1 2018-01-29 Richard Biener <rguenther@suse.de>
2 Kelvin Nilsen <kelvin@gcc.gnu.org>
5 * tree-vect-stmts.c (vectorizable_call): Don't call
6 targetm.vectorize_builtin_md_vectorized_function if callee is
9 2018-01-22 Carl Love <cel@us.ibm.com>
11 * doc/extend.tex: Fix typo in second arg in
12 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
14 2018-01-29 Richard Biener <rguenther@suse.de>
16 PR tree-optimization/84086
17 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
18 (flush_ssaname_freelist): When SSA names were released reset
21 2018-01-29 Richard Biener <rguenther@suse.de>
23 PR tree-optimization/84057
24 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
25 removed paths when removing edges.
27 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
29 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
30 -mfunction-return=@var{choice}.
32 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
35 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
37 (layout::print_source_line): Likewise.
38 (test_get_line_width_without_trailing_whitespace): Add test cases.
40 2018-01-27 Jakub Jelinek <jakub@redhat.com>
43 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
46 2018-01-26 Jim Wilson <jimw@sifive.com>
48 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
50 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
53 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
55 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
56 and CMP + SUB-immediate -> SUBS.
58 2018-01-26 Martin Sebor <msebor@redhat.com>
60 PR tree-optimization/83896
61 * tree-ssa-strlen.c (get_string_len): Rename...
62 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
63 Avoid assuming length is constant.
64 (handle_char_store): Use HOST_WIDE_INT for string length.
66 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
69 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
70 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
72 2018-01-26 Richard Biener <rguenther@suse.de>
74 PR rtl-optimization/84003
75 * dse.c (record_store): Only record redundant stores when
76 the earlier store aliases at least all accesses the later one does.
78 2018-01-26 Jakub Jelinek <jakub@redhat.com>
80 PR rtl-optimization/83985
81 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
82 REG_CFA_RESTORE insns.
83 (delete_unmarked_insns): Don't ignore separate shrink wrapping
84 REG_CFA_RESTORE insns here.
87 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
88 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
90 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
92 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
93 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
95 (arc_override_options): Likewise.
96 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
98 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
100 * config/arc/arc.h (TARGET_DBNZ): Define.
101 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
102 properly set the tune attribute.
103 (dbnz): Use TARGET_DBNZ guard.
104 * config/arc/arc.opt (mtune): Add core3 option.
106 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
108 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
109 recognize new pic like addresses.
110 (arc_delegitimize_address): Clean up.
112 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
114 * config/arc/arc-arches.def: Option mrf16 valid for all
116 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
117 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
118 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
119 * config/arc/arc-tables.opt: Regenerate.
120 * config/arc/arc.c (arc_conditional_register_usage): Handle
121 reduced register file case.
122 (arc_file_start): Set must have build attributes.
123 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
125 * config/arc/arc.opt (mrf16): Add new option.
126 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
127 * config/arc/genmultilib.awk: Handle new mrf16 option.
128 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
129 * config/arc/t-multilib: Regenerate.
130 * doc/invoke.texi (ARC Options): Document mrf16 option.
132 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
134 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
135 * config/arc/arc.c (arc_handle_secure_attribute): New function.
136 (arc_attribute_table): Add 'secure_call' attribute.
137 (arc_print_operand): Print secure call operand.
138 (arc_function_ok_for_sibcall): Don't optimize tail calls when
140 (arc_is_secure_call_p): New function. * config/arc/arc.md
141 (call_i): Add support for sjli instruction.
142 (call_value_i): Likewise.
143 * config/arc/constraints.md (Csc): New constraint.
145 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
146 John Eric Martin <John.Martin@emmicro-us.com>
148 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
149 * config/arc/arc.c (_arc_jli_section): New struct.
150 (arc_jli_section): New type.
151 (rc_jli_sections): New static variable.
152 (arc_handle_jli_attribute): New function.
153 (arc_attribute_table): Add jli_always and jli_fixed attribute.
154 (arc_file_end): New function.
155 (TARGET_ASM_FILE_END): Define.
156 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
157 (arc_add_jli_section): New function.
158 (jli_call_scan): Likewise.
159 (arc_reorg): Call jli_call_scan.
160 (arc_output_addsi): Remove 'S' from printing asm operand.
161 (arc_is_jli_call_p): New function.
162 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
164 (movhi_insn): Likewise.
165 (movsi_insn): Likewise.
166 (movsi_set_cc_insn): Likewise.
167 (loadqi_update): Likewise.
168 (load_zeroextendqisi_update): Likewise.
169 (load_signextendqisi_update): Likewise.
170 (loadhi_update): Likewise.
171 (load_zeroextendhisi_update): Likewise.
172 (load_signextendhisi_update): Likewise.
173 (loadsi_update): Likewise.
174 (loadsf_update): Likewise.
175 (movsicc_insn): Likewise.
176 (bset_insn): Likewise.
177 (bxor_insn): Likewise.
178 (bclr_insn): Likewise.
179 (bmsk_insn): Likewise.
180 (bicsi3_insn): Likewise.
181 (cmpsi_cc_c_insn): Likewise.
182 (movsi_ne): Likewise.
183 (movsi_cond_exec): Likewise.
184 (clrsbsi2): Likewise.
195 (call_i): Remove 'S' asm letter, add jli instruction.
196 (call_value_i): Likewise.
197 * config/arc/arc.op (mjli-always): New option.
198 * config/arc/constraints.md (Cji): New constraint.
199 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
201 (subsf3_fpx): Likewise.
202 (mulsf3_fpx): Likewise.
203 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
205 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
207 * doc/invoke.texi (ARC): Document mjli-always option.
209 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
211 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
212 and use incw and decw where possible
213 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
215 2018-01-26 Richard Biener <rguenther@suse.de>
217 PR tree-optimization/81082
218 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
219 association if it requires casting to unsigned.
220 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
221 from fold_plusminus_mult_expr to catch important cases late when
222 range info is available.
224 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
226 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
227 * configure.ac (hidden_linkonce): New test.
228 * configure: Regenerate.
229 * config.in: Regenerate.
231 2018-01-26 Julia Koval <julia.koval@intel.com>
233 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
234 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
235 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
236 _mm_mask_bitshuffle_epi64_mask): Fix type.
237 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
238 USI_FTYPE_V4DI_V4DI_USI): Remove.
239 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
240 __builtin_ia32_vpshufbitqmb256_mask,
241 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
242 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
243 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
245 2018-01-26 Alan Modra <amodra@gmail.com>
248 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
249 UNSPEC_VBPERMQ. Sort other unspecs.
251 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
253 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
255 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
258 * predict.c (drop_profile): Do not push/pop cfun; update also
260 (handle_missing_profiles): Fix logic looking for zero profiles.
262 2018-01-25 Jakub Jelinek <jakub@redhat.com>
265 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
266 on functions with #pragma omp declare simd or functions with simd
268 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
269 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
270 Remove trailing \n from warning_at calls.
272 2018-01-25 Tom de Vries <tom@codesourcery.com>
275 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
276 for neutered workers.
278 2018-01-24 Joseph Myers <joseph@codesourcery.com>
281 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
282 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
284 2017-01-08 Jeff Law <law@redhat.com>
287 * i386.c (get_probe_interval): Move to earlier point.
288 (ix86_compute_frame_layout): If -fstack-clash-protection and
289 the frame is larger than the probe interval, then use pushes
290 to save registers rather than reg->mem moves.
291 (ix86_expand_prologue): Remove conditional for int_registers_saved
294 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
297 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
298 min/max for never referenced object.
300 2018-01-24 Jakub Jelinek <jakub@redhat.com>
303 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
305 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
306 attributes from DECL_ATTRIBUTES (decl) without affecting
307 DECL_ATTRIBUTES (current_function_decl).
308 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
309 functions with non-NULL DECL_ABSTRACT_ORIGIN.
311 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
313 PR tree-optimization/83979
314 * fold-const.c (fold_comparison): Use constant_boolean_node
315 instead of boolean_{true,false}_node.
317 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
319 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
322 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
324 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
325 Simplify the clause that sets the length attribute.
326 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
327 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
328 clause that sets the length attribute.
329 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
331 2018-01-24 Tom de Vries <tom@codesourcery.com>
334 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
335 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
336 Add strict parameter.
337 (prevent_branch_around_nothing): Insert dummy insn between branch to
338 label and label with no ptx insn inbetween.
339 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
341 2018-01-24 Tom de Vries <tom@codesourcery.com>
344 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
345 for neutered threads in warp.
346 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
348 2018-01-24 Richard Biener <rguenther@suse.de>
350 PR tree-optimization/83176
351 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
354 2018-01-24 Richard Biener <rguenther@suse.de>
356 PR tree-optimization/82819
357 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
358 code generating pluses that are no-ops in the target precision.
360 2018-01-24 Richard Biener <rguenther@suse.de>
363 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
365 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
367 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
368 to merge probabilities.
369 * predict.c (probably_never_executed): Also mark as cold functions
370 with global 0 profile and guessed local profile.
371 * profile-count.c (profile_probability::combine_with_count): New
373 * profile-count.h (profile_probability::operator*,
374 profile_probability::operator*=, profile_probability::operator/,
375 profile_probability::operator/=): Reduce precision to adjusted
376 and set value to guessed on contradictory divisions.
377 (profile_probability::combine_with_freq): Remove.
378 (profile_probability::combine_wiht_count): Declare.
379 (profile_count::force_nonzero):: Set to adjusted.
380 (profile_count::probability_in):: Set quality to adjusted.
381 * tree-ssa-tail-merge.c (replace_block_by): Use
384 2018-01-23 Andrew Waterman <andrew@sifive.com>
385 Jim Wilson <jimw@sifive.com>
387 * config/riscv/riscv.c (riscv_stack_boundary): New.
388 (riscv_option_override): Set riscv_stack_boundary. Handle
389 riscv_preferred_stack_boundary_arg.
390 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
391 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
392 (STACK_BOUNDARY): Set to riscv_stack_boundary.
393 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
394 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
395 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
397 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
400 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
401 of struct ix86_frame.
402 (ix86_expand_epilogue): Likewise. Add a local variable for
403 the reg_save_offset field in struct ix86_frame.
405 2018-01-23 Bin Cheng <bin.cheng@arm.com>
407 PR tree-optimization/82604
408 * tree-loop-distribution.c (enum partition_kind): New enum item
409 PKIND_PARTIAL_MEMSET.
410 (partition_builtin_p): Support above new enum item.
411 (generate_code_for_partition): Ditto.
412 (compute_access_range): Differentiate cases that equality can be
413 proven at all loops, the innermost loops or no loops.
414 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
415 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
416 (finalize_partitions, distribute_loop): Don't fuse partition of
417 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
418 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
421 2018-01-23 Martin Liska <mliska@suse.cz>
423 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
424 order to ignore the predictor.
425 (PRED_POLYMORPHIC_CALL): Likewise.
426 (PRED_RECURSIVE_CALL): Likewise.
428 2018-01-23 Martin Liska <mliska@suse.cz>
430 * tree-profile.c (tree_profiling): Print function header to
431 aware reader which function we are working on.
432 * value-prof.c (gimple_find_values_to_profile): Do not print
433 not interesting value histograms.
435 2018-01-23 Martin Liska <mliska@suse.cz>
437 * profile-count.h (enum profile_quality): Add
438 profile_uninitialized as the first value. Do not number values
439 as they are zero based.
440 (profile_count::verify): Update sanity check.
441 (profile_probability::verify): Likewise.
443 2018-01-23 Nathan Sidwell <nathan@acm.org>
445 * doc/invoke.texi (ffor-scope): Deprecate.
447 2018-01-23 David Malcolm <dmalcolm@redhat.com>
449 PR tree-optimization/83510
450 * domwalk.c (set_all_edges_as_executable): New function.
451 (dom_walker::dom_walker): Convert bool param
452 "skip_unreachable_blocks" to enum reachability. Move setup of
453 edge flags to set_all_edges_as_executable and only do it when
454 reachability is REACHABLE_BLOCKS.
455 * domwalk.h (enum dom_walker::reachability): New enum.
456 (dom_walker::dom_walker): Convert bool param
457 "skip_unreachable_blocks" to enum reachability.
458 (set_all_edges_as_executable): New decl.
459 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
460 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
462 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
463 but converting true to REACHABLE_BLOCKS.
464 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
466 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
467 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
468 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
470 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
471 if check_all_array_refs will be called.
473 2018-01-23 David Malcolm <dmalcolm@redhat.com>
475 * tree.c (selftest::test_location_wrappers): Add more test
478 2018-01-23 David Malcolm <dmalcolm@redhat.com>
480 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
481 (selftest::test_bit_in_range): Likewise.
483 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
486 * doc/sourcebuild.texi (vect_float): Say that the selector
487 only describes the situation when -funsafe-math-optimizations is on.
488 (vect_float_strict): Document.
490 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
492 PR tree-optimization/83965
493 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
494 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
495 instead of checking only for a reduction.
496 (vect_recog_widen_sum_pattern): Likewise.
498 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
500 * predict.c (probably_never_executed): Only use precise profile info.
501 (compute_function_frequency): Skip after inlining hack since we now
502 have quality checking.
504 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
506 * profile-count.h (profile_probability::very_unlikely,
507 profile_probability::unlikely, profile_probability::even): Set
508 precision to guessed.
510 2018-01-23 Richard Biener <rguenther@suse.de>
512 PR tree-optimization/83963
513 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
514 Properly terminate dominator walk when crossing the exit edge not
515 when visiting its source block.
517 2018-01-23 Jakub Jelinek <jakub@redhat.com>
520 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
521 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
523 2018-01-22 Jakub Jelinek <jakub@redhat.com>
525 PR tree-optimization/83957
526 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
527 semicolon after for body surrounded by braces.
529 PR tree-optimization/83081
530 * profile-count.h (profile_probability::split): New method.
531 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
532 Use profile_probability::split.
533 (do_compare_rtx_and_jump): Fix adjustment of probabilities
534 when splitting a single conditional jump into 2.
536 2018-01-22 David Malcolm <dmalcolm@redhat.com>
538 PR tree-optimization/69452
539 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
542 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
544 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
545 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
546 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
548 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
550 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
551 * config/rl78/rl78.md: New define_expand "movdi"
552 * config/rl78/rl78.c: New function definition rl78_split_movdi
554 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
557 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
559 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
560 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
561 128-bit to produce an UNSPEC move to get the double word with the
562 signbit and then a shift directly to do signbit.
563 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
564 implementation with a new version that just does either a direct
565 move or a regular move. Move memory interface to separate insns.
566 Move insns so they are next to the expander.
567 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
568 with signbit move. Split big and little endian case.
569 (signbit<mode>2_dm_mem_le): Likewise.
570 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
571 (signbit<mode>2_dm2): Likewise.
573 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
575 * config/rl78/rl78.md: New define_expand "anddi3".
577 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
579 * config/rl78/rl78.md: New define_expand "umindi3".
581 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
583 * config/rl78/rl78.md: New define_expand "smindi3".
585 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
587 * config/rl78/rl78.md: New define_expand "smaxdi3".
589 2018-01-22 Carl Love <cel@us.ibm.com>
591 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
592 LVX_V1TI): Add macro expansion.
593 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
594 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
595 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
596 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
597 Change check to determine if the instruction is a byte reversing
598 entry. Fix typo in comment.
599 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
600 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
601 Add def_builtin calls for new builtins.
602 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
603 Add define_insn expansion.
605 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
607 * config/rl78/rl78.md: New define_expand "umaxdi3".
609 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
611 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
612 for non-QImode registers
614 2018-01-22 Richard Biener <rguenther@suse.de>
616 PR tree-optimization/83963
617 * graphite-scop-detection.c (scop_detection::get_sese): Delay
618 including the loop exit block.
619 (scop_detection::merge_sese): Likewise.
620 (scop_detection::add_scop): Do it here instead.
622 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
624 * doc/sourcebuild.texi (arm_softfloat): Document.
626 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
629 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
630 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
631 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
633 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
634 David Edelsohn <dje.gcc@gmail.com>
637 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
638 Change "crset eq" to "crset 2".
639 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
640 (*call_indirect_aix<mode>_nospec): Likewise.
641 (*call_value_indirect_aix<mode>_nospec): Likewise.
642 (*call_indirect_elfv2<mode>_nospec): Likewise.
643 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
644 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
645 change assembly output from . to $.
646 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
647 (indirect_jump<mode>_nospec): Change assembly output from . to $.
648 (*tablejump<mode>_internal1_nospec): Likewise.
650 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
653 * config/sh/sh_optimize_sett_clrt.cc:
654 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
656 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
658 PR tree-optimization/83940
659 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
660 offset_dt to vect_constant_def rather than vect_unknown_def_type.
661 (vect_check_load_store_mask): Add a mask_dt_out parameter and
662 use it to pass back the definition type.
663 (vect_check_store_rhs): Likewise rhs_dt_out.
664 (vect_build_gather_load_calls): Add a mask_dt argument and use
665 it instead of a call to vect_is_simple_use.
666 (vectorizable_store): Update calls to vect_check_load_store_mask
667 and vect_check_store_rhs. Use the dt returned by the latter instead
668 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
669 instead of calls to vect_is_simple_use. Pass the scalar rather
670 than the vector operand to vect_is_simple_use when handling
671 second and subsequent copies of an rhs value.
672 (vectorizable_load): Update calls to vect_check_load_store_mask
673 and vect_build_gather_load_calls. Use the cached mask_dt and
674 gs_info.offset_dt instead of calls to vect_is_simple_use.
676 2018-01-20 Jakub Jelinek <jakub@redhat.com>
679 * tree-emutls.c: Include gimplify.h.
680 (lower_emutls_2): New function.
681 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
682 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
683 it before further processing.
686 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
687 UINTVAL (trueop1) instead of INTVAL (op1).
689 2018-01-19 Jakub Jelinek <jakub@redhat.com>
693 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
694 INCOMING_FRAME_SP_OFFSET if not defined.
695 (scan_trace): Add ENTRY argument. If true and
696 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
697 emit a note to adjust the CFA offset.
698 (create_cfi_notes): Adjust scan_trace callers.
699 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
700 INCOMING_FRAME_SP_OFFSET in the CIE.
701 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
702 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
704 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
705 * doc/tm.texi: Regenerated.
707 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
709 PR rtl-optimization/83147
710 * lra-constraints.c (remove_inheritance_pseudos): Use
711 lra_substitute_pseudo_within_insn.
713 2018-01-19 Tom de Vries <tom@codesourcery.com>
714 Cesar Philippidis <cesar@codesourcery.com>
717 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
719 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
722 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
723 spaces for function labels.
725 2018-01-19 Martin Liska <mliska@suse.cz>
727 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
728 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
729 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
730 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
731 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
732 (PRED_CONST_RETURN): Change from 69 to 65.
733 (PRED_NULL_RETURN): Change from 91 to 71.
734 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
735 (PRED_LOOP_GUARD): Change from 66 to 73.
737 2018-01-19 Martin Liska <mliska@suse.cz>
739 * predict.c (predict_insn_def): Add new assert.
740 (struct branch_predictor): Change type to signed integer.
741 (test_prediction_value_range): Amend test to cover
743 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
744 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
745 (PRED_LOOP_ITERATIONS_MAX): Likewise.
746 (PRED_LOOP_IV_COMPARE): Likewise.
747 * predict.h (PROB_UNINITIALIZED): Define new constant.
749 2018-01-19 Martin Liska <mliska@suse.cz>
751 * predict.c (dump_prediction): Add new format for
752 analyze_brprob.py script which is enabled with -details
754 * profile-count.h (precise_p): New function.
756 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
758 PR tree-optimization/83922
759 * tree-vect-loop.c (vect_verify_full_masking): Return false if
760 there are no statements that need masking.
761 (vect_active_double_reduction_p): New function.
762 (vect_analyze_loop_operations): Use it when handling phis that
763 are not in the loop header.
765 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
767 PR tree-optimization/83914
768 * tree-vect-loop.c (vectorizable_induction): Don't convert
769 init_expr or apply the peeling adjustment for inductions
770 that are nested within the vectorized loop.
772 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
774 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
777 2018-01-18 Jakub Jelinek <jakub@redhat.com>
781 * function.h (gimplify_parameters): Add gimple_seq * argument.
782 * function.c: Include gimple.h and options.h.
783 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
784 for the added local temporaries if needed.
785 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
786 if there are any parameter cleanups, wrap whole body into a
787 try/finally with the cleanups.
789 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
792 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
793 Use GET_MODE_CLASS for scalar floating point.
795 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
799 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
800 Fix call of call_cgraph_insertion_hooks.
802 2018-01-18 Martin Sebor <msebor@redhat.com>
804 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
806 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
809 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
812 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
815 * common.opt: (-ffile-prefix-map): New option.
