1 2018-01-27 Jakub Jelinek <jakub@redhat.com>
4 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
7 2018-01-26 Jim Wilson <jimw@sifive.com>
9 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
11 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
14 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
16 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
17 and CMP + SUB-immediate -> SUBS.
19 2018-01-26 Martin Sebor <msebor@redhat.com>
21 PR tree-optimization/83896
22 * tree-ssa-strlen.c (get_string_len): Rename...
23 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
24 Avoid assuming length is constant.
25 (handle_char_store): Use HOST_WIDE_INT for string length.
27 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
30 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
31 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
33 2018-01-26 Richard Biener <rguenther@suse.de>
35 PR rtl-optimization/84003
36 * dse.c (record_store): Only record redundant stores when
37 the earlier store aliases at least all accesses the later one does.
39 2018-01-26 Jakub Jelinek <jakub@redhat.com>
41 PR rtl-optimization/83985
42 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
43 REG_CFA_RESTORE insns.
44 (delete_unmarked_insns): Don't ignore separate shrink wrapping
45 REG_CFA_RESTORE insns here.
48 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
49 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
51 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
53 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
54 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
56 (arc_override_options): Likewise.
57 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
59 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
61 * config/arc/arc.h (TARGET_DBNZ): Define.
62 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
63 properly set the tune attribute.
64 (dbnz): Use TARGET_DBNZ guard.
65 * config/arc/arc.opt (mtune): Add core3 option.
67 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
69 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
70 recognize new pic like addresses.
71 (arc_delegitimize_address): Clean up.
73 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
75 * config/arc/arc-arches.def: Option mrf16 valid for all
77 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
78 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
79 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
80 * config/arc/arc-tables.opt: Regenerate.
81 * config/arc/arc.c (arc_conditional_register_usage): Handle
82 reduced register file case.
83 (arc_file_start): Set must have build attributes.
84 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
86 * config/arc/arc.opt (mrf16): Add new option.
87 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
88 * config/arc/genmultilib.awk: Handle new mrf16 option.
89 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
90 * config/arc/t-multilib: Regenerate.
91 * doc/invoke.texi (ARC Options): Document mrf16 option.
93 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
95 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
96 * config/arc/arc.c (arc_handle_secure_attribute): New function.
97 (arc_attribute_table): Add 'secure_call' attribute.
98 (arc_print_operand): Print secure call operand.
99 (arc_function_ok_for_sibcall): Don't optimize tail calls when
101 (arc_is_secure_call_p): New function. * config/arc/arc.md
102 (call_i): Add support for sjli instruction.
103 (call_value_i): Likewise.
104 * config/arc/constraints.md (Csc): New constraint.
106 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
107 John Eric Martin <John.Martin@emmicro-us.com>
109 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
110 * config/arc/arc.c (_arc_jli_section): New struct.
111 (arc_jli_section): New type.
112 (rc_jli_sections): New static variable.
113 (arc_handle_jli_attribute): New function.
114 (arc_attribute_table): Add jli_always and jli_fixed attribute.
115 (arc_file_end): New function.
116 (TARGET_ASM_FILE_END): Define.
117 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
118 (arc_add_jli_section): New function.
119 (jli_call_scan): Likewise.
120 (arc_reorg): Call jli_call_scan.
121 (arc_output_addsi): Remove 'S' from printing asm operand.
122 (arc_is_jli_call_p): New function.
123 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
125 (movhi_insn): Likewise.
126 (movsi_insn): Likewise.
127 (movsi_set_cc_insn): Likewise.
128 (loadqi_update): Likewise.
129 (load_zeroextendqisi_update): Likewise.
130 (load_signextendqisi_update): Likewise.
131 (loadhi_update): Likewise.
132 (load_zeroextendhisi_update): Likewise.
133 (load_signextendhisi_update): Likewise.
134 (loadsi_update): Likewise.
135 (loadsf_update): Likewise.
136 (movsicc_insn): Likewise.
137 (bset_insn): Likewise.
138 (bxor_insn): Likewise.
139 (bclr_insn): Likewise.
140 (bmsk_insn): Likewise.
141 (bicsi3_insn): Likewise.
142 (cmpsi_cc_c_insn): Likewise.
143 (movsi_ne): Likewise.
144 (movsi_cond_exec): Likewise.
145 (clrsbsi2): Likewise.
156 (call_i): Remove 'S' asm letter, add jli instruction.
157 (call_value_i): Likewise.
158 * config/arc/arc.op (mjli-always): New option.
159 * config/arc/constraints.md (Cji): New constraint.
160 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
162 (subsf3_fpx): Likewise.
163 (mulsf3_fpx): Likewise.
164 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
166 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
168 * doc/invoke.texi (ARC): Document mjli-always option.
170 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
172 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
173 and use incw and decw where possible
174 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
176 2018-01-26 Richard Biener <rguenther@suse.de>
178 PR tree-optimization/81082
179 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
180 association if it requires casting to unsigned.
181 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
182 from fold_plusminus_mult_expr to catch important cases late when
183 range info is available.
185 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
187 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
188 * configure.ac (hidden_linkonce): New test.
189 * configure: Regenerate.
190 * config.in: Regenerate.
192 2018-01-26 Julia Koval <julia.koval@intel.com>
194 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
195 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
196 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
197 _mm_mask_bitshuffle_epi64_mask): Fix type.
198 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
199 USI_FTYPE_V4DI_V4DI_USI): Remove.
200 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
201 __builtin_ia32_vpshufbitqmb256_mask,
202 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
203 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
204 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
206 2018-01-26 Alan Modra <amodra@gmail.com>
209 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
210 UNSPEC_VBPERMQ. Sort other unspecs.
212 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
214 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
216 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
219 * predict.c (drop_profile): Do not push/pop cfun; update also
221 (handle_missing_profiles): Fix logic looking for zero profiles.
223 2018-01-25 Jakub Jelinek <jakub@redhat.com>
226 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
227 on functions with #pragma omp declare simd or functions with simd
229 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
230 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
231 Remove trailing \n from warning_at calls.
233 2018-01-25 Tom de Vries <tom@codesourcery.com>
236 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
237 for neutered workers.
239 2018-01-24 Joseph Myers <joseph@codesourcery.com>
242 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
243 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
245 2017-01-08 Jeff Law <law@redhat.com>
248 * i386.c (get_probe_interval): Move to earlier point.
249 (ix86_compute_frame_layout): If -fstack-clash-protection and
250 the frame is larger than the probe interval, then use pushes
251 to save registers rather than reg->mem moves.
252 (ix86_expand_prologue): Remove conditional for int_registers_saved
255 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
258 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
259 min/max for never referenced object.
261 2018-01-24 Jakub Jelinek <jakub@redhat.com>
264 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
266 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
267 attributes from DECL_ATTRIBUTES (decl) without affecting
268 DECL_ATTRIBUTES (current_function_decl).
269 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
270 functions with non-NULL DECL_ABSTRACT_ORIGIN.
272 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
274 PR tree-optimization/83979
275 * fold-const.c (fold_comparison): Use constant_boolean_node
276 instead of boolean_{true,false}_node.
278 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
280 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
283 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
285 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
286 Simplify the clause that sets the length attribute.
287 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
288 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
289 clause that sets the length attribute.
290 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
292 2018-01-24 Tom de Vries <tom@codesourcery.com>
295 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
296 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
297 Add strict parameter.
298 (prevent_branch_around_nothing): Insert dummy insn between branch to
299 label and label with no ptx insn inbetween.
300 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
302 2018-01-24 Tom de Vries <tom@codesourcery.com>
305 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
306 for neutered threads in warp.
307 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
309 2018-01-24 Richard Biener <rguenther@suse.de>
311 PR tree-optimization/83176
312 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
315 2018-01-24 Richard Biener <rguenther@suse.de>
317 PR tree-optimization/82819
318 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
319 code generating pluses that are no-ops in the target precision.
321 2018-01-24 Richard Biener <rguenther@suse.de>
324 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
326 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
328 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
329 to merge probabilities.
330 * predict.c (probably_never_executed): Also mark as cold functions
331 with global 0 profile and guessed local profile.
332 * profile-count.c (profile_probability::combine_with_count): New
334 * profile-count.h (profile_probability::operator*,
335 profile_probability::operator*=, profile_probability::operator/,
336 profile_probability::operator/=): Reduce precision to adjusted
337 and set value to guessed on contradictory divisions.
338 (profile_probability::combine_with_freq): Remove.
339 (profile_probability::combine_wiht_count): Declare.
340 (profile_count::force_nonzero):: Set to adjusted.
341 (profile_count::probability_in):: Set quality to adjusted.
342 * tree-ssa-tail-merge.c (replace_block_by): Use
345 2018-01-23 Andrew Waterman <andrew@sifive.com>
346 Jim Wilson <jimw@sifive.com>
348 * config/riscv/riscv.c (riscv_stack_boundary): New.
349 (riscv_option_override): Set riscv_stack_boundary. Handle
350 riscv_preferred_stack_boundary_arg.
351 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
352 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
353 (STACK_BOUNDARY): Set to riscv_stack_boundary.
354 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
355 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
356 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
358 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
361 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
362 of struct ix86_frame.
363 (ix86_expand_epilogue): Likewise. Add a local variable for
364 the reg_save_offset field in struct ix86_frame.
366 2018-01-23 Bin Cheng <bin.cheng@arm.com>
368 PR tree-optimization/82604
369 * tree-loop-distribution.c (enum partition_kind): New enum item
370 PKIND_PARTIAL_MEMSET.
371 (partition_builtin_p): Support above new enum item.
372 (generate_code_for_partition): Ditto.
373 (compute_access_range): Differentiate cases that equality can be
374 proven at all loops, the innermost loops or no loops.
375 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
376 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
377 (finalize_partitions, distribute_loop): Don't fuse partition of
378 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
379 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
382 2018-01-23 Martin Liska <mliska@suse.cz>
384 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
385 order to ignore the predictor.
386 (PRED_POLYMORPHIC_CALL): Likewise.
387 (PRED_RECURSIVE_CALL): Likewise.
389 2018-01-23 Martin Liska <mliska@suse.cz>
391 * tree-profile.c (tree_profiling): Print function header to
392 aware reader which function we are working on.
393 * value-prof.c (gimple_find_values_to_profile): Do not print
394 not interesting value histograms.
396 2018-01-23 Martin Liska <mliska@suse.cz>
398 * profile-count.h (enum profile_quality): Add
399 profile_uninitialized as the first value. Do not number values
400 as they are zero based.
401 (profile_count::verify): Update sanity check.
402 (profile_probability::verify): Likewise.
404 2018-01-23 Nathan Sidwell <nathan@acm.org>
406 * doc/invoke.texi (ffor-scope): Deprecate.
408 2018-01-23 David Malcolm <dmalcolm@redhat.com>
410 PR tree-optimization/83510
411 * domwalk.c (set_all_edges_as_executable): New function.
412 (dom_walker::dom_walker): Convert bool param
413 "skip_unreachable_blocks" to enum reachability. Move setup of
414 edge flags to set_all_edges_as_executable and only do it when
415 reachability is REACHABLE_BLOCKS.
416 * domwalk.h (enum dom_walker::reachability): New enum.
417 (dom_walker::dom_walker): Convert bool param
418 "skip_unreachable_blocks" to enum reachability.
419 (set_all_edges_as_executable): New decl.
420 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
421 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
423 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
424 but converting true to REACHABLE_BLOCKS.
425 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
427 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
428 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
429 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
431 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
432 if check_all_array_refs will be called.
434 2018-01-23 David Malcolm <dmalcolm@redhat.com>
436 * tree.c (selftest::test_location_wrappers): Add more test
439 2018-01-23 David Malcolm <dmalcolm@redhat.com>
441 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
442 (selftest::test_bit_in_range): Likewise.
444 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
447 * doc/sourcebuild.texi (vect_float): Say that the selector
448 only describes the situation when -funsafe-math-optimizations is on.
449 (vect_float_strict): Document.
451 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
453 PR tree-optimization/83965
454 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
455 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
456 instead of checking only for a reduction.
457 (vect_recog_widen_sum_pattern): Likewise.
459 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
461 * predict.c (probably_never_executed): Only use precise profile info.
462 (compute_function_frequency): Skip after inlining hack since we now
463 have quality checking.
465 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
467 * profile-count.h (profile_probability::very_unlikely,
468 profile_probability::unlikely, profile_probability::even): Set
469 precision to guessed.
471 2018-01-23 Richard Biener <rguenther@suse.de>
473 PR tree-optimization/83963
474 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
475 Properly terminate dominator walk when crossing the exit edge not
476 when visiting its source block.
478 2018-01-23 Jakub Jelinek <jakub@redhat.com>
481 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
482 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
484 2018-01-22 Jakub Jelinek <jakub@redhat.com>
486 PR tree-optimization/83957
487 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
488 semicolon after for body surrounded by braces.
490 PR tree-optimization/83081
491 * profile-count.h (profile_probability::split): New method.
492 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
493 Use profile_probability::split.
494 (do_compare_rtx_and_jump): Fix adjustment of probabilities
495 when splitting a single conditional jump into 2.
497 2018-01-22 David Malcolm <dmalcolm@redhat.com>
499 PR tree-optimization/69452
500 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
503 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
505 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
506 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
507 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
509 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
511 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
512 * config/rl78/rl78.md: New define_expand "movdi"
513 * config/rl78/rl78.c: New function definition rl78_split_movdi
515 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
518 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
520 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
521 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
522 128-bit to produce an UNSPEC move to get the double word with the
523 signbit and then a shift directly to do signbit.
524 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
525 implementation with a new version that just does either a direct
526 move or a regular move. Move memory interface to separate insns.
527 Move insns so they are next to the expander.
528 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
529 with signbit move. Split big and little endian case.
530 (signbit<mode>2_dm_mem_le): Likewise.
531 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
532 (signbit<mode>2_dm2): Likewise.
534 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
536 * config/rl78/rl78.md: New define_expand "anddi3".
538 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
540 * config/rl78/rl78.md: New define_expand "umindi3".
542 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
544 * config/rl78/rl78.md: New define_expand "smindi3".
546 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
548 * config/rl78/rl78.md: New define_expand "smaxdi3".
550 2018-01-22 Carl Love <cel@us.ibm.com>
552 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
553 LVX_V1TI): Add macro expansion.
554 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
555 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
556 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
557 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
558 Change check to determine if the instruction is a byte reversing
559 entry. Fix typo in comment.
560 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
561 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
562 Add def_builtin calls for new builtins.
563 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
564 Add define_insn expansion.
566 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
568 * config/rl78/rl78.md: New define_expand "umaxdi3".
570 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
572 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
573 for non-QImode registers
575 2018-01-22 Richard Biener <rguenther@suse.de>
577 PR tree-optimization/83963
578 * graphite-scop-detection.c (scop_detection::get_sese): Delay
579 including the loop exit block.
580 (scop_detection::merge_sese): Likewise.
581 (scop_detection::add_scop): Do it here instead.
583 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
585 * doc/sourcebuild.texi (arm_softfloat): Document.
587 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
590 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
591 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
592 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
594 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
595 David Edelsohn <dje.gcc@gmail.com>
598 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
599 Change "crset eq" to "crset 2".
600 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
601 (*call_indirect_aix<mode>_nospec): Likewise.
602 (*call_value_indirect_aix<mode>_nospec): Likewise.
603 (*call_indirect_elfv2<mode>_nospec): Likewise.
604 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
605 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
606 change assembly output from . to $.
607 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
608 (indirect_jump<mode>_nospec): Change assembly output from . to $.
609 (*tablejump<mode>_internal1_nospec): Likewise.
611 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
614 * config/sh/sh_optimize_sett_clrt.cc:
615 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
617 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
619 PR tree-optimization/83940
620 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
621 offset_dt to vect_constant_def rather than vect_unknown_def_type.
622 (vect_check_load_store_mask): Add a mask_dt_out parameter and
623 use it to pass back the definition type.
624 (vect_check_store_rhs): Likewise rhs_dt_out.
625 (vect_build_gather_load_calls): Add a mask_dt argument and use
626 it instead of a call to vect_is_simple_use.
627 (vectorizable_store): Update calls to vect_check_load_store_mask
628 and vect_check_store_rhs. Use the dt returned by the latter instead
629 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
630 instead of calls to vect_is_simple_use. Pass the scalar rather
631 than the vector operand to vect_is_simple_use when handling
632 second and subsequent copies of an rhs value.
633 (vectorizable_load): Update calls to vect_check_load_store_mask
634 and vect_build_gather_load_calls. Use the cached mask_dt and
635 gs_info.offset_dt instead of calls to vect_is_simple_use.
637 2018-01-20 Jakub Jelinek <jakub@redhat.com>
640 * tree-emutls.c: Include gimplify.h.
641 (lower_emutls_2): New function.
642 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
643 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
644 it before further processing.
647 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
648 UINTVAL (trueop1) instead of INTVAL (op1).
650 2018-01-19 Jakub Jelinek <jakub@redhat.com>
654 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
655 INCOMING_FRAME_SP_OFFSET if not defined.
656 (scan_trace): Add ENTRY argument. If true and
657 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
658 emit a note to adjust the CFA offset.
659 (create_cfi_notes): Adjust scan_trace callers.
660 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
661 INCOMING_FRAME_SP_OFFSET in the CIE.
662 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
663 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
665 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
666 * doc/tm.texi: Regenerated.
668 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
670 PR rtl-optimization/83147
671 * lra-constraints.c (remove_inheritance_pseudos): Use
672 lra_substitute_pseudo_within_insn.
674 2018-01-19 Tom de Vries <tom@codesourcery.com>
675 Cesar Philippidis <cesar@codesourcery.com>
678 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
680 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
683 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
684 spaces for function labels.
686 2018-01-19 Martin Liska <mliska@suse.cz>
688 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
689 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
690 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
691 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
692 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
693 (PRED_CONST_RETURN): Change from 69 to 65.
694 (PRED_NULL_RETURN): Change from 91 to 71.
695 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
696 (PRED_LOOP_GUARD): Change from 66 to 73.
698 2018-01-19 Martin Liska <mliska@suse.cz>
700 * predict.c (predict_insn_def): Add new assert.
701 (struct branch_predictor): Change type to signed integer.
702 (test_prediction_value_range): Amend test to cover
704 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
705 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
706 (PRED_LOOP_ITERATIONS_MAX): Likewise.
707 (PRED_LOOP_IV_COMPARE): Likewise.
708 * predict.h (PROB_UNINITIALIZED): Define new constant.
710 2018-01-19 Martin Liska <mliska@suse.cz>
712 * predict.c (dump_prediction): Add new format for
713 analyze_brprob.py script which is enabled with -details
715 * profile-count.h (precise_p): New function.
717 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
719 PR tree-optimization/83922
720 * tree-vect-loop.c (vect_verify_full_masking): Return false if
721 there are no statements that need masking.
722 (vect_active_double_reduction_p): New function.
723 (vect_analyze_loop_operations): Use it when handling phis that
724 are not in the loop header.
726 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
728 PR tree-optimization/83914
729 * tree-vect-loop.c (vectorizable_induction): Don't convert
730 init_expr or apply the peeling adjustment for inductions
731 that are nested within the vectorized loop.
733 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
735 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
738 2018-01-18 Jakub Jelinek <jakub@redhat.com>
742 * function.h (gimplify_parameters): Add gimple_seq * argument.
743 * function.c: Include gimple.h and options.h.
744 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
745 for the added local temporaries if needed.
746 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
747 if there are any parameter cleanups, wrap whole body into a
748 try/finally with the cleanups.
750 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
753 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
754 Use GET_MODE_CLASS for scalar floating point.
756 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
760 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
761 Fix call of call_cgraph_insertion_hooks.
763 2018-01-18 Martin Sebor <msebor@redhat.com>
765 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
767 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
770 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
773 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
776 * common.opt: (-ffile-prefix-map): New option.
777 * opts.c (common_handle_option): Defer it.
778 * opts-global.c (handle_common_deferred_options): Handle it.
779 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
780 * file-prefix-map.h: New file.
781 (remap_debug_filename, add_debug_prefix_map): ...here.
782 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
783 * final.c (debug_prefix_map, add_debug_prefix_map
784 remap_debug_filename): Move to...
785 * file-prefix-map.c: New file.
786 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
787 generalize, get rid of alloca(), use strrchr() instead of strchr().
788 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
789 Implement in terms of add_prefix_map().
790 (remap_macro_filename, remap_debug_filename): Implement in term of
792 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
793 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
794 * dbxout.c: Include file-prefix-map.h.
795 * varasm.c: Likewise.
796 * vmsdbgout.c: Likewise.
797 * xcoffout.c: Likewise.
798 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
799 * doc/cppopts.texi (-fmacro-prefix-map): Document.
800 * doc/invoke.texi (-ffile-prefix-map): Document.
801 (-fdebug-prefix-map): Update description.
803 2018-01-18 Martin Liska <mliska@suse.cz>
805 * config/i386/i386.c (indirect_thunk_name): Document that also
807 (output_indirect_thunk): Document why both instructions
808 (pause and lfence) are generated.
810 2018-01-18 Richard Biener <rguenther@suse.de>
812 PR tree-optimization/83887
813 * graphite-scop-detection.c
814 (scop_detection::get_nearest_dom_with_single_entry): Remove.
815 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
816 (scop_detection::merge_sese): Re-implement with a flood-fill
817 algorithm that properly finds a SESE region if it exists.
819 2018-01-18 Jakub Jelinek <jakub@redhat.com>
822 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
823 pointer_diff optimizations use view_convert instead of convert.
825 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
827 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
828 Generate different code for -mno-speculate-indirect-jumps.
829 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
830 (*call_indirect_aix<mode>): Disable for
831 -mno-speculate-indirect-jumps.
832 (*call_indirect_aix<mode>_nospec): New define_insn.
833 (*call_value_indirect_aix<mode>): Disable for
834 -mno-speculate-indirect-jumps.
835 (*call_value_indirect_aix<mode>_nospec): New define_insn.
836 (*sibcall_nonlocal_sysv<mode>): Generate different code for
837 -mno-speculate-indirect-jumps.
838 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
840 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
842 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
843 long double type, set the flags for noting the default long double
844 type, even if we don't pass or return a long double type.
846 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
849 * ipa-inline.c (flatten_function): Do not overwrite final inlining
852 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
854 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
855 support for merge[hl].
856 (fold_mergehl_helper): New helper function.
857 (tree-vector-builder.h): New #include for tree_vector_builder usage.
858 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
859 (altivec_vmrglw_direct): Add xxmrglw insn.
861 2018-01-17 Andrew Waterman <andrew@sifive.com>
863 * config/riscv/riscv.c (riscv_conditional_register_usage): If
864 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
866 2018-01-17 David Malcolm <dmalcolm@redhat.com>
869 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
870 call the lto_location_cache before reading the
871 DECL_SOURCE_LOCATION of the types.
873 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
874 Richard Sandiford <richard.sandiford@linaro.org>
876 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
877 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
878 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
879 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
880 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
882 * config/aarch64/constraints.md (aarch64_movti_operand):
884 * config/aarch64/predicates.md (Uti): Add new constraint.
886 2018-01-17 Carl Love <cel@us.ibm.com>
887 * config/rs6000/vsx.md (define_expand xl_len_r,
888 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
889 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
891 (define_expand, define_insn): Move the shift left from the
892 define_insn to the define_expand for lxvl and stxvl instructions.
893 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
894 and XL_LEN_R definitions to PURE.
896 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
898 * config/i386/i386.c (indirect_thunk_name): Declare regno
899 as unsigned int. Compare regno with INVALID_REGNUM.
900 (output_indirect_thunk): Ditto.
901 (output_indirect_thunk_function): Ditto.
902 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
903 in the call to output_indirect_thunk_function.
905 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
908 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
909 rather than the size of inner_type to determine the stack slot size
910 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
912 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
915 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
918 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
920 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
921 endian Linux systems to optionally enable multilibs for selecting
922 the long double type if the user configured an explicit type.
923 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
924 have no long double multilibs if not defined.
925 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
926 warn if the user used -mabi={ieee,ibm}longdouble and we built
927 multilibs for long double.
928 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
929 appropriate multilib option.
930 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
932 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
933 for building long double multilibs.
934 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
936 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
938 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
941 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
943 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
946 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
949 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
952 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
954 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
955 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
957 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
959 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
960 different rtl trees depending on TARGET_64BIT.
961 (rs6000_gen_lvx): Likewise.
963 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
965 * config/visium/visium.md (nop): Tweak comment.
966 (hazard_nop): Likewise.
968 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
970 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
971 -mspeculate-indirect-jumps.
972 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
973 for -mno-speculate-indirect-jumps.
974 (*call_indirect_elfv2<mode>_nospec): New define_insn.
975 (*call_value_indirect_elfv2<mode>): Disable for
976 -mno-speculate-indirect-jumps.
977 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
978 (indirect_jump): Emit different RTL for
979 -mno-speculate-indirect-jumps.
980 (*indirect_jump<mode>): Disable for
981 -mno-speculate-indirect-jumps.
982 (*indirect_jump<mode>_nospec): New define_insn.
983 (tablejump): Emit different RTL for
984 -mno-speculate-indirect-jumps.
985 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
986 (tablejumpsi_nospec): New define_expand.
987 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
988 (tablejumpdi_nospec): New define_expand.
989 (*tablejump<mode>_internal1): Disable for
990 -mno-speculate-indirect-jumps.
991 (*tablejump<mode>_internal1_nospec): New define_insn.
992 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
995 2018-01-16 Artyom Skrobov tyomitch@gmail.com
997 * caller-save.c (insert_save): Drop unnecessary parameter. All
1000 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1001 Richard Biener <rguenth@suse.de>
1004 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1005 return early, inline manually is_gimple_sizepos. Make sure if we
1006 call gimplify_expr we don't end up with a gimple constant.
1007 * tree.c (variably_modified_type_p): Don't return true for
1008 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1009 * gimplify.h (is_gimple_sizepos): Remove.
1011 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1013 PR tree-optimization/83857
1014 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1015 vectorizable_live_operation for pure SLP statements.
1016 (vectorizable_live_operation): Handle PHIs.
1018 2018-01-16 Richard Biener <rguenther@suse.de>
1020 PR tree-optimization/83867
1021 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1022 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1024 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1027 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1028 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1029 If off is not INTEGER_CST, issue a may not be aligned warning
1030 rather than isn't aligned. Use isn%'t rather than isn't.
1031 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1033 <case MULT_EXPR>: Improve the case when bottom and one of the
1034 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1035 operand, in that case check if the other operand is multiple of
1036 bottom divided by the INTEGER_CST operand.
1038 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1041 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1042 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1043 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1044 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1045 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1046 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1047 (pa_function_arg_size): New function.
1049 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1051 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1052 in a separate statement.
1054 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1056 PR tree-optimization/83847
1057 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1058 group gathers and scatters.
1060 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1062 PR rtl-optimization/86620
1063 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1065 PR rtl-optimization/83213
1066 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1067 to last if both are JUMP_INSNs.
1069 PR tree-optimization/83843
1070 * gimple-ssa-store-merging.c
1071 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1072 store_immediate_info for bswap/nop orig_stores.
1074 2018-01-15 Andrew Waterman <andrew@sifive.com>
1076 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1078 <UDIV>: Increase cost if !TARGET_DIV.
1080 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1082 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1083 (define_attr "cr_logical_3op"): New.
1084 (cceq_ior_compare): Adjust.
1085 (cceq_ior_compare_complement): Adjust.
1086 (*cceq_rev_compare): Adjust.
1087 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1088 (is_cracked_insn): Adjust.
1089 (insn_must_be_first_in_group): Adjust.
1090 * config/rs6000/40x.md: Adjust.
1091 * config/rs6000/440.md: Adjust.
1092 * config/rs6000/476.md: Adjust.
1093 * config/rs6000/601.md: Adjust.
1094 * config/rs6000/603.md: Adjust.
1095 * config/rs6000/6xx.md: Adjust.
1096 * config/rs6000/7450.md: Adjust.
1097 * config/rs6000/7xx.md: Adjust.
1098 * config/rs6000/8540.md: Adjust.
1099 * config/rs6000/cell.md: Adjust.
1100 * config/rs6000/e300c2c3.md: Adjust.
1101 * config/rs6000/e500mc.md: Adjust.
1102 * config/rs6000/e500mc64.md: Adjust.
1103 * config/rs6000/e5500.md: Adjust.
1104 * config/rs6000/e6500.md: Adjust.
1105 * config/rs6000/mpc.md: Adjust.
1106 * config/rs6000/power4.md: Adjust.
1107 * config/rs6000/power5.md: Adjust.
1108 * config/rs6000/power6.md: Adjust.
1109 * config/rs6000/power7.md: Adjust.
1110 * config/rs6000/power8.md: Adjust.
1111 * config/rs6000/power9.md: Adjust.
1112 * config/rs6000/rs64.md: Adjust.
1113 * config/rs6000/titan.md: Adjust.
1115 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1117 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1118 ix86_indirect_branch_register logic.
1120 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1122 * config/i386/constraints.md (Bs): Update
1123 ix86_indirect_branch_register check. Don't check
1124 ix86_indirect_branch_register with GOT_memory_operand.
1126 * config/i386/predicates.md (GOT_memory_operand): Don't check
1127 ix86_indirect_branch_register here.
1128 (GOT32_symbol_operand): Likewise.
1130 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1132 * config/i386/predicates.md (constant_call_address_operand):
1133 Rewrite ix86_indirect_branch_register logic.
1134 (sibcall_insn_operand): Likewise.
1136 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1138 * config/i386/constraints.md (Bs): Replace
1139 ix86_indirect_branch_thunk_register with
1140 ix86_indirect_branch_register.
1142 * config/i386/i386.md (indirect_jump): Likewise.
1143 (tablejump): Likewise.
1144 (*sibcall_memory): Likewise.
1145 (*sibcall_value_memory): Likewise.
1146 Peepholes of indirect call and jump via memory: Likewise.
1147 * config/i386/i386.opt: Likewise.
1148 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1149 (GOT_memory_operand): Likewise.
1150 (call_insn_operand): Likewise.
1151 (sibcall_insn_operand): Likewise.
1152 (GOT32_symbol_operand): Likewise.
1154 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1157 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1158 type rather than type addr's type points to.
1159 (expand_omp_atomic_mutex): Likewise.
1160 (expand_omp_atomic): Likewise.
1162 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1165 * config/i386/i386.c (output_indirect_thunk_function): Use
1166 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1167 for __x86_return_thunk.
1169 2018-01-15 Richard Biener <rguenther@suse.de>
1172 * expmed.c (extract_bit_field_1): Fix typo.
1174 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1177 * config/arm/iterators.md (VF): New mode iterator.
1178 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1179 Remove integer-related logic from pattern.
1180 (neon_vabd<mode>_3): Likewise.
1182 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1185 * common.opt (fstrict-overflow): No longer an alias.
1186 (fwrapv-pointer): New option.
1187 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1188 also for pointer types based on flag_wrapv_pointer.
1189 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1190 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1191 opts->x_flag_wrapv got set.
1192 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1193 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1194 POINTER_TYPE_OVERFLOW_UNDEFINED.
1195 * match.pd: Likewise in address comparison pattern.
1196 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1198 2018-01-15 Richard Biener <rguenther@suse.de>
1201 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1202 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1203 Reset type names to their identifier if their TYPE_DECL doesn't
1204 have linkage (and thus is used for ODR and devirt).
1205 (save_debug_info_for_decl): Remove.
1206 (save_debug_info_for_type): Likewise.
1207 (add_tree_to_fld_list): Adjust.
1208 * tree-pretty-print.c (dump_generic_node): Make dumping of
1209 type names more robust.
1211 2018-01-15 Richard Biener <rguenther@suse.de>
1213 * BASE-VER: Bump to 8.0.1.
1215 2018-01-14 Martin Sebor <msebor@redhat.com>
1218 * builtins.c (check_access): Avoid warning when the no-warning bit
1221 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1223 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1224 * ira-color (allocno_hard_regs_compare): Likewise.
1226 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1229 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1230 Use .pushsection/.popsection.
1232 2018-01-14 Martin Sebor <msebor@redhat.com>
1235 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1237 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1239 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1240 entry from extra_headers.
1241 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1242 extra_headers, make the list bitwise identical to the i?86-*-* one.
1244 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1246 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1247 -mcmodel=large with -mindirect-branch=thunk,
1248 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1249 -mfunction-return=thunk-extern.
1250 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1251 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1252 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1254 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1256 * config/i386/i386.c (print_reg): Print the name of the full
1257 integer register without '%'.
1258 (ix86_print_operand): Handle 'V'.
1259 * doc/extend.texi: Document 'V' modifier.
1261 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1263 * config/i386/constraints.md (Bs): Disallow memory operand for
1264 -mindirect-branch-register.
1266 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1267 (GOT_memory_operand): Likewise.
1268 (call_insn_operand): Likewise.
1269 (sibcall_insn_operand): Likewise.
1270 (GOT32_symbol_operand): Likewise.
1271 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1272 for -mindirect-branch-register.
1273 (tablejump): Likewise.
1274 (*sibcall_memory): Likewise.
1275 (*sibcall_value_memory): Likewise.
1276 Disallow peepholes of indirect call and jump via memory for
1277 -mindirect-branch-register.
1278 (*call_pop): Replace m with Bw.
1279 (*call_value_pop): Likewise.
1280 (*sibcall_pop_memory): Replace m with Bs.
1281 * config/i386/i386.opt (mindirect-branch-register): New option.
1282 * doc/invoke.texi: Document -mindirect-branch-register option.
1284 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1286 * config/i386/i386-protos.h (ix86_output_function_return): New.
1287 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1288 set function_return_type.
1289 (indirect_thunk_name): Add ret_p to indicate thunk for function
1291 (output_indirect_thunk_function): Pass false to
1292 indirect_thunk_name.
1293 (ix86_output_indirect_branch_via_reg): Likewise.
1294 (ix86_output_indirect_branch_via_push): Likewise.
1295 (output_indirect_thunk_function): Create alias for function
1296 return thunk if regno < 0.
1297 (ix86_output_function_return): New function.
1298 (ix86_handle_fndecl_attribute): Handle function_return.
1299 (ix86_attribute_table): Add function_return.
1300 * config/i386/i386.h (machine_function): Add
1301 function_return_type.
1302 * config/i386/i386.md (simple_return_internal): Use
1303 ix86_output_function_return.
1304 (simple_return_internal_long): Likewise.
1305 * config/i386/i386.opt (mfunction-return=): New option.
1306 (indirect_branch): Mention -mfunction-return=.
1307 * doc/extend.texi: Document function_return function attribute.
1308 * doc/invoke.texi: Document -mfunction-return= option.
1310 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1312 * config/i386/i386-opts.h (indirect_branch): New.
1313 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1314 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1315 with local indirect jump when converting indirect call and jump.
1316 (ix86_set_indirect_branch_type): New.
1317 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1318 (indirectlabelno): New.
1319 (indirect_thunk_needed): Likewise.
1320 (indirect_thunk_bnd_needed): Likewise.
1321 (indirect_thunks_used): Likewise.
1322 (indirect_thunks_bnd_used): Likewise.
1323 (INDIRECT_LABEL): Likewise.
1324 (indirect_thunk_name): Likewise.
1325 (output_indirect_thunk): Likewise.
1326 (output_indirect_thunk_function): Likewise.
1327 (ix86_output_indirect_branch_via_reg): Likewise.
1328 (ix86_output_indirect_branch_via_push): Likewise.
1329 (ix86_output_indirect_branch): Likewise.
1330 (ix86_output_indirect_jmp): Likewise.
1331 (ix86_code_end): Call output_indirect_thunk_function if needed.
1332 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1334 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1335 (ix86_attribute_table): Add indirect_branch.
1336 * config/i386/i386.h (machine_function): Add indirect_branch_type
1337 and has_local_indirect_jump.
1338 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1340 (tablejump): Likewise.
1341 (*indirect_jump): Use ix86_output_indirect_jmp.
1342 (*tablejump_1): Likewise.
1343 (simple_return_indirect_internal): Likewise.
1344 * config/i386/i386.opt (mindirect-branch=): New option.
1345 (indirect_branch): New.
1348 (thunk-inline): Likewise.
1349 (thunk-extern): Likewise.
1350 * doc/extend.texi: Document indirect_branch function attribute.
1351 * doc/invoke.texi: Document -mindirect-branch= option.
1353 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1356 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1358 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1360 * ipa-inline.c (want_inline_small_function_p): Return false if
1361 inlining has already failed with CIF_FINAL_ERROR.
1362 (update_caller_keys): Call want_inline_small_function_p before
1364 (update_callee_keys): Likewise.
1366 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1368 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1370 (rs6000_quadword_masked_address_p): Likewise.
1371 (quad_aligned_load_p): Likewise.
1372 (quad_aligned_store_p): Likewise.
1373 (const_load_sequence_p): Add comment to describe the outer-most loop.
1374 (mimic_memory_attributes_and_flags): New function.
1375 (rs6000_gen_stvx): Likewise.
1376 (replace_swapped_aligned_store): Likewise.
1377 (rs6000_gen_lvx): Likewise.
1378 (replace_swapped_aligned_load): Likewise.
1379 (replace_swapped_load_constant): Capitalize argument name in
1380 comment describing this function.
1381 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1382 and stores that access quad-word aligned addresses and replace
1383 with stvx or lvx instructions when appropriate.
1384 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1385 New function prototype.
1386 (rs6000_quadword_masked_address_p): Likewise.
1387 (rs6000_gen_lvx): Likewise.
1388 (rs6000_gen_stvx): Likewise.
1389 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1390 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1391 when memory address is aligned.
1392 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1393 this split to select lvx instruction when memory address is aligned.
1394 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1395 instruction when memory address is aligned.
1396 (*vsx_le_perm_load_v16qi): Likewise.
1397 (four unnamed splitters): Modify to select the stvx instruction
1398 when memory is aligned.
1400 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1402 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1403 which appears in the queue multiple times.
1405 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1406 Alan Hayward <alan.hayward@arm.com>
1407 David Sherwood <david.sherwood@arm.com>
1409 * tree-vectorizer.h (vec_lower_bound): New structure.
1410 (_loop_vec_info): Add check_nonzero and lower_bounds.
1411 (LOOP_VINFO_CHECK_NONZERO): New macro.
1412 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1413 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1414 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1415 fields. Make seg_len the distance travelled, not including the
1417 (dr_direction_indicator): Declare.
1418 (dr_zero_step_indicator): Likewise.
1419 (dr_known_forward_stride_p): Likewise.
1420 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1422 (runtime_alias_check_p): Allow runtime alias checks with
1424 (operator ==): Compare access_size and align.
1425 (prune_runtime_alias_test_list): Rework for new distinction between
1426 the access_size and seg_len.
1427 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1429 (get_segment_min_max): New function.
1430 (create_intersect_range_checks): Use it.
1431 (dr_step_indicator): New function.
1432 (dr_direction_indicator): Likewise.
1433 (dr_zero_step_indicator): Likewise.
1434 (dr_known_forward_stride_p): Likewise.
1435 * tree-loop-distribution.c (data_ref_segment_size): Return
1436 DR_STEP * (niters - 1).
1437 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1439 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1440 (vect_preserves_scalar_order_p): New function, split out from...
1441 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1442 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1443 (vect_vfa_access_size): New function.
1444 (vect_vfa_align): Likewise.
1445 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1446 (dump_lower_bound): New function.
1447 (vect_check_lower_bound): Likewise.
1448 (vect_small_gap_p): Likewise.
1449 (vectorizable_with_step_bound_p): Likewise.
1450 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1451 depencies if the vectorization factor is 1. Convert the checks
1452 for nonzero steps into checks on the bounds of DR_STEP. Try using
1453 a bunds check for variable steps if the minimum required step is
1454 relatively small. Update calls to the dr_with_seg_len
1455 constructor and to vect_compile_time_alias.
1456 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1458 (vect_loop_versioning): Call it.
1459 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1461 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1463 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1464 Alan Hayward <alan.hayward@arm.com>
1465 David Sherwood <david.sherwood@arm.com>
1467 * doc/sourcebuild.texi (vect_scatter_store): Document.
1468 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1470 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1472 * genopinit.c (main): Add supports_vec_scatter_store and
1473 supports_vec_scatter_store_cached to target_optabs.
1474 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1475 IFN_MASK_SCATTER_STORE.
1476 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1478 * internal-fn.h (internal_store_fn_p): Declare.
1479 (internal_fn_stored_value_index): Likewise.
1480 * internal-fn.c (scatter_store_direct): New macro.
1481 (expand_scatter_store_optab_fn): New function.
1482 (direct_scatter_store_optab_supported_p): New macro.
1483 (internal_store_fn_p): New function.
1484 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1485 IFN_MASK_SCATTER_STORE.
1486 (internal_fn_mask_index): Likewise.
1487 (internal_fn_stored_value_index): New function.
1488 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1490 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1491 * optabs-query.c (supports_vec_scatter_store_p): New function.
1492 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1493 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1494 true for scatter stores.
1495 (vect_gather_scatter_fn_p): Handle scatter stores too.
1496 (vect_check_gather_scatter): Consider using scatter stores if
1497 supports_vec_scatter_store_p.
1498 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1500 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1501 internal_fn_stored_value_index.
1502 (check_load_store_masking): Handle scatter stores too.
1503 (vect_get_store_rhs): Make public.
1504 (vectorizable_call): Use internal_store_fn_p.
1505 (vectorizable_store): Handle scatter store internal functions.
1506 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1507 when deciding whether the end of the group has been reached.
1508 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1509 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1510 (mask_scatter_store<mode>): New insns.
1512 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1513 Alan Hayward <alan.hayward@arm.com>
1514 David Sherwood <david.sherwood@arm.com>
1516 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1517 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1518 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1520 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1521 Use vect_truncate_gather_scatter_offset if we can't treat the
1522 operation as a normal gather load or scatter store.
1523 (get_group_load_store_type): Take the gather_scatter_info
1524 as argument. Try using a gather load or scatter store for
1525 single-element groups.
1526 (get_load_store_type): Update calls to get_group_load_store_type
1527 and vect_use_strided_gather_scatters_p.
1529 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1530 Alan Hayward <alan.hayward@arm.com>
1531 David Sherwood <david.sherwood@arm.com>
1533 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1534 optional tree argument.
1535 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1537 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1538 but continue to use the current value as a fallback.
1539 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1540 to compare the updates.
1541 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1542 (get_load_store_type): Use it when handling a strided access.
1543 (vect_get_strided_load_store_ops): New function.
1544 (vect_get_data_ptr_increment): Likewise.
1545 (vectorizable_load): Handle strided gather loads. Always pass
1546 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1548 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1549 Alan Hayward <alan.hayward@arm.com>
1550 David Sherwood <david.sherwood@arm.com>
1552 * doc/md.texi (gather_load@var{m}): Document.
1553 (mask_gather_load@var{m}): Likewise.
1554 * genopinit.c (main): Add supports_vec_gather_load and
1555 supports_vec_gather_load_cached to target_optabs.
1556 * optabs-tree.c (init_tree_optimization_optabs): Use
1557 ggc_cleared_alloc to allocate target_optabs.
1558 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1559 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1561 * internal-fn.h (internal_load_fn_p): Declare.
1562 (internal_gather_scatter_fn_p): Likewise.
1563 (internal_fn_mask_index): Likewise.
1564 (internal_gather_scatter_fn_supported_p): Likewise.
1565 * internal-fn.c (gather_load_direct): New macro.
1566 (expand_gather_load_optab_fn): New function.
1567 (direct_gather_load_optab_supported_p): New macro.
1568 (direct_internal_fn_optab): New function.
1569 (internal_load_fn_p): Likewise.
1570 (internal_gather_scatter_fn_p): Likewise.
1571 (internal_fn_mask_index): Likewise.
1572 (internal_gather_scatter_fn_supported_p): Likewise.
1573 * optabs-query.c (supports_at_least_one_mode_p): New function.
1574 (supports_vec_gather_load_p): Likewise.
1575 * optabs-query.h (supports_vec_gather_load_p): Declare.
1576 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1577 and memory_type field.
1578 (NUM_PATTERNS): Bump to 15.
1579 * tree-vect-data-refs.c: Include internal-fn.h.
1580 (vect_gather_scatter_fn_p): New function.
1581 (vect_describe_gather_scatter_call): Likewise.
1582 (vect_check_gather_scatter): Try using internal functions for
1583 gather loads. Recognize existing calls to a gather load function.
1584 (vect_analyze_data_refs): Consider using gather loads if
1585 supports_vec_gather_load_p.
1586 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1587 (vect_get_gather_scatter_offset_type): Likewise.
1588 (vect_convert_mask_for_vectype): Likewise.
1589 (vect_add_conversion_to_patterm): Likewise.
1590 (vect_try_gather_scatter_pattern): Likewise.
1591 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1592 (vect_vect_recog_func_ptrs): Add it.
1593 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1594 internal_fn_mask_index and internal_gather_scatter_fn_p.
1595 (check_load_store_masking): Take the gather_scatter_info as an
1596 argument and handle gather loads.
1597 (vect_get_gather_scatter_ops): New function.
1598 (vectorizable_call): Check internal_load_fn_p.
1599 (vectorizable_load): Likewise. Handle gather load internal
1601 (vectorizable_store): Update call to check_load_store_masking.
1602 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1603 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1604 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1605 (aarch64_gather_scale_operand_d): New predicates.
1606 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1607 (mask_gather_load<mode>): New insns.
1609 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1610 Alan Hayward <alan.hayward@arm.com>
1611 David Sherwood <david.sherwood@arm.com>
1613 * optabs.def (fold_left_plus_optab): New optab.
1614 * doc/md.texi (fold_left_plus_@var{m}): Document.
1615 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1616 * internal-fn.c (fold_left_direct): Define.
1617 (expand_fold_left_optab_fn): Likewise.
1618 (direct_fold_left_optab_supported_p): Likewise.
1619 * fold-const-call.c (fold_const_fold_left): New function.
1620 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1621 * tree-parloops.c (valid_reduction_p): New function.
1622 (gather_scalar_reductions): Use it.
1623 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1624 (vect_finish_replace_stmt): Declare.
1625 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1626 (needs_fold_left_reduction_p): New function, split out from...
1627 (vect_is_simple_reduction): ...here. Accept reductions that
1628 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1629 (vect_force_simple_reduction): Also store the reduction type in
1630 the assignment's STMT_VINFO_REDUC_TYPE.
1631 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1632 (merge_with_identity): New function.
1633 (vect_expand_fold_left): Likewise.
1634 (vectorize_fold_left_reduction): Likewise.
1635 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1636 scalar phi in place for it. Check for target support and reject
1637 cases that would reassociate the operation. Defer the transform
1638 phase to vectorize_fold_left_reduction.
1639 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1640 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1641 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1643 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1645 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1646 call to ifc_temp_var.
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 * target.def (legitimize_address_displacement): Take the original
1653 offset as a poly_int.
1654 * targhooks.h (default_legitimize_address_displacement): Update
1656 * targhooks.c (default_legitimize_address_displacement): Likewise.
1657 * doc/tm.texi: Regenerate.
1658 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1659 as an argument, moving assert of ad->disp == ad->disp_term to...
1660 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1661 Try calling targetm.legitimize_address_displacement before expanding
1662 the address rather than afterwards, and adjust for the new interface.
1663 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1664 Match the new hook interface. Handle SVE addresses.
1665 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1668 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1670 * Makefile.in (OBJS): Add early-remat.o.
1671 * target.def (select_early_remat_modes): New hook.
1672 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1673 * doc/tm.texi: Regenerate.
1674 * targhooks.h (default_select_early_remat_modes): Declare.
1675 * targhooks.c (default_select_early_remat_modes): New function.
1676 * timevar.def (TV_EARLY_REMAT): New timevar.
1677 * passes.def (pass_early_remat): New pass.
1678 * tree-pass.h (make_pass_early_remat): Declare.
1679 * early-remat.c: New file.
1680 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1682 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1684 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1685 Alan Hayward <alan.hayward@arm.com>
1686 David Sherwood <david.sherwood@arm.com>
1688 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1689 vfm1 with a bound_epilog parameter.
1690 (vect_do_peeling): Update calls accordingly, and move the prologue
1691 call earlier in the function. Treat the base bound_epilog as 0 for
1692 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1693 this base when peeling for gaps.
1694 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1695 with fully-masked loops.
1696 (vect_estimate_min_profitable_iters): Handle the single peeled
1697 iteration in that case.
1699 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1700 Alan Hayward <alan.hayward@arm.com>
1701 David Sherwood <david.sherwood@arm.com>
1703 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1704 single-element interleaving even if the size is not a power of 2.
1705 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1706 accesses for single-element interleaving if the group size is
1709 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1710 Alan Hayward <alan.hayward@arm.com>
1711 David Sherwood <david.sherwood@arm.com>
1713 * doc/md.texi (fold_extract_last_@var{m}): Document.
1714 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1715 * optabs.def (fold_extract_last_optab): New optab.
1716 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1717 * internal-fn.c (fold_extract_direct): New macro.
1718 (expand_fold_extract_optab_fn): Likewise.
1719 (direct_fold_extract_optab_supported_p): Likewise.
1720 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1721 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1722 EXTRACT_LAST_REDUCTION.
1723 (get_initial_def_for_reduction): Do not create an initial vector
1724 for EXTRACT_LAST_REDUCTION reductions.
1725 (vectorizable_reduction): Leave the scalar phi in place for
1726 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1727 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1728 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1729 transform phase to vectorizable_condition.
1730 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1732 (vect_finish_stmt_generation): ...here.
1733 (vect_finish_replace_stmt): New function.
1734 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1735 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1737 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1739 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1740 Alan Hayward <alan.hayward@arm.com>
1741 David Sherwood <david.sherwood@arm.com>
1743 * doc/md.texi (extract_last_@var{m}): Document.
1744 * optabs.def (extract_last_optab): New optab.
1745 * internal-fn.def (EXTRACT_LAST): New internal function.
1746 * internal-fn.c (cond_unary_direct): New macro.
1747 (expand_cond_unary_optab_fn): Likewise.
1748 (direct_cond_unary_optab_supported_p): Likewise.
1749 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1750 loops using EXTRACT_LAST.
1751 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1752 (extract_last_<mode>): ...this optab.
1753 (vec_extract<mode><Vel>): Update accordingly.
1755 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1756 Alan Hayward <alan.hayward@arm.com>
1757 David Sherwood <david.sherwood@arm.com>
1759 * target.def (empty_mask_is_expensive): New hook.
1760 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
1761 * doc/tm.texi: Regenerate.
1762 * targhooks.h (default_empty_mask_is_expensive): Declare.
1763 * targhooks.c (default_empty_mask_is_expensive): New function.
1764 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
1765 if the target says that empty masks are expensive.
1766 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
1768 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
1770 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1771 Alan Hayward <alan.hayward@arm.com>
1772 David Sherwood <david.sherwood@arm.com>
1774 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
1775 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
1776 (vect_use_loop_mask_for_alignment_p): New function.
1777 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
1778 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
1779 niters_skip argument. Make sure that the first niters_skip elements
1780 of the first iteration are inactive.
1781 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
1782 Update call to vect_set_loop_masks_directly.
1783 (get_misalign_in_elems): New function, split out from...
1784 (vect_gen_prolog_loop_niters): ...here.
1785 (vect_update_init_of_dr): Take a code argument that specifies whether
1786 the adjustment should be added or subtracted.
1787 (vect_update_init_of_drs): Likewise.
1788 (vect_prepare_for_masked_peels): New function.
1789 (vect_do_peeling): Skip prologue peeling if we're using a mask
1790 instead. Update call to vect_update_inits_of_drs.
1791 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1793 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
1794 alignment. Do not include the number of peeled iterations in
1795 the minimum threshold in that case.
1796 (vectorizable_induction): Adjust the start value down by
1797 LOOP_VINFO_MASK_SKIP_NITERS iterations.
1798 (vect_transform_loop): Call vect_prepare_for_masked_peels.
1799 Take the number of skipped iterations into account when calculating
1801 * tree-vect-stmts.c (vect_gen_while_not): New function.
1803 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1804 Alan Hayward <alan.hayward@arm.com>
1805 David Sherwood <david.sherwood@arm.com>
1807 * doc/sourcebuild.texi (vect_fully_masked): Document.
1808 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
1810 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
1812 (vect_analyze_loop_2): ...here. Don't check the vectorization
1813 factor against the number of loop iterations if the loop is
1816 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1817 Alan Hayward <alan.hayward@arm.com>
1818 David Sherwood <david.sherwood@arm.com>
1820 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
1821 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
1822 (dump_groups): Update accordingly.
1823 (iv_use::mem_type): New member variable.
1824 (address_p): New function.
1825 (record_use): Add a mem_type argument and initialize the new
1827 (record_group_use): Add a mem_type argument. Use address_p.
1828 Remove obsolete null checks of base_object. Update call to record_use.
1829 (find_interesting_uses_op): Update call to record_group_use.
1830 (find_interesting_uses_cond): Likewise.
1831 (find_interesting_uses_address): Likewise.
1832 (get_mem_type_for_internal_fn): New function.
1833 (find_address_like_use): Likewise.
1834 (find_interesting_uses_stmt): Try find_address_like_use before
1835 calling find_interesting_uses_op.
1836 (addr_offset_valid_p): Use the iv mem_type field as the type
1837 of the addressed memory.
1838 (add_autoinc_candidates): Likewise.
1839 (get_address_cost): Likewise.
1840 (split_small_address_groups_p): Use address_p.
1841 (split_address_groups): Likewise.
1842 (add_iv_candidate_for_use): Likewise.
1843 (autoinc_possible_for_pair): Likewise.
1844 (rewrite_groups): Likewise.
1845 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
1846 (determine_group_iv_cost): Update after split of USE_ADDRESS.
1847 (get_alias_ptr_type_for_ptr_address): New function.
1848 (rewrite_use_address): Rewrite address uses in calls that were
1849 identified by find_address_like_use.
1851 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1852 Alan Hayward <alan.hayward@arm.com>
1853 David Sherwood <david.sherwood@arm.com>
1855 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
1857 * gimple-expr.h (is_gimple_addressable: Likewise.
1858 * gimple-expr.c (is_gimple_address): Likewise.
1859 * internal-fn.c (expand_call_mem_ref): New function.
1860 (expand_mask_load_optab_fn): Use it.
1861 (expand_mask_store_optab_fn): Likewise.
1863 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1864 Alan Hayward <alan.hayward@arm.com>
1865 David Sherwood <david.sherwood@arm.com>
1867 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
1868 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
1869 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
1870 (cond_umax@var{mode}): Document.
1871 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
1872 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
1873 (cond_umin_optab, cond_umax_optab): New optabs.
1874 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
1875 (COND_IOR, COND_XOR): New internal functions.
1876 * internal-fn.h (get_conditional_internal_fn): Declare.
1877 * internal-fn.c (cond_binary_direct): New macro.
1878 (expand_cond_binary_optab_fn): Likewise.
1879 (direct_cond_binary_optab_supported_p): Likewise.
1880 (get_conditional_internal_fn): New function.
1881 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
1882 Cope with reduction statements that are vectorized as calls rather
1884 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
1885 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
1886 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
1887 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
1888 (UNSPEC_COND_EOR): New unspecs.
1889 (optab): Add mappings for them.
1890 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
1891 (sve_int_op, sve_fp_op): New int attributes.
1893 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1894 Alan Hayward <alan.hayward@arm.com>
1895 David Sherwood <david.sherwood@arm.com>
1897 * optabs.def (while_ult_optab): New optab.
1898 * doc/md.texi (while_ult@var{m}@var{n}): Document.
1899 * internal-fn.def (WHILE_ULT): New internal function.
1900 * internal-fn.h (direct_internal_fn_supported_p): New override
1901 that takes two types as argument.
1902 * internal-fn.c (while_direct): New macro.
1903 (expand_while_optab_fn): New function.
1904 (convert_optab_supported_p): Likewise.
1905 (direct_while_optab_supported_p): New macro.
1906 * wide-int.h (wi::udiv_ceil): New function.
1907 * tree-vectorizer.h (rgroup_masks): New structure.
1908 (vec_loop_masks): New typedef.
1909 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
1911 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
1912 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
1913 (vect_max_vf): New function.
1914 (slpeel_make_loop_iterate_ntimes): Delete.
1915 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
1916 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
1917 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
1918 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
1919 internal-fn.h, stor-layout.h and optabs-query.h.
1920 (vect_set_loop_mask): New function.
1921 (add_preheader_seq): Likewise.
1922 (add_header_seq): Likewise.
1923 (interleave_supported_p): Likewise.
1924 (vect_maybe_permute_loop_masks): Likewise.
1925 (vect_set_loop_masks_directly): Likewise.
1926 (vect_set_loop_condition_masked): Likewise.
1927 (vect_set_loop_condition_unmasked): New function, split out from
1928 slpeel_make_loop_iterate_ntimes.
1929 (slpeel_make_loop_iterate_ntimes): Rename to..
1930 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
1931 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
1932 (vect_do_peeling): Update call accordingly.
1933 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
1935 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1936 mask_compare_type, can_fully_mask_p and fully_masked_p.
1937 (release_vec_loop_masks): New function.
1938 (_loop_vec_info): Use it to free the loop masks.
1939 (can_produce_all_loop_masks_p): New function.
1940 (vect_get_max_nscalars_per_iter): Likewise.
1941 (vect_verify_full_masking): Likewise.
1942 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
1943 retries, and free the mask rgroups before retrying. Check loop-wide
1944 reasons for disallowing fully-masked loops. Make the final decision
1945 about whether use a fully-masked loop or not.
1946 (vect_estimate_min_profitable_iters): Do not assume that peeling
1947 for the number of iterations will be needed for fully-masked loops.
1948 (vectorizable_reduction): Disable fully-masked loops.
1949 (vectorizable_live_operation): Likewise.
1950 (vect_halve_mask_nunits): New function.
1951 (vect_double_mask_nunits): Likewise.
1952 (vect_record_loop_mask): Likewise.
1953 (vect_get_loop_mask): Likewise.
1954 (vect_transform_loop): Handle the case in which the final loop
1955 iteration might handle a partial vector. Call vect_set_loop_condition
1956 instead of slpeel_make_loop_iterate_ntimes.
1957 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
1958 (check_load_store_masking): New function.
1959 (prepare_load_store_mask): Likewise.
1960 (vectorizable_store): Handle fully-masked loops.
1961 (vectorizable_load): Likewise.
1962 (supportable_widening_operation): Use vect_halve_mask_nunits for
1964 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
1965 (vect_gen_while): New function.
1966 * config/aarch64/aarch64.md (umax<mode>3): New expander.
1967 (aarch64_uqdec<mode>): New insn.
1969 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1970 Alan Hayward <alan.hayward@arm.com>
1971 David Sherwood <david.sherwood@arm.com>
1973 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
1974 (reduc_xor_scal_optab): New optabs.
1975 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
1976 (reduc_xor_scal_@var{m}): Document.
1977 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
1978 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
1980 * fold-const-call.c (fold_const_call): Handle them.
1981 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
1982 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
1983 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
1984 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
1985 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
1986 (UNSPEC_XORV): New unspecs.
1987 (optab): Add entries for them.
1988 (BITWISEV): New int iterator.
1989 (bit_reduc_op): New int attributes.
1991 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1992 Alan Hayward <alan.hayward@arm.com>
1993 David Sherwood <david.sherwood@arm.com>
1995 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
1996 * internal-fn.def (VEC_SHL_INSERT): New internal function.
1997 * optabs.def (vec_shl_insert_optab): New optab.
1998 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
1999 (duplicate_and_interleave): Likewise.
2000 * tree-vect-loop.c: Include internal-fn.h.
2001 (neutral_op_for_slp_reduction): New function, split out from
2002 get_initial_defs_for_reduction.
2003 (get_initial_def_for_reduction): Handle option 2 for variable-length
2004 vectors by loading the neutral value into a vector and then shifting
2005 the initial value into element 0.
2006 (get_initial_defs_for_reduction): Replace the code argument with
2007 the neutral value calculated by neutral_op_for_slp_reduction.
2008 Use gimple_build_vector for constant-length vectors.
2009 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2010 but the first group_size elements have a neutral value.
2011 Use duplicate_and_interleave otherwise.
2012 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2013 Update call to get_initial_defs_for_reduction. Handle SLP
2014 reductions for variable-length vectors by creating one vector
2015 result for each scalar result, with the elements associated
2016 with other scalar results stubbed out with the neutral value.
2017 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2018 Require IFN_VEC_SHL_INSERT for double reductions on
2019 variable-length vectors, or SLP reductions that have
2020 a neutral value. Require can_duplicate_and_interleave_p
2021 support for variable-length unchained SLP reductions if there
2022 is no neutral value, such as for MIN/MAX reductions. Also require
2023 the number of vector elements to be a multiple of the number of
2024 SLP statements when doing variable-length unchained SLP reductions.
2025 Update call to vect_create_epilog_for_reduction.
2026 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2027 and remove initial values.
2028 (duplicate_and_interleave): Make public.
2029 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2030 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2032 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2033 Alan Hayward <alan.hayward@arm.com>
2034 David Sherwood <david.sherwood@arm.com>
2036 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2037 (can_duplicate_and_interleave_p): New function.
2038 (vect_get_and_check_slp_defs): Take the vector of statements
2039 rather than just the current one. Remove excess parentheses.
2040 Restriction rejectinon of vect_constant_def and vect_external_def
2041 for variable-length vectors to boolean types, or types for which
2042 can_duplicate_and_interleave_p is false.
2043 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2044 (duplicate_and_interleave): New function.
2045 (vect_get_constant_vectors): Use gimple_build_vector for
2046 constant-length vectors and suitable variable-length constant
2047 vectors. Use duplicate_and_interleave for other variable-length
2048 vectors. Don't defer the update when inserting new statements.
2050 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2051 Alan Hayward <alan.hayward@arm.com>
2052 David Sherwood <david.sherwood@arm.com>
2054 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2055 min_profitable_iters doesn't go negative.
2057 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2058 Alan Hayward <alan.hayward@arm.com>
2059 David Sherwood <david.sherwood@arm.com>
2061 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2062 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2063 * optabs.def (vec_mask_load_lanes_optab): New optab.
2064 (vec_mask_store_lanes_optab): Likewise.
2065 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2066 (MASK_STORE_LANES): Likewise.
2067 * internal-fn.c (mask_load_lanes_direct): New macro.
2068 (mask_store_lanes_direct): Likewise.
2069 (expand_mask_load_optab_fn): Handle masked operations.
2070 (expand_mask_load_lanes_optab_fn): New macro.
2071 (expand_mask_store_optab_fn): Handle masked operations.
2072 (expand_mask_store_lanes_optab_fn): New macro.
2073 (direct_mask_load_lanes_optab_supported_p): Likewise.
2074 (direct_mask_store_lanes_optab_supported_p): Likewise.
2075 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2077 (vect_load_lanes_supported): Likewise.
2078 * tree-vect-data-refs.c (strip_conversion): New function.
2079 (can_group_stmts_p): Likewise.
2080 (vect_analyze_data_ref_accesses): Use it instead of checking
2081 for a pair of assignments.
2082 (vect_store_lanes_supported): Take a masked_p parameter.
2083 (vect_load_lanes_supported): Likewise.
2084 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2085 vect_store_lanes_supported and vect_load_lanes_supported.
2086 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2087 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2088 parameter. Don't allow gaps for masked accesses.
2089 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2090 and vect_load_lanes_supported.
2091 (get_load_store_type): Take a masked_p parameter and update
2092 call to get_group_load_store_type.
2093 (vectorizable_store): Update call to get_load_store_type.
2094 Handle IFN_MASK_STORE_LANES.
2095 (vectorizable_load): Update call to get_load_store_type.
2096 Handle IFN_MASK_LOAD_LANES.
2098 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2099 Alan Hayward <alan.hayward@arm.com>
2100 David Sherwood <david.sherwood@arm.com>
2102 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2104 * config/aarch64/aarch64-protos.h
2105 (aarch64_sve_struct_memory_operand_p): Declare.
2106 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2107 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2108 (VPRED, vpred): Handle SVE structure modes.
2109 * config/aarch64/constraints.md (Utx): New constraint.
2110 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2111 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2112 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2113 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2114 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2115 structure modes. Split into pieces after RA.
2116 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2117 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2119 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2120 SVE structure modes.
2121 (aarch64_classify_address): Likewise.
2122 (sizetochar): Move earlier in file.
2123 (aarch64_print_operand): Handle SVE register lists.
2124 (aarch64_array_mode): New function.
2125 (aarch64_sve_struct_memory_operand_p): Likewise.
2126 (TARGET_ARRAY_MODE): Redefine.
2128 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2129 Alan Hayward <alan.hayward@arm.com>
2130 David Sherwood <david.sherwood@arm.com>
2132 * target.def (array_mode): New target hook.
2133 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2134 * doc/tm.texi: Regenerate.
2135 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2136 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2137 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2139 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2142 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2143 Alan Hayward <alan.hayward@arm.com>
2144 David Sherwood <david.sherwood@arm.com>
2146 * fold-const.c (fold_binary_loc): Check the argument types
2147 rather than the result type when testing for a vector operation.
2149 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2151 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2152 * doc/tm.texi: Regenerate.
2154 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2155 Alan Hayward <alan.hayward@arm.com>
2156 David Sherwood <david.sherwood@arm.com>
2158 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2159 (sve): Document new AArch64 extension.
2160 * doc/md.texi (w): Extend the description of the AArch64
2161 constraint to include SVE vectors.
2162 (Upl, Upa): Document new AArch64 predicate constraints.
2163 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2165 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2166 (msve-vector-bits=): New option.
2167 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2168 SVE when these are disabled.
2169 (sve): New extension.
2170 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2171 modes. Adjust their number of units based on aarch64_sve_vg.
2172 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2173 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2174 aarch64_addr_query_type.
2175 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2176 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2177 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2178 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2179 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2180 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2181 (aarch64_simd_imm_zero_p): Delete.
2182 (aarch64_check_zero_based_sve_index_immediate): Declare.
2183 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2184 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2185 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2186 (aarch64_sve_float_mul_immediate_p): Likewise.
2187 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2189 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2190 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2191 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2192 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2193 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2194 (aarch64_regmode_natural_size): Likewise.
2195 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2196 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2198 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2199 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2200 for VG and the SVE predicate registers.
2201 (V_ALIASES): Add a "z"-prefixed alias.
2202 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2203 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2204 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2205 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2206 (REG_CLASS_NAMES): Add entries for them.
2207 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2208 and the predicate registers.
2209 (aarch64_sve_vg): Declare.
2210 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2211 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2212 (REGMODE_NATURAL_SIZE): Define.
2213 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2215 * config/aarch64/aarch64.c: Include cfgrtl.h.
2216 (simd_immediate_info): Add a constructor for series vectors,
2217 and an associated step field.
2218 (aarch64_sve_vg): New variable.
2219 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2220 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2221 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2222 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2223 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2224 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2225 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2226 (aarch64_get_mask_mode): New functions.
2227 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2228 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2229 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2230 predicate modes and predicate registers. Explicitly restrict
2231 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2232 to store a vector mode if it is recognized by
2233 aarch64_classify_vector_mode.
2234 (aarch64_regmode_natural_size): New function.
2235 (aarch64_hard_regno_caller_save_mode): Return the original mode
2237 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2238 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2239 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2240 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2242 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2243 does not overlap dest if the function is frame-related. Handle
2245 (aarch64_split_add_offset): New function.
2246 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2247 them aarch64_add_offset.
2248 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2249 and update call to aarch64_sub_sp.
2250 (aarch64_add_cfa_expression): New function.
2251 (aarch64_expand_prologue): Pass extra temporary registers to the
2252 functions above. Handle the case in which we need to emit new
2253 DW_CFA_expressions for registers that were originally saved
2254 relative to the stack pointer, but now have to be expressed
2255 relative to the frame pointer.
2256 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2258 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2259 IP0 and IP1 values for SVE frames.
2260 (aarch64_expand_vec_series): New function.
2261 (aarch64_expand_sve_widened_duplicate): Likewise.
2262 (aarch64_expand_sve_const_vector): Likewise.
2263 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2264 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2265 into the register, rather than emitting a SET directly.
2266 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2267 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2268 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2269 (offset_9bit_signed_scaled_p): New functions.
2270 (aarch64_replicate_bitmask_imm): New function.
2271 (aarch64_bitmask_imm): Use it.
2272 (aarch64_cannot_force_const_mem): Reject expressions involving
2273 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2274 (aarch64_classify_index): Handle SVE indices, by requiring
2275 a plain register index with a scale that matches the element size.
2276 (aarch64_classify_address): Handle SVE addresses. Assert that
2277 the mode of the address is VOIDmode or an integer mode.
2278 Update call to aarch64_classify_symbol.
2279 (aarch64_classify_symbolic_expression): Update call to
2280 aarch64_classify_symbol.
2281 (aarch64_const_vec_all_in_range_p): New function.
2282 (aarch64_print_vector_float_operand): Likewise.
2283 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2284 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2285 and the FP immediates 1.0 and 0.5.
2286 (aarch64_print_address_internal): Handle SVE addresses.
2287 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2288 (aarch64_regno_regclass): Handle predicate registers.
2289 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2291 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2292 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2293 (aarch64_convert_sve_vector_bits): New function.
2294 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2295 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2297 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2298 Handle SVE vector and predicate modes. Accept VL-based constants
2299 that need only one temporary register, and VL offsets that require
2300 no temporary registers.
2301 (aarch64_conditional_register_usage): Mark the predicate registers
2302 as fixed if SVE isn't available.
2303 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2304 Return true for SVE vector and predicate modes.
2305 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2306 rather than an unsigned int. Handle SVE modes.
2307 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2309 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2311 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2312 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2313 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2314 (aarch64_sve_float_mul_immediate_p): New functions.
2315 (aarch64_sve_valid_immediate): New function.
2316 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2317 Explicitly reject structure modes. Check for INDEX constants.
2318 Handle PTRUE and PFALSE constants.
2319 (aarch64_check_zero_based_sve_index_immediate): New function.
2320 (aarch64_simd_imm_zero_p): Delete.
2321 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2322 vector modes. Accept constants in the range of CNT[BHWD].
2323 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2324 ask for an Advanced SIMD mode.
2325 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2326 (aarch64_simd_vector_alignment): Handle SVE predicates.
2327 (aarch64_vectorize_preferred_vector_alignment): New function.
2328 (aarch64_simd_vector_alignment_reachable): Use it instead of
2330 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2331 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2333 (MAX_VECT_LEN): Delete.
2334 (expand_vec_perm_d): Add a vec_flags field.
2335 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2336 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2337 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2339 (aarch64_evpc_rev): Rename to...
2340 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2341 (aarch64_evpc_rev_global): New function.
2342 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2343 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2345 (aarch64_evpc_sve_tbl): New function.
2346 (aarch64_expand_vec_perm_const_1): Update after rename of
2347 aarch64_evpc_rev. Handle SVE permutes too, trying
2348 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2349 than aarch64_evpc_tbl.
2350 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2351 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2352 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2353 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2354 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2355 (aarch64_expand_sve_vcond): New functions.
2356 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2357 of aarch64_vector_mode_p.
2358 (aarch64_dwarf_poly_indeterminate_value): New function.
2359 (aarch64_compute_pressure_classes): Likewise.
2360 (aarch64_can_change_mode_class): Likewise.
2361 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2362 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2363 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2364 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2365 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2366 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2367 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2368 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2370 (Dn, Dl, Dr): Accept const as well as const_vector.
2371 (Dz): Likewise. Compare against CONST0_RTX.
2372 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2373 of "vector" where appropriate.
2374 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2375 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2376 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2377 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2378 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2379 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2380 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2381 (v_int_equiv): Extend to SVE modes.
2382 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2384 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2385 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2386 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2387 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2388 (SVE_COND_FP_CMP): New int iterators.
2389 (perm_hilo): Handle the new unpack unspecs.
2390 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2392 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2393 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2394 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2395 (aarch64_equality_operator, aarch64_constant_vector_operand)
2396 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2397 (aarch64_sve_nonimmediate_operand): Likewise.
2398 (aarch64_sve_general_operand): Likewise.
2399 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2400 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2401 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2402 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2403 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2404 (aarch64_sve_float_arith_immediate): Likewise.
2405 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2406 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2407 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2408 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2409 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2410 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2411 (aarch64_sve_float_arith_operand): Likewise.
2412 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2413 (aarch64_sve_float_mul_operand): Likewise.
2414 (aarch64_sve_vec_perm_operand): Likewise.
2415 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2416 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2417 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2418 as well as const_vector.
2419 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2420 in file. Use CONST0_RTX and CONSTM1_RTX.
2421 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2422 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2423 Use aarch64_simd_imm_zero.
2424 * config/aarch64/aarch64-sve.md: New file.
2425 * config/aarch64/aarch64.md: Include it.
2426 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2427 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2428 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2429 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2430 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2431 (sve): New attribute.
2432 (enabled): Disable instructions with the sve attribute unless
2434 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2435 aarch64_expand_mov_immediate.
2436 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2437 CNT[BHSD] immediates.
2438 (movti): Split CONST_POLY_INT moves into two halves.
2439 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2440 Split additions that need a temporary here if the destination
2441 is the stack pointer.
2442 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2443 (*add<mode>3_poly_1): New instruction.
2444 (set_clobber_cc): New expander.
2446 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2448 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2449 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2450 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2451 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2452 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2453 Change innermode from fixed_mode_size to machine_mode.
2454 (simplify_subreg): Update call accordingly. Handle a constant-sized
2455 subreg of a variable-length CONST_VECTOR.
2457 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2458 Alan Hayward <alan.hayward@arm.com>
2459 David Sherwood <david.sherwood@arm.com>
2461 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2462 (add_offset_to_base): New function, split out from...
2463 (create_mem_ref): ...here. When handling a scale other than 1,
2464 check first whether the address is valid without the offset.
2465 Add it into the base if so, leaving the index and scale as-is.
2467 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2470 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2471 fold_for_warn before checking if arg2 is INTEGER_CST.
2473 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2475 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2476 (store_multiple_operation): Delete.
2477 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2478 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2479 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2480 guarded by TARGET_STRING.
2481 (rs6000_output_load_multiple): Delete.
2482 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2483 OPTION_MASK_STRING / TARGET_STRING handling.
2484 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2485 (const rs6000_opt_masks) <"string">: Change mask to 0.
2486 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2487 (MASK_STRING): Delete.
2488 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2490 (load_multiple): Delete.
2497 (store_multiple): Delete.
2504 (movmemsi_8reg): Delete.
2505 (corresponding unnamed define_insn): Delete.
2506 (movmemsi_6reg): Delete.
2507 (corresponding unnamed define_insn): Delete.
2508 (movmemsi_4reg): Delete.
2509 (corresponding unnamed define_insn): Delete.
2510 (movmemsi_2reg): Delete.
2511 (corresponding unnamed define_insn): Delete.
2512 (movmemsi_1reg): Delete.
2513 (corresponding unnamed define_insn): Delete.
2514 * config/rs6000/rs6000.opt (mno-string): New.
2515 (mstring): Replace by deprecation warning stub.
2516 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2518 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2520 * regrename.c (regrename_do_replace): If replacing the same
2521 reg multiple times, try to reuse last created gen_raw_REG.
2524 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2525 main to workaround a bug in GDB.
2527 2018-01-12 Tom de Vries <tom@codesourcery.com>
2530 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2532 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2534 PR rtl-optimization/80481
2535 * ira-color.c (get_cap_member): New function.
2536 (allocnos_conflict_by_live_ranges_p): Use it.
2537 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2538 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2540 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2543 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2544 (*saddl_se_1): Ditto.
2546 (*ssubl_se_1): Ditto.
2548 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2550 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2551 rather than wi::to_widest for DR_INITs.
2552 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2553 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2554 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2556 (vect_analyze_group_access_1): Note that here.
2558 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2560 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2561 polynomial type sizes.
2563 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2565 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2566 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2567 (gimple_add_tmp_var): Likewise.
2569 2018-01-12 Martin Liska <mliska@suse.cz>
2571 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2572 (gimple_alloc_sizes): Likewise.
2573 (dump_gimple_statistics): Use PRIu64 in printf format.
2574 * gimple.h: Change uint64_t to int.
2576 2018-01-12 Martin Liska <mliska@suse.cz>
2578 * tree-core.h: Use uint64_t instead of int.
2579 * tree.c (tree_node_counts): Likewise.
2580 (tree_node_sizes): Likewise.
2581 (dump_tree_statistics): Use PRIu64 in printf format.
2583 2018-01-12 Martin Liska <mliska@suse.cz>
2585 * Makefile.in: As qsort_chk is implemented in vec.c, add
2586 vec.o to linkage of gencfn-macros.
2587 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2588 passing the info to record_node_allocation_statistics.
2589 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2591 * ggc-common.c (struct ggc_usage): Add operator== and use
2592 it in operator< and compare function.
2593 * mem-stats.h (struct mem_usage): Likewise.
2594 * vec.c (struct vec_usage): Remove operator< and compare
2595 function. Can be simply inherited.
2597 2018-01-12 Martin Jambor <mjambor@suse.cz>
2600 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2601 * tree-ssa-math-opts.c: Include domwalk.h.
2602 (convert_mult_to_fma_1): New function.
2603 (fma_transformation_info): New type.
2604 (fma_deferring_state): Likewise.
2605 (cancel_fma_deferring): New function.
2606 (result_of_phi): Likewise.
2607 (last_fma_candidate_feeds_initial_phi): Likewise.
2608 (convert_mult_to_fma): Added deferring logic, split actual
2609 transformation to convert_mult_to_fma_1.
2610 (math_opts_dom_walker): New type.
2611 (math_opts_dom_walker::after_dom_children): New method, body moved
2612 here from pass_optimize_widening_mul::execute, added deferring logic
2614 (pass_optimize_widening_mul::execute): Moved most of code to
2615 math_opts_dom_walker::after_dom_children.
2616 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2617 * config/i386/i386.c (ix86_option_override_internal): Added
2618 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2620 2018-01-12 Richard Biener <rguenther@suse.de>
2623 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2624 inline instance vars.
2626 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2629 * config/rx/rx.c (rx_is_restricted_memory_address):
2632 2018-01-12 Richard Biener <rguenther@suse.de>
2634 PR tree-optimization/80846
2635 * target.def (split_reduction): New target hook.
2636 * targhooks.c (default_split_reduction): New function.
2637 * targhooks.h (default_split_reduction): Declare.
2638 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2639 target requests first reduce vectors by combining low and high
2641 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2642 (get_vectype_for_scalar_type_and_size): Export.
2643 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2644 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2645 * doc/tm.texi: Regenerate.
2646 * config/i386/i386.c (ix86_split_reduction): Implement
2647 TARGET_VECTORIZE_SPLIT_REDUCTION.
2649 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2652 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2653 in PIC mode except for TARGET_VXWORKS_RTP.
2654 * config/sparc/sparc.c: Include cfgrtl.h.
2655 (TARGET_INIT_PIC_REG): Define.
2656 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2657 (sparc_pic_register_p): New predicate.
2658 (sparc_legitimate_address_p): Use it.
2659 (sparc_legitimize_pic_address): Likewise.
2660 (sparc_delegitimize_address): Likewise.
2661 (sparc_mode_dependent_address_p): Likewise.
2662 (gen_load_pcrel_sym): Remove 4th parameter.
2663 (load_got_register): Adjust call to above. Remove obsolete stuff.
2664 (sparc_expand_prologue): Do not call load_got_register here.
2665 (sparc_flat_expand_prologue): Likewise.
2666 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2667 (sparc_use_pseudo_pic_reg): New function.
2668 (sparc_init_pic_reg): Likewise.
2669 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2670 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2672 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2674 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2675 Add item for branch_cost.
2677 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2679 PR rtl-optimization/83565
2680 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2681 not extend the result to a larger mode for rotate operations.
2682 (num_sign_bit_copies1): Likewise.
2684 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2687 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2689 Use values-Xc.o for -pedantic.
2690 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2692 2018-01-12 Martin Liska <mliska@suse.cz>
2695 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2697 (possible_polymorphic_call_targets): Use it.
2698 (ipa_devirt): Likewise.
2700 2018-01-12 Martin Liska <mliska@suse.cz>
2702 * profile-count.h (enum profile_quality): Use 0 as invalid
2703 enum value of profile_quality.
2705 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2707 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2708 -mext-string options.
2710 2018-01-12 Richard Biener <rguenther@suse.de>
2712 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2713 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2714 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2716 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2718 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2720 * configure.ac (--with-long-double-format): Add support for the
2721 configuration option to change the default long double format on
2723 * config.gcc (powerpc*-linux*-*): Likewise.
2724 * configure: Regenerate.
2725 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2726 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2727 used without modification.
2729 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2731 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2732 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2733 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2734 MISC_BUILTIN_SPEC_BARRIER.
2735 (rs6000_init_builtins): Likewise.
2736 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2738 (speculation_barrier): New define_insn.
2739 * doc/extend.texi: Document __builtin_speculation_barrier.
2741 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2744 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2745 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2746 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2748 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2749 integral modes instead of "ss" and "sd".
2750 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2751 vectors with 32-bit and 64-bit elements.
2752 (vecdupssescalarmodesuffix): New mode attribute.
2753 (vec_dup<mode>): Use it.
2755 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2758 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2759 frame if argument is passed on stack.
2761 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2764 * ree.c (combine_reaching_defs): Optimize also
2765 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
2766 reg2=any_extend(exp); reg1=reg2;, formatting fix.
2768 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2771 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
2773 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2776 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
2777 after they are computed.
2779 2018-01-11 Bin Cheng <bin.cheng@arm.com>
2781 PR tree-optimization/83695
2782 * gimple-loop-linterchange.cc
2783 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
2784 reset cached scev information after interchange.
2785 (pass_linterchange::execute): Remove call to scev_reset_htab.
2787 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2789 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
2790 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
2791 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
2792 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
2793 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
2794 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
2795 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
2796 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
2797 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
2798 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
2799 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
2800 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
2801 (V_lane_reg): Likewise.
2802 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
2804 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
2805 (vfmal_lane_low<mode>_intrinsic,
2806 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
2807 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
2808 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
2809 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
2810 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
2811 vfmsl_lane_high<mode>_intrinsic): New define_insns.
2813 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2815 * config/arm/arm-cpus.in (fp16fml): New feature.
2816 (ALL_SIMD): Add fp16fml.
2817 (armv8.2-a): Add fp16fml as an option.
2818 (armv8.3-a): Likewise.
2819 (armv8.4-a): Add fp16fml as part of fp16.
2820 * config/arm/arm.h (TARGET_FP16FML): Define.
2821 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
2823 * config/arm/arm-modes.def (V2HF): Define.
2824 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
2825 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
2826 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
2827 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
2828 vfmsl_low, vfmsl_high): New set of builtins.
2829 * config/arm/iterators.md (PLUSMINUS): New code iterator.
2830 (vfml_op): New code attribute.
2831 (VFMLHALVES): New int iterator.
2832 (VFML, VFMLSEL): New mode attributes.
2833 (V_reg): Define mapping for V2HF.
2834 (V_hi, V_lo): New mode attributes.
2835 (VF_constraint): Likewise.
2836 (vfml_half, vfml_half_selector): New int attributes.
2837 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
2839 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
2840 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
2842 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
2843 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
2844 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
2845 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
2847 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
2848 Document new effective target and option set.
2850 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2852 * config/arm/arm-cpus.in (armv8_4): New feature.
2853 (ARMv8_4a): New fgroup.
2854 (armv8.4-a): New arch.
2855 * config/arm/arm-tables.opt: Regenerate.
2856 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
2857 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
2858 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
2859 Add matching rules for -march=armv8.4-a and extensions.
2860 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
2862 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
2865 * config/rx/rx.md (BW): New mode attribute.
2866 (sync_lock_test_and_setsi): Add mode suffix to insn output.
2868 2018-01-11 Richard Biener <rguenther@suse.de>
2870 PR tree-optimization/83435
2871 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
2872 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
2873 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
2875 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2876 Alan Hayward <alan.hayward@arm.com>
2877 David Sherwood <david.sherwood@arm.com>
2879 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
2881 (aarch64_classify_address): Initialize it. Track polynomial offsets.
2882 (aarch64_print_address_internal): Use it to check for a zero offset.
2884 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2885 Alan Hayward <alan.hayward@arm.com>
2886 David Sherwood <david.sherwood@arm.com>
2888 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
2889 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
2890 Return a poly_int64 rather than a HOST_WIDE_INT.
2891 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
2892 rather than a HOST_WIDE_INT.
2893 * config/aarch64/aarch64.h (aarch64_frame): Protect with
2894 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
2895 hard_fp_offset, frame_size, initial_adjust, callee_offset and
2896 final_offset from HOST_WIDE_INT to poly_int64.
2897 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
2898 to_constant when getting the number of units in an Advanced SIMD
2900 (aarch64_builtin_vectorized_function): Check for a constant number
2902 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
2904 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
2905 attribute instead of GET_MODE_NUNITS.
2906 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
2907 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
2908 GET_MODE_SIZE for fixed-size registers.
2909 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
2910 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
2911 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
2912 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
2913 (aarch64_print_operand, aarch64_print_address_internal)
2914 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
2915 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
2916 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
2917 Handle polynomial GET_MODE_SIZE.
2918 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
2919 wider than SImode without modification.
2920 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
2921 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
2922 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
2923 passing and returning SVE modes.
2924 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
2925 rather than GEN_INT.
2926 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
2927 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
2928 (aarch64_allocate_and_probe_stack_space): Likewise.
2929 (aarch64_layout_frame): Cope with polynomial offsets.
2930 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
2931 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
2933 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
2934 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
2935 poly_int64 rather than a HOST_WIDE_INT.
2936 (aarch64_get_separate_components, aarch64_process_components)
2937 (aarch64_expand_prologue, aarch64_expand_epilogue)
2938 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
2939 (aarch64_anchor_offset): New function, split out from...
2940 (aarch64_legitimize_address): ...here.
2941 (aarch64_builtin_vectorization_cost): Handle polynomial
2942 TYPE_VECTOR_SUBPARTS.
2943 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
2945 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
2946 number of elements from the PARALLEL rather than the mode.
2947 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
2948 rather than GET_MODE_BITSIZE.
2949 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
2950 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
2951 (aarch64_expand_vec_perm_const_1): Handle polynomial
2952 d->perm.length () and d->perm elements.
2953 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
2954 Apply to_constant to d->perm elements.
2955 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
2956 polynomial CONST_VECTOR_NUNITS.
2957 (aarch64_move_pointer): Take amount as a poly_int64 rather
2959 (aarch64_progress_pointer): Avoid temporary variable.
2960 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
2961 the mode attribute instead of GET_MODE.
2963 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2964 Alan Hayward <alan.hayward@arm.com>
2965 David Sherwood <david.sherwood@arm.com>
2967 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
2968 x exists before using it.
2969 (aarch64_add_constant_internal): Rename to...
2970 (aarch64_add_offset_1): ...this. Replace regnum with separate
2971 src and dest rtxes. Handle the case in which they're different,
2972 including when the offset is zero. Replace scratchreg with an rtx.
2973 Use 2 additions if there is no spare register into which we can
2974 move a 16-bit constant.
2975 (aarch64_add_constant): Delete.
2976 (aarch64_add_offset): Replace reg with separate src and dest
2977 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
2978 Use aarch64_add_offset_1.
2979 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
2980 an rtx rather than an int. Take the delta as a poly_int64
2981 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
2982 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
2983 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
2984 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
2985 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
2987 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
2988 aarch64_add_constant.
2990 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2992 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
2993 Use scalar_float_mode.
2995 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2997 * config/aarch64/aarch64-simd.md
2998 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
2999 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3000 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3001 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3002 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3003 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3004 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3005 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3006 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3007 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3009 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3012 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3013 targ_options->x_arm_arch_string is non NULL.
3015 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3017 * config/aarch64/aarch64.h
3018 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3020 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3023 * expmed.c (emit_store_flag_force): Swap if const op0
3024 and change VOIDmode to mode of op0.
3026 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3028 PR rtl-optimization/83761
3029 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3030 than bytes to mode_for_size.
3032 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3035 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3036 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3039 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3042 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3043 when in layout mode.
3044 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3045 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3048 2018-01-10 Michael Collison <michael.collison@arm.com>
3050 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3051 * config/aarch64/aarch64-option-extension.def: Add
3052 AARCH64_OPT_EXTENSION of 'fp16fml'.
3053 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3054 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3055 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3056 * config/aarch64/constraints.md (Ui7): New constraint.
3057 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3058 (VFMLA_SEL_W): Ditto.
3061 (VFMLA16_LOW): New int iterator.
3062 (VFMLA16_HIGH): Ditto.
3063 (UNSPEC_FMLAL): New unspec.
3064 (UNSPEC_FMLSL): Ditto.
3065 (UNSPEC_FMLAL2): Ditto.
3066 (UNSPEC_FMLSL2): Ditto.
3067 (f16mac): New code attribute.
3068 * config/aarch64/aarch64-simd-builtins.def
3069 (aarch64_fmlal_lowv2sf): Ditto.
3070 (aarch64_fmlsl_lowv2sf): Ditto.
3071 (aarch64_fmlalq_lowv4sf): Ditto.
3072 (aarch64_fmlslq_lowv4sf): Ditto.
3073 (aarch64_fmlal_highv2sf): Ditto.
3074 (aarch64_fmlsl_highv2sf): Ditto.
3075 (aarch64_fmlalq_highv4sf): Ditto.
3076 (aarch64_fmlslq_highv4sf): Ditto.
3077 (aarch64_fmlal_lane_lowv2sf): Ditto.
3078 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3079 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3080 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3081 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3082 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3083 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3084 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3085 (aarch64_fmlal_lane_highv2sf): Ditto.
3086 (aarch64_fmlsl_lane_highv2sf): Ditto.
3087 (aarch64_fmlal_laneq_highv2sf): Ditto.
3088 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3089 (aarch64_fmlalq_lane_highv4sf): Ditto.
3090 (aarch64_fmlsl_lane_highv4sf): Ditto.
3091 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3092 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3093 * config/aarch64/aarch64-simd.md:
3094 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3095 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3096 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3097 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3098 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3099 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3100 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3101 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3102 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3103 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3104 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3105 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3106 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3107 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3108 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3109 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3110 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3111 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3112 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3113 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3114 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3115 (vfmlsl_low_u32): Ditto.
3116 (vfmlalq_low_u32): Ditto.
3117 (vfmlslq_low_u32): Ditto.
3118 (vfmlal_high_u32): Ditto.
3119 (vfmlsl_high_u32): Ditto.
3120 (vfmlalq_high_u32): Ditto.
3121 (vfmlslq_high_u32): Ditto.
3122 (vfmlal_lane_low_u32): Ditto.
3123 (vfmlsl_lane_low_u32): Ditto.
3124 (vfmlal_laneq_low_u32): Ditto.
3125 (vfmlsl_laneq_low_u32): Ditto.
3126 (vfmlalq_lane_low_u32): Ditto.
3127 (vfmlslq_lane_low_u32): Ditto.
3128 (vfmlalq_laneq_low_u32): Ditto.
3129 (vfmlslq_laneq_low_u32): Ditto.
3130 (vfmlal_lane_high_u32): Ditto.
3131 (vfmlsl_lane_high_u32): Ditto.
3132 (vfmlal_laneq_high_u32): Ditto.
3133 (vfmlsl_laneq_high_u32): Ditto.
3134 (vfmlalq_lane_high_u32): Ditto.
3135 (vfmlslq_lane_high_u32): Ditto.
3136 (vfmlalq_laneq_high_u32): Ditto.
3137 (vfmlslq_laneq_high_u32): Ditto.
3138 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3139 (AARCH64_FL_FOR_ARCH8_4): New.
3140 (AARCH64_ISA_F16FML): New ISA flag.
3141 (TARGET_F16FML): New feature flag for fp16fml.
3142 (doc/invoke.texi): Document new fp16fml option.
3144 2018-01-10 Michael Collison <michael.collison@arm.com>
3146 * config/aarch64/aarch64-builtins.c:
3147 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3148 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3149 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3150 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3151 (AARCH64_ISA_SHA3): New ISA flag.
3152 (TARGET_SHA3): New feature flag for sha3.
3153 * config/aarch64/iterators.md (sha512_op): New int attribute.
3154 (CRYPTO_SHA512): New int iterator.
3155 (UNSPEC_SHA512H): New unspec.
3156 (UNSPEC_SHA512H2): Ditto.
3157 (UNSPEC_SHA512SU0): Ditto.
3158 (UNSPEC_SHA512SU1): Ditto.
3159 * config/aarch64/aarch64-simd-builtins.def
3160 (aarch64_crypto_sha512hqv2di): New builtin.
3161 (aarch64_crypto_sha512h2qv2di): Ditto.
3162 (aarch64_crypto_sha512su0qv2di): Ditto.
3163 (aarch64_crypto_sha512su1qv2di): Ditto.
3164 (aarch64_eor3qv8hi): Ditto.
3165 (aarch64_rax1qv2di): Ditto.
3166 (aarch64_xarqv2di): Ditto.
3167 (aarch64_bcaxqv8hi): Ditto.
3168 * config/aarch64/aarch64-simd.md:
3169 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3170 (aarch64_crypto_sha512su0qv2di): Ditto.
3171 (aarch64_crypto_sha512su1qv2di): Ditto.
3172 (aarch64_eor3qv8hi): Ditto.
3173 (aarch64_rax1qv2di): Ditto.
3174 (aarch64_xarqv2di): Ditto.
3175 (aarch64_bcaxqv8hi): Ditto.
3176 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3177 (vsha512h2q_u64): Ditto.
3178 (vsha512su0q_u64): Ditto.
3179 (vsha512su1q_u64): Ditto.
3180 (veor3q_u16): Ditto.
3181 (vrax1q_u64): Ditto.
3183 (vbcaxq_u16): Ditto.
3184 * config/arm/types.md (crypto_sha512): New type attribute.
3185 (crypto_sha3): Ditto.
3186 (doc/invoke.texi): Document new sha3 option.
3188 2018-01-10 Michael Collison <michael.collison@arm.com>
3190 * config/aarch64/aarch64-builtins.c:
3191 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3192 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3193 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3194 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3195 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3196 (AARCH64_ISA_SM4): New ISA flag.
3197 (TARGET_SM4): New feature flag for sm4.
3198 * config/aarch64/aarch64-simd-builtins.def
3199 (aarch64_sm3ss1qv4si): Ditto.
3200 (aarch64_sm3tt1aq4si): Ditto.
3201 (aarch64_sm3tt1bq4si): Ditto.
3202 (aarch64_sm3tt2aq4si): Ditto.
3203 (aarch64_sm3tt2bq4si): Ditto.
3204 (aarch64_sm3partw1qv4si): Ditto.
3205 (aarch64_sm3partw2qv4si): Ditto.
3206 (aarch64_sm4eqv4si): Ditto.
3207 (aarch64_sm4ekeyqv4si): Ditto.
3208 * config/aarch64/aarch64-simd.md:
3209 (aarch64_sm3ss1qv4si): Ditto.
3210 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3211 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3212 (aarch64_sm4eqv4si): Ditto.
3213 (aarch64_sm4ekeyqv4si): Ditto.
3214 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3215 (sm3part_op): Ditto.
3216 (CRYPTO_SM3TT): Ditto.
3217 (CRYPTO_SM3PART): Ditto.
3218 (UNSPEC_SM3SS1): New unspec.
3219 (UNSPEC_SM3TT1A): Ditto.
3220 (UNSPEC_SM3TT1B): Ditto.
3221 (UNSPEC_SM3TT2A): Ditto.
3222 (UNSPEC_SM3TT2B): Ditto.
3223 (UNSPEC_SM3PARTW1): Ditto.
3224 (UNSPEC_SM3PARTW2): Ditto.
3225 (UNSPEC_SM4E): Ditto.
3226 (UNSPEC_SM4EKEY): Ditto.
3227 * config/aarch64/constraints.md (Ui2): New constraint.
3228 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3229 * config/arm/types.md (crypto_sm3): New type attribute.
3230 (crypto_sm4): Ditto.
3231 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3232 (vsm3tt1aq_u32): Ditto.
3233 (vsm3tt1bq_u32): Ditto.
3234 (vsm3tt2aq_u32): Ditto.
3235 (vsm3tt2bq_u32): Ditto.
3236 (vsm3partw1q_u32): Ditto.
3237 (vsm3partw2q_u32): Ditto.
3238 (vsm4eq_u32): Ditto.
3239 (vsm4ekeyq_u32): Ditto.
3240 (doc/invoke.texi): Document new sm4 option.
3242 2018-01-10 Michael Collison <michael.collison@arm.com>
3244 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3245 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3246 (AARCH64_FL_FOR_ARCH8_4): New.
3247 (AARCH64_FL_V8_4): New flag.
3248 (doc/invoke.texi): Document new armv8.4-a option.
3250 2018-01-10 Michael Collison <michael.collison@arm.com>
3252 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3253 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3254 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3255 * config/aarch64/aarch64-option-extension.def: Add
3256 AARCH64_OPT_EXTENSION of 'sha2'.
3257 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3258 (crypto): Disable sha2 and aes if crypto disabled.
3259 (crypto): Enable aes and sha2 if enabled.
3260 (simd): Disable sha2 and aes if simd disabled.
3261 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3263 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3264 (TARGET_SHA2): New feature flag for sha2.
3265 (TARGET_AES): New feature flag for aes.
3266 * config/aarch64/aarch64-simd.md:
3267 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3268 conditional on TARGET_AES.
3269 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3270 (aarch64_crypto_sha1hsi): Make pattern conditional
3272 (aarch64_crypto_sha1hv4si): Ditto.
3273 (aarch64_be_crypto_sha1hv4si): Ditto.
3274 (aarch64_crypto_sha1su1v4si): Ditto.
3275 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3276 (aarch64_crypto_sha1su0v4si): Ditto.
3277 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3278 (aarch64_crypto_sha256su0v4si): Ditto.
3279 (aarch64_crypto_sha256su1v4si): Ditto.
3280 (doc/invoke.texi): Document new aes and sha2 options.
3282 2018-01-10 Martin Sebor <msebor@redhat.com>
3284 PR tree-optimization/83781
3285 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3288 2018-01-11 Martin Sebor <msebor@gmail.com>
3289 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3291 PR tree-optimization/83501
3292 PR tree-optimization/81703
3294 * tree-ssa-strlen.c (get_string_cst): Rename...
3295 (get_string_len): ...to this. Handle global constants.
3296 (handle_char_store): Adjust.
3298 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3299 Jim Wilson <jimw@sifive.com>
3301 * config/riscv/riscv-protos.h (riscv_output_return): New.
3302 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3303 (riscv_attribute_table, riscv_output_return),
3304 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3305 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3306 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3308 (riscv_expand_prologue): Add early return for naked function.
3309 (riscv_expand_epilogue): Likewise.
3310 (riscv_function_ok_for_sibcall): Return false for naked function.
3311 (riscv_set_current_function): New.
3312 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3313 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3314 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3315 * doc/extend.texi (RISC-V Function Attributes): New.
3317 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3319 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3320 check for 128-bit long double before checking TCmode.
3321 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3322 128-bit long doubles before checking TFmode or TCmode.
3323 (FLOAT128_IBM_P): Likewise.
3325 2018-01-10 Martin Sebor <msebor@redhat.com>
3327 PR tree-optimization/83671
3328 * builtins.c (c_strlen): Unconditionally return zero for the empty
3330 Use -Warray-bounds for warnings.
3331 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3332 for non-constant array indices with COMPONENT_REF, arrays of
3333 arrays, and pointers to arrays.
3334 (gimple_fold_builtin_strlen): Determine and set length range for
3335 non-constant character arrays.
3337 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3340 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3343 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3345 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3347 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3350 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3351 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3352 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3353 indexed_or_indirect_operand predicate.
3354 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3355 (*vsx_le_perm_load_v8hi): Likewise.
3356 (*vsx_le_perm_load_v16qi): Likewise.
3357 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3358 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3359 (*vsx_le_perm_store_v8hi): Likewise.
3360 (*vsx_le_perm_store_v16qi): Likewise.
3361 (eight unnamed splitters): Likewise.
3363 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3365 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3366 * config/rs6000/emmintrin.h: Likewise.
3367 * config/rs6000/mmintrin.h: Likewise.
3368 * config/rs6000/xmmintrin.h: Likewise.
3370 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3373 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3375 * tree.c (tree_nop_conversion): Return true for location wrapper
3377 (maybe_wrap_with_location): New function.
3378 (selftest::check_strip_nops): New function.
3379 (selftest::test_location_wrappers): New function.
3380 (selftest::tree_c_tests): Call it.
3381 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3382 (maybe_wrap_with_location): New decl.
3383 (EXPR_LOCATION_WRAPPER_P): New macro.
3384 (location_wrapper_p): New inline function.
3385 (tree_strip_any_location_wrapper): New inline function.
3387 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3390 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3391 stack_realign_offset for the largest alignment of stack slot
3393 (ix86_find_max_used_stack_alignment): New function.
3394 (ix86_finalize_stack_frame_flags): Use it. Set
3395 max_used_stack_alignment if we don't realign stack.
3396 * config/i386/i386.h (machine_function): Add
3397 max_used_stack_alignment.
3399 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3401 * config/arm/arm.opt (-mbranch-cost): New option.
3402 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3405 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3408 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3409 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3411 2018-01-10 Richard Biener <rguenther@suse.de>
3414 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3415 early out so it also covers the case where we have a non-NULL
3418 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3420 PR tree-optimization/83753
3421 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3422 for non-strided grouped accesses if the number of elements is 1.
3424 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3427 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3428 * i386.h (TARGET_USE_GATHER): Define.
3429 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3431 2018-01-10 Martin Liska <mliska@suse.cz>
3434 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3435 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3437 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3438 CLEANUP_NO_PARTITIONING is not set.
3440 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3442 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3443 for vectors, as a partial revert of r254296.
3444 * rtl.h (const_vec_p): Delete.
3445 (const_vec_duplicate_p): Don't test for vector CONSTs.
3446 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3447 * expmed.c (make_tree): Likewise.
3450 * common.md (E, F): Use CONSTANT_P instead of checking for
3452 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3453 checking for CONST_VECTOR.
3455 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3458 * predict.c (force_edge_cold): Handle in more sane way edges
3461 2018-01-09 Carl Love <cel@us.ibm.com>
3463 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3465 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3466 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3467 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3468 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3469 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3470 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3471 * config/rs6000/rs6000-protos.h: Add extern defition for
3472 rs6000_generate_float2_double_code.
3473 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3475 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3476 (float2_v2df): Add define_expand.
3478 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3481 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3482 op_mode in the force_to_mode call.
3484 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3486 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3487 instead of checking each element individually.
3488 (aarch64_evpc_uzp): Likewise.
3489 (aarch64_evpc_zip): Likewise.
3490 (aarch64_evpc_ext): Likewise.
3491 (aarch64_evpc_rev): Likewise.
3492 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3493 instead of checking each element individually. Return true without
3495 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3496 whether all selected elements come from the same input, instead of
3497 checking each element individually. Remove calls to gen_rtx_REG,
3498 start_sequence and end_sequence and instead assert that no rtl is
3501 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3503 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3504 order of HIGH and CONST checks.
3506 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3508 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3509 if the destination isn't an SSA_NAME.
3511 2018-01-09 Richard Biener <rguenther@suse.de>
3513 PR tree-optimization/83668
3514 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3516 (canonicalize_loop_form): ... here, renamed from ...
3517 (canonicalize_loop_closed_ssa_form): ... this and amended to
3518 swap successor edges for loop exit blocks to make us use
3519 the RPO order we need for initial schedule generation.
3521 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3523 PR tree-optimization/64811
3524 * match.pd: When optimizing comparisons with Inf, avoid
3525 introducing or losing exceptions from comparisons with NaN.
3527 2018-01-09 Martin Liska <mliska@suse.cz>
3530 * asan.c (shadow_mem_size): Add gcc_assert.
3532 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3534 Don't save registers in main().
3537 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3538 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3539 * config/avr/avr.c (avr_set_current_function): Don't error if
3540 naked, OS_task or OS_main are specified at the same time.
3541 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3543 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3545 * common/config/avr/avr-common.c (avr_option_optimization_table):
3546 Switch on -mmain-is-OS_task for optimizing compilations.
3548 2018-01-09 Richard Biener <rguenther@suse.de>
3550 PR tree-optimization/83572
3551 * graphite.c: Include cfganal.h.
3552 (graphite_transform_loops): Connect infinite loops to exit
3553 and remove fake edges at the end.
3555 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3557 * ipa-inline.c (edge_badness): Revert accidental checkin.
3559 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3562 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3563 symbols; not inline clones.
3565 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3568 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3569 hard registers. Formatting fixes.
3571 PR preprocessor/83722
3572 * gcc.c (try_generate_repro): Pass
3573 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3574 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3577 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3578 Kito Cheng <kito.cheng@gmail.com>
3580 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3581 (riscv_leaf_function_p): Delete.
3582 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3584 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3586 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3588 (do_ifelse): New function.
3589 (do_isel): New function.
3590 (do_sub3): New function.
3591 (do_add3): New function.
3592 (do_load_mask_compare): New function.
3593 (do_overlap_load_compare): New function.
3594 (expand_compare_loop): New function.
3595 (expand_block_compare): Call expand_compare_loop() when appropriate.
3596 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3598 (-mblock-compare-inline-loop-limit): New option.
3600 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3603 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3604 Reverse order of second and third operands in first alternative.
3605 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3606 of first and second elements in UNSPEC_VPERMR vector.
3607 (altivec_expand_vec_perm_le): Likewise.
3609 2017-01-08 Jeff Law <law@redhat.com>
3611 PR rtl-optimizatin/81308
3612 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3613 (process_switch): If group_case_labels makes a change, then set
3615 (pass_convert_switch::execute): If a switch is converted, then
3616 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3618 PR rtl-optimization/81308
3619 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3622 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3624 PR target/83663 - Revert r255946
3625 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3626 generation for cases where splatting a value is not useful.
3627 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3628 across a vec_duplicate and a paradoxical subreg forming a vector
3629 mode to a vec_concat.
3631 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3633 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3634 -march=armv8.3-a variants.
3635 * config/arm/t-multilib: Likewise.
3636 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3638 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3640 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3642 (cceq_ior_compare_complement): Give it a name so I can use it, and
3643 change boolean_or_operator predicate to boolean_operator so it can
3644 be used to generate a crand.
3645 (eqne): New code iterator.
3646 (bd/bd_neg): New code_attrs.
3647 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3648 a single define_insn.
3649 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3650 decrement (bdnzt/bdnzf/bdzt/bdzf).
3651 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3652 with the new names of the branch decrement patterns, and added the
3653 names of the branch decrement conditional patterns.
3655 2018-01-08 Richard Biener <rguenther@suse.de>
3657 PR tree-optimization/83563
3658 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3661 2018-01-08 Richard Biener <rguenther@suse.de>
3664 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3666 2018-01-08 Richard Biener <rguenther@suse.de>
3668 PR tree-optimization/83685
3669 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3670 references to abnormals.
3672 2018-01-08 Richard Biener <rguenther@suse.de>
3675 * dwarf2out.c (output_indirect_strings): Handle empty
3676 skeleton_debug_str_hash.
3677 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3679 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3681 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3682 (emit_store_direct): Likewise.
3683 (arc_trampoline_adjust_address): Likewise.
3684 (arc_asm_trampoline_template): New function.
3685 (arc_initialize_trampoline): Use asm_trampoline_template.
3686 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3687 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3688 * config/arc/arc.md (flush_icache): Delete pattern.
3690 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3692 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3693 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3696 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3699 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3700 by not USED_FOR_TARGET.
3701 (make_pass_resolve_sw_modes): Likewise.
3703 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3705 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3708 2018-01-08 Richard Biener <rguenther@suse.de>
3711 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3713 2018-01-08 Richard Biener <rguenther@suse.de>
3716 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3718 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3721 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3722 basic blocks with a small number of successors.
3723 (convert_control_dep_chain_into_preds): Improve handling of
3725 (dump_predicates): Split apart into...
3726 (dump_pred_chain): ...here...
3727 (dump_pred_info): ...and here.
3728 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3729 (can_chain_union_be_invalidated_p): Improve check for invalidation
3731 (uninit_uses_cannot_happen): Avoid unnecessary if
3732 convert_control_dep_chain_into_preds yielded nothing.
3734 2018-01-06 Martin Sebor <msebor@redhat.com>
3736 PR tree-optimization/83640
3737 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3738 subtracting negative offset from size.
3739 (builtin_access::overlap): Adjust offset bounds of the access to fall
3740 within the size of the object if possible.
3742 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3744 PR rtl-optimization/83699
3745 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3746 extract_bit_field_as_subreg to cases in which the extracted
3747 value is also a vector.
3749 * lra-constraints.c (process_alt_operands): Test for the equivalence
3750 substitutions when detecting a possible reload cycle.
3752 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3755 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3756 by default if flag_selective_schedling{,2}. Formatting fixes.
3758 PR rtl-optimization/83682
3759 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
3760 if it has non-VECTOR_MODE element mode.
3761 (vec_duplicate_p): Likewise.
3764 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
3765 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
3767 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3770 * config/i386/i386-builtin.def
3771 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
3772 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
3773 Require also OPTION_MASK_ISA_AVX512F in addition to
3774 OPTION_MASK_ISA_GFNI.
3775 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
3776 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
3777 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
3778 to OPTION_MASK_ISA_GFNI.
3779 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
3780 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
3781 OPTION_MASK_ISA_AVX512BW.
3782 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
3783 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
3784 addition to OPTION_MASK_ISA_GFNI.
3785 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
3786 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
3787 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
3788 to OPTION_MASK_ISA_GFNI.
3789 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
3790 a requirement for all ISAs rather than any of them with a few
3792 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
3794 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
3795 bitmasks to be enabled with 3 exceptions, instead of requiring any
3796 enabled ISA with lots of exceptions.
3797 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
3798 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
3799 Change avx512bw in isa attribute to avx512f.
3800 * config/i386/sgxintrin.h: Add license boilerplate.
3801 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
3802 to __AVX512F__ and __AVX512VL to __AVX512VL__.
3803 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
3804 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
3806 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
3807 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
3808 temporarily sse2 rather than sse if not enabled already.
3811 * config/i386/sse.md (VI248_VLBW): Rename to ...
3812 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
3813 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
3814 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
3815 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
3816 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
3817 mode iterator instead of VI248_VLBW.
3819 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
3821 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
3822 (record_modified): Skip clobbers; add debug output.
3823 (param_change_prob): Use sreal frequencies.
3825 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3827 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
3828 punt for user-aligned variables.
3830 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3832 * tree-chrec.c (chrec_contains_symbols): Return true for
3835 2018-01-05 Sudakshina Das <sudi.das@arm.com>
3838 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
3839 of (x|y) == x for BICS pattern.
3841 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3843 PR tree-optimization/83605
3844 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
3845 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
3848 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
3850 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
3851 * config/epiphany/rtems.h: New file.
3853 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3854 Uros Bizjak <ubizjak@gmail.com>
3857 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
3858 QIreg_operand instead of register_operand predicate.
3859 * config/i386/i386.c (ix86_rop_should_change_byte_p,
3860 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
3861 comments instead of -fmitigate[-_]rop.
3863 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3866 * cgraphunit.c (symbol_table::compile): Switch to text_section
3867 before calling assembly_start debug hook.
3868 * run-rtl-passes.c (run_rtl_passes): Likewise.
3871 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3873 * tree-vrp.c (extract_range_from_binary_expr_1): Check
3874 range_int_cst_p rather than !symbolic_range_p before calling
3875 extract_range_from_multiplicative_op_1.
3877 2017-01-04 Jeff Law <law@redhat.com>
3879 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
3880 redundant test in assertion.
3882 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3884 * doc/rtl.texi: Document machine_mode wrapper classes.
3886 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3888 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
3891 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3893 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
3894 the VEC_PERM_EXPR fold to fail.
3896 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3899 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
3900 to switched_sections.
3902 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3905 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
3908 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
3911 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
3912 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
3914 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3917 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
3918 is BLKmode and bitpos not zero or mode change is needed.
3920 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3923 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
3926 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
3929 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
3930 instead of MULT rtx. Update all corresponding splitters.
3932 (*ssub<modesuffix>): Ditto.
3934 (*cmp_sadd_di): Update split patterns.
3935 (*cmp_sadd_si): Ditto.
3936 (*cmp_sadd_sidi): Ditto.
3937 (*cmp_ssub_di): Ditto.
3938 (*cmp_ssub_si): Ditto.
3939 (*cmp_ssub_sidi): Ditto.
3940 * config/alpha/predicates.md (const23_operand): New predicate.
3941 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
3942 Look for ASHIFT, not MULT inner operand.
3943 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
3945 2018-01-04 Martin Liska <mliska@suse.cz>
3947 PR gcov-profile/83669
3948 * gcov.c (output_intermediate_file): Add version to intermediate
3950 * doc/gcov.texi: Document new field 'version' in intermediate
3951 file format. Fix location of '-k' option of gcov command.
3953 2018-01-04 Martin Liska <mliska@suse.cz>
3956 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
3958 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3960 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
3962 2018-01-03 Martin Sebor <msebor@redhat.com>
3964 PR tree-optimization/83655
3965 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
3966 checking calls with invalid arguments.
3968 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3970 * tree-vect-stmts.c (vect_get_store_rhs): New function.
3971 (vectorizable_mask_load_store): Delete.
3972 (vectorizable_call): Return false for masked loads and stores.
3973 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
3974 instead of gimple_assign_rhs1.
3975 (vectorizable_load): Handle IFN_MASK_LOAD.
3976 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
3978 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3980 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
3982 (vectorizable_mask_load_store): ...here.
3983 (vectorizable_load): ...and here.
3985 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3987 * tree-vect-stmts.c (vect_build_all_ones_mask)
3988 (vect_build_zero_merge_argument): New functions, split out from...
3989 (vectorizable_load): ...here.
3991 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3993 * tree-vect-stmts.c (vect_check_store_rhs): New function,
3995 (vectorizable_mask_load_store): ...here.
3996 (vectorizable_store): ...and here.
3998 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4000 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4002 (vectorizable_mask_load_store): ...here.
4004 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4006 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4007 (vect_model_store_cost): Take a vec_load_store_type instead of a
4009 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4010 (vect_model_store_cost): Take a vec_load_store_type instead of a
4012 (vectorizable_mask_load_store): Update accordingly.
4013 (vectorizable_store): Likewise.
4014 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4016 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4018 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4019 IFN_MASK_LOAD calls here rather than...
4020 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4022 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4023 Alan Hayward <alan.hayward@arm.com>
4024 David Sherwood <david.sherwood@arm.com>
4026 * expmed.c (extract_bit_field_1): For vector extracts,
4027 fall back to extract_bit_field_as_subreg if vec_extract
4030 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4031 Alan Hayward <alan.hayward@arm.com>
4032 David Sherwood <david.sherwood@arm.com>
4034 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4035 they are variable or constant sized.
4036 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4037 slots for constant-sized data.
4039 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4040 Alan Hayward <alan.hayward@arm.com>
4041 David Sherwood <david.sherwood@arm.com>
4043 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4044 handling COND_EXPRs with boolean comparisons, try to find a better
4045 basis for the mask type than the boolean itself.
4047 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4049 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4050 is calculated and how it can be overridden.
4051 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4052 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4054 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4057 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4058 Alan Hayward <alan.hayward@arm.com>
4059 David Sherwood <david.sherwood@arm.com>
4061 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4062 Remove the mode argument.
4063 (aarch64_simd_valid_immediate): Remove the mode and inverse
4065 * config/aarch64/iterators.md (bitsize): New iterator.
4066 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4067 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4068 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4069 aarch64_simd_valid_immediate.
4070 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4071 (aarch64_reg_or_bic_imm): Likewise.
4072 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4073 with an insn_type enum and msl with a modifier_type enum.
4074 Replace element_width with a scalar_mode. Change the shift
4075 to unsigned int. Add constructors for scalar_float_mode and
4076 scalar_int_mode elements.
4077 (aarch64_vect_float_const_representable_p): Delete.
4078 (aarch64_can_const_movi_rtx_p)
4079 (aarch64_simd_scalar_immediate_valid_for_move)
4080 (aarch64_simd_make_constant): Update call to
4081 aarch64_simd_valid_immediate.
4082 (aarch64_advsimd_valid_immediate_hs): New function.
4083 (aarch64_advsimd_valid_immediate): Likewise.
4084 (aarch64_simd_valid_immediate): Remove mode and inverse
4085 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4086 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4087 and aarch64_float_const_representable_p on the result.
4088 (aarch64_output_simd_mov_immediate): Remove mode argument.
4089 Update call to aarch64_simd_valid_immediate and use of
4090 simd_immediate_info.
4091 (aarch64_output_scalar_simd_mov_immediate): Update call
4094 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4095 Alan Hayward <alan.hayward@arm.com>
4096 David Sherwood <david.sherwood@arm.com>
4098 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4099 (mode_nunits): Likewise CONST_MODE_NUNITS.
4100 * machmode.def (ADJUST_NUNITS): Document.
4101 * genmodes.c (mode_data::need_nunits_adj): New field.
4102 (blank_mode): Update accordingly.
4103 (adj_nunits): New variable.
4104 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4106 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4107 listed in adj_nunits.
4108 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4109 listed in adj_nunits. Don't emit case statements for such modes.
4110 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4111 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4112 nothing if adj_nunits is nonnull.
4113 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4114 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4115 (emit_mode_fbit): Update use of print_maybe_const_decl.
4116 (emit_move_size): Likewise. Treat the array as non-const
4118 (emit_mode_adjustments): Handle adj_nunits.
4120 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4122 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4123 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4124 (VECTOR_MODES): Use it.
4125 (make_vector_modes): Take the prefix as an argument.
4127 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4128 Alan Hayward <alan.hayward@arm.com>
4129 David Sherwood <david.sherwood@arm.com>
4131 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4132 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4133 for MODE_VECTOR_BOOL.
4134 * machmode.def (VECTOR_BOOL_MODE): Document.
4135 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4136 (make_vector_bool_mode): New function.
4137 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4139 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4140 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4142 * stor-layout.c (int_mode_for_mode): Likewise.
4143 * tree.c (build_vector_type_for_mode): Likewise.
4144 * varasm.c (output_constant_pool_2): Likewise.
4145 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4146 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4147 for MODE_VECTOR_BOOL.
4148 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4149 of mode class checks.
4150 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4151 instead of a list of mode class checks.
4152 (expand_vector_scalar_condition): Likewise.
4153 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4155 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4156 Alan Hayward <alan.hayward@arm.com>
4157 David Sherwood <david.sherwood@arm.com>
4159 * machmode.h (mode_size): Change from unsigned short to
4161 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4162 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4163 or if measurement_type is not polynomial.
4164 (fixed_size_mode::includes_p): Check for constant-sized modes.
4165 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4166 return a poly_uint16 rather than an unsigned short.
4167 (emit_mode_size): Change the type of mode_size from unsigned short
4168 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4169 (emit_mode_adjustments): Cope with polynomial vector sizes.
4170 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4172 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4174 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4175 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4176 * caller-save.c (setup_save_areas): Likewise.
4177 (replace_reg_with_saved_mem): Likewise.
4178 * calls.c (emit_library_call_value_1): Likewise.
4179 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4180 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4181 (gen_lowpart_for_combine): Likewise.
4182 * convert.c (convert_to_integer_1): Likewise.
4183 * cse.c (equiv_constant, cse_insn): Likewise.
4184 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4185 (cselib_subst_to_values): Likewise.
4186 * dce.c (word_dce_process_block): Likewise.
4187 * df-problems.c (df_word_lr_mark_ref): Likewise.
4188 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4189 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4190 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4191 (rtl_for_decl_location): Likewise.
4192 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4193 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4194 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4195 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4196 (expand_expr_real_1): Likewise.
4197 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4198 (pad_below): Likewise.
4199 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4200 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4201 * ira.c (get_subreg_tracking_sizes): Likewise.
4202 * ira-build.c (ira_create_allocno_objects): Likewise.
4203 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4204 (ira_sort_regnos_for_alter_reg): Likewise.
4205 * ira-costs.c (record_operand_costs): Likewise.
4206 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4207 (resolve_simple_move): Likewise.
4208 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4209 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4210 (lra_constraints): Likewise.
4211 (CONST_POOL_OK_P): Reject variable-sized modes.
4212 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4213 (add_pseudo_to_slot, lra_spill): Likewise.
4214 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4215 * optabs-query.c (get_best_extraction_insn): Likewise.
4216 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4217 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4218 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4219 * recog.c (offsettable_address_addr_space_p): Likewise.
4220 * regcprop.c (maybe_mode_change): Likewise.
4221 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4222 * regrename.c (build_def_use): Likewise.
4223 * regstat.c (dump_reg_info): Likewise.
4224 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4225 (find_reloads, find_reloads_subreg_address): Likewise.
4226 * reload1.c (eliminate_regs_1): Likewise.
4227 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4228 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4229 (simplify_binary_operation_1, simplify_subreg): Likewise.
4230 * targhooks.c (default_function_arg_padding): Likewise.
4231 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4232 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4233 (verify_gimple_assign_ternary): Likewise.
4234 * tree-inline.c (estimate_move_cost): Likewise.
4235 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4236 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4237 (get_address_cost_ainc): Likewise.
4238 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4239 (vect_supportable_dr_alignment): Likewise.
4240 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4241 (vectorizable_reduction): Likewise.
4242 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4243 (vectorizable_operation, vectorizable_load): Likewise.
4244 * tree.c (build_same_sized_truth_vector_type): Likewise.
4245 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4246 * var-tracking.c (emit_note_insn_var_location): Likewise.
4247 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4248 (ADDR_VEC_ALIGN): Likewise.
4250 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4251 Alan Hayward <alan.hayward@arm.com>
4252 David Sherwood <david.sherwood@arm.com>
4254 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4256 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4257 or if measurement_type is polynomial.
4258 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4259 * combine.c (make_extraction): Likewise.
4260 * dse.c (find_shift_sequence): Likewise.
4261 * dwarf2out.c (mem_loc_descriptor): Likewise.
4262 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4263 (extract_bit_field, extract_low_bits): Likewise.
4264 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4265 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4266 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4267 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4268 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4269 * reload.c (find_reloads): Likewise.
4270 * reload1.c (alter_reg): Likewise.
4271 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4272 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4273 * tree-if-conv.c (predicate_mem_writes): Likewise.
4274 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4275 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4276 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4277 * valtrack.c (dead_debug_insert_temp): Likewise.
4278 * varasm.c (mergeable_constant_section): Likewise.
4279 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4281 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4282 Alan Hayward <alan.hayward@arm.com>
4283 David Sherwood <david.sherwood@arm.com>
4285 * expr.c (expand_assignment): Cope with polynomial mode sizes
4286 when assigning to a CONCAT.
4288 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4289 Alan Hayward <alan.hayward@arm.com>
4290 David Sherwood <david.sherwood@arm.com>
4292 * machmode.h (mode_precision): Change from unsigned short to
4294 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4296 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4297 or if measurement_type is not polynomial.
4298 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4299 in which the mode is already known to be a scalar_int_mode.
4300 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4301 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4303 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4304 for GET_MODE_PRECISION.
4305 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4306 for GET_MODE_PRECISION.
4307 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4309 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4310 (expand_field_assignment, make_extraction): Likewise.
4311 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4312 (get_last_value): Likewise.
4313 * convert.c (convert_to_integer_1): Likewise.
4314 * cse.c (cse_insn): Likewise.
4315 * expr.c (expand_expr_real_1): Likewise.
4316 * lra-constraints.c (simplify_operand_subreg): Likewise.
4317 * optabs-query.c (can_atomic_load_p): Likewise.
4318 * optabs.c (expand_atomic_load): Likewise.
4319 (expand_atomic_store): Likewise.
4320 * ree.c (combine_reaching_defs): Likewise.
4321 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4322 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4323 * tree.h (type_has_mode_precision_p): Likewise.
4324 * ubsan.c (instrument_si_overflow): Likewise.
4326 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4327 Alan Hayward <alan.hayward@arm.com>
4328 David Sherwood <david.sherwood@arm.com>
4330 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4331 polynomial numbers of units.
4332 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4333 (valid_vector_subparts_p): New function.
4334 (build_vector_type): Remove temporary shim and take the number
4335 of units as a poly_uint64 rather than an int.
4336 (build_opaque_vector_type): Take the number of units as a
4337 poly_uint64 rather than an int.
4338 * tree.c (build_vector_from_ctor): Handle polynomial
4339 TYPE_VECTOR_SUBPARTS.
4340 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4341 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4342 (build_vector_from_val): If the number of units is variable,
4343 use build_vec_duplicate_cst for constant operands and
4344 VEC_DUPLICATE_EXPR otherwise.
4345 (make_vector_type): Remove temporary is_constant ().
4346 (build_vector_type, build_opaque_vector_type): Take the number of
4347 units as a poly_uint64 rather than an int.
4348 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4350 * cfgexpand.c (expand_debug_expr): Likewise.
4351 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4352 (store_constructor, expand_expr_real_1): Likewise.
4353 (const_scalar_mask_from_tree): Likewise.
4354 * fold-const-call.c (fold_const_reduction): Likewise.
4355 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4356 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4357 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4358 (fold_relational_const): Likewise.
4359 (native_interpret_vector): Likewise. Change the size from an
4360 int to an unsigned int.
4361 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4362 TYPE_VECTOR_SUBPARTS.
4363 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4364 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4365 duplicating a non-constant operand into a variable-length vector.
4366 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4367 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4368 * ipa-icf.c (sem_variable::equals): Likewise.
4369 * match.pd: Likewise.
4370 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4371 * print-tree.c (print_node): Likewise.
4372 * stor-layout.c (layout_type): Likewise.
4373 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4374 * tree-cfg.c (verify_gimple_comparison): Likewise.
4375 (verify_gimple_assign_binary): Likewise.
4376 (verify_gimple_assign_ternary): Likewise.
4377 (verify_gimple_assign_single): Likewise.
4378 * tree-pretty-print.c (dump_generic_node): Likewise.
4379 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4380 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4381 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4382 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4383 (vect_shift_permute_load_chain): Likewise.
4384 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4385 (expand_vector_condition, optimize_vector_constructor): Likewise.
4386 (lower_vec_perm, get_compute_type): Likewise.
4387 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4388 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4389 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4390 (vect_recog_mask_conversion_pattern): Likewise.
4391 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4392 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4393 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4394 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4395 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4396 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4397 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4398 (supportable_widening_operation): Likewise.
4399 (supportable_narrowing_operation): Likewise.
4400 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4402 * varasm.c (output_constant): Likewise.
4404 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4405 Alan Hayward <alan.hayward@arm.com>
4406 David Sherwood <david.sherwood@arm.com>
4408 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4409 so that both the length == 3 and length != 3 cases set up their
4410 own permute vectors. Add comments explaining why we know the
4411 number of elements is constant.
4412 (vect_permute_load_chain): Likewise.
4414 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4415 Alan Hayward <alan.hayward@arm.com>
4416 David Sherwood <david.sherwood@arm.com>
4418 * machmode.h (mode_nunits): Change from unsigned char to
4420 (ONLY_FIXED_SIZE_MODES): New macro.
4421 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4422 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4423 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4425 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4426 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4427 or if measurement_type is not polynomial.
4428 * genmodes.c (ZERO_COEFFS): New macro.
4429 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4431 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4432 Use ZERO_COEFFS when emitting initializers.
4433 * data-streamer.h (bp_pack_poly_value): New function.
4434 (bp_unpack_poly_value): Likewise.
4435 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4436 for GET_MODE_NUNITS.
4437 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4438 for GET_MODE_NUNITS.
4439 * tree.c (make_vector_type): Remove temporary shim and make
4440 the real function take the number of units as a poly_uint64
4442 (build_vector_type_for_mode): Handle polynomial nunits.
4443 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4444 * emit-rtl.c (const_vec_series_p_1): Likewise.
4445 (gen_rtx_CONST_VECTOR): Likewise.
4446 * fold-const.c (test_vec_duplicate_folding): Likewise.
4447 * genrecog.c (validate_pattern): Likewise.
4448 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4449 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4450 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4451 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4452 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4453 * rtlanal.c (subreg_get_info): Likewise.
4454 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4455 (vect_grouped_load_supported): Likewise.
4456 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4457 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4458 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4459 (simplify_const_unary_operation, simplify_binary_operation_1)
4460 (simplify_const_binary_operation, simplify_ternary_operation)
4461 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4462 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4463 instead of CONST_VECTOR_NUNITS.
4464 * varasm.c (output_constant_pool_2): Likewise.
4465 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4466 explicit-encoded elements in the XVEC for variable-length vectors.
4468 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4470 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4472 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4473 Alan Hayward <alan.hayward@arm.com>
4474 David Sherwood <david.sherwood@arm.com>
4476 * coretypes.h (fixed_size_mode): Declare.
4477 (fixed_size_mode_pod): New typedef.
4478 * builtins.h (target_builtins::x_apply_args_mode)
4479 (target_builtins::x_apply_result_mode): Change type to
4480 fixed_size_mode_pod.
4481 * builtins.c (apply_args_size, apply_result_size, result_vector)
4482 (expand_builtin_apply_args_1, expand_builtin_apply)
4483 (expand_builtin_return): Update accordingly.
4485 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4487 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4488 * cselib.c (cselib_hash_rtx): Likewise.
4489 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4490 CONST_VECTOR encoding.
4492 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4493 Jeff Law <law@redhat.com>
4496 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4497 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4498 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4499 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4502 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4503 explicitly probe *sp in a noreturn function if there were any callee
4504 register saves or frame pointer is needed.
4506 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4509 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4510 BLKmode for ternary, binary or unary expressions.
4513 * var-tracking.c (delete_vta_debug_insn): New inline function.
4514 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4515 insns from get_insns () to NULL instead of each bb separately.
4516 Use delete_vta_debug_insn. No longer static.
4517 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4518 delete_vta_debug_insns callers.
4519 * rtl.h (delete_vta_debug_insns): Declare.
4520 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4521 instead of variable_tracking_main.
4523 2018-01-03 Martin Sebor <msebor@redhat.com>
4525 PR tree-optimization/83603
4526 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4527 arguments past the endof the argument list in functions declared
4528 without a prototype.
4529 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4530 Avoid checking when arguments are null.
4532 2018-01-03 Martin Sebor <msebor@redhat.com>
4535 * doc/extend.texi (attribute const): Fix a typo.
4536 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4537 issuing -Wsuggest-attribute for void functions.
4539 2018-01-03 Martin Sebor <msebor@redhat.com>
4541 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4542 offset_int::from instead of wide_int::to_shwi.
4543 (maybe_diag_overlap): Remove assertion.
4544 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4545 * gimple-ssa-sprintf.c (format_directive): Same.
4546 (parse_directive): Same.
4547 (sprintf_dom_walker::compute_format_length): Same.
4548 (try_substitute_return_value): Same.
4550 2017-01-03 Jeff Law <law@redhat.com>
4553 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4554 non-constant residual for zero at runtime and avoid probing in
4555 that case. Reorganize code for trailing problem to mirror handling
4558 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4560 PR tree-optimization/83501
4561 * tree-ssa-strlen.c (get_string_cst): New.
4562 (handle_char_store): Call get_string_cst.
4564 2018-01-03 Martin Liska <mliska@suse.cz>
4566 PR tree-optimization/83593
4567 * tree-ssa-strlen.c: Include tree-cfg.h.
4568 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4569 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4570 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4572 (strlen_dom_walker::before_dom_children): Call
4573 gimple_purge_dead_eh_edges. Dump tranformation with details
4575 (strlen_dom_walker::before_dom_children): Update call by adding
4576 new argument cleanup_eh.
4577 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4579 2018-01-03 Martin Liska <mliska@suse.cz>
4582 * cif-code.def (VARIADIC_THUNK): New enum value.
4583 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4586 2018-01-03 Jan Beulich <jbeulich@suse.com>
4588 * sse.md (mov<mode>_internal): Tighten condition for when to use
4589 vmovdqu<ssescalarsize> for TI and OI modes.
4591 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4593 Update copyright years.
4595 2018-01-03 Martin Liska <mliska@suse.cz>
4598 * ipa-visibility.c (function_and_variable_visibility): Skip
4599 functions with noipa attribure.
4601 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4603 * gcc.c (process_command): Update copyright notice dates.
4604 * gcov-dump.c (print_version): Ditto.
4605 * gcov.c (print_version): Ditto.
4606 * gcov-tool.c (print_version): Ditto.
4607 * gengtype.c (create_file): Ditto.
4608 * doc/cpp.texi: Bump @copying's copyright year.
4609 * doc/cppinternals.texi: Ditto.
4610 * doc/gcc.texi: Ditto.
4611 * doc/gccint.texi: Ditto.
4612 * doc/gcov.texi: Ditto.
4613 * doc/install.texi: Ditto.
4614 * doc/invoke.texi: Ditto.
4616 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4618 * vector-builder.h (vector_builder::m_full_nelts): Change from
4619 unsigned int to poly_uint64.
4620 (vector_builder::full_nelts): Update prototype accordingly.
4621 (vector_builder::new_vector): Likewise.
4622 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4623 (vector_builder::operator ==): Likewise.
4624 (vector_builder::finalize): Likewise.
4625 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4626 Take the number of elements as a poly_uint64 rather than an
4628 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4629 from unsigned int to poly_uint64.
4630 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4631 (vec_perm_indices::new_vector): Likewise.
4632 (vec_perm_indices::length): Likewise.
4633 (vec_perm_indices::nelts_per_input): Likewise.
4634 (vec_perm_indices::input_nelts): Likewise.
4635 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4636 number of elements per input as a poly_uint64 rather than an
4637 unsigned int. Use the original encoding for variable-length
4638 vectors, rather than clamping each individual element.
4639 For the second and subsequent elements in each pattern,
4640 clamp the step and base before clamping their sum.
4641 (vec_perm_indices::series_p): Handle polynomial element counts.
4642 (vec_perm_indices::all_in_range_p): Likewise.
4643 (vec_perm_indices_to_tree): Likewise.
4644 (vec_perm_indices_to_rtx): Likewise.
4645 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4646 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4647 (tree_vector_builder::new_binary_operation): Handle polynomial
4648 element counts. Return false if we need to know the number
4649 of elements at compile time.
4650 * fold-const.c (fold_vec_perm): Punt if the number of elements
4651 isn't known at compile time.
4653 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4655 * vec-perm-indices.h (vec_perm_builder): Change element type
4656 from HOST_WIDE_INT to poly_int64.
4657 (vec_perm_indices::element_type): Update accordingly.
4658 (vec_perm_indices::clamp): Handle polynomial element_types.
4659 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4660 (vec_perm_indices::all_in_range_p): Likewise.
4661 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4663 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4664 polynomial vec_perm_indices element types.
4665 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4666 * fold-const.c (fold_vec_perm): Likewise.
4667 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4668 * tree-vect-generic.c (lower_vec_perm): Likewise.
4669 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4670 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4671 element type to HOST_WIDE_INT.
4673 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4674 Alan Hayward <alan.hayward@arm.com>
4675 David Sherwood <david.sherwood@arm.com>
4677 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4678 rather than an int. Use plus_constant.
4679 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4680 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4682 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4683 Alan Hayward <alan.hayward@arm.com>
4684 David Sherwood <david.sherwood@arm.com>
4686 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4687 a HOST_WIDE_INT to a poly_int64.
4689 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4690 Alan Hayward <alan.hayward@arm.com>
4691 David Sherwood <david.sherwood@arm.com>
4693 * calls.c (load_register_parameters): Cope with polynomial
4694 mode sizes. Require a constant size for BLKmode parameters
4695 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4696 forces a parameter to be padded at the lsb end in order to
4697 fill a complete number of words, require the parameter size
4698 to be ordered wrt UNITS_PER_WORD.
4700 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4701 Alan Hayward <alan.hayward@arm.com>
4702 David Sherwood <david.sherwood@arm.com>
4704 * reload1.c (spill_stack_slot_width): Change element type
4705 from unsigned int to poly_uint64_pod.
4706 (alter_reg): Treat mode sizes as polynomial.
4708 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4709 Alan Hayward <alan.hayward@arm.com>
4710 David Sherwood <david.sherwood@arm.com>
4712 * reload.c (complex_word_subreg_p): New function.
4713 (reload_inner_reg_of_subreg, push_reload): Use it.
4715 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4716 Alan Hayward <alan.hayward@arm.com>
4717 David Sherwood <david.sherwood@arm.com>
4719 * lra-constraints.c (process_alt_operands): Reject matched
4720 operands whose sizes aren't ordered.
4721 (match_reload): Refer to this check here.
4723 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4724 Alan Hayward <alan.hayward@arm.com>
4725 David Sherwood <david.sherwood@arm.com>
4727 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4728 that the mode size is in the set {1, 2, 4, 8, 16}.
4730 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4731 Alan Hayward <alan.hayward@arm.com>
4732 David Sherwood <david.sherwood@arm.com>
4734 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4735 Use plus_constant instead of gen_rtx_PLUS.
4737 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4738 Alan Hayward <alan.hayward@arm.com>
4739 David Sherwood <david.sherwood@arm.com>
4741 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4742 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4743 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4744 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4745 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4746 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4747 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4748 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4749 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4750 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4752 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4753 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4754 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4755 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4756 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4757 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4758 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4759 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
4760 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
4761 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
4763 * expr.c (emit_move_resolve_push): Treat the input and result
4764 of PUSH_ROUNDING as a poly_int64.
4765 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
4766 (emit_push_insn): Likewise.
4767 * lra-eliminations.c (mark_not_eliminable): Likewise.
4768 * recog.c (push_operand): Likewise.
4769 * reload1.c (elimination_effects): Likewise.
4770 * rtlanal.c (nonzero_bits1): Likewise.
4771 * calls.c (store_one_arg): Likewise. Require the padding to be
4772 known at compile time.
4774 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4775 Alan Hayward <alan.hayward@arm.com>
4776 David Sherwood <david.sherwood@arm.com>
4778 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
4779 Use plus_constant instead of gen_rtx_PLUS.
4781 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4782 Alan Hayward <alan.hayward@arm.com>
4783 David Sherwood <david.sherwood@arm.com>
4785 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
4788 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4789 Alan Hayward <alan.hayward@arm.com>
4790 David Sherwood <david.sherwood@arm.com>
4792 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
4793 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
4794 via stack temporaries. Treat the mode size as polynomial too.
4796 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4797 Alan Hayward <alan.hayward@arm.com>
4798 David Sherwood <david.sherwood@arm.com>
4800 * expr.c (expand_expr_real_2): When handling conversions involving
4801 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
4802 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
4803 as a poly_uint64 too.
4805 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4806 Alan Hayward <alan.hayward@arm.com>
4807 David Sherwood <david.sherwood@arm.com>
4809 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
4811 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4812 Alan Hayward <alan.hayward@arm.com>
4813 David Sherwood <david.sherwood@arm.com>
4815 * combine.c (can_change_dest_mode): Handle polynomial
4816 REGMODE_NATURAL_SIZE.
4817 * expmed.c (store_bit_field_1): Likewise.
4818 * expr.c (store_constructor): Likewise.
4819 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
4820 and polynomial REGMODE_NATURAL_SIZE.
4821 (gen_lowpart_common): Likewise.
4822 * reginfo.c (record_subregs_of_mode): Likewise.
4823 * rtlanal.c (read_modify_subreg_p): Likewise.
4825 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4826 Alan Hayward <alan.hayward@arm.com>
4827 David Sherwood <david.sherwood@arm.com>
4829 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
4830 numbers of elements.
4832 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4833 Alan Hayward <alan.hayward@arm.com>
4834 David Sherwood <david.sherwood@arm.com>
4836 * match.pd: Cope with polynomial numbers of vector elements.
4838 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4839 Alan Hayward <alan.hayward@arm.com>
4840 David Sherwood <david.sherwood@arm.com>
4842 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
4843 in a POINTER_PLUS_EXPR.
4845 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4846 Alan Hayward <alan.hayward@arm.com>
4847 David Sherwood <david.sherwood@arm.com>
4849 * omp-simd-clone.c (simd_clone_subparts): New function.
4850 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
4851 (ipa_simd_modify_function_body): Likewise.
4853 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4854 Alan Hayward <alan.hayward@arm.com>
4855 David Sherwood <david.sherwood@arm.com>
4857 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
4858 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
4859 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
4860 (expand_vector_condition, vector_element): Likewise.
4861 (subparts_gt): New function.
4862 (get_compute_type): Use subparts_gt.
4863 (count_type_subparts): Delete.
4864 (expand_vector_operations_1): Use subparts_gt instead of
4865 count_type_subparts.
4867 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4868 Alan Hayward <alan.hayward@arm.com>
4869 David Sherwood <david.sherwood@arm.com>
4871 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
4872 (vect_compile_time_alias): ...this new function. Do the calculation
4873 on poly_ints rather than trees.
4874 (vect_prune_runtime_alias_test_list): Update call accordingly.
4876 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4877 Alan Hayward <alan.hayward@arm.com>
4878 David Sherwood <david.sherwood@arm.com>
4880 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
4882 (vect_schedule_slp_instance): Likewise.
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 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
4889 constant and extern definitions for variable-length vectors.
4890 (vect_get_constant_vectors): Note that the number of units
4891 is known to be constant.
4893 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4894 Alan Hayward <alan.hayward@arm.com>
4895 David Sherwood <david.sherwood@arm.com>
4897 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
4898 of units as polynomial. Choose between WIDE and NARROW based
4901 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4902 Alan Hayward <alan.hayward@arm.com>
4903 David Sherwood <david.sherwood@arm.com>
4905 * tree-vect-stmts.c (simd_clone_subparts): New function.
4906 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
4908 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4909 Alan Hayward <alan.hayward@arm.com>
4910 David Sherwood <david.sherwood@arm.com>
4912 * tree-vect-stmts.c (vectorizable_call): Treat the number of
4913 vectors as polynomial. Use build_index_vector for
4916 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4917 Alan Hayward <alan.hayward@arm.com>
4918 David Sherwood <david.sherwood@arm.com>
4920 * tree-vect-stmts.c (get_load_store_type): Treat the number of
4921 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
4922 for variable-length vectors.
4923 (vectorizable_mask_load_store): Treat the number of units as
4924 polynomial, asserting that it is constant if the condition has
4925 already been enforced.
4926 (vectorizable_store, vectorizable_load): Likewise.
4928 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4929 Alan Hayward <alan.hayward@arm.com>
4930 David Sherwood <david.sherwood@arm.com>
4932 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
4933 of units as polynomial. Punt if we can't tell at compile time
4934 which vector contains the final result.
4936 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4937 Alan Hayward <alan.hayward@arm.com>
4938 David Sherwood <david.sherwood@arm.com>
4940 * tree-vect-loop.c (vectorizable_induction): Treat the number
4941 of units as polynomial. Punt on SLP inductions. Use an integer
4942 VEC_SERIES_EXPR for variable-length integer reductions. Use a
4943 cast of such a series for variable-length floating-point
4946 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4947 Alan Hayward <alan.hayward@arm.com>
4948 David Sherwood <david.sherwood@arm.com>
4950 * tree.h (build_index_vector): Declare.
4951 * tree.c (build_index_vector): New function.
4952 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
4953 of units as polynomial, forcibly converting it to a constant if
4954 vectorizable_reduction has already enforced the condition.
4955 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
4956 to create a {1,2,3,...} vector.
4957 (vectorizable_reduction): Treat the number of units as polynomial.
4958 Choose vectype_in based on the largest scalar element size rather
4959 than the smallest number of units. Enforce the restrictions
4962 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4963 Alan Hayward <alan.hayward@arm.com>
4964 David Sherwood <david.sherwood@arm.com>
4966 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
4967 number of units as polynomial.
4969 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4970 Alan Hayward <alan.hayward@arm.com>
4971 David Sherwood <david.sherwood@arm.com>
4973 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
4974 * target.def (autovectorize_vector_sizes): Return the vector sizes
4975 by pointer, using vector_sizes rather than a bitmask.
4976 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
4977 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
4978 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
4980 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
4981 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
4982 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
4983 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
4984 * omp-general.c (omp_max_vf): Likewise.
4985 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4986 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
4987 * tree-vect-loop.c (vect_analyze_loop): Likewise.
4988 * tree-vect-slp.c (vect_slp_bb): Likewise.
4989 * doc/tm.texi: Regenerate.
4990 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
4992 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
4993 the vector size as a poly_uint64 rather than an unsigned int.
4994 (current_vector_size): Change from an unsigned int to a poly_uint64.
4995 (get_vectype_for_scalar_type): Update accordingly.
4996 * tree.h (build_truth_vector_type): Take the size and number of
4997 units as a poly_uint64 rather than an unsigned int.
4998 (build_vector_type): Add a temporary overload that takes
4999 the number of units as a poly_uint64 rather than an unsigned int.
5000 * tree.c (make_vector_type): Likewise.
5001 (build_truth_vector_type): Take the number of units as a poly_uint64
5002 rather than an unsigned int.
5004 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5005 Alan Hayward <alan.hayward@arm.com>
5006 David Sherwood <david.sherwood@arm.com>
5008 * target.def (get_mask_mode): Take the number of units and length
5009 as poly_uint64s rather than unsigned ints.
5010 * targhooks.h (default_get_mask_mode): Update accordingly.
5011 * targhooks.c (default_get_mask_mode): Likewise.
5012 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5013 * doc/tm.texi: Regenerate.
5015 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5016 Alan Hayward <alan.hayward@arm.com>
5017 David Sherwood <david.sherwood@arm.com>
5019 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5020 * omp-general.c (omp_max_vf): Likewise.
5021 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5022 (expand_omp_simd): Handle polynomial safelen.
5023 * omp-low.c (omplow_simd_context): Add a default constructor.
5024 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5025 (lower_rec_simd_input_clauses): Update accordingly.
5026 (lower_rec_input_clauses): Likewise.
5028 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5029 Alan Hayward <alan.hayward@arm.com>
5030 David Sherwood <david.sherwood@arm.com>
5032 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5033 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5034 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5035 (vect_analyze_slp_cost): Likewise.
5036 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5037 (vect_model_load_cost): Likewise.
5039 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5040 Alan Hayward <alan.hayward@arm.com>
5041 David Sherwood <david.sherwood@arm.com>
5043 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5044 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5045 from an unsigned int * to a poly_uint64_pod *.
5046 (calculate_unrolling_factor): New function.
5047 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5049 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5050 Alan Hayward <alan.hayward@arm.com>
5051 David Sherwood <david.sherwood@arm.com>
5053 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5054 from an unsigned int to a poly_uint64.
5055 (_loop_vec_info::slp_unrolling_factor): Likewise.
5056 (_loop_vec_info::vectorization_factor): Change from an int
5058 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5059 (vect_get_num_vectors): New function.
5060 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5061 (vect_get_num_copies): Use vect_get_num_vectors.
5062 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5063 to an unsigned int *.
5064 (vect_analyze_data_refs): Change min_vf from an int * to a
5066 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5067 than an unsigned HOST_WIDE_INT.
5068 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5069 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5070 to an unsigned int *.
5071 (vect_analyze_data_ref_dependences): Likewise.
5072 (vect_compute_data_ref_alignment): Handle polynomial vf.
5073 (vect_enhance_data_refs_alignment): Likewise.
5074 (vect_prune_runtime_alias_test_list): Likewise.
5075 (vect_shift_permute_load_chain): Likewise.
5076 (vect_supportable_dr_alignment): Likewise.
5077 (dependence_distance_ge_vf): Take the vectorization factor as a
5078 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5079 (vect_analyze_data_refs): Change min_vf from an int * to a
5081 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5082 vfm1 as a poly_uint64 rather than an int. Make the same change
5083 for the returned bound_scalar.
5084 (vect_gen_vector_loop_niters): Handle polynomial vf.
5085 (vect_do_peeling): Likewise. Update call to
5086 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5087 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5089 * tree-vect-loop.c (vect_determine_vectorization_factor)
5090 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5091 (vect_get_known_peeling_cost): Likewise.
5092 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5093 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5094 (vect_transform_loop): Likewise. Use the lowest possible VF when
5095 updating the upper bounds of the loop.
5096 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5098 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5099 polynomial unroll factors.
5100 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5101 (vect_make_slp_decision): Likewise.
5102 (vect_supported_load_permutation_p): Likewise, and polynomial
5104 (vect_analyze_slp_cost): Handle polynomial vf.
5105 (vect_slp_analyze_node_operations): Likewise.
5106 (vect_slp_analyze_bb_1): Likewise.
5107 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5108 than an unsigned HOST_WIDE_INT.
5109 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5110 (vectorizable_load): Handle polynomial vf.
5111 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5113 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5115 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5116 Alan Hayward <alan.hayward@arm.com>
5117 David Sherwood <david.sherwood@arm.com>
5119 * match.pd: Handle bit operations involving three constants
5120 and try to fold one pair.
5122 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5124 * tree-vect-loop-manip.c: Include gimple-fold.h.
5125 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5126 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5127 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5128 Add a path that uses a step of VF instead of 1, but disable it
5130 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5131 and niters_no_overflow parameters. Update calls to
5132 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5133 Create a new SSA name if the latter choses to use a ste other
5134 than zero, and return it via niters_vector_mult_vf_var.
5135 * tree-vect-loop.c (vect_transform_loop): Update calls to
5136 vect_do_peeling, vect_gen_vector_loop_niters and
5137 slpeel_make_loop_iterate_ntimes.
5138 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5139 (vect_gen_vector_loop_niters): Update declarations after above changes.
5141 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5143 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5144 128-bit round to integer instructions.
5145 (ceil<mode>2): Likewise.
5146 (btrunc<mode>2): Likewise.
5147 (round<mode>2): Likewise.
5149 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5151 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5152 unaligned VSX load/store on P8/P9.
5153 (expand_block_clear): Allow the use of unaligned VSX
5154 load/store on P8/P9.
5156 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5158 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5160 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5161 swap associated with both a load and a store.
5163 2018-01-02 Andrew Waterman <andrew@sifive.com>
5165 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5166 * config/riscv/riscv.md (clear_cache): Use it.
5168 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5170 * web.c: Remove out-of-date comment.
5172 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5174 * expr.c (fixup_args_size_notes): Check that any existing
5175 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5176 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5177 (emit_single_push_insn): ...here.
5179 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5181 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5182 (const_vector_encoded_nelts): New function.
5183 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5184 (const_vector_int_elt, const_vector_elt): Declare.
5185 * emit-rtl.c (const_vector_int_elt_1): New function.
5186 (const_vector_elt): Likewise.
5187 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5188 of CONST_VECTOR_ELT.
5190 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5192 * expr.c: Include rtx-vector-builder.h.
5193 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5194 directly on the tree encoding.
5195 (const_vector_from_tree): Likewise.
5196 * optabs.c: Include rtx-vector-builder.h.
5197 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5198 sequence of "u" values.
5199 * vec-perm-indices.c: Include rtx-vector-builder.h.
5200 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5201 directly on the vec_perm_indices encoding.
5203 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5205 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5206 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5207 * rtx-vector-builder.h: New file.
5208 * rtx-vector-builder.c: Likewise.
5209 * rtl.h (rtx_def::u2): Add a const_vector field.
5210 (CONST_VECTOR_NPATTERNS): New macro.
5211 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5212 (CONST_VECTOR_DUPLICATE_P): Likewise.
5213 (CONST_VECTOR_STEPPED_P): Likewise.
5214 (CONST_VECTOR_ENCODED_ELT): Likewise.
5215 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5216 (unwrap_const_vec_duplicate): Likewise.
5217 (const_vec_series_p): Check for a non-duplicated vector encoding.
5218 Say that the function only returns true for integer vectors.
5219 * emit-rtl.c: Include rtx-vector-builder.h.
5220 (gen_const_vec_duplicate_1): Delete.
5221 (gen_const_vector): Call gen_const_vec_duplicate instead of
5222 gen_const_vec_duplicate_1.
5223 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5224 (gen_const_vec_duplicate): Use rtx_vector_builder.
5225 (gen_const_vec_series): Likewise.
5226 (gen_rtx_CONST_VECTOR): Likewise.
5227 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5228 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5229 Build a new vector rather than modifying a CONST_VECTOR in-place.
5230 (handle_special_swappables): Update call accordingly.
5231 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5232 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5233 Build a