the copyright
5627 notice and this notice are preserved.