816 * opts.c (common_handle_option): Defer it.
817 * opts-global.c (handle_common_deferred_options): Handle it.
818 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
819 * file-prefix-map.h: New file.
820 (remap_debug_filename, add_debug_prefix_map): ...here.
821 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
822 * final.c (debug_prefix_map, add_debug_prefix_map
823 remap_debug_filename): Move to...
824 * file-prefix-map.c: New file.
825 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
826 generalize, get rid of alloca(), use strrchr() instead of strchr().
827 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
828 Implement in terms of add_prefix_map().
829 (remap_macro_filename, remap_debug_filename): Implement in term of
831 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
832 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
833 * dbxout.c: Include file-prefix-map.h.
834 * varasm.c: Likewise.
835 * vmsdbgout.c: Likewise.
836 * xcoffout.c: Likewise.
837 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
838 * doc/cppopts.texi (-fmacro-prefix-map): Document.
839 * doc/invoke.texi (-ffile-prefix-map): Document.
840 (-fdebug-prefix-map): Update description.
842 2018-01-18 Martin Liska <mliska@suse.cz>
844 * config/i386/i386.c (indirect_thunk_name): Document that also
846 (output_indirect_thunk): Document why both instructions
847 (pause and lfence) are generated.
849 2018-01-18 Richard Biener <rguenther@suse.de>
851 PR tree-optimization/83887
852 * graphite-scop-detection.c
853 (scop_detection::get_nearest_dom_with_single_entry): Remove.
854 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
855 (scop_detection::merge_sese): Re-implement with a flood-fill
856 algorithm that properly finds a SESE region if it exists.
858 2018-01-18 Jakub Jelinek <jakub@redhat.com>
861 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
862 pointer_diff optimizations use view_convert instead of convert.
864 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
866 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
867 Generate different code for -mno-speculate-indirect-jumps.
868 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
869 (*call_indirect_aix<mode>): Disable for
870 -mno-speculate-indirect-jumps.
871 (*call_indirect_aix<mode>_nospec): New define_insn.
872 (*call_value_indirect_aix<mode>): Disable for
873 -mno-speculate-indirect-jumps.
874 (*call_value_indirect_aix<mode>_nospec): New define_insn.
875 (*sibcall_nonlocal_sysv<mode>): Generate different code for
876 -mno-speculate-indirect-jumps.
877 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
879 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
881 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
882 long double type, set the flags for noting the default long double
883 type, even if we don't pass or return a long double type.
885 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
888 * ipa-inline.c (flatten_function): Do not overwrite final inlining
891 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
893 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
894 support for merge[hl].
895 (fold_mergehl_helper): New helper function.
896 (tree-vector-builder.h): New #include for tree_vector_builder usage.
897 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
898 (altivec_vmrglw_direct): Add xxmrglw insn.
900 2018-01-17 Andrew Waterman <andrew@sifive.com>
902 * config/riscv/riscv.c (riscv_conditional_register_usage): If
903 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
905 2018-01-17 David Malcolm <dmalcolm@redhat.com>
908 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
909 call the lto_location_cache before reading the
910 DECL_SOURCE_LOCATION of the types.
912 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
913 Richard Sandiford <richard.sandiford@linaro.org>
915 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
916 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
917 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
918 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
919 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
921 * config/aarch64/constraints.md (aarch64_movti_operand):
923 * config/aarch64/predicates.md (Uti): Add new constraint.
925 2018-01-17 Carl Love <cel@us.ibm.com>
926 * config/rs6000/vsx.md (define_expand xl_len_r,
927 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
928 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
930 (define_expand, define_insn): Move the shift left from the
931 define_insn to the define_expand for lxvl and stxvl instructions.
932 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
933 and XL_LEN_R definitions to PURE.
935 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
937 * config/i386/i386.c (indirect_thunk_name): Declare regno
938 as unsigned int. Compare regno with INVALID_REGNUM.
939 (output_indirect_thunk): Ditto.
940 (output_indirect_thunk_function): Ditto.
941 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
942 in the call to output_indirect_thunk_function.
944 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
947 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
948 rather than the size of inner_type to determine the stack slot size
949 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
951 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
954 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
957 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
959 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
960 endian Linux systems to optionally enable multilibs for selecting
961 the long double type if the user configured an explicit type.
962 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
963 have no long double multilibs if not defined.
964 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
965 warn if the user used -mabi={ieee,ibm}longdouble and we built
966 multilibs for long double.
967 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
968 appropriate multilib option.
969 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
971 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
972 for building long double multilibs.
973 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
975 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
977 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
980 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
982 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
985 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
988 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
991 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
993 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
994 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
996 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
998 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
999 different rtl trees depending on TARGET_64BIT.
1000 (rs6000_gen_lvx): Likewise.
1002 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1004 * config/visium/visium.md (nop): Tweak comment.
1005 (hazard_nop): Likewise.
1007 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1009 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1010 -mspeculate-indirect-jumps.
1011 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1012 for -mno-speculate-indirect-jumps.
1013 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1014 (*call_value_indirect_elfv2<mode>): Disable for
1015 -mno-speculate-indirect-jumps.
1016 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1017 (indirect_jump): Emit different RTL for
1018 -mno-speculate-indirect-jumps.
1019 (*indirect_jump<mode>): Disable for
1020 -mno-speculate-indirect-jumps.
1021 (*indirect_jump<mode>_nospec): New define_insn.
1022 (tablejump): Emit different RTL for
1023 -mno-speculate-indirect-jumps.
1024 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1025 (tablejumpsi_nospec): New define_expand.
1026 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1027 (tablejumpdi_nospec): New define_expand.
1028 (*tablejump<mode>_internal1): Disable for
1029 -mno-speculate-indirect-jumps.
1030 (*tablejump<mode>_internal1_nospec): New define_insn.
1031 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1034 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1036 * caller-save.c (insert_save): Drop unnecessary parameter. All
1039 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1040 Richard Biener <rguenth@suse.de>
1043 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1044 return early, inline manually is_gimple_sizepos. Make sure if we
1045 call gimplify_expr we don't end up with a gimple constant.
1046 * tree.c (variably_modified_type_p): Don't return true for
1047 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1048 * gimplify.h (is_gimple_sizepos): Remove.
1050 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1052 PR tree-optimization/83857
1053 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1054 vectorizable_live_operation for pure SLP statements.
1055 (vectorizable_live_operation): Handle PHIs.
1057 2018-01-16 Richard Biener <rguenther@suse.de>
1059 PR tree-optimization/83867
1060 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1061 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1063 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1066 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1067 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1068 If off is not INTEGER_CST, issue a may not be aligned warning
1069 rather than isn't aligned. Use isn%'t rather than isn't.
1070 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1072 <case MULT_EXPR>: Improve the case when bottom and one of the
1073 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1074 operand, in that case check if the other operand is multiple of
1075 bottom divided by the INTEGER_CST operand.
1077 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1080 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1081 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1082 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1083 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1084 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1085 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1086 (pa_function_arg_size): New function.
1088 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1090 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1091 in a separate statement.
1093 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1095 PR tree-optimization/83847
1096 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1097 group gathers and scatters.
1099 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1101 PR rtl-optimization/86620
1102 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1104 PR rtl-optimization/83213
1105 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1106 to last if both are JUMP_INSNs.
1108 PR tree-optimization/83843
1109 * gimple-ssa-store-merging.c
1110 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1111 store_immediate_info for bswap/nop orig_stores.
1113 2018-01-15 Andrew Waterman <andrew@sifive.com>
1115 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1117 <UDIV>: Increase cost if !TARGET_DIV.
1119 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1121 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1122 (define_attr "cr_logical_3op"): New.
1123 (cceq_ior_compare): Adjust.
1124 (cceq_ior_compare_complement): Adjust.
1125 (*cceq_rev_compare): Adjust.
1126 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1127 (is_cracked_insn): Adjust.
1128 (insn_must_be_first_in_group): Adjust.
1129 * config/rs6000/40x.md: Adjust.
1130 * config/rs6000/440.md: Adjust.
1131 * config/rs6000/476.md: Adjust.
1132 * config/rs6000/601.md: Adjust.
1133 * config/rs6000/603.md: Adjust.
1134 * config/rs6000/6xx.md: Adjust.
1135 * config/rs6000/7450.md: Adjust.
1136 * config/rs6000/7xx.md: Adjust.
1137 * config/rs6000/8540.md: Adjust.
1138 * config/rs6000/cell.md: Adjust.
1139 * config/rs6000/e300c2c3.md: Adjust.
1140 * config/rs6000/e500mc.md: Adjust.
1141 * config/rs6000/e500mc64.md: Adjust.
1142 * config/rs6000/e5500.md: Adjust.
1143 * config/rs6000/e6500.md: Adjust.
1144 * config/rs6000/mpc.md: Adjust.
1145 * config/rs6000/power4.md: Adjust.
1146 * config/rs6000/power5.md: Adjust.
1147 * config/rs6000/power6.md: Adjust.
1148 * config/rs6000/power7.md: Adjust.
1149 * config/rs6000/power8.md: Adjust.
1150 * config/rs6000/power9.md: Adjust.
1151 * config/rs6000/rs64.md: Adjust.
1152 * config/rs6000/titan.md: Adjust.
1154 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1156 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1157 ix86_indirect_branch_register logic.
1159 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1161 * config/i386/constraints.md (Bs): Update
1162 ix86_indirect_branch_register check. Don't check
1163 ix86_indirect_branch_register with GOT_memory_operand.
1165 * config/i386/predicates.md (GOT_memory_operand): Don't check
1166 ix86_indirect_branch_register here.
1167 (GOT32_symbol_operand): Likewise.
1169 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1171 * config/i386/predicates.md (constant_call_address_operand):
1172 Rewrite ix86_indirect_branch_register logic.
1173 (sibcall_insn_operand): Likewise.
1175 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1177 * config/i386/constraints.md (Bs): Replace
1178 ix86_indirect_branch_thunk_register with
1179 ix86_indirect_branch_register.
1181 * config/i386/i386.md (indirect_jump): Likewise.
1182 (tablejump): Likewise.
1183 (*sibcall_memory): Likewise.
1184 (*sibcall_value_memory): Likewise.
1185 Peepholes of indirect call and jump via memory: Likewise.
1186 * config/i386/i386.opt: Likewise.
1187 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1188 (GOT_memory_operand): Likewise.
1189 (call_insn_operand): Likewise.
1190 (sibcall_insn_operand): Likewise.
1191 (GOT32_symbol_operand): Likewise.
1193 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1196 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1197 type rather than type addr's type points to.
1198 (expand_omp_atomic_mutex): Likewise.
1199 (expand_omp_atomic): Likewise.
1201 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1204 * config/i386/i386.c (output_indirect_thunk_function): Use
1205 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1206 for __x86_return_thunk.
1208 2018-01-15 Richard Biener <rguenther@suse.de>
1211 * expmed.c (extract_bit_field_1): Fix typo.
1213 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1216 * config/arm/iterators.md (VF): New mode iterator.
1217 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1218 Remove integer-related logic from pattern.
1219 (neon_vabd<mode>_3): Likewise.
1221 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1224 * common.opt (fstrict-overflow): No longer an alias.
1225 (fwrapv-pointer): New option.
1226 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1227 also for pointer types based on flag_wrapv_pointer.
1228 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1229 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1230 opts->x_flag_wrapv got set.
1231 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1232 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1233 POINTER_TYPE_OVERFLOW_UNDEFINED.
1234 * match.pd: Likewise in address comparison pattern.
1235 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1237 2018-01-15 Richard Biener <rguenther@suse.de>
1240 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1241 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1242 Reset type names to their identifier if their TYPE_DECL doesn't
1243 have linkage (and thus is used for ODR and devirt).
1244 (save_debug_info_for_decl): Remove.
1245 (save_debug_info_for_type): Likewise.
1246 (add_tree_to_fld_list): Adjust.
1247 * tree-pretty-print.c (dump_generic_node): Make dumping of
1248 type names more robust.
1250 2018-01-15 Richard Biener <rguenther@suse.de>
1252 * BASE-VER: Bump to 8.0.1.
1254 2018-01-14 Martin Sebor <msebor@redhat.com>
1257 * builtins.c (check_access): Avoid warning when the no-warning bit
1260 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1262 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1263 * ira-color (allocno_hard_regs_compare): Likewise.
1265 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1268 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1269 Use .pushsection/.popsection.
1271 2018-01-14 Martin Sebor <msebor@redhat.com>
1274 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1276 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1278 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1279 entry from extra_headers.
1280 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1281 extra_headers, make the list bitwise identical to the i?86-*-* one.
1283 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1285 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1286 -mcmodel=large with -mindirect-branch=thunk,
1287 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1288 -mfunction-return=thunk-extern.
1289 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1290 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1291 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1293 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1295 * config/i386/i386.c (print_reg): Print the name of the full
1296 integer register without '%'.
1297 (ix86_print_operand): Handle 'V'.
1298 * doc/extend.texi: Document 'V' modifier.
1300 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1302 * config/i386/constraints.md (Bs): Disallow memory operand for
1303 -mindirect-branch-register.
1305 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1306 (GOT_memory_operand): Likewise.
1307 (call_insn_operand): Likewise.
1308 (sibcall_insn_operand): Likewise.
1309 (GOT32_symbol_operand): Likewise.
1310 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1311 for -mindirect-branch-register.
1312 (tablejump): Likewise.
1313 (*sibcall_memory): Likewise.
1314 (*sibcall_value_memory): Likewise.
1315 Disallow peepholes of indirect call and jump via memory for
1316 -mindirect-branch-register.
1317 (*call_pop): Replace m with Bw.
1318 (*call_value_pop): Likewise.
1319 (*sibcall_pop_memory): Replace m with Bs.
1320 * config/i386/i386.opt (mindirect-branch-register): New option.
1321 * doc/invoke.texi: Document -mindirect-branch-register option.
1323 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1325 * config/i386/i386-protos.h (ix86_output_function_return): New.
1326 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1327 set function_return_type.
1328 (indirect_thunk_name): Add ret_p to indicate thunk for function
1330 (output_indirect_thunk_function): Pass false to
1331 indirect_thunk_name.
1332 (ix86_output_indirect_branch_via_reg): Likewise.
1333 (ix86_output_indirect_branch_via_push): Likewise.
1334 (output_indirect_thunk_function): Create alias for function
1335 return thunk if regno < 0.
1336 (ix86_output_function_return): New function.
1337 (ix86_handle_fndecl_attribute): Handle function_return.
1338 (ix86_attribute_table): Add function_return.
1339 * config/i386/i386.h (machine_function): Add
1340 function_return_type.
1341 * config/i386/i386.md (simple_return_internal): Use
1342 ix86_output_function_return.
1343 (simple_return_internal_long): Likewise.
1344 * config/i386/i386.opt (mfunction-return=): New option.
1345 (indirect_branch): Mention -mfunction-return=.
1346 * doc/extend.texi: Document function_return function attribute.
1347 * doc/invoke.texi: Document -mfunction-return= option.
1349 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1351 * config/i386/i386-opts.h (indirect_branch): New.
1352 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1353 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1354 with local indirect jump when converting indirect call and jump.
1355 (ix86_set_indirect_branch_type): New.
1356 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1357 (indirectlabelno): New.
1358 (indirect_thunk_needed): Likewise.
1359 (indirect_thunk_bnd_needed): Likewise.
1360 (indirect_thunks_used): Likewise.
1361 (indirect_thunks_bnd_used): Likewise.
1362 (INDIRECT_LABEL): Likewise.
1363 (indirect_thunk_name): Likewise.
1364 (output_indirect_thunk): Likewise.
1365 (output_indirect_thunk_function): Likewise.
1366 (ix86_output_indirect_branch_via_reg): Likewise.
1367 (ix86_output_indirect_branch_via_push): Likewise.
1368 (ix86_output_indirect_branch): Likewise.
1369 (ix86_output_indirect_jmp): Likewise.
1370 (ix86_code_end): Call output_indirect_thunk_function if needed.
1371 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1373 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1374 (ix86_attribute_table): Add indirect_branch.
1375 * config/i386/i386.h (machine_function): Add indirect_branch_type
1376 and has_local_indirect_jump.
1377 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1379 (tablejump): Likewise.
1380 (*indirect_jump): Use ix86_output_indirect_jmp.
1381 (*tablejump_1): Likewise.
1382 (simple_return_indirect_internal): Likewise.
1383 * config/i386/i386.opt (mindirect-branch=): New option.
1384 (indirect_branch): New.
1387 (thunk-inline): Likewise.
1388 (thunk-extern): Likewise.
1389 * doc/extend.texi: Document indirect_branch function attribute.
1390 * doc/invoke.texi: Document -mindirect-branch= option.
1392 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1395 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1397 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1399 * ipa-inline.c (want_inline_small_function_p): Return false if
1400 inlining has already failed with CIF_FINAL_ERROR.
1401 (update_caller_keys): Call want_inline_small_function_p before
1403 (update_callee_keys): Likewise.
1405 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1407 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1409 (rs6000_quadword_masked_address_p): Likewise.
1410 (quad_aligned_load_p): Likewise.
1411 (quad_aligned_store_p): Likewise.
1412 (const_load_sequence_p): Add comment to describe the outer-most loop.
1413 (mimic_memory_attributes_and_flags): New function.
1414 (rs6000_gen_stvx): Likewise.
1415 (replace_swapped_aligned_store): Likewise.
1416 (rs6000_gen_lvx): Likewise.
1417 (replace_swapped_aligned_load): Likewise.
1418 (replace_swapped_load_constant): Capitalize argument name in
1419 comment describing this function.
1420 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1421 and stores that access quad-word aligned addresses and replace
1422 with stvx or lvx instructions when appropriate.
1423 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1424 New function prototype.
1425 (rs6000_quadword_masked_address_p): Likewise.
1426 (rs6000_gen_lvx): Likewise.
1427 (rs6000_gen_stvx): Likewise.
1428 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1429 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1430 when memory address is aligned.
1431 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1432 this split to select lvx instruction when memory address is aligned.
1433 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1434 instruction when memory address is aligned.
1435 (*vsx_le_perm_load_v16qi): Likewise.
1436 (four unnamed splitters): Modify to select the stvx instruction
1437 when memory is aligned.
1439 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1441 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1442 which appears in the queue multiple times.
1444 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1445 Alan Hayward <alan.hayward@arm.com>
1446 David Sherwood <david.sherwood@arm.com>
1448 * tree-vectorizer.h (vec_lower_bound): New structure.
1449 (_loop_vec_info): Add check_nonzero and lower_bounds.
1450 (LOOP_VINFO_CHECK_NONZERO): New macro.
1451 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1452 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1453 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1454 fields. Make seg_len the distance travelled, not including the
1456 (dr_direction_indicator): Declare.
1457 (dr_zero_step_indicator): Likewise.
1458 (dr_known_forward_stride_p): Likewise.
1459 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1461 (runtime_alias_check_p): Allow runtime alias checks with
1463 (operator ==): Compare access_size and align.
1464 (prune_runtime_alias_test_list): Rework for new distinction between
1465 the access_size and seg_len.
1466 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1468 (get_segment_min_max): New function.
1469 (create_intersect_range_checks): Use it.
1470 (dr_step_indicator): New function.
1471 (dr_direction_indicator): Likewise.
1472 (dr_zero_step_indicator): Likewise.
1473 (dr_known_forward_stride_p): Likewise.
1474 * tree-loop-distribution.c (data_ref_segment_size): Return
1475 DR_STEP * (niters - 1).
1476 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1478 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1479 (vect_preserves_scalar_order_p): New function, split out from...
1480 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1481 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1482 (vect_vfa_access_size): New function.
1483 (vect_vfa_align): Likewise.
1484 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1485 (dump_lower_bound): New function.
1486 (vect_check_lower_bound): Likewise.
1487 (vect_small_gap_p): Likewise.
1488 (vectorizable_with_step_bound_p): Likewise.
1489 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1490 depencies if the vectorization factor is 1. Convert the checks
1491 for nonzero steps into checks on the bounds of DR_STEP. Try using
1492 a bunds check for variable steps if the minimum required step is
1493 relatively small. Update calls to the dr_with_seg_len
1494 constructor and to vect_compile_time_alias.
1495 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1497 (vect_loop_versioning): Call it.
1498 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1500 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1502 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1503 Alan Hayward <alan.hayward@arm.com>
1504 David Sherwood <david.sherwood@arm.com>
1506 * doc/sourcebuild.texi (vect_scatter_store): Document.
1507 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1509 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1511 * genopinit.c (main): Add supports_vec_scatter_store and
1512 supports_vec_scatter_store_cached to target_optabs.
1513 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1514 IFN_MASK_SCATTER_STORE.
1515 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1517 * internal-fn.h (internal_store_fn_p): Declare.
1518 (internal_fn_stored_value_index): Likewise.
1519 * internal-fn.c (scatter_store_direct): New macro.
1520 (expand_scatter_store_optab_fn): New function.
1521 (direct_scatter_store_optab_supported_p): New macro.
1522 (internal_store_fn_p): New function.
1523 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1524 IFN_MASK_SCATTER_STORE.
1525 (internal_fn_mask_index): Likewise.
1526 (internal_fn_stored_value_index): New function.
1527 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1529 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1530 * optabs-query.c (supports_vec_scatter_store_p): New function.
1531 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1532 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1533 true for scatter stores.
1534 (vect_gather_scatter_fn_p): Handle scatter stores too.
1535 (vect_check_gather_scatter): Consider using scatter stores if
1536 supports_vec_scatter_store_p.
1537 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1539 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1540 internal_fn_stored_value_index.
1541 (check_load_store_masking): Handle scatter stores too.
1542 (vect_get_store_rhs): Make public.
1543 (vectorizable_call): Use internal_store_fn_p.
1544 (vectorizable_store): Handle scatter store internal functions.
1545 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1546 when deciding whether the end of the group has been reached.
1547 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1548 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1549 (mask_scatter_store<mode>): New insns.
1551 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1552 Alan Hayward <alan.hayward@arm.com>
1553 David Sherwood <david.sherwood@arm.com>
1555 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1556 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1557 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1559 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1560 Use vect_truncate_gather_scatter_offset if we can't treat the
1561 operation as a normal gather load or scatter store.
1562 (get_group_load_store_type): Take the gather_scatter_info
1563 as argument. Try using a gather load or scatter store for
1564 single-element groups.
1565 (get_load_store_type): Update calls to get_group_load_store_type
1566 and vect_use_strided_gather_scatters_p.
1568 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1569 Alan Hayward <alan.hayward@arm.com>
1570 David Sherwood <david.sherwood@arm.com>
1572 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1573 optional tree argument.
1574 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1576 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1577 but continue to use the current value as a fallback.
1578 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1579 to compare the updates.
1580 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1581 (get_load_store_type): Use it when handling a strided access.
1582 (vect_get_strided_load_store_ops): New function.
1583 (vect_get_data_ptr_increment): Likewise.
1584 (vectorizable_load): Handle strided gather loads. Always pass
1585 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1587 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1588 Alan Hayward <alan.hayward@arm.com>
1589 David Sherwood <david.sherwood@arm.com>
1591 * doc/md.texi (gather_load@var{m}): Document.
1592 (mask_gather_load@var{m}): Likewise.
1593 * genopinit.c (main): Add supports_vec_gather_load and
1594 supports_vec_gather_load_cached to target_optabs.
1595 * optabs-tree.c (init_tree_optimization_optabs): Use
1596 ggc_cleared_alloc to allocate target_optabs.
1597 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1598 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1600 * internal-fn.h (internal_load_fn_p): Declare.
1601 (internal_gather_scatter_fn_p): Likewise.
1602 (internal_fn_mask_index): Likewise.
1603 (internal_gather_scatter_fn_supported_p): Likewise.
1604 * internal-fn.c (gather_load_direct): New macro.
1605 (expand_gather_load_optab_fn): New function.
1606 (direct_gather_load_optab_supported_p): New macro.
1607 (direct_internal_fn_optab): New function.
1608 (internal_load_fn_p): Likewise.
1609 (internal_gather_scatter_fn_p): Likewise.
1610 (internal_fn_mask_index): Likewise.
1611 (internal_gather_scatter_fn_supported_p): Likewise.
1612 * optabs-query.c (supports_at_least_one_mode_p): New function.
1613 (supports_vec_gather_load_p): Likewise.
1614 * optabs-query.h (supports_vec_gather_load_p): Declare.
1615 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1616 and memory_type field.
1617 (NUM_PATTERNS): Bump to 15.
1618 * tree-vect-data-refs.c: Include internal-fn.h.
1619 (vect_gather_scatter_fn_p): New function.
1620 (vect_describe_gather_scatter_call): Likewise.
1621 (vect_check_gather_scatter): Try using internal functions for
1622 gather loads. Recognize existing calls to a gather load function.
1623 (vect_analyze_data_refs): Consider using gather loads if
1624 supports_vec_gather_load_p.
1625 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1626 (vect_get_gather_scatter_offset_type): Likewise.
1627 (vect_convert_mask_for_vectype): Likewise.
1628 (vect_add_conversion_to_patterm): Likewise.
1629 (vect_try_gather_scatter_pattern): Likewise.
1630 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1631 (vect_vect_recog_func_ptrs): Add it.
1632 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1633 internal_fn_mask_index and internal_gather_scatter_fn_p.
1634 (check_load_store_masking): Take the gather_scatter_info as an
1635 argument and handle gather loads.
1636 (vect_get_gather_scatter_ops): New function.
1637 (vectorizable_call): Check internal_load_fn_p.
1638 (vectorizable_load): Likewise. Handle gather load internal
1640 (vectorizable_store): Update call to check_load_store_masking.
1641 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1642 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1643 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1644 (aarch64_gather_scale_operand_d): New predicates.
1645 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1646 (mask_gather_load<mode>): New insns.
1648 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1649 Alan Hayward <alan.hayward@arm.com>
1650 David Sherwood <david.sherwood@arm.com>
1652 * optabs.def (fold_left_plus_optab): New optab.
1653 * doc/md.texi (fold_left_plus_@var{m}): Document.
1654 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1655 * internal-fn.c (fold_left_direct): Define.
1656 (expand_fold_left_optab_fn): Likewise.
1657 (direct_fold_left_optab_supported_p): Likewise.
1658 * fold-const-call.c (fold_const_fold_left): New function.
1659 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1660 * tree-parloops.c (valid_reduction_p): New function.
1661 (gather_scalar_reductions): Use it.
1662 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1663 (vect_finish_replace_stmt): Declare.
1664 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1665 (needs_fold_left_reduction_p): New function, split out from...
1666 (vect_is_simple_reduction): ...here. Accept reductions that
1667 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1668 (vect_force_simple_reduction): Also store the reduction type in
1669 the assignment's STMT_VINFO_REDUC_TYPE.
1670 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1671 (merge_with_identity): New function.
1672 (vect_expand_fold_left): Likewise.
1673 (vectorize_fold_left_reduction): Likewise.
1674 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1675 scalar phi in place for it. Check for target support and reject
1676 cases that would reassociate the operation. Defer the transform
1677 phase to vectorize_fold_left_reduction.
1678 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1679 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1680 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1682 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1684 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1685 call to ifc_temp_var.
1687 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1688 Alan Hayward <alan.hayward@arm.com>
1689 David Sherwood <david.sherwood@arm.com>
1691 * target.def (legitimize_address_displacement): Take the original
1692 offset as a poly_int.
1693 * targhooks.h (default_legitimize_address_displacement): Update
1695 * targhooks.c (default_legitimize_address_displacement): Likewise.
1696 * doc/tm.texi: Regenerate.
1697 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1698 as an argument, moving assert of ad->disp == ad->disp_term to...
1699 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1700 Try calling targetm.legitimize_address_displacement before expanding
1701 the address rather than afterwards, and adjust for the new interface.
1702 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1703 Match the new hook interface. Handle SVE addresses.
1704 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1707 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1709 * Makefile.in (OBJS): Add early-remat.o.
1710 * target.def (select_early_remat_modes): New hook.
1711 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1712 * doc/tm.texi: Regenerate.
1713 * targhooks.h (default_select_early_remat_modes): Declare.
1714 * targhooks.c (default_select_early_remat_modes): New function.
1715 * timevar.def (TV_EARLY_REMAT): New timevar.
1716 * passes.def (pass_early_remat): New pass.
1717 * tree-pass.h (make_pass_early_remat): Declare.
1718 * early-remat.c: New file.
1719 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1721 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1723 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1724 Alan Hayward <alan.hayward@arm.com>
1725 David Sherwood <david.sherwood@arm.com>
1727 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1728 vfm1 with a bound_epilog parameter.
1729 (vect_do_peeling): Update calls accordingly, and move the prologue
1730 call earlier in the function. Treat the base bound_epilog as 0 for
1731 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1732 this base when peeling for gaps.
1733 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1734 with fully-masked loops.
1735 (vect_estimate_min_profitable_iters): Handle the single peeled
1736 iteration in that case.
1738 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1739 Alan Hayward <alan.hayward@arm.com>
1740 David Sherwood <david.sherwood@arm.com>
1742 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1743 single-element interleaving even if the size is not a power of 2.
1744 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1745 accesses for single-element interleaving if the group size is
1748 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1749 Alan Hayward <alan.hayward@arm.com>
1750 David Sherwood <david.sherwood@arm.com>
1752 * doc/md.texi (fold_extract_last_@var{m}): Document.
1753 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1754 * optabs.def (fold_extract_last_optab): New optab.
1755 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1756 * internal-fn.c (fold_extract_direct): New macro.
1757 (expand_fold_extract_optab_fn): Likewise.
1758 (direct_fold_extract_optab_supported_p): Likewise.
1759 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1760 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1761 EXTRACT_LAST_REDUCTION.
1762 (get_initial_def_for_reduction): Do not create an initial vector
1763 for EXTRACT_LAST_REDUCTION reductions.
1764 (vectorizable_reduction): Leave the scalar phi in place for
1765 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1766 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1767 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1768 transform phase to vectorizable_condition.
1769 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1771 (vect_finish_stmt_generation): ...here.
1772 (vect_finish_replace_stmt): New function.
1773 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1774 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1776 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1778 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1779 Alan Hayward <alan.hayward@arm.com>
1780 David Sherwood <david.sherwood@arm.com>
1782 * doc/md.texi (extract_last_@var{m}): Document.
1783 * optabs.def (extract_last_optab): New optab.
1784 * internal-fn.def (EXTRACT_LAST): New internal function.
1785 * internal-fn.c (cond_unary_direct): New macro.
1786 (expand_cond_unary_optab_fn): Likewise.
1787 (direct_cond_unary_optab_supported_p): Likewise.
1788 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1789 loops using EXTRACT_LAST.
1790 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1791 (extract_last_<mode>): ...this optab.
1792 (vec_extract<mode><Vel>): Update accordingly.
1794 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1795 Alan Hayward <alan.hayward@arm.com>
1796 David Sherwood <david.sherwood@arm.com>
1798 * target.def (empty_mask_is_expensive): New hook.
1799 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
1800 * doc/tm.texi: Regenerate.
1801 * targhooks.h (default_empty_mask_is_expensive): Declare.
1802 * targhooks.c (default_empty_mask_is_expensive): New function.
1803 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
1804 if the target says that empty masks are expensive.
1805 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
1807 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
1809 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1810 Alan Hayward <alan.hayward@arm.com>
1811 David Sherwood <david.sherwood@arm.com>
1813 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
1814 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
1815 (vect_use_loop_mask_for_alignment_p): New function.
1816 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
1817 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
1818 niters_skip argument. Make sure that the first niters_skip elements
1819 of the first iteration are inactive.
1820 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
1821 Update call to vect_set_loop_masks_directly.
1822 (get_misalign_in_elems): New function, split out from...
1823 (vect_gen_prolog_loop_niters): ...here.
1824 (vect_update_init_of_dr): Take a code argument that specifies whether
1825 the adjustment should be added or subtracted.
1826 (vect_update_init_of_drs): Likewise.
1827 (vect_prepare_for_masked_peels): New function.
1828 (vect_do_peeling): Skip prologue peeling if we're using a mask
1829 instead. Update call to vect_update_inits_of_drs.
1830 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1832 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
1833 alignment. Do not include the number of peeled iterations in
1834 the minimum threshold in that case.
1835 (vectorizable_induction): Adjust the start value down by
1836 LOOP_VINFO_MASK_SKIP_NITERS iterations.
1837 (vect_transform_loop): Call vect_prepare_for_masked_peels.
1838 Take the number of skipped iterations into account when calculating
1840 * tree-vect-stmts.c (vect_gen_while_not): New function.
1842 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1843 Alan Hayward <alan.hayward@arm.com>
1844 David Sherwood <david.sherwood@arm.com>
1846 * doc/sourcebuild.texi (vect_fully_masked): Document.
1847 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
1849 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
1851 (vect_analyze_loop_2): ...here. Don't check the vectorization
1852 factor against the number of loop iterations if the loop is
1855 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1856 Alan Hayward <alan.hayward@arm.com>
1857 David Sherwood <david.sherwood@arm.com>
1859 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
1860 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
1861 (dump_groups): Update accordingly.
1862 (iv_use::mem_type): New member variable.
1863 (address_p): New function.
1864 (record_use): Add a mem_type argument and initialize the new
1866 (record_group_use): Add a mem_type argument. Use address_p.
1867 Remove obsolete null checks of base_object. Update call to record_use.
1868 (find_interesting_uses_op): Update call to record_group_use.
1869 (find_interesting_uses_cond): Likewise.
1870 (find_interesting_uses_address): Likewise.
1871 (get_mem_type_for_internal_fn): New function.
1872 (find_address_like_use): Likewise.
1873 (find_interesting_uses_stmt): Try find_address_like_use before
1874 calling find_interesting_uses_op.
1875 (addr_offset_valid_p): Use the iv mem_type field as the type
1876 of the addressed memory.
1877 (add_autoinc_candidates): Likewise.
1878 (get_address_cost): Likewise.
1879 (split_small_address_groups_p): Use address_p.
1880 (split_address_groups): Likewise.
1881 (add_iv_candidate_for_use): Likewise.
1882 (autoinc_possible_for_pair): Likewise.
1883 (rewrite_groups): Likewise.
1884 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
1885 (determine_group_iv_cost): Update after split of USE_ADDRESS.
1886 (get_alias_ptr_type_for_ptr_address): New function.
1887 (rewrite_use_address): Rewrite address uses in calls that were
1888 identified by find_address_like_use.
1890 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1891 Alan Hayward <alan.hayward@arm.com>
1892 David Sherwood <david.sherwood@arm.com>
1894 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
1896 * gimple-expr.h (is_gimple_addressable: Likewise.
1897 * gimple-expr.c (is_gimple_address): Likewise.
1898 * internal-fn.c (expand_call_mem_ref): New function.
1899 (expand_mask_load_optab_fn): Use it.
1900 (expand_mask_store_optab_fn): Likewise.
1902 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1903 Alan Hayward <alan.hayward@arm.com>
1904 David Sherwood <david.sherwood@arm.com>
1906 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
1907 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
1908 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
1909 (cond_umax@var{mode}): Document.
1910 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
1911 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
1912 (cond_umin_optab, cond_umax_optab): New optabs.
1913 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
1914 (COND_IOR, COND_XOR): New internal functions.
1915 * internal-fn.h (get_conditional_internal_fn): Declare.
1916 * internal-fn.c (cond_binary_direct): New macro.
1917 (expand_cond_binary_optab_fn): Likewise.
1918 (direct_cond_binary_optab_supported_p): Likewise.
1919 (get_conditional_internal_fn): New function.
1920 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
1921 Cope with reduction statements that are vectorized as calls rather
1923 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
1924 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
1925 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
1926 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
1927 (UNSPEC_COND_EOR): New unspecs.
1928 (optab): Add mappings for them.
1929 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
1930 (sve_int_op, sve_fp_op): New int attributes.
1932 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1933 Alan Hayward <alan.hayward@arm.com>
1934 David Sherwood <david.sherwood@arm.com>
1936 * optabs.def (while_ult_optab): New optab.
1937 * doc/md.texi (while_ult@var{m}@var{n}): Document.
1938 * internal-fn.def (WHILE_ULT): New internal function.
1939 * internal-fn.h (direct_internal_fn_supported_p): New override
1940 that takes two types as argument.
1941 * internal-fn.c (while_direct): New macro.
1942 (expand_while_optab_fn): New function.
1943 (convert_optab_supported_p): Likewise.
1944 (direct_while_optab_supported_p): New macro.
1945 * wide-int.h (wi::udiv_ceil): New function.
1946 * tree-vectorizer.h (rgroup_masks): New structure.
1947 (vec_loop_masks): New typedef.
1948 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
1950 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
1951 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
1952 (vect_max_vf): New function.
1953 (slpeel_make_loop_iterate_ntimes): Delete.
1954 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
1955 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
1956 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
1957 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
1958 internal-fn.h, stor-layout.h and optabs-query.h.
1959 (vect_set_loop_mask): New function.
1960 (add_preheader_seq): Likewise.
1961 (add_header_seq): Likewise.
1962 (interleave_supported_p): Likewise.
1963 (vect_maybe_permute_loop_masks): Likewise.
1964 (vect_set_loop_masks_directly): Likewise.
1965 (vect_set_loop_condition_masked): Likewise.
1966 (vect_set_loop_condition_unmasked): New function, split out from
1967 slpeel_make_loop_iterate_ntimes.
1968 (slpeel_make_loop_iterate_ntimes): Rename to..
1969 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
1970 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
1971 (vect_do_peeling): Update call accordingly.
1972 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
1974 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1975 mask_compare_type, can_fully_mask_p and fully_masked_p.
1976 (release_vec_loop_masks): New function.
1977 (_loop_vec_info): Use it to free the loop masks.
1978 (can_produce_all_loop_masks_p): New function.
1979 (vect_get_max_nscalars_per_iter): Likewise.
1980 (vect_verify_full_masking): Likewise.
1981 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
1982 retries, and free the mask rgroups before retrying. Check loop-wide
1983 reasons for disallowing fully-masked loops. Make the final decision
1984 about whether use a fully-masked loop or not.
1985 (vect_estimate_min_profitable_iters): Do not assume that peeling
1986 for the number of iterations will be needed for fully-masked loops.
1987 (vectorizable_reduction): Disable fully-masked loops.
1988 (vectorizable_live_operation): Likewise.
1989 (vect_halve_mask_nunits): New function.
1990 (vect_double_mask_nunits): Likewise.
1991 (vect_record_loop_mask): Likewise.
1992 (vect_get_loop_mask): Likewise.
1993 (vect_transform_loop): Handle the case in which the final loop
1994 iteration might handle a partial vector. Call vect_set_loop_condition
1995 instead of slpeel_make_loop_iterate_ntimes.
1996 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
1997 (check_load_store_masking): New function.
1998 (prepare_load_store_mask): Likewise.
1999 (vectorizable_store): Handle fully-masked loops.
2000 (vectorizable_load): Likewise.
2001 (supportable_widening_operation): Use vect_halve_mask_nunits for
2003 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2004 (vect_gen_while): New function.
2005 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2006 (aarch64_uqdec<mode>): New insn.
2008 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2009 Alan Hayward <alan.hayward@arm.com>
2010 David Sherwood <david.sherwood@arm.com>
2012 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2013 (reduc_xor_scal_optab): New optabs.
2014 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2015 (reduc_xor_scal_@var{m}): Document.
2016 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2017 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2019 * fold-const-call.c (fold_const_call): Handle them.
2020 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2021 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2022 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2023 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2024 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2025 (UNSPEC_XORV): New unspecs.
2026 (optab): Add entries for them.
2027 (BITWISEV): New int iterator.
2028 (bit_reduc_op): New int attributes.
2030 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2031 Alan Hayward <alan.hayward@arm.com>
2032 David Sherwood <david.sherwood@arm.com>
2034 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2035 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2036 * optabs.def (vec_shl_insert_optab): New optab.
2037 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2038 (duplicate_and_interleave): Likewise.
2039 * tree-vect-loop.c: Include internal-fn.h.
2040 (neutral_op_for_slp_reduction): New function, split out from
2041 get_initial_defs_for_reduction.
2042 (get_initial_def_for_reduction): Handle option 2 for variable-length
2043 vectors by loading the neutral value into a vector and then shifting
2044 the initial value into element 0.
2045 (get_initial_defs_for_reduction): Replace the code argument with
2046 the neutral value calculated by neutral_op_for_slp_reduction.
2047 Use gimple_build_vector for constant-length vectors.
2048 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2049 but the first group_size elements have a neutral value.
2050 Use duplicate_and_interleave otherwise.
2051 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2052 Update call to get_initial_defs_for_reduction. Handle SLP
2053 reductions for variable-length vectors by creating one vector
2054 result for each scalar result, with the elements associated
2055 with other scalar results stubbed out with the neutral value.
2056 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2057 Require IFN_VEC_SHL_INSERT for double reductions on
2058 variable-length vectors, or SLP reductions that have
2059 a neutral value. Require can_duplicate_and_interleave_p
2060 support for variable-length unchained SLP reductions if there
2061 is no neutral value, such as for MIN/MAX reductions. Also require
2062 the number of vector elements to be a multiple of the number of
2063 SLP statements when doing variable-length unchained SLP reductions.
2064 Update call to vect_create_epilog_for_reduction.
2065 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2066 and remove initial values.
2067 (duplicate_and_interleave): Make public.
2068 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2069 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2071 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2072 Alan Hayward <alan.hayward@arm.com>
2073 David Sherwood <david.sherwood@arm.com>
2075 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2076 (can_duplicate_and_interleave_p): New function.
2077 (vect_get_and_check_slp_defs): Take the vector of statements
2078 rather than just the current one. Remove excess parentheses.
2079 Restriction rejectinon of vect_constant_def and vect_external_def
2080 for variable-length vectors to boolean types, or types for which
2081 can_duplicate_and_interleave_p is false.
2082 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2083 (duplicate_and_interleave): New function.
2084 (vect_get_constant_vectors): Use gimple_build_vector for
2085 constant-length vectors and suitable variable-length constant
2086 vectors. Use duplicate_and_interleave for other variable-length
2087 vectors. Don't defer the update when inserting new statements.
2089 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2090 Alan Hayward <alan.hayward@arm.com>
2091 David Sherwood <david.sherwood@arm.com>
2093 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2094 min_profitable_iters doesn't go negative.
2096 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2097 Alan Hayward <alan.hayward@arm.com>
2098 David Sherwood <david.sherwood@arm.com>
2100 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2101 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2102 * optabs.def (vec_mask_load_lanes_optab): New optab.
2103 (vec_mask_store_lanes_optab): Likewise.
2104 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2105 (MASK_STORE_LANES): Likewise.
2106 * internal-fn.c (mask_load_lanes_direct): New macro.
2107 (mask_store_lanes_direct): Likewise.
2108 (expand_mask_load_optab_fn): Handle masked operations.
2109 (expand_mask_load_lanes_optab_fn): New macro.
2110 (expand_mask_store_optab_fn): Handle masked operations.
2111 (expand_mask_store_lanes_optab_fn): New macro.
2112 (direct_mask_load_lanes_optab_supported_p): Likewise.
2113 (direct_mask_store_lanes_optab_supported_p): Likewise.
2114 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2116 (vect_load_lanes_supported): Likewise.
2117 * tree-vect-data-refs.c (strip_conversion): New function.
2118 (can_group_stmts_p): Likewise.
2119 (vect_analyze_data_ref_accesses): Use it instead of checking
2120 for a pair of assignments.
2121 (vect_store_lanes_supported): Take a masked_p parameter.
2122 (vect_load_lanes_supported): Likewise.
2123 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2124 vect_store_lanes_supported and vect_load_lanes_supported.
2125 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2126 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2127 parameter. Don't allow gaps for masked accesses.
2128 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2129 and vect_load_lanes_supported.
2130 (get_load_store_type): Take a masked_p parameter and update
2131 call to get_group_load_store_type.
2132 (vectorizable_store): Update call to get_load_store_type.
2133 Handle IFN_MASK_STORE_LANES.
2134 (vectorizable_load): Update call to get_load_store_type.
2135 Handle IFN_MASK_LOAD_LANES.
2137 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2138 Alan Hayward <alan.hayward@arm.com>
2139 David Sherwood <david.sherwood@arm.com>
2141 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2143 * config/aarch64/aarch64-protos.h
2144 (aarch64_sve_struct_memory_operand_p): Declare.
2145 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2146 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2147 (VPRED, vpred): Handle SVE structure modes.
2148 * config/aarch64/constraints.md (Utx): New constraint.
2149 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2150 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2151 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2152 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2153 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2154 structure modes. Split into pieces after RA.
2155 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2156 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2158 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2159 SVE structure modes.
2160 (aarch64_classify_address): Likewise.
2161 (sizetochar): Move earlier in file.
2162 (aarch64_print_operand): Handle SVE register lists.
2163 (aarch64_array_mode): New function.
2164 (aarch64_sve_struct_memory_operand_p): Likewise.
2165 (TARGET_ARRAY_MODE): Redefine.
2167 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2168 Alan Hayward <alan.hayward@arm.com>
2169 David Sherwood <david.sherwood@arm.com>
2171 * target.def (array_mode): New target hook.
2172 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2173 * doc/tm.texi: Regenerate.
2174 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2175 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2176 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2178 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2181 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2182 Alan Hayward <alan.hayward@arm.com>
2183 David Sherwood <david.sherwood@arm.com>
2185 * fold-const.c (fold_binary_loc): Check the argument types
2186 rather than the result type when testing for a vector operation.
2188 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2190 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2191 * doc/tm.texi: Regenerate.
2193 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2194 Alan Hayward <alan.hayward@arm.com>
2195 David Sherwood <david.sherwood@arm.com>
2197 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2198 (sve): Document new AArch64 extension.
2199 * doc/md.texi (w): Extend the description of the AArch64
2200 constraint to include SVE vectors.
2201 (Upl, Upa): Document new AArch64 predicate constraints.
2202 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2204 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2205 (msve-vector-bits=): New option.
2206 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2207 SVE when these are disabled.
2208 (sve): New extension.
2209 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2210 modes. Adjust their number of units based on aarch64_sve_vg.
2211 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2212 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2213 aarch64_addr_query_type.
2214 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2215 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2216 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2217 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2218 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2219 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2220 (aarch64_simd_imm_zero_p): Delete.
2221 (aarch64_check_zero_based_sve_index_immediate): Declare.
2222 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2223 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2224 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2225 (aarch64_sve_float_mul_immediate_p): Likewise.
2226 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2228 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2229 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2230 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2231 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2232 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2233 (aarch64_regmode_natural_size): Likewise.
2234 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2235 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2237 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2238 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2239 for VG and the SVE predicate registers.
2240 (V_ALIASES): Add a "z"-prefixed alias.
2241 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2242 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2243 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2244 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2245 (REG_CLASS_NAMES): Add entries for them.
2246 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2247 and the predicate registers.
2248 (aarch64_sve_vg): Declare.
2249 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2250 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2251 (REGMODE_NATURAL_SIZE): Define.
2252 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2254 * config/aarch64/aarch64.c: Include cfgrtl.h.
2255 (simd_immediate_info): Add a constructor for series vectors,
2256 and an associated step field.
2257 (aarch64_sve_vg): New variable.
2258 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2259 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2260 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2261 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2262 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2263 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2264 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2265 (aarch64_get_mask_mode): New functions.
2266 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2267 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2268 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2269 predicate modes and predicate registers. Explicitly restrict
2270 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2271 to store a vector mode if it is recognized by
2272 aarch64_classify_vector_mode.
2273 (aarch64_regmode_natural_size): New function.
2274 (aarch64_hard_regno_caller_save_mode): Return the original mode
2276 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2277 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2278 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2279 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2281 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2282 does not overlap dest if the function is frame-related. Handle
2284 (aarch64_split_add_offset): New function.
2285 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2286 them aarch64_add_offset.
2287 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2288 and update call to aarch64_sub_sp.
2289 (aarch64_add_cfa_expression): New function.
2290 (aarch64_expand_prologue): Pass extra temporary registers to the
2291 functions above. Handle the case in which we need to emit new
2292 DW_CFA_expressions for registers that were originally saved
2293 relative to the stack pointer, but now have to be expressed
2294 relative to the frame pointer.
2295 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2297 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2298 IP0 and IP1 values for SVE frames.
2299 (aarch64_expand_vec_series): New function.
2300 (aarch64_expand_sve_widened_duplicate): Likewise.
2301 (aarch64_expand_sve_const_vector): Likewise.
2302 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2303 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2304 into the register, rather than emitting a SET directly.
2305 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2306 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2307 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2308 (offset_9bit_signed_scaled_p): New functions.
2309 (aarch64_replicate_bitmask_imm): New function.
2310 (aarch64_bitmask_imm): Use it.
2311 (aarch64_cannot_force_const_mem): Reject expressions involving
2312 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2313 (aarch64_classify_index): Handle SVE indices, by requiring
2314 a plain register index with a scale that matches the element size.
2315 (aarch64_classify_address): Handle SVE addresses. Assert that
2316 the mode of the address is VOIDmode or an integer mode.
2317 Update call to aarch64_classify_symbol.
2318 (aarch64_classify_symbolic_expression): Update call to
2319 aarch64_classify_symbol.
2320 (aarch64_const_vec_all_in_range_p): New function.
2321 (aarch64_print_vector_float_operand): Likewise.
2322 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2323 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2324 and the FP immediates 1.0 and 0.5.
2325 (aarch64_print_address_internal): Handle SVE addresses.
2326 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2327 (aarch64_regno_regclass): Handle predicate registers.
2328 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2330 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2331 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2332 (aarch64_convert_sve_vector_bits): New function.
2333 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2334 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2336 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2337 Handle SVE vector and predicate modes. Accept VL-based constants
2338 that need only one temporary register, and VL offsets that require
2339 no temporary registers.
2340 (aarch64_conditional_register_usage): Mark the predicate registers
2341 as fixed if SVE isn't available.
2342 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2343 Return true for SVE vector and predicate modes.
2344 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2345 rather than an unsigned int. Handle SVE modes.
2346 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2348 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2350 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2351 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2352 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2353 (aarch64_sve_float_mul_immediate_p): New functions.
2354 (aarch64_sve_valid_immediate): New function.
2355 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2356 Explicitly reject structure modes. Check for INDEX constants.
2357 Handle PTRUE and PFALSE constants.
2358 (aarch64_check_zero_based_sve_index_immediate): New function.
2359 (aarch64_simd_imm_zero_p): Delete.
2360 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2361 vector modes. Accept constants in the range of CNT[BHWD].
2362 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2363 ask for an Advanced SIMD mode.
2364 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2365 (aarch64_simd_vector_alignment): Handle SVE predicates.
2366 (aarch64_vectorize_preferred_vector_alignment): New function.
2367 (aarch64_simd_vector_alignment_reachable): Use it instead of
2369 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2370 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2372 (MAX_VECT_LEN): Delete.
2373 (expand_vec_perm_d): Add a vec_flags field.
2374 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2375 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2376 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2378 (aarch64_evpc_rev): Rename to...
2379 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2380 (aarch64_evpc_rev_global): New function.
2381 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2382 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2384 (aarch64_evpc_sve_tbl): New function.
2385 (aarch64_expand_vec_perm_const_1): Update after rename of
2386 aarch64_evpc_rev. Handle SVE permutes too, trying
2387 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2388 than aarch64_evpc_tbl.
2389 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2390 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2391 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2392 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2393 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2394 (aarch64_expand_sve_vcond): New functions.
2395 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2396 of aarch64_vector_mode_p.
2397 (aarch64_dwarf_poly_indeterminate_value): New function.
2398 (aarch64_compute_pressure_classes): Likewise.
2399 (aarch64_can_change_mode_class): Likewise.
2400 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2401 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2402 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2403 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2404 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2405 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2406 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2407 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2409 (Dn, Dl, Dr): Accept const as well as const_vector.
2410 (Dz): Likewise. Compare against CONST0_RTX.
2411 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2412 of "vector" where appropriate.
2413 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2414 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2415 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2416 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2417 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2418 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2419 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2420 (v_int_equiv): Extend to SVE modes.
2421 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2423 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2424 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2425 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2426 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2427 (SVE_COND_FP_CMP): New int iterators.
2428 (perm_hilo): Handle the new unpack unspecs.
2429 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2431 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2432 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2433 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2434 (aarch64_equality_operator, aarch64_constant_vector_operand)
2435 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2436 (aarch64_sve_nonimmediate_operand): Likewise.
2437 (aarch64_sve_general_operand): Likewise.
2438 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2439 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2440 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2441 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2442 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2443 (aarch64_sve_float_arith_immediate): Likewise.
2444 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2445 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2446 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2447 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2448 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2449 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2450 (aarch64_sve_float_arith_operand): Likewise.
2451 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2452 (aarch64_sve_float_mul_operand): Likewise.
2453 (aarch64_sve_vec_perm_operand): Likewise.
2454 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2455 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2456 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2457 as well as const_vector.
2458 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2459 in file. Use CONST0_RTX and CONSTM1_RTX.
2460 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2461 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2462 Use aarch64_simd_imm_zero.
2463 * config/aarch64/aarch64-sve.md: New file.
2464 * config/aarch64/aarch64.md: Include it.
2465 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2466 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2467 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2468 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2469 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2470 (sve): New attribute.
2471 (enabled): Disable instructions with the sve attribute unless
2473 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2474 aarch64_expand_mov_immediate.
2475 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2476 CNT[BHSD] immediates.
2477 (movti): Split CONST_POLY_INT moves into two halves.
2478 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2479 Split additions that need a temporary here if the destination
2480 is the stack pointer.
2481 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2482 (*add<mode>3_poly_1): New instruction.
2483 (set_clobber_cc): New expander.
2485 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2487 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2488 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2489 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2490 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2491 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2492 Change innermode from fixed_mode_size to machine_mode.
2493 (simplify_subreg): Update call accordingly. Handle a constant-sized
2494 subreg of a variable-length CONST_VECTOR.
2496 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2497 Alan Hayward <alan.hayward@arm.com>
2498 David Sherwood <david.sherwood@arm.com>
2500 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2501 (add_offset_to_base): New function, split out from...
2502 (create_mem_ref): ...here. When handling a scale other than 1,
2503 check first whether the address is valid without the offset.
2504 Add it into the base if so, leaving the index and scale as-is.
2506 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2509 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2510 fold_for_warn before checking if arg2 is INTEGER_CST.
2512 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2514 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2515 (store_multiple_operation): Delete.
2516 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2517 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2518 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2519 guarded by TARGET_STRING.
2520 (rs6000_output_load_multiple): Delete.
2521 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2522 OPTION_MASK_STRING / TARGET_STRING handling.
2523 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2524 (const rs6000_opt_masks) <"string">: Change mask to 0.
2525 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2526 (MASK_STRING): Delete.
2527 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2529 (load_multiple): Delete.
2536 (store_multiple): Delete.
2543 (movmemsi_8reg): Delete.
2544 (corresponding unnamed define_insn): Delete.
2545 (movmemsi_6reg): Delete.
2546 (corresponding unnamed define_insn): Delete.
2547 (movmemsi_4reg): Delete.
2548 (corresponding unnamed define_insn): Delete.
2549 (movmemsi_2reg): Delete.
2550 (corresponding unnamed define_insn): Delete.
2551 (movmemsi_1reg): Delete.
2552 (corresponding unnamed define_insn): Delete.
2553 * config/rs6000/rs6000.opt (mno-string): New.
2554 (mstring): Replace by deprecation warning stub.
2555 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2557 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2559 * regrename.c (regrename_do_replace): If replacing the same
2560 reg multiple times, try to reuse last created gen_raw_REG.
2563 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2564 main to workaround a bug in GDB.
2566 2018-01-12 Tom de Vries <tom@codesourcery.com>
2569 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2571 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2573 PR rtl-optimization/80481
2574 * ira-color.c (get_cap_member): New function.
2575 (allocnos_conflict_by_live_ranges_p): Use it.
2576 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2577 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2579 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2582 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2583 (*saddl_se_1): Ditto.
2585 (*ssubl_se_1): Ditto.
2587 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2589 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2590 rather than wi::to_widest for DR_INITs.
2591 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2592 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2593 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2595 (vect_analyze_group_access_1): Note that here.
2597 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2599 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2600 polynomial type sizes.
2602 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2604 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2605 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2606 (gimple_add_tmp_var): Likewise.
2608 2018-01-12 Martin Liska <mliska@suse.cz>
2610 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2611 (gimple_alloc_sizes): Likewise.
2612 (dump_gimple_statistics): Use PRIu64 in printf format.
2613 * gimple.h: Change uint64_t to int.
2615 2018-01-12 Martin Liska <mliska@suse.cz>
2617 * tree-core.h: Use uint64_t instead of int.
2618 * tree.c (tree_node_counts): Likewise.
2619 (tree_node_sizes): Likewise.
2620 (dump_tree_statistics): Use PRIu64 in printf format.
2622 2018-01-12 Martin Liska <mliska@suse.cz>
2624 * Makefile.in: As qsort_chk is implemented in vec.c, add
2625 vec.o to linkage of gencfn-macros.
2626 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2627 passing the info to record_node_allocation_statistics.
2628 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2630 * ggc-common.c (struct ggc_usage): Add operator== and use
2631 it in operator< and compare function.
2632 * mem-stats.h (struct mem_usage): Likewise.
2633 * vec.c (struct vec_usage): Remove operator< and compare
2634 function. Can be simply inherited.
2636 2018-01-12 Martin Jambor <mjambor@suse.cz>
2639 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2640 * tree-ssa-math-opts.c: Include domwalk.h.
2641 (convert_mult_to_fma_1): New function.
2642 (fma_transformation_info): New type.
2643 (fma_deferring_state): Likewise.
2644 (cancel_fma_deferring): New function.
2645 (result_of_phi): Likewise.
2646 (last_fma_candidate_feeds_initial_phi): Likewise.
2647 (convert_mult_to_fma): Added deferring logic, split actual
2648 transformation to convert_mult_to_fma_1.
2649 (math_opts_dom_walker): New type.
2650 (math_opts_dom_walker::after_dom_children): New method, body moved
2651 here from pass_optimize_widening_mul::execute, added deferring logic
2653 (pass_optimize_widening_mul::execute): Moved most of code to
2654 math_opts_dom_walker::after_dom_children.
2655 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2656 * config/i386/i386.c (ix86_option_override_internal): Added
2657 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2659 2018-01-12 Richard Biener <rguenther@suse.de>
2662 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2663 inline instance vars.
2665 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2668 * config/rx/rx.c (rx_is_restricted_memory_address):
2671 2018-01-12 Richard Biener <rguenther@suse.de>
2673 PR tree-optimization/80846
2674 * target.def (split_reduction): New target hook.
2675 * targhooks.c (default_split_reduction): New function.
2676 * targhooks.h (default_split_reduction): Declare.
2677 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2678 target requests first reduce vectors by combining low and high
2680 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2681 (get_vectype_for_scalar_type_and_size): Export.
2682 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2683 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2684 * doc/tm.texi: Regenerate.
2685 * config/i386/i386.c (ix86_split_reduction): Implement
2686 TARGET_VECTORIZE_SPLIT_REDUCTION.
2688 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2691 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2692 in PIC mode except for TARGET_VXWORKS_RTP.
2693 * config/sparc/sparc.c: Include cfgrtl.h.
2694 (TARGET_INIT_PIC_REG): Define.
2695 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2696 (sparc_pic_register_p): New predicate.
2697 (sparc_legitimate_address_p): Use it.
2698 (sparc_legitimize_pic_address): Likewise.
2699 (sparc_delegitimize_address): Likewise.
2700 (sparc_mode_dependent_address_p): Likewise.
2701 (gen_load_pcrel_sym): Remove 4th parameter.
2702 (load_got_register): Adjust call to above. Remove obsolete stuff.
2703 (sparc_expand_prologue): Do not call load_got_register here.
2704 (sparc_flat_expand_prologue): Likewise.
2705 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2706 (sparc_use_pseudo_pic_reg): New function.
2707 (sparc_init_pic_reg): Likewise.
2708 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2709 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2711 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2713 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2714 Add item for branch_cost.
2716 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2718 PR rtl-optimization/83565
2719 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2720 not extend the result to a larger mode for rotate operations.
2721 (num_sign_bit_copies1): Likewise.
2723 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2726 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2728 Use values-Xc.o for -pedantic.
2729 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2731 2018-01-12 Martin Liska <mliska@suse.cz>
2734 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2736 (possible_polymorphic_call_targets): Use it.
2737 (ipa_devirt): Likewise.
2739 2018-01-12 Martin Liska <mliska@suse.cz>
2741 * profile-count.h (enum profile_quality): Use 0 as invalid
2742 enum value of profile_quality.
2744 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2746 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2747 -mext-string options.
2749 2018-01-12 Richard Biener <rguenther@suse.de>
2751 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2752 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2753 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2755 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2757 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2759 * configure.ac (--with-long-double-format): Add support for the
2760 configuration option to change the default long double format on
2762 * config.gcc (powerpc*-linux*-*): Likewise.
2763 * configure: Regenerate.
2764 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2765 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2766 used without modification.
2768 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2770 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2771 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2772 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2773 MISC_BUILTIN_SPEC_BARRIER.
2774 (rs6000_init_builtins): Likewise.
2775 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2777 (speculation_barrier): New define_insn.
2778 * doc/extend.texi: Document __builtin_speculation_barrier.
2780 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2783 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2784 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2785 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2787 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2788 integral modes instead of "ss" and "sd".
2789 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2790 vectors with 32-bit and 64-bit elements.
2791 (vecdupssescalarmodesuffix): New mode attribute.
2792 (vec_dup<mode>): Use it.
2794 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2797 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2798 frame if argument is passed on stack.
2800 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2803 * ree.c (combine_reaching_defs): Optimize also
2804 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
2805 reg2=any_extend(exp); reg1=reg2;, formatting fix.
2807 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2810 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
2812 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2815 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
2816 after they are computed.
2818 2018-01-11 Bin Cheng <bin.cheng@arm.com>
2820 PR tree-optimization/83695
2821 * gimple-loop-linterchange.cc
2822 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
2823 reset cached scev information after interchange.
2824 (pass_linterchange::execute): Remove call to scev_reset_htab.
2826 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2828 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
2829 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
2830 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
2831 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
2832 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
2833 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
2834 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
2835 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
2836 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
2837 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
2838 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
2839 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
2840 (V_lane_reg): Likewise.
2841 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
2843 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
2844 (vfmal_lane_low<mode>_intrinsic,
2845 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
2846 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
2847 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
2848 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
2849 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
2850 vfmsl_lane_high<mode>_intrinsic): New define_insns.
2852 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2854 * config/arm/arm-cpus.in (fp16fml): New feature.
2855 (ALL_SIMD): Add fp16fml.
2856 (armv8.2-a): Add fp16fml as an option.
2857 (armv8.3-a): Likewise.
2858 (armv8.4-a): Add fp16fml as part of fp16.
2859 * config/arm/arm.h (TARGET_FP16FML): Define.
2860 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
2862 * config/arm/arm-modes.def (V2HF): Define.
2863 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
2864 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
2865 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
2866 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
2867 vfmsl_low, vfmsl_high): New set of builtins.
2868 * config/arm/iterators.md (PLUSMINUS): New code iterator.
2869 (vfml_op): New code attribute.
2870 (VFMLHALVES): New int iterator.
2871 (VFML, VFMLSEL): New mode attributes.
2872 (V_reg): Define mapping for V2HF.
2873 (V_hi, V_lo): New mode attributes.
2874 (VF_constraint): Likewise.
2875 (vfml_half, vfml_half_selector): New int attributes.
2876 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
2878 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
2879 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
2881 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
2882 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
2883 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
2884 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
2886 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
2887 Document new effective target and option set.
2889 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2891 * config/arm/arm-cpus.in (armv8_4): New feature.
2892 (ARMv8_4a): New fgroup.
2893 (armv8.4-a): New arch.
2894 * config/arm/arm-tables.opt: Regenerate.
2895 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
2896 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
2897 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
2898 Add matching rules for -march=armv8.4-a and extensions.
2899 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
2901 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
2904 * config/rx/rx.md (BW): New mode attribute.
2905 (sync_lock_test_and_setsi): Add mode suffix to insn output.
2907 2018-01-11 Richard Biener <rguenther@suse.de>
2909 PR tree-optimization/83435
2910 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
2911 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
2912 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
2914 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2915 Alan Hayward <alan.hayward@arm.com>
2916 David Sherwood <david.sherwood@arm.com>
2918 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
2920 (aarch64_classify_address): Initialize it. Track polynomial offsets.
2921 (aarch64_print_address_internal): Use it to check for a zero offset.
2923 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2924 Alan Hayward <alan.hayward@arm.com>
2925 David Sherwood <david.sherwood@arm.com>
2927 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
2928 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
2929 Return a poly_int64 rather than a HOST_WIDE_INT.
2930 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
2931 rather than a HOST_WIDE_INT.
2932 * config/aarch64/aarch64.h (aarch64_frame): Protect with
2933 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
2934 hard_fp_offset, frame_size, initial_adjust, callee_offset and
2935 final_offset from HOST_WIDE_INT to poly_int64.
2936 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
2937 to_constant when getting the number of units in an Advanced SIMD
2939 (aarch64_builtin_vectorized_function): Check for a constant number
2941 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
2943 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
2944 attribute instead of GET_MODE_NUNITS.
2945 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
2946 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
2947 GET_MODE_SIZE for fixed-size registers.
2948 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
2949 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
2950 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
2951 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
2952 (aarch64_print_operand, aarch64_print_address_internal)
2953 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
2954 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
2955 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
2956 Handle polynomial GET_MODE_SIZE.
2957 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
2958 wider than SImode without modification.
2959 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
2960 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
2961 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
2962 passing and returning SVE modes.
2963 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
2964 rather than GEN_INT.
2965 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
2966 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
2967 (aarch64_allocate_and_probe_stack_space): Likewise.
2968 (aarch64_layout_frame): Cope with polynomial offsets.
2969 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
2970 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
2972 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
2973 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
2974 poly_int64 rather than a HOST_WIDE_INT.
2975 (aarch64_get_separate_components, aarch64_process_components)
2976 (aarch64_expand_prologue, aarch64_expand_epilogue)
2977 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
2978 (aarch64_anchor_offset): New function, split out from...
2979 (aarch64_legitimize_address): ...here.
2980 (aarch64_builtin_vectorization_cost): Handle polynomial
2981 TYPE_VECTOR_SUBPARTS.
2982 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
2984 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
2985 number of elements from the PARALLEL rather than the mode.
2986 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
2987 rather than GET_MODE_BITSIZE.
2988 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
2989 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
2990 (aarch64_expand_vec_perm_const_1): Handle polynomial
2991 d->perm.length () and d->perm elements.
2992 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
2993 Apply to_constant to d->perm elements.
2994 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
2995 polynomial CONST_VECTOR_NUNITS.
2996 (aarch64_move_pointer): Take amount as a poly_int64 rather
2998 (aarch64_progress_pointer): Avoid temporary variable.
2999 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3000 the mode attribute instead of GET_MODE.
3002 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3003 Alan Hayward <alan.hayward@arm.com>
3004 David Sherwood <david.sherwood@arm.com>
3006 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3007 x exists before using it.
3008 (aarch64_add_constant_internal): Rename to...
3009 (aarch64_add_offset_1): ...this. Replace regnum with separate
3010 src and dest rtxes. Handle the case in which they're different,
3011 including when the offset is zero. Replace scratchreg with an rtx.
3012 Use 2 additions if there is no spare register into which we can
3013 move a 16-bit constant.
3014 (aarch64_add_constant): Delete.
3015 (aarch64_add_offset): Replace reg with separate src and dest
3016 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3017 Use aarch64_add_offset_1.
3018 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3019 an rtx rather than an int. Take the delta as a poly_int64
3020 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3021 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3022 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3023 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3024 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3026 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3027 aarch64_add_constant.
3029 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3031 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3032 Use scalar_float_mode.
3034 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3036 * config/aarch64/aarch64-simd.md
3037 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3038 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3039 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3040 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3041 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3042 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3043 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3044 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3045 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3046 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3048 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3051 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3052 targ_options->x_arm_arch_string is non NULL.
3054 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3056 * config/aarch64/aarch64.h
3057 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3059 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3062 * expmed.c (emit_store_flag_force): Swap if const op0
3063 and change VOIDmode to mode of op0.
3065 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3067 PR rtl-optimization/83761
3068 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3069 than bytes to mode_for_size.
3071 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3074 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3075 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3078 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3081 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3082 when in layout mode.
3083 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3084 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3087 2018-01-10 Michael Collison <michael.collison@arm.com>
3089 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3090 * config/aarch64/aarch64-option-extension.def: Add
3091 AARCH64_OPT_EXTENSION of 'fp16fml'.
3092 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3093 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3094 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3095 * config/aarch64/constraints.md (Ui7): New constraint.
3096 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3097 (VFMLA_SEL_W): Ditto.
3100 (VFMLA16_LOW): New int iterator.
3101 (VFMLA16_HIGH): Ditto.
3102 (UNSPEC_FMLAL): New unspec.
3103 (UNSPEC_FMLSL): Ditto.
3104 (UNSPEC_FMLAL2): Ditto.
3105 (UNSPEC_FMLSL2): Ditto.
3106 (f16mac): New code attribute.
3107 * config/aarch64/aarch64-simd-builtins.def
3108 (aarch64_fmlal_lowv2sf): Ditto.
3109 (aarch64_fmlsl_lowv2sf): Ditto.
3110 (aarch64_fmlalq_lowv4sf): Ditto.
3111 (aarch64_fmlslq_lowv4sf): Ditto.
3112 (aarch64_fmlal_highv2sf): Ditto.
3113 (aarch64_fmlsl_highv2sf): Ditto.
3114 (aarch64_fmlalq_highv4sf): Ditto.
3115 (aarch64_fmlslq_highv4sf): Ditto.
3116 (aarch64_fmlal_lane_lowv2sf): Ditto.
3117 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3118 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3119 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3120 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3121 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3122 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3123 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3124 (aarch64_fmlal_lane_highv2sf): Ditto.
3125 (aarch64_fmlsl_lane_highv2sf): Ditto.
3126 (aarch64_fmlal_laneq_highv2sf): Ditto.
3127 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3128 (aarch64_fmlalq_lane_highv4sf): Ditto.
3129 (aarch64_fmlsl_lane_highv4sf): Ditto.
3130 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3131 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3132 * config/aarch64/aarch64-simd.md:
3133 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3134 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3135 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3136 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3137 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3138 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3139 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3140 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3141 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3142 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3143 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3144 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3145 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3146 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3147 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3148 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3149 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3150 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3151 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3152 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3153 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3154 (vfmlsl_low_u32): Ditto.
3155 (vfmlalq_low_u32): Ditto.
3156 (vfmlslq_low_u32): Ditto.
3157 (vfmlal_high_u32): Ditto.
3158 (vfmlsl_high_u32): Ditto.
3159 (vfmlalq_high_u32): Ditto.
3160 (vfmlslq_high_u32): Ditto.
3161 (vfmlal_lane_low_u32): Ditto.
3162 (vfmlsl_lane_low_u32): Ditto.
3163 (vfmlal_laneq_low_u32): Ditto.
3164 (vfmlsl_laneq_low_u32): Ditto.
3165 (vfmlalq_lane_low_u32): Ditto.
3166 (vfmlslq_lane_low_u32): Ditto.
3167 (vfmlalq_laneq_low_u32): Ditto.
3168 (vfmlslq_laneq_low_u32): Ditto.
3169 (vfmlal_lane_high_u32): Ditto.
3170 (vfmlsl_lane_high_u32): Ditto.
3171 (vfmlal_laneq_high_u32): Ditto.
3172 (vfmlsl_laneq_high_u32): Ditto.
3173 (vfmlalq_lane_high_u32): Ditto.
3174 (vfmlslq_lane_high_u32): Ditto.
3175 (vfmlalq_laneq_high_u32): Ditto.
3176 (vfmlslq_laneq_high_u32): Ditto.
3177 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3178 (AARCH64_FL_FOR_ARCH8_4): New.
3179 (AARCH64_ISA_F16FML): New ISA flag.
3180 (TARGET_F16FML): New feature flag for fp16fml.
3181 (doc/invoke.texi): Document new fp16fml option.
3183 2018-01-10 Michael Collison <michael.collison@arm.com>
3185 * config/aarch64/aarch64-builtins.c:
3186 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3187 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3188 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3189 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3190 (AARCH64_ISA_SHA3): New ISA flag.
3191 (TARGET_SHA3): New feature flag for sha3.
3192 * config/aarch64/iterators.md (sha512_op): New int attribute.
3193 (CRYPTO_SHA512): New int iterator.
3194 (UNSPEC_SHA512H): New unspec.
3195 (UNSPEC_SHA512H2): Ditto.
3196 (UNSPEC_SHA512SU0): Ditto.
3197 (UNSPEC_SHA512SU1): Ditto.
3198 * config/aarch64/aarch64-simd-builtins.def
3199 (aarch64_crypto_sha512hqv2di): New builtin.
3200 (aarch64_crypto_sha512h2qv2di): Ditto.
3201 (aarch64_crypto_sha512su0qv2di): Ditto.
3202 (aarch64_crypto_sha512su1qv2di): Ditto.
3203 (aarch64_eor3qv8hi): Ditto.
3204 (aarch64_rax1qv2di): Ditto.
3205 (aarch64_xarqv2di): Ditto.
3206 (aarch64_bcaxqv8hi): Ditto.
3207 * config/aarch64/aarch64-simd.md:
3208 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3209 (aarch64_crypto_sha512su0qv2di): Ditto.
3210 (aarch64_crypto_sha512su1qv2di): Ditto.
3211 (aarch64_eor3qv8hi): Ditto.
3212 (aarch64_rax1qv2di): Ditto.
3213 (aarch64_xarqv2di): Ditto.
3214 (aarch64_bcaxqv8hi): Ditto.
3215 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3216 (vsha512h2q_u64): Ditto.
3217 (vsha512su0q_u64): Ditto.
3218 (vsha512su1q_u64): Ditto.
3219 (veor3q_u16): Ditto.
3220 (vrax1q_u64): Ditto.
3222 (vbcaxq_u16): Ditto.
3223 * config/arm/types.md (crypto_sha512): New type attribute.
3224 (crypto_sha3): Ditto.
3225 (doc/invoke.texi): Document new sha3 option.
3227 2018-01-10 Michael Collison <michael.collison@arm.com>
3229 * config/aarch64/aarch64-builtins.c:
3230 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3231 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3232 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3233 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3234 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3235 (AARCH64_ISA_SM4): New ISA flag.
3236 (TARGET_SM4): New feature flag for sm4.
3237 * config/aarch64/aarch64-simd-builtins.def
3238 (aarch64_sm3ss1qv4si): Ditto.
3239 (aarch64_sm3tt1aq4si): Ditto.
3240 (aarch64_sm3tt1bq4si): Ditto.
3241 (aarch64_sm3tt2aq4si): Ditto.
3242 (aarch64_sm3tt2bq4si): Ditto.
3243 (aarch64_sm3partw1qv4si): Ditto.
3244 (aarch64_sm3partw2qv4si): Ditto.
3245 (aarch64_sm4eqv4si): Ditto.
3246 (aarch64_sm4ekeyqv4si): Ditto.
3247 * config/aarch64/aarch64-simd.md:
3248 (aarch64_sm3ss1qv4si): Ditto.
3249 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3250 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3251 (aarch64_sm4eqv4si): Ditto.
3252 (aarch64_sm4ekeyqv4si): Ditto.
3253 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3254 (sm3part_op): Ditto.
3255 (CRYPTO_SM3TT): Ditto.
3256 (CRYPTO_SM3PART): Ditto.
3257 (UNSPEC_SM3SS1): New unspec.
3258 (UNSPEC_SM3TT1A): Ditto.
3259 (UNSPEC_SM3TT1B): Ditto.
3260 (UNSPEC_SM3TT2A): Ditto.
3261 (UNSPEC_SM3TT2B): Ditto.
3262 (UNSPEC_SM3PARTW1): Ditto.
3263 (UNSPEC_SM3PARTW2): Ditto.
3264 (UNSPEC_SM4E): Ditto.
3265 (UNSPEC_SM4EKEY): Ditto.
3266 * config/aarch64/constraints.md (Ui2): New constraint.
3267 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3268 * config/arm/types.md (crypto_sm3): New type attribute.
3269 (crypto_sm4): Ditto.
3270 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3271 (vsm3tt1aq_u32): Ditto.
3272 (vsm3tt1bq_u32): Ditto.
3273 (vsm3tt2aq_u32): Ditto.
3274 (vsm3tt2bq_u32): Ditto.
3275 (vsm3partw1q_u32): Ditto.
3276 (vsm3partw2q_u32): Ditto.
3277 (vsm4eq_u32): Ditto.
3278 (vsm4ekeyq_u32): Ditto.
3279 (doc/invoke.texi): Document new sm4 option.
3281 2018-01-10 Michael Collison <michael.collison@arm.com>
3283 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3284 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3285 (AARCH64_FL_FOR_ARCH8_4): New.
3286 (AARCH64_FL_V8_4): New flag.
3287 (doc/invoke.texi): Document new armv8.4-a option.
3289 2018-01-10 Michael Collison <michael.collison@arm.com>
3291 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3292 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3293 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3294 * config/aarch64/aarch64-option-extension.def: Add
3295 AARCH64_OPT_EXTENSION of 'sha2'.
3296 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3297 (crypto): Disable sha2 and aes if crypto disabled.
3298 (crypto): Enable aes and sha2 if enabled.
3299 (simd): Disable sha2 and aes if simd disabled.
3300 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3302 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3303 (TARGET_SHA2): New feature flag for sha2.
3304 (TARGET_AES): New feature flag for aes.
3305 * config/aarch64/aarch64-simd.md:
3306 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3307 conditional on TARGET_AES.
3308 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3309 (aarch64_crypto_sha1hsi): Make pattern conditional
3311 (aarch64_crypto_sha1hv4si): Ditto.
3312 (aarch64_be_crypto_sha1hv4si): Ditto.
3313 (aarch64_crypto_sha1su1v4si): Ditto.
3314 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3315 (aarch64_crypto_sha1su0v4si): Ditto.
3316 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3317 (aarch64_crypto_sha256su0v4si): Ditto.
3318 (aarch64_crypto_sha256su1v4si): Ditto.
3319 (doc/invoke.texi): Document new aes and sha2 options.
3321 2018-01-10 Martin Sebor <msebor@redhat.com>
3323 PR tree-optimization/83781
3324 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3327 2018-01-11 Martin Sebor <msebor@gmail.com>
3328 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3330 PR tree-optimization/83501
3331 PR tree-optimization/81703
3333 * tree-ssa-strlen.c (get_string_cst): Rename...
3334 (get_string_len): ...to this. Handle global constants.
3335 (handle_char_store): Adjust.
3337 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3338 Jim Wilson <jimw@sifive.com>
3340 * config/riscv/riscv-protos.h (riscv_output_return): New.
3341 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3342 (riscv_attribute_table, riscv_output_return),
3343 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3344 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3345 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3347 (riscv_expand_prologue): Add early return for naked function.
3348 (riscv_expand_epilogue): Likewise.
3349 (riscv_function_ok_for_sibcall): Return false for naked function.
3350 (riscv_set_current_function): New.
3351 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3352 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3353 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3354 * doc/extend.texi (RISC-V Function Attributes): New.
3356 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3358 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3359 check for 128-bit long double before checking TCmode.
3360 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3361 128-bit long doubles before checking TFmode or TCmode.
3362 (FLOAT128_IBM_P): Likewise.
3364 2018-01-10 Martin Sebor <msebor@redhat.com>
3366 PR tree-optimization/83671
3367 * builtins.c (c_strlen): Unconditionally return zero for the empty
3369 Use -Warray-bounds for warnings.
3370 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3371 for non-constant array indices with COMPONENT_REF, arrays of
3372 arrays, and pointers to arrays.
3373 (gimple_fold_builtin_strlen): Determine and set length range for
3374 non-constant character arrays.
3376 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3379 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3382 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3384 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3386 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3389 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3390 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3391 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3392 indexed_or_indirect_operand predicate.
3393 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3394 (*vsx_le_perm_load_v8hi): Likewise.
3395 (*vsx_le_perm_load_v16qi): Likewise.
3396 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3397 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3398 (*vsx_le_perm_store_v8hi): Likewise.
3399 (*vsx_le_perm_store_v16qi): Likewise.
3400 (eight unnamed splitters): Likewise.
3402 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3404 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3405 * config/rs6000/emmintrin.h: Likewise.
3406 * config/rs6000/mmintrin.h: Likewise.
3407 * config/rs6000/xmmintrin.h: Likewise.
3409 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3412 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3414 * tree.c (tree_nop_conversion): Return true for location wrapper
3416 (maybe_wrap_with_location): New function.
3417 (selftest::check_strip_nops): New function.
3418 (selftest::test_location_wrappers): New function.
3419 (selftest::tree_c_tests): Call it.
3420 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3421 (maybe_wrap_with_location): New decl.
3422 (EXPR_LOCATION_WRAPPER_P): New macro.
3423 (location_wrapper_p): New inline function.
3424 (tree_strip_any_location_wrapper): New inline function.
3426 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3429 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3430 stack_realign_offset for the largest alignment of stack slot
3432 (ix86_find_max_used_stack_alignment): New function.
3433 (ix86_finalize_stack_frame_flags): Use it. Set
3434 max_used_stack_alignment if we don't realign stack.
3435 * config/i386/i386.h (machine_function): Add
3436 max_used_stack_alignment.
3438 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3440 * config/arm/arm.opt (-mbranch-cost): New option.
3441 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3444 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3447 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3448 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3450 2018-01-10 Richard Biener <rguenther@suse.de>
3453 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3454 early out so it also covers the case where we have a non-NULL
3457 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3459 PR tree-optimization/83753
3460 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3461 for non-strided grouped accesses if the number of elements is 1.
3463 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3466 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3467 * i386.h (TARGET_USE_GATHER): Define.
3468 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3470 2018-01-10 Martin Liska <mliska@suse.cz>
3473 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3474 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3476 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3477 CLEANUP_NO_PARTITIONING is not set.
3479 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3481 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3482 for vectors, as a partial revert of r254296.
3483 * rtl.h (const_vec_p): Delete.
3484 (const_vec_duplicate_p): Don't test for vector CONSTs.
3485 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3486 * expmed.c (make_tree): Likewise.
3489 * common.md (E, F): Use CONSTANT_P instead of checking for
3491 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3492 checking for CONST_VECTOR.
3494 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3497 * predict.c (force_edge_cold): Handle in more sane way edges
3500 2018-01-09 Carl Love <cel@us.ibm.com>
3502 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3504 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3505 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3506 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3507 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3508 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3509 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3510 * config/rs6000/rs6000-protos.h: Add extern defition for
3511 rs6000_generate_float2_double_code.
3512 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3514 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3515 (float2_v2df): Add define_expand.
3517 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3520 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3521 op_mode in the force_to_mode call.
3523 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3525 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3526 instead of checking each element individually.
3527 (aarch64_evpc_uzp): Likewise.
3528 (aarch64_evpc_zip): Likewise.
3529 (aarch64_evpc_ext): Likewise.
3530 (aarch64_evpc_rev): Likewise.
3531 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3532 instead of checking each element individually. Return true without
3534 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3535 whether all selected elements come from the same input, instead of
3536 checking each element individually. Remove calls to gen_rtx_REG,
3537 start_sequence and end_sequence and instead assert that no rtl is
3540 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3542 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3543 order of HIGH and CONST checks.
3545 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3547 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3548 if the destination isn't an SSA_NAME.
3550 2018-01-09 Richard Biener <rguenther@suse.de>
3552 PR tree-optimization/83668
3553 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3555 (canonicalize_loop_form): ... here, renamed from ...
3556 (canonicalize_loop_closed_ssa_form): ... this and amended to
3557 swap successor edges for loop exit blocks to make us use
3558 the RPO order we need for initial schedule generation.
3560 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3562 PR tree-optimization/64811
3563 * match.pd: When optimizing comparisons with Inf, avoid
3564 introducing or losing exceptions from comparisons with NaN.
3566 2018-01-09 Martin Liska <mliska@suse.cz>
3569 * asan.c (shadow_mem_size): Add gcc_assert.
3571 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3573 Don't save registers in main().
3576 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3577 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3578 * config/avr/avr.c (avr_set_current_function): Don't error if
3579 naked, OS_task or OS_main are specified at the same time.
3580 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3582 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3584 * common/config/avr/avr-common.c (avr_option_optimization_table):
3585 Switch on -mmain-is-OS_task for optimizing compilations.
3587 2018-01-09 Richard Biener <rguenther@suse.de>
3589 PR tree-optimization/83572
3590 * graphite.c: Include cfganal.h.
3591 (graphite_transform_loops): Connect infinite loops to exit
3592 and remove fake edges at the end.
3594 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3596 * ipa-inline.c (edge_badness): Revert accidental checkin.
3598 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3601 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3602 symbols; not inline clones.
3604 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3607 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3608 hard registers. Formatting fixes.
3610 PR preprocessor/83722
3611 * gcc.c (try_generate_repro): Pass
3612 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3613 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3616 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3617 Kito Cheng <kito.cheng@gmail.com>
3619 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3620 (riscv_leaf_function_p): Delete.
3621 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3623 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3625 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3627 (do_ifelse): New function.
3628 (do_isel): New function.
3629 (do_sub3): New function.
3630 (do_add3): New function.
3631 (do_load_mask_compare): New function.
3632 (do_overlap_load_compare): New function.
3633 (expand_compare_loop): New function.
3634 (expand_block_compare): Call expand_compare_loop() when appropriate.
3635 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3637 (-mblock-compare-inline-loop-limit): New option.
3639 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3642 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3643 Reverse order of second and third operands in first alternative.
3644 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3645 of first and second elements in UNSPEC_VPERMR vector.
3646 (altivec_expand_vec_perm_le): Likewise.
3648 2017-01-08 Jeff Law <law@redhat.com>
3650 PR rtl-optimizatin/81308
3651 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3652 (process_switch): If group_case_labels makes a change, then set
3654 (pass_convert_switch::execute): If a switch is converted, then
3655 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3657 PR rtl-optimization/81308
3658 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3661 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3663 PR target/83663 - Revert r255946
3664 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3665 generation for cases where splatting a value is not useful.
3666 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3667 across a vec_duplicate and a paradoxical subreg forming a vector
3668 mode to a vec_concat.
3670 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3672 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3673 -march=armv8.3-a variants.
3674 * config/arm/t-multilib: Likewise.
3675 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3677 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3679 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3681 (cceq_ior_compare_complement): Give it a name so I can use it, and
3682 change boolean_or_operator predicate to boolean_operator so it can
3683 be used to generate a crand.
3684 (eqne): New code iterator.
3685 (bd/bd_neg): New code_attrs.
3686 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3687 a single define_insn.
3688 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3689 decrement (bdnzt/bdnzf/bdzt/bdzf).
3690 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3691 with the new names of the branch decrement patterns, and added the
3692 names of the branch decrement conditional patterns.
3694 2018-01-08 Richard Biener <rguenther@suse.de>
3696 PR tree-optimization/83563
3697 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3700 2018-01-08 Richard Biener <rguenther@suse.de>
3703 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3705 2018-01-08 Richard Biener <rguenther@suse.de>
3707 PR tree-optimization/83685
3708 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3709 references to abnormals.
3711 2018-01-08 Richard Biener <rguenther@suse.de>
3714 * dwarf2out.c (output_indirect_strings): Handle empty
3715 skeleton_debug_str_hash.
3716 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3718 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3720 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3721 (emit_store_direct): Likewise.
3722 (arc_trampoline_adjust_address): Likewise.
3723 (arc_asm_trampoline_template): New function.
3724 (arc_initialize_trampoline): Use asm_trampoline_template.
3725 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3726 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3727 * config/arc/arc.md (flush_icache): Delete pattern.
3729 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3731 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3732 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3735 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3738 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3739 by not USED_FOR_TARGET.
3740 (make_pass_resolve_sw_modes): Likewise.
3742 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3744 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3747 2018-01-08 Richard Biener <rguenther@suse.de>
3750 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3752 2018-01-08 Richard Biener <rguenther@suse.de>
3755 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3757 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3760 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3761 basic blocks with a small number of successors.
3762 (convert_control_dep_chain_into_preds): Improve handling of
3764 (dump_predicates): Split apart into...
3765 (dump_pred_chain): ...here...
3766 (dump_pred_info): ...and here.
3767 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3768 (can_chain_union_be_invalidated_p): Improve check for invalidation
3770 (uninit_uses_cannot_happen): Avoid unnecessary if
3771 convert_control_dep_chain_into_preds yielded nothing.
3773 2018-01-06 Martin Sebor <msebor@redhat.com>
3775 PR tree-optimization/83640
3776 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3777 subtracting negative offset from size.
3778 (builtin_access::overlap): Adjust offset bounds of the access to fall
3779 within the size of the object if possible.
3781 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3783 PR rtl-optimization/83699
3784 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3785 extract_bit_field_as_subreg to cases in which the extracted
3786 value is also a vector.
3788 * lra-constraints.c (process_alt_operands): Test for the equivalence
3789 substitutions when detecting a possible reload cycle.
3791 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3794 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3795 by default if flag_selective_schedling{,2}. Formatting fixes.
3797 PR rtl-optimization/83682
3798 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
3799 if it has non-VECTOR_MODE element mode.
3800 (vec_duplicate_p): Likewise.
3803 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
3804 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
3806 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3809 * config/i386/i386-builtin.def
3810 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
3811 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
3812 Require also OPTION_MASK_ISA_AVX512F in addition to
3813 OPTION_MASK_ISA_GFNI.
3814 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
3815 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
3816 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
3817 to OPTION_MASK_ISA_GFNI.
3818 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
3819 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
3820 OPTION_MASK_ISA_AVX512BW.
3821 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
3822 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
3823 addition to OPTION_MASK_ISA_GFNI.
3824 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
3825 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
3826 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
3827 to OPTION_MASK_ISA_GFNI.
3828 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
3829 a requirement for all ISAs rather than any of them with a few
3831 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
3833 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
3834 bitmasks to be enabled with 3 exceptions, instead of requiring any
3835 enabled ISA with lots of exceptions.
3836 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
3837 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
3838 Change avx512bw in isa attribute to avx512f.
3839 * config/i386/sgxintrin.h: Add license boilerplate.
3840 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
3841 to __AVX512F__ and __AVX512VL to __AVX512VL__.
3842 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
3843 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
3845 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
3846 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
3847 temporarily sse2 rather than sse if not enabled already.
3850 * config/i386/sse.md (VI248_VLBW): Rename to ...
3851 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
3852 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
3853 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
3854 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
3855 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
3856 mode iterator instead of VI248_VLBW.
3858 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
3860 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
3861 (record_modified): Skip clobbers; add debug output.
3862 (param_change_prob): Use sreal frequencies.
3864 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3866 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
3867 punt for user-aligned variables.
3869 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3871 * tree-chrec.c (chrec_contains_symbols): Return true for
3874 2018-01-05 Sudakshina Das <sudi.das@arm.com>
3877 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
3878 of (x|y) == x for BICS pattern.
3880 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3882 PR tree-optimization/83605
3883 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
3884 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
3887 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
3889 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
3890 * config/epiphany/rtems.h: New file.
3892 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3893 Uros Bizjak <ubizjak@gmail.com>
3896 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
3897 QIreg_operand instead of register_operand predicate.
3898 * config/i386/i386.c (ix86_rop_should_change_byte_p,
3899 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
3900 comments instead of -fmitigate[-_]rop.
3902 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3905 * cgraphunit.c (symbol_table::compile): Switch to text_section
3906 before calling assembly_start debug hook.
3907 * run-rtl-passes.c (run_rtl_passes): Likewise.
3910 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3912 * tree-vrp.c (extract_range_from_binary_expr_1): Check
3913 range_int_cst_p rather than !symbolic_range_p before calling
3914 extract_range_from_multiplicative_op_1.
3916 2017-01-04 Jeff Law <law@redhat.com>
3918 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
3919 redundant test in assertion.
3921 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3923 * doc/rtl.texi: Document machine_mode wrapper classes.
3925 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3927 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
3930 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3932 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
3933 the VEC_PERM_EXPR fold to fail.
3935 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3938 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
3939 to switched_sections.
3941 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3944 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
3947 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
3950 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
3951 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
3953 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3956 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
3957 is BLKmode and bitpos not zero or mode change is needed.
3959 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3962 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
3965 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
3968 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
3969 instead of MULT rtx. Update all corresponding splitters.
3971 (*ssub<modesuffix>): Ditto.
3973 (*cmp_sadd_di): Update split patterns.
3974 (*cmp_sadd_si): Ditto.
3975 (*cmp_sadd_sidi): Ditto.
3976 (*cmp_ssub_di): Ditto.
3977 (*cmp_ssub_si): Ditto.
3978 (*cmp_ssub_sidi): Ditto.
3979 * config/alpha/predicates.md (const23_operand): New predicate.
3980 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
3981 Look for ASHIFT, not MULT inner operand.
3982 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
3984 2018-01-04 Martin Liska <mliska@suse.cz>
3986 PR gcov-profile/83669
3987 * gcov.c (output_intermediate_file): Add version to intermediate
3989 * doc/gcov.texi: Document new field 'version' in intermediate
3990 file format. Fix location of '-k' option of gcov command.
3992 2018-01-04 Martin Liska <mliska@suse.cz>
3995 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
3997 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3999 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4001 2018-01-03 Martin Sebor <msebor@redhat.com>
4003 PR tree-optimization/83655
4004 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4005 checking calls with invalid arguments.
4007 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4009 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4010 (vectorizable_mask_load_store): Delete.
4011 (vectorizable_call): Return false for masked loads and stores.
4012 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4013 instead of gimple_assign_rhs1.
4014 (vectorizable_load): Handle IFN_MASK_LOAD.
4015 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4017 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4019 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4021 (vectorizable_mask_load_store): ...here.
4022 (vectorizable_load): ...and here.
4024 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4026 * tree-vect-stmts.c (vect_build_all_ones_mask)
4027 (vect_build_zero_merge_argument): New functions, split out from...
4028 (vectorizable_load): ...here.
4030 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4032 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4034 (vectorizable_mask_load_store): ...here.
4035 (vectorizable_store): ...and here.
4037 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4039 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4041 (vectorizable_mask_load_store): ...here.
4043 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4045 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4046 (vect_model_store_cost): Take a vec_load_store_type instead of a
4048 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4049 (vect_model_store_cost): Take a vec_load_store_type instead of a
4051 (vectorizable_mask_load_store): Update accordingly.
4052 (vectorizable_store): Likewise.
4053 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4055 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4057 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4058 IFN_MASK_LOAD calls here rather than...
4059 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4061 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4062 Alan Hayward <alan.hayward@arm.com>
4063 David Sherwood <david.sherwood@arm.com>
4065 * expmed.c (extract_bit_field_1): For vector extracts,
4066 fall back to extract_bit_field_as_subreg if vec_extract
4069 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4070 Alan Hayward <alan.hayward@arm.com>
4071 David Sherwood <david.sherwood@arm.com>
4073 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4074 they are variable or constant sized.
4075 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4076 slots for constant-sized data.
4078 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4079 Alan Hayward <alan.hayward@arm.com>
4080 David Sherwood <david.sherwood@arm.com>
4082 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4083 handling COND_EXPRs with boolean comparisons, try to find a better
4084 basis for the mask type than the boolean itself.
4086 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4088 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4089 is calculated and how it can be overridden.
4090 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4091 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4093 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4096 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4097 Alan Hayward <alan.hayward@arm.com>
4098 David Sherwood <david.sherwood@arm.com>
4100 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4101 Remove the mode argument.
4102 (aarch64_simd_valid_immediate): Remove the mode and inverse
4104 * config/aarch64/iterators.md (bitsize): New iterator.
4105 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4106 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4107 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4108 aarch64_simd_valid_immediate.
4109 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4110 (aarch64_reg_or_bic_imm): Likewise.
4111 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4112 with an insn_type enum and msl with a modifier_type enum.
4113 Replace element_width with a scalar_mode. Change the shift
4114 to unsigned int. Add constructors for scalar_float_mode and
4115 scalar_int_mode elements.
4116 (aarch64_vect_float_const_representable_p): Delete.
4117 (aarch64_can_const_movi_rtx_p)
4118 (aarch64_simd_scalar_immediate_valid_for_move)
4119 (aarch64_simd_make_constant): Update call to
4120 aarch64_simd_valid_immediate.
4121 (aarch64_advsimd_valid_immediate_hs): New function.
4122 (aarch64_advsimd_valid_immediate): Likewise.
4123 (aarch64_simd_valid_immediate): Remove mode and inverse
4124 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4125 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4126 and aarch64_float_const_representable_p on the result.
4127 (aarch64_output_simd_mov_immediate): Remove mode argument.
4128 Update call to aarch64_simd_valid_immediate and use of
4129 simd_immediate_info.
4130 (aarch64_output_scalar_simd_mov_immediate): Update call
4133 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4134 Alan Hayward <alan.hayward@arm.com>
4135 David Sherwood <david.sherwood@arm.com>
4137 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4138 (mode_nunits): Likewise CONST_MODE_NUNITS.
4139 * machmode.def (ADJUST_NUNITS): Document.
4140 * genmodes.c (mode_data::need_nunits_adj): New field.
4141 (blank_mode): Update accordingly.
4142 (adj_nunits): New variable.
4143 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4145 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4146 listed in adj_nunits.
4147 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4148 listed in adj_nunits. Don't emit case statements for such modes.
4149 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4150 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4151 nothing if adj_nunits is nonnull.
4152 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4153 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4154 (emit_mode_fbit): Update use of print_maybe_const_decl.
4155 (emit_move_size): Likewise. Treat the array as non-const
4157 (emit_mode_adjustments): Handle adj_nunits.
4159 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4161 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4162 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4163 (VECTOR_MODES): Use it.
4164 (make_vector_modes): Take the prefix as an argument.
4166 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4167 Alan Hayward <alan.hayward@arm.com>
4168 David Sherwood <david.sherwood@arm.com>
4170 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4171 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4172 for MODE_VECTOR_BOOL.
4173 * machmode.def (VECTOR_BOOL_MODE): Document.
4174 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4175 (make_vector_bool_mode): New function.
4176 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4178 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4179 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4181 * stor-layout.c (int_mode_for_mode): Likewise.
4182 * tree.c (build_vector_type_for_mode): Likewise.
4183 * varasm.c (output_constant_pool_2): Likewise.
4184 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4185 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4186 for MODE_VECTOR_BOOL.
4187 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4188 of mode class checks.
4189 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4190 instead of a list of mode class checks.
4191 (expand_vector_scalar_condition): Likewise.
4192 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4194 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4195 Alan Hayward <alan.hayward@arm.com>
4196 David Sherwood <david.sherwood@arm.com>
4198 * machmode.h (mode_size): Change from unsigned short to
4200 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4201 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4202 or if measurement_type is not polynomial.
4203 (fixed_size_mode::includes_p): Check for constant-sized modes.
4204 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4205 return a poly_uint16 rather than an unsigned short.
4206 (emit_mode_size): Change the type of mode_size from unsigned short
4207 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4208 (emit_mode_adjustments): Cope with polynomial vector sizes.
4209 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4211 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4213 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4214 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4215 * caller-save.c (setup_save_areas): Likewise.
4216 (replace_reg_with_saved_mem): Likewise.
4217 * calls.c (emit_library_call_value_1): Likewise.
4218 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4219 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4220 (gen_lowpart_for_combine): Likewise.
4221 * convert.c (convert_to_integer_1): Likewise.
4222 * cse.c (equiv_constant, cse_insn): Likewise.
4223 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4224 (cselib_subst_to_values): Likewise.
4225 * dce.c (word_dce_process_block): Likewise.
4226 * df-problems.c (df_word_lr_mark_ref): Likewise.
4227 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4228 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4229 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4230 (rtl_for_decl_location): Likewise.
4231 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4232 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4233 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4234 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4235 (expand_expr_real_1): Likewise.
4236 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4237 (pad_below): Likewise.
4238 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4239 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4240 * ira.c (get_subreg_tracking_sizes): Likewise.
4241 * ira-build.c (ira_create_allocno_objects): Likewise.
4242 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4243 (ira_sort_regnos_for_alter_reg): Likewise.
4244 * ira-costs.c (record_operand_costs): Likewise.
4245 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4246 (resolve_simple_move): Likewise.
4247 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4248 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4249 (lra_constraints): Likewise.
4250 (CONST_POOL_OK_P): Reject variable-sized modes.
4251 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4252 (add_pseudo_to_slot, lra_spill): Likewise.
4253 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4254 * optabs-query.c (get_best_extraction_insn): Likewise.
4255 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4256 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4257 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4258 * recog.c (offsettable_address_addr_space_p): Likewise.
4259 * regcprop.c (maybe_mode_change): Likewise.
4260 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4261 * regrename.c (build_def_use): Likewise.
4262 * regstat.c (dump_reg_info): Likewise.
4263 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4264 (find_reloads, find_reloads_subreg_address): Likewise.
4265 * reload1.c (eliminate_regs_1): Likewise.
4266 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4267 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4268 (simplify_binary_operation_1, simplify_subreg): Likewise.
4269 * targhooks.c (default_function_arg_padding): Likewise.
4270 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4271 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4272 (verify_gimple_assign_ternary): Likewise.
4273 * tree-inline.c (estimate_move_cost): Likewise.
4274 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4275 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4276 (get_address_cost_ainc): Likewise.
4277 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4278 (vect_supportable_dr_alignment): Likewise.
4279 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4280 (vectorizable_reduction): Likewise.
4281 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4282 (vectorizable_operation, vectorizable_load): Likewise.
4283 * tree.c (build_same_sized_truth_vector_type): Likewise.
4284 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4285 * var-tracking.c (emit_note_insn_var_location): Likewise.
4286 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4287 (ADDR_VEC_ALIGN): Likewise.
4289 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4290 Alan Hayward <alan.hayward@arm.com>
4291 David Sherwood <david.sherwood@arm.com>
4293 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4295 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4296 or if measurement_type is polynomial.
4297 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4298 * combine.c (make_extraction): Likewise.
4299 * dse.c (find_shift_sequence): Likewise.
4300 * dwarf2out.c (mem_loc_descriptor): Likewise.
4301 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4302 (extract_bit_field, extract_low_bits): Likewise.
4303 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4304 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4305 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4306 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4307 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4308 * reload.c (find_reloads): Likewise.
4309 * reload1.c (alter_reg): Likewise.
4310 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4311 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4312 * tree-if-conv.c (predicate_mem_writes): Likewise.
4313 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4314 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4315 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4316 * valtrack.c (dead_debug_insert_temp): Likewise.
4317 * varasm.c (mergeable_constant_section): Likewise.
4318 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4320 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4321 Alan Hayward <alan.hayward@arm.com>
4322 David Sherwood <david.sherwood@arm.com>
4324 * expr.c (expand_assignment): Cope with polynomial mode sizes
4325 when assigning to a CONCAT.
4327 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4328 Alan Hayward <alan.hayward@arm.com>
4329 David Sherwood <david.sherwood@arm.com>
4331 * machmode.h (mode_precision): Change from unsigned short to
4333 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4335 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4336 or if measurement_type is not polynomial.
4337 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4338 in which the mode is already known to be a scalar_int_mode.
4339 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4340 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4342 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4343 for GET_MODE_PRECISION.
4344 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4345 for GET_MODE_PRECISION.
4346 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4348 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4349 (expand_field_assignment, make_extraction): Likewise.
4350 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4351 (get_last_value): Likewise.
4352 * convert.c (convert_to_integer_1): Likewise.
4353 * cse.c (cse_insn): Likewise.
4354 * expr.c (expand_expr_real_1): Likewise.
4355 * lra-constraints.c (simplify_operand_subreg): Likewise.
4356 * optabs-query.c (can_atomic_load_p): Likewise.
4357 * optabs.c (expand_atomic_load): Likewise.
4358 (expand_atomic_store): Likewise.
4359 * ree.c (combine_reaching_defs): Likewise.
4360 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4361 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4362 * tree.h (type_has_mode_precision_p): Likewise.
4363 * ubsan.c (instrument_si_overflow): Likewise.
4365 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4366 Alan Hayward <alan.hayward@arm.com>
4367 David Sherwood <david.sherwood@arm.com>
4369 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4370 polynomial numbers of units.
4371 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4372 (valid_vector_subparts_p): New function.
4373 (build_vector_type): Remove temporary shim and take the number
4374 of units as a poly_uint64 rather than an int.
4375 (build_opaque_vector_type): Take the number of units as a
4376 poly_uint64 rather than an int.
4377 * tree.c (build_vector_from_ctor): Handle polynomial
4378 TYPE_VECTOR_SUBPARTS.
4379 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4380 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4381 (build_vector_from_val): If the number of units is variable,
4382 use build_vec_duplicate_cst for constant operands and
4383 VEC_DUPLICATE_EXPR otherwise.
4384 (make_vector_type): Remove temporary is_constant ().
4385 (build_vector_type, build_opaque_vector_type): Take the number of
4386 units as a poly_uint64 rather than an int.
4387 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4389 * cfgexpand.c (expand_debug_expr): Likewise.
4390 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4391 (store_constructor, expand_expr_real_1): Likewise.
4392 (const_scalar_mask_from_tree): Likewise.
4393 * fold-const-call.c (fold_const_reduction): Likewise.
4394 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4395 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4396 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4397 (fold_relational_const): Likewise.
4398 (native_interpret_vector): Likewise. Change the size from an
4399 int to an unsigned int.
4400 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4401 TYPE_VECTOR_SUBPARTS.
4402 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4403 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4404 duplicating a non-constant operand into a variable-length vector.
4405 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4406 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4407 * ipa-icf.c (sem_variable::equals): Likewise.
4408 * match.pd: Likewise.
4409 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4410 * print-tree.c (print_node): Likewise.
4411 * stor-layout.c (layout_type): Likewise.
4412 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4413 * tree-cfg.c (verify_gimple_comparison): Likewise.
4414 (verify_gimple_assign_binary): Likewise.
4415 (verify_gimple_assign_ternary): Likewise.
4416 (verify_gimple_assign_single): Likewise.
4417 * tree-pretty-print.c (dump_generic_node): Likewise.
4418 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4419 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4420 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4421 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4422 (vect_shift_permute_load_chain): Likewise.
4423 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4424 (expand_vector_condition, optimize_vector_constructor): Likewise.
4425 (lower_vec_perm, get_compute_type): Likewise.
4426 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4427 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4428 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4429 (vect_recog_mask_conversion_pattern): Likewise.
4430 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4431 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4432 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4433 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4434 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4435 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4436 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4437 (supportable_widening_operation): Likewise.
4438 (supportable_narrowing_operation): Likewise.
4439 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4441 * varasm.c (output_constant): Likewise.
4443 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4444 Alan Hayward <alan.hayward@arm.com>
4445 David Sherwood <david.sherwood@arm.com>
4447 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4448 so that both the length == 3 and length != 3 cases set up their
4449 own permute vectors. Add comments explaining why we know the
4450 number of elements is constant.
4451 (vect_permute_load_chain): Likewise.
4453 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4454 Alan Hayward <alan.hayward@arm.com>
4455 David Sherwood <david.sherwood@arm.com>
4457 * machmode.h (mode_nunits): Change from unsigned char to
4459 (ONLY_FIXED_SIZE_MODES): New macro.
4460 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4461 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4462 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4464 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4465 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4466 or if measurement_type is not polynomial.
4467 * genmodes.c (ZERO_COEFFS): New macro.
4468 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4470 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4471 Use ZERO_COEFFS when emitting initializers.
4472 * data-streamer.h (bp_pack_poly_value): New function.
4473 (bp_unpack_poly_value): Likewise.
4474 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4475 for GET_MODE_NUNITS.
4476 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4477 for GET_MODE_NUNITS.
4478 * tree.c (make_vector_type): Remove temporary shim and make
4479 the real function take the number of units as a poly_uint64
4481 (build_vector_type_for_mode): Handle polynomial nunits.
4482 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4483 * emit-rtl.c (const_vec_series_p_1): Likewise.
4484 (gen_rtx_CONST_VECTOR): Likewise.
4485 * fold-const.c (test_vec_duplicate_folding): Likewise.
4486 * genrecog.c (validate_pattern): Likewise.
4487 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4488 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4489 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4490 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4491 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4492 * rtlanal.c (subreg_get_info): Likewise.
4493 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4494 (vect_grouped_load_supported): Likewise.
4495 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4496 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4497 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4498 (simplify_const_unary_operation, simplify_binary_operation_1)
4499 (simplify_const_binary_operation, simplify_ternary_operation)
4500 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4501 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4502 instead of CONST_VECTOR_NUNITS.
4503 * varasm.c (output_constant_pool_2): Likewise.
4504 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4505 explicit-encoded elements in the XVEC for variable-length vectors.
4507 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4509 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4511 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4512 Alan Hayward <alan.hayward@arm.com>
4513 David Sherwood <david.sherwood@arm.com>
4515 * coretypes.h (fixed_size_mode): Declare.
4516 (fixed_size_mode_pod): New typedef.
4517 * builtins.h (target_builtins::x_apply_args_mode)
4518 (target_builtins::x_apply_result_mode): Change type to
4519 fixed_size_mode_pod.
4520 * builtins.c (apply_args_size, apply_result_size, result_vector)
4521 (expand_builtin_apply_args_1, expand_builtin_apply)
4522 (expand_builtin_return): Update accordingly.
4524 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4526 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4527 * cselib.c (cselib_hash_rtx): Likewise.
4528 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4529 CONST_VECTOR encoding.
4531 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4532 Jeff Law <law@redhat.com>
4535 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4536 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4537 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4538 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4541 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4542 explicitly probe *sp in a noreturn function if there were any callee
4543 register saves or frame pointer is needed.
4545 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4548 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4549 BLKmode for ternary, binary or unary expressions.
4552 * var-tracking.c (delete_vta_debug_insn): New inline function.
4553 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4554 insns from get_insns () to NULL instead of each bb separately.
4555 Use delete_vta_debug_insn. No longer static.
4556 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4557 delete_vta_debug_insns callers.
4558 * rtl.h (delete_vta_debug_insns): Declare.
4559 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4560 instead of variable_tracking_main.
4562 2018-01-03 Martin Sebor <msebor@redhat.com>
4564 PR tree-optimization/83603
4565 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4566 arguments past the endof the argument list in functions declared
4567 without a prototype.
4568 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4569 Avoid checking when arguments are null.
4571 2018-01-03 Martin Sebor <msebor@redhat.com>
4574 * doc/extend.texi (attribute const): Fix a typo.
4575 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4576 issuing -Wsuggest-attribute for void functions.
4578 2018-01-03 Martin Sebor <msebor@redhat.com>
4580 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4581 offset_int::from instead of wide_int::to_shwi.
4582 (maybe_diag_overlap): Remove assertion.
4583 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4584 * gimple-ssa-sprintf.c (format_directive): Same.
4585 (parse_directive): Same.
4586 (sprintf_dom_walker::compute_format_length): Same.
4587 (try_substitute_return_value): Same.
4589 2017-01-03 Jeff Law <law@redhat.com>
4592 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4593 non-constant residual for zero at runtime and avoid probing in
4594 that case. Reorganize code for trailing problem to mirror handling
4597 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4599 PR tree-optimization/83501
4600 * tree-ssa-strlen.c (get_string_cst): New.
4601 (handle_char_store): Call get_string_cst.
4603 2018-01-03 Martin Liska <mliska@suse.cz>
4605 PR tree-optimization/83593
4606 * tree-ssa-strlen.c: Include tree-cfg.h.
4607 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4608 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4609 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4611 (strlen_dom_walker::before_dom_children): Call
4612 gimple_purge_dead_eh_edges. Dump tranformation with details
4614 (strlen_dom_walker::before_dom_children): Update call by adding
4615 new argument cleanup_eh.
4616 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4618 2018-01-03 Martin Liska <mliska@suse.cz>
4621 * cif-code.def (VARIADIC_THUNK): New enum value.
4622 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4625 2018-01-03 Jan Beulich <jbeulich@suse.com>
4627 * sse.md (mov<mode>_internal): Tighten condition for when to use
4628 vmovdqu<ssescalarsize> for TI and OI modes.
4630 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4632 Update copyright years.
4634 2018-01-03 Martin Liska <mliska@suse.cz>
4637 * ipa-visibility.c (function_and_variable_visibility): Skip
4638 functions with noipa attribure.
4640 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4642 * gcc.c (process_command): Update copyright notice dates.
4643 * gcov-dump.c (print_version): Ditto.
4644 * gcov.c (print_version): Ditto.
4645 * gcov-tool.c (print_version): Ditto.
4646 * gengtype.c (create_file): Ditto.
4647 * doc/cpp.texi: Bump @copying's copyright year.
4648 * doc/cppinternals.texi: Ditto.
4649 * doc/gcc.texi: Ditto.
4650 * doc/gccint.texi: Ditto.
4651 * doc/gcov.texi: Ditto.
4652 * doc/install.texi: Ditto.
4653 * doc/invoke.texi: Ditto.
4655 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4657 * vector-builder.h (vector_builder::m_full_nelts): Change from
4658 unsigned int to poly_uint64.
4659 (vector_builder::full_nelts): Update prototype accordingly.
4660 (vector_builder::new_vector): Likewise.
4661 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4662 (vector_builder::operator ==): Likewise.
4663 (vector_builder::finalize): Likewise.
4664 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4665 Take the number of elements as a poly_uint64 rather than an
4667 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4668 from unsigned int to poly_uint64.
4669 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4670 (vec_perm_indices::new_vector): Likewise.
4671 (vec_perm_indices::length): Likewise.
4672 (vec_perm_indices::nelts_per_input): Likewise.
4673 (vec_perm_indices::input_nelts): Likewise.
4674 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4675 number of elements per input as a poly_uint64 rather than an
4676 unsigned int. Use the original encoding for variable-length
4677 vectors, rather than clamping each individual element.
4678 For the second and subsequent elements in each pattern,
4679 clamp the step and base before clamping their sum.
4680 (vec_perm_indices::series_p): Handle polynomial element counts.
4681 (vec_perm_indices::all_in_range_p): Likewise.
4682 (vec_perm_indices_to_tree): Likewise.
4683 (vec_perm_indices_to_rtx): Likewise.
4684 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4685 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4686 (tree_vector_builder::new_binary_operation): Handle polynomial
4687 element counts. Return false if we need to know the number
4688 of elements at compile time.
4689 * fold-const.c (fold_vec_perm): Punt if the number of elements
4690 isn't known at compile time.
4692 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4694 * vec-perm-indices.h (vec_perm_builder): Change element type
4695 from HOST_WIDE_INT to poly_int64.
4696 (vec_perm_indices::element_type): Update accordingly.
4697 (vec_perm_indices::clamp): Handle polynomial element_types.
4698 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4699 (vec_perm_indices::all_in_range_p): Likewise.
4700 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4702 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4703 polynomial vec_perm_indices element types.
4704 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4705 * fold-const.c (fold_vec_perm): Likewise.
4706 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4707 * tree-vect-generic.c (lower_vec_perm): Likewise.
4708 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4709 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4710 element type to HOST_WIDE_INT.
4712 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4713 Alan Hayward <alan.hayward@arm.com>
4714 David Sherwood <david.sherwood@arm.com>
4716 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4717 rather than an int. Use plus_constant.
4718 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4719 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4721 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4722 Alan Hayward <alan.hayward@arm.com>
4723 David Sherwood <david.sherwood@arm.com>
4725 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4726 a HOST_WIDE_INT to a poly_int64.
4728 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4729 Alan Hayward <alan.hayward@arm.com>
4730 David Sherwood <david.sherwood@arm.com>
4732 * calls.c (load_register_parameters): Cope with polynomial
4733 mode sizes. Require a constant size for BLKmode parameters
4734 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4735 forces a parameter to be padded at the lsb end in order to
4736 fill a complete number of words, require the parameter size
4737 to be ordered wrt UNITS_PER_WORD.
4739 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4740 Alan Hayward <alan.hayward@arm.com>
4741 David Sherwood <david.sherwood@arm.com>
4743 * reload1.c (spill_stack_slot_width): Change element type
4744 from unsigned int to poly_uint64_pod.
4745 (alter_reg): Treat mode sizes as polynomial.
4747 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4748 Alan Hayward <alan.hayward@arm.com>
4749 David Sherwood <david.sherwood@arm.com>
4751 * reload.c (complex_word_subreg_p): New function.
4752 (reload_inner_reg_of_subreg, push_reload): Use it.
4754 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4755 Alan Hayward <alan.hayward@arm.com>
4756 David Sherwood <david.sherwood@arm.com>
4758 * lra-constraints.c (process_alt_operands): Reject matched
4759 operands whose sizes aren't ordered.
4760 (match_reload): Refer to this check here.
4762 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4763 Alan Hayward <alan.hayward@arm.com>
4764 David Sherwood <david.sherwood@arm.com>
4766 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4767 that the mode size is in the set {1, 2, 4, 8, 16}.
4769 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4770 Alan Hayward <alan.hayward@arm.com>
4771 David Sherwood <david.sherwood@arm.com>
4773 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4774 Use plus_constant instead of gen_rtx_PLUS.
4776 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4777 Alan Hayward <alan.hayward@arm.com>
4778 David Sherwood <david.sherwood@arm.com>
4780 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4781 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4782 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4783 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4784 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4785 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4786 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4787 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4788 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4789 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4791 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4792 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4793 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4794 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4795 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4796 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4797 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4798 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
4799 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
4800 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
4802 * expr.c (emit_move_resolve_push): Treat the input and result
4803 of PUSH_ROUNDING as a poly_int64.
4804 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
4805 (emit_push_insn): Likewise.
4806 * lra-eliminations.c (mark_not_eliminable): Likewise.
4807 * recog.c (push_operand): Likewise.
4808 * reload1.c (elimination_effects): Likewise.
4809 * rtlanal.c (nonzero_bits1): Likewise.
4810 * calls.c (store_one_arg): Likewise. Require the padding to be
4811 known at compile time.
4813 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4814 Alan Hayward <alan.hayward@arm.com>
4815 David Sherwood <david.sherwood@arm.com>
4817 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
4818 Use plus_constant instead of gen_rtx_PLUS.
4820 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4821 Alan Hayward <alan.hayward@arm.com>
4822 David Sherwood <david.sherwood@arm.com>
4824 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
4827 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4828 Alan Hayward <alan.hayward@arm.com>
4829 David Sherwood <david.sherwood@arm.com>
4831 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
4832 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
4833 via stack temporaries. Treat the mode size as polynomial too.
4835 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4836 Alan Hayward <alan.hayward@arm.com>
4837 David Sherwood <david.sherwood@arm.com>
4839 * expr.c (expand_expr_real_2): When handling conversions involving
4840 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
4841 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
4842 as a poly_uint64 too.
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 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
4850 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4851 Alan Hayward <alan.hayward@arm.com>
4852 David Sherwood <david.sherwood@arm.com>
4854 * combine.c (can_change_dest_mode): Handle polynomial
4855 REGMODE_NATURAL_SIZE.
4856 * expmed.c (store_bit_field_1): Likewise.
4857 * expr.c (store_constructor): Likewise.
4858 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
4859 and polynomial REGMODE_NATURAL_SIZE.
4860 (gen_lowpart_common): Likewise.
4861 * reginfo.c (record_subregs_of_mode): Likewise.
4862 * rtlanal.c (read_modify_subreg_p): Likewise.
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 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
4869 numbers of elements.
4871 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4872 Alan Hayward <alan.hayward@arm.com>
4873 David Sherwood <david.sherwood@arm.com>
4875 * match.pd: Cope with polynomial numbers of vector elements.
4877 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4878 Alan Hayward <alan.hayward@arm.com>
4879 David Sherwood <david.sherwood@arm.com>
4881 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
4882 in a POINTER_PLUS_EXPR.
4884 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4885 Alan Hayward <alan.hayward@arm.com>
4886 David Sherwood <david.sherwood@arm.com>
4888 * omp-simd-clone.c (simd_clone_subparts): New function.
4889 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
4890 (ipa_simd_modify_function_body): Likewise.
4892 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4893 Alan Hayward <alan.hayward@arm.com>
4894 David Sherwood <david.sherwood@arm.com>
4896 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
4897 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
4898 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
4899 (expand_vector_condition, vector_element): Likewise.
4900 (subparts_gt): New function.
4901 (get_compute_type): Use subparts_gt.
4902 (count_type_subparts): Delete.
4903 (expand_vector_operations_1): Use subparts_gt instead of
4904 count_type_subparts.
4906 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4907 Alan Hayward <alan.hayward@arm.com>
4908 David Sherwood <david.sherwood@arm.com>
4910 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
4911 (vect_compile_time_alias): ...this new function. Do the calculation
4912 on poly_ints rather than trees.
4913 (vect_prune_runtime_alias_test_list): Update call accordingly.
4915 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4916 Alan Hayward <alan.hayward@arm.com>
4917 David Sherwood <david.sherwood@arm.com>
4919 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
4921 (vect_schedule_slp_instance): Likewise.
4923 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4924 Alan Hayward <alan.hayward@arm.com>
4925 David Sherwood <david.sherwood@arm.com>
4927 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
4928 constant and extern definitions for variable-length vectors.
4929 (vect_get_constant_vectors): Note that the number of units
4930 is known to be constant.
4932 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4933 Alan Hayward <alan.hayward@arm.com>
4934 David Sherwood <david.sherwood@arm.com>
4936 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
4937 of units as polynomial. Choose between WIDE and NARROW based
4940 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4941 Alan Hayward <alan.hayward@arm.com>
4942 David Sherwood <david.sherwood@arm.com>
4944 * tree-vect-stmts.c (simd_clone_subparts): New function.
4945 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
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-stmts.c (vectorizable_call): Treat the number of
4952 vectors as polynomial. Use build_index_vector for
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-stmts.c (get_load_store_type): Treat the number of
4960 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
4961 for variable-length vectors.
4962 (vectorizable_mask_load_store): Treat the number of units as
4963 polynomial, asserting that it is constant if the condition has
4964 already been enforced.
4965 (vectorizable_store, vectorizable_load): Likewise.
4967 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4968 Alan Hayward <alan.hayward@arm.com>
4969 David Sherwood <david.sherwood@arm.com>
4971 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
4972 of units as polynomial. Punt if we can't tell at compile time
4973 which vector contains the final result.
4975 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4976 Alan Hayward <alan.hayward@arm.com>
4977 David Sherwood <david.sherwood@arm.com>
4979 * tree-vect-loop.c (vectorizable_induction): Treat the number
4980 of units as polynomial. Punt on SLP inductions. Use an integer
4981 VEC_SERIES_EXPR for variable-length integer reductions. Use a
4982 cast of such a series for variable-length floating-point
4985 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4986 Alan Hayward <alan.hayward@arm.com>
4987 David Sherwood <david.sherwood@arm.com>
4989 * tree.h (build_index_vector): Declare.
4990 * tree.c (build_index_vector): New function.
4991 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
4992 of units as polynomial, forcibly converting it to a constant if
4993 vectorizable_reduction has already enforced the condition.
4994 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
4995 to create a {1,2,3,...} vector.
4996 (vectorizable_reduction): Treat the number of units as polynomial.
4997 Choose vectype_in based on the largest scalar element size rather
4998 than the smallest number of units. Enforce the restrictions
5001 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5002 Alan Hayward <alan.hayward@arm.com>
5003 David Sherwood <david.sherwood@arm.com>
5005 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5006 number of units as polynomial.
5008 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5009 Alan Hayward <alan.hayward@arm.com>
5010 David Sherwood <david.sherwood@arm.com>
5012 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5013 * target.def (autovectorize_vector_sizes): Return the vector sizes
5014 by pointer, using vector_sizes rather than a bitmask.
5015 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5016 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5017 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5019 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5020 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5021 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5022 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5023 * omp-general.c (omp_max_vf): Likewise.
5024 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5025 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5026 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5027 * tree-vect-slp.c (vect_slp_bb): Likewise.
5028 * doc/tm.texi: Regenerate.
5029 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5031 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5032 the vector size as a poly_uint64 rather than an unsigned int.
5033 (current_vector_size): Change from an unsigned int to a poly_uint64.
5034 (get_vectype_for_scalar_type): Update accordingly.
5035 * tree.h (build_truth_vector_type): Take the size and number of
5036 units as a poly_uint64 rather than an unsigned int.
5037 (build_vector_type): Add a temporary overload that takes
5038 the number of units as a poly_uint64 rather than an unsigned int.
5039 * tree.c (make_vector_type): Likewise.
5040 (build_truth_vector_type): Take the number of units as a poly_uint64
5041 rather than an unsigned int.
5043 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5044 Alan Hayward <alan.hayward@arm.com>
5045 David Sherwood <david.sherwood@arm.com>
5047 * target.def (get_mask_mode): Take the number of units and length
5048 as poly_uint64s rather than unsigned ints.
5049 * targhooks.h (default_get_mask_mode): Update accordingly.
5050 * targhooks.c (default_get_mask_mode): Likewise.
5051 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5052 * doc/tm.texi: Regenerate.
5054 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5055 Alan Hayward <alan.hayward@arm.com>
5056 David Sherwood <david.sherwood@arm.com>
5058 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5059 * omp-general.c (omp_max_vf): Likewise.
5060 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5061 (expand_omp_simd): Handle polynomial safelen.
5062 * omp-low.c (omplow_simd_context): Add a default constructor.
5063 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5064 (lower_rec_simd_input_clauses): Update accordingly.
5065 (lower_rec_input_clauses): Likewise.
5067 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5068 Alan Hayward <alan.hayward@arm.com>
5069 David Sherwood <david.sherwood@arm.com>
5071 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5072 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5073 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5074 (vect_analyze_slp_cost): Likewise.
5075 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5076 (vect_model_load_cost): Likewise.
5078 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5079 Alan Hayward <alan.hayward@arm.com>
5080 David Sherwood <david.sherwood@arm.com>
5082 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5083 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5084 from an unsigned int * to a poly_uint64_pod *.
5085 (calculate_unrolling_factor): New function.
5086 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5088 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5089 Alan Hayward <alan.hayward@arm.com>
5090 David Sherwood <david.sherwood@arm.com>
5092 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5093 from an unsigned int to a poly_uint64.
5094 (_loop_vec_info::slp_unrolling_factor): Likewise.
5095 (_loop_vec_info::vectorization_factor): Change from an int
5097 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5098 (vect_get_num_vectors): New function.
5099 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5100 (vect_get_num_copies): Use vect_get_num_vectors.
5101 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5102 to an unsigned int *.
5103 (vect_analyze_data_refs): Change min_vf from an int * to a
5105 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5106 than an unsigned HOST_WIDE_INT.
5107 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5108 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5109 to an unsigned int *.
5110 (vect_analyze_data_ref_dependences): Likewise.
5111 (vect_compute_data_ref_alignment): Handle polynomial vf.
5112 (vect_enhance_data_refs_alignment): Likewise.
5113 (vect_prune_runtime_alias_test_list): Likewise.
5114 (vect_shift_permute_load_chain): Likewise.
5115 (vect_supportable_dr_alignment): Likewise.
5116 (dependence_distance_ge_vf): Take the vectorization factor as a
5117 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5118 (vect_analyze_data_refs): Change min_vf from an int * to a
5120 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5121 vfm1 as a poly_uint64 rather than an int. Make the same change
5122 for the returned bound_scalar.
5123 (vect_gen_vector_loop_niters): Handle polynomial vf.
5124 (vect_do_peeling): Likewise. Update call to
5125 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5126 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5128 * tree-vect-loop.c (vect_determine_vectorization_factor)
5129 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5130 (vect_get_known_peeling_cost): Likewise.
5131 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5132 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5133 (vect_transform_loop): Likewise. Use the lowest possible VF when
5134 updating the upper bounds of the loop.
5135 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5137 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5138 polynomial unroll factors.
5139 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5140 (vect_make_slp_decision): Likewise.
5141 (vect_supported_load_permutation_p): Likewise, and polynomial
5143 (vect_analyze_slp_cost): Handle polynomial vf.
5144 (vect_slp_analyze_node_operations): Likewise.
5145 (vect_slp_analyze_bb_1): Likewise.
5146 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5147 than an unsigned HOST_WIDE_INT.
5148 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5149 (vectorizable_load): Handle polynomial vf.
5150 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5152 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5154 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5155 Alan Hayward <alan.hayward@arm.com>
5156 David Sherwood <david.sherwood@arm.com>
5158 * match.pd: Handle bit operations involving three constants
5159 and try to fold one pair.
5161 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5163 * tree-vect-loop-manip.c: Include gimple-fold.h.
5164 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5165 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5166 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5167 Add a path that uses a step of VF instead of 1, but disable it
5169 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5170 and niters_no_overflow parameters. Update calls to
5171 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5172 Create a new SSA name if the latter choses to use a ste other
5173 than zero, and return it via niters_vector_mult_vf_var.
5174 * tree-vect-loop.c (vect_transform_loop): Update calls to
5175 vect_do_peeling, vect_gen_vector_loop_niters and
5176 slpeel_make_loop_iterate_ntimes.
5177 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5178 (vect_gen_vector_loop_niters): Update declarations after above changes.
5180 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5182 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5183 128-bit round to integer instructions.
5184 (ceil<mode>2): Likewise.
5185 (btrunc<mode>2): Likewise.
5186 (round<mode>2): Likewise.
5188 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5190 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5191 unaligned VSX load/store on P8/P9.
5192 (expand_block_clear): Allow the use of unaligned VSX
5193 load/store on P8/P9.
5195 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5197 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5199 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5200 swap associated with both a load and a store.
5202 2018-01-02 Andrew Waterman <andrew@sifive.com>
5204 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5205 * config/riscv/riscv.md (clear_cache): Use it.
5207 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5209 * web.c: Remove out-of-date comment.
5211 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5213 * expr.c (fixup_args_size_notes): Check that any existing
5214 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5215 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5216 (emit_single_push_insn): ...here.
5218 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5220 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5221 (const_vector_encoded_nelts): New function.
5222 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5223 (const_vector_int_elt, const_vector_elt): Declare.
5224 * emit-rtl.c (const_vector_int_elt_1): New function.
5225 (const_vector_elt): Likewise.
5226 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5227 of CONST_VECTOR_ELT.
5229 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5231 * expr.c: Include rtx-vector-builder.h.
5232 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5233 directly on the tree encoding.
5234 (const_vector_from_tree): Likewise.
5235 * optabs.c: Include rtx-vector-builder.h.
5236 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5237 sequence of "u" values.
5238 * vec-perm-indices.c: Include rtx-vector-builder.h.
5239 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5240 directly on the vec_perm_indices encoding.
5242 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5244 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5245 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5246 * rtx-vector-builder.h: New file.
5247 * rtx-vector-builder.c: Likewise.
5248 * rtl.h (rtx_def::u2): Add a const_vector field.
5249 (CONST_VECTOR_NPATTERNS): New macro.
5250 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5251 (CONST_VECTOR_DUPLICATE_P): Likewise.
5252 (CONST_VECTOR_STEPPED_P): Likewise.
5253 (CONST_VECTOR_ENCODED_ELT): Likewise.
5254 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5255 (unwrap_const_vec_duplicate): Likewise.
5256 (const_vec_series_p): Check for a non-duplicated vector encoding.
5257 Say that the function only returns true for integer vectors.
5258 * emit-rtl.c: Include rtx-vector-builder.h.
5259 (gen_const_vec_duplicate_1): Delete.
5260 (gen_const_vector): Call gen_const_vec_duplicate instead of
5261 gen_const_vec_duplicate_1.
5262 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5263 (gen_const_vec_duplicate): Use rtx_vector_builder.
5264 (gen_const_vec_series): Likewise.
5265 (gen_rtx_CONST_VECTOR): Likewise.
5266 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5267 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5268 Build a new vector rather than modifying a CONST_VECTOR in-place.
5269 (handle_special_swappables): Update call accordingly.
5270 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5271 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5272 Build a new vector rather than modifying a CONST_VECTOR in-place.
5273 (handle_special_swappables): Update call accordingly.
5275 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5277 * simplify-rtx.c (simplify_const_binary_operation): Use
5278 CONST_VECTOR_ELT instead of XVECEXP.
5280 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5282 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5283 the selector elements to be different from the data elements
5284 if the selector is a VECTOR_CST.
5285 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5286 ssizetype for the selector.
5288 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5290 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5291 before testing each element individually.
5292 * tree-vect-generic.c (lower_vec_perm): Likewise.
5294 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5296 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5297 * selftest-run-tests.c (selftest::run_tests): Call it.
5298 * vector-builder.h (vector_builder::operator ==): New function.
5299 (vector_builder::operator !=): Likewise.
5300 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5301 (vec_perm_indices::all_from_input_p): New function.
5302 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5303 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5304 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5305 instead of reading the VECTOR_CST directly. Detect whether both
5306 vector inputs are the same before constructing the vec_perm_indices,
5307 and update the number of inputs argument accordingly. Use the
5308 utility functions added above. Only construct sel2 if we need to.
5310 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5312 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5313 the broadcast of the low byte.
5314 (expand_mult_highpart): Use an explicit encoding for the permutes.
5315 * optabs-query.c (can_mult_highpart_p): Likewise.
5316 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5317 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5318 (vectorizable_bswap): Likewise.
5319 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5320 explicit encoding for the power-of-2 permutes.
5321 (vect_permute_store_chain): Likewise.
5322 (vect_grouped_load_supported): Likewise.
5323 (vect_permute_load_chain): Likewise.
5325 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5327 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5328 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5329 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5330 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5331 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5332 (vect_gen_perm_mask_any): Likewise.
5334 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5336 * int-vector-builder.h: New file.
5337 * vec-perm-indices.h: Include int-vector-builder.h.
5338 (vec_perm_indices): Redefine as an int_vector_builder.
5339 (auto_vec_perm_indices): Delete.
5340 (vec_perm_builder): Redefine as a stand-alone class.
5341 (vec_perm_indices::vec_perm_indices): New function.
5342 (vec_perm_indices::clamp): Likewise.
5343 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5344 (vec_perm_indices::new_vector): New function.
5345 (vec_perm_indices::new_expanded_vector): Update for new
5346 vec_perm_indices class.
5347 (vec_perm_indices::rotate_inputs): New function.
5348 (vec_perm_indices::all_in_range_p): Operate directly on the
5349 encoded form, without computing elided elements.
5350 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5351 encoding. Update for new vec_perm_indices class.
5352 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5353 the given vec_perm_builder.
5354 (expand_vec_perm_var): Update vec_perm_builder constructor.
5355 (expand_mult_highpart): Use vec_perm_builder instead of
5356 auto_vec_perm_indices.
5357 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5358 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5359 or double series encoding as appropriate.
5360 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5361 vec_perm_indices instead of auto_vec_perm_indices.
5362 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5363 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5364 (vect_permute_store_chain): Likewise.
5365 (vect_grouped_load_supported): Likewise.
5366 (vect_permute_load_chain): Likewise.
5367 (vect_shift_permute_load_chain): Likewise.
5368 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5369 (vect_transform_slp_perm_load): Likewise.
5370 (vect_schedule_slp_instance): Likewise.
5371 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5372 (vectorizable_mask_load_store): Likewise.
5373 (vectorizable_bswap): Likewise.
5374 (vectorizable_store): Likewise.
5375 (vectorizable_load): Likewise.
5376 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5377 vec_perm_indices instead of auto_vec_perm_indices. Use
5378 tree_to_vec_perm_builder to read the vector from a tree.
5379 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5380 vec_perm_builder instead of a vec_perm_indices.
5381 (have_whole_vector_shift): Use vec_perm_builder and
5382 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5383 truncation to calc_vec_perm_mask_for_shift.
5384 (vect_create_epilog_for_reduction): Likewise.
5385 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5386 from auto_vec_perm_indices to vec_perm_indices.
5387 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5388 instead of changing individual elements.
5389 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5390 the vector in d.perm.
5391 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5392 from auto_vec_perm_indices to vec_perm_indices.
5393 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5394 instead of changing individual elements.
5395 (arm_vectorize_vec_perm_const): Use new_vector to install
5396 the vector in d.perm.
5397 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5398 Update vec_perm_builder constructor.
5399 (rs6000_expand_interleave): Likewise.
5400 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5401 (rs6000_expand_interleave): Likewise.
5403 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5405 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5406 to qimode could truncate the indices.
5407 * optabs.c (expand_vec_perm_var): Likewise.
5409 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5411 * Makefile.in (OBJS): Add vec-perm-indices.o.
5412 * vec-perm-indices.h: New file.
5413 * vec-perm-indices.c: Likewise.
5414 * target.h (vec_perm_indices): Replace with a forward class
5416 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5417 * optabs.h: Include vec-perm-indices.h.
5418 (expand_vec_perm): Delete.
5419 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5420 (expand_vec_perm_const): Declare.
5421 * target.def (vec_perm_const_ok): Replace with...
5422 (vec_perm_const): ...this new hook.
5423 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5424 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5425 * doc/tm.texi: Regenerate.
5426 * optabs.def (vec_perm_const): Delete.
5427 * doc/md.texi (vec_perm_const): Likewise.
5428 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5429 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5430 expand_vec_perm for constant permutation vectors. Assert that
5431 the mode of variable permutation vectors is the integer equivalent
5432 of the mode that is being permuted.
5433 * optabs-query.h (selector_fits_mode_p): Declare.
5434 * optabs-query.c: Include vec-perm-indices.h.
5435 (selector_fits_mode_p): New function.
5436 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5437 is defined, instead of checking whether the vec_perm_const_optab
5438 exists. Use targetm.vectorize.vec_perm_const instead of
5439 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5440 fit in the vector mode before using a variable permute.
5441 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5442 vec_perm_indices instead of an rtx.
5443 (expand_vec_perm): Replace with...
5444 (expand_vec_perm_const): ...this new function. Take the selector
5445 as a vec_perm_indices rather than an rtx. Also take the mode of
5446 the selector. Update call to shift_amt_for_vec_perm_mask.
5447 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5448 Use vec_perm_indices::new_expanded_vector to expand the original
5449 selector into bytes. Check whether the indices fit in the vector
5450 mode before using a variable permute.
5451 (expand_vec_perm_var): Make global.
5452 (expand_mult_highpart): Use expand_vec_perm_const.
5453 * fold-const.c: Includes vec-perm-indices.h.
5454 * tree-ssa-forwprop.c: Likewise.
5455 * tree-vect-data-refs.c: Likewise.
5456 * tree-vect-generic.c: Likewise.
5457 * tree-vect-loop.c: Likewise.
5458 * tree-vect-slp.c: Likewise.
5459 * tree-vect-stmts.c: Likewise.
5460 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5462 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5463 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5464 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5465 (aarch64_vectorize_vec_perm_const): ...this new function.
5466 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5467 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5468 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5469 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5470 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5471 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5472 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5474 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5475 check for NEON modes.
5476 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5477 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5478 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5479 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5481 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5482 the old VEC_PERM_CONST conditions.
5483 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5484 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5485 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5486 (ia64_vectorize_vec_perm_const_ok): Merge into...
5487 (ia64_vectorize_vec_perm_const): ...this new function.
5488 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5489 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5490 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5491 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5492 * config/mips/mips.c (mips_expand_vec_perm_const)
5493 (mips_vectorize_vec_perm_const_ok): Merge into...
5494 (mips_vectorize_vec_perm_const): ...this new function.
5495 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5496 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5497 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5498 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5499 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5500 (rs6000_expand_vec_perm_const): Delete.
5501 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5503 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5504 (altivec_expand_vec_perm_const_le): Take each operand individually.
5505 Operate on constant selectors rather than rtxes.
5506 (altivec_expand_vec_perm_const): Likewise. Update call to
5507 altivec_expand_vec_perm_const_le.
5508 (rs6000_expand_vec_perm_const): Delete.
5509 (rs6000_vectorize_vec_perm_const_ok): Delete.
5510 (rs6000_vectorize_vec_perm_const): New function.
5511 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5512 an element count and rtx array.
5513 (rs6000_expand_extract_even): Update call accordingly.
5514 (rs6000_expand_interleave): Likewise.
5515 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5516 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5517 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5518 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5519 (rs6000_expand_vec_perm_const): Delete.
5520 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5521 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5522 (altivec_expand_vec_perm_const_le): Take each operand individually.
5523 Operate on constant selectors rather than rtxes.
5524 (altivec_expand_vec_perm_const): Likewise. Update call to
5525 altivec_expand_vec_perm_const_le.
5526 (rs6000_expand_vec_perm_const): Delete.
5527 (rs6000_vectorize_vec_perm_const_ok): Delete.
5528 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5529 reference to the SPE evmerge intructions.
5530 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5531 an element count and rtx array.
5532 (rs6000_expand_extract_even): Update call accordingly.
5533 (rs6000_expand_interleave): Likewise.
5534 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5535 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5537 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5539 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5541 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5542 vector mode and that that mode matches the mode of the data
5544 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5545 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5546 directly using expand_vec_perm_1 when forcing selectors into
5548 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5550 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5552 * optabs-query.h (can_vec_perm_p): Delete.
5553 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5554 * optabs-query.c (can_vec_perm_p): Split into...
5555 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5556 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5557 particular selector is valid.
5558 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5559 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5560 (vect_grouped_load_supported): Likewise.
5561 (vect_shift_permute_load_chain): Likewise.
5562 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5563 (vect_transform_slp_perm_load): Likewise.
5564 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5565 (vectorizable_bswap): Likewise.
5566 (vect_gen_perm_mask_checked): Likewise.
5567 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5568 implementations of variable permutation vectors into account
5569 when deciding which selector to use.
5570 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5571 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5572 with a false third argument.
5573 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5574 to test whether the constant selector is valid and can_vec_perm_var_p
5575 to test whether a variable selector is valid.
5577 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5579 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5580 * optabs-query.c (can_vec_perm_p): Likewise.
5581 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5582 instead of vec_perm_indices.
5583 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5584 (vect_gen_perm_mask_checked): Likewise,
5585 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5586 (vect_gen_perm_mask_checked): Likewise,
5588 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5590 * optabs-query.h (qimode_for_vec_perm): Declare.
5591 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5592 (qimode_for_vec_perm): ...this new function.
5593 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5595 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5597 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5598 does not have a conditional at the top.
5600 2018-01-02 Richard Biener <rguenther@suse.de>
5602 * ipa-inline.c (big_speedup_p): Fix expression.
5604 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5607 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5610 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5614 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5615 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5616 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5617 cond_taken_branch_cost 3->4.
5619 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5621 PR tree-optimization/83581
5622 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5623 TODO_cleanup_cfg if any changes have been made.
5626 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5627 convert_modes if target mode has the right side, but different mode
5631 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5632 last argument when extracting from CONCAT. If either from_real or
5633 from_imag is NULL, use expansion through memory. If result is not
5634 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5635 the parts directly to inner mode, if even that fails, use expansion
5639 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5640 check for bswap in mode rather than HImode and use that in expand_unop
5643 Copyright (C) 2018 Free Software Foundation, Inc.
5645 Copying and distribution of this file, with or without modification,
5646 are permitted in any medium without royalty provided the copyright
5647 notice and this notice are preserved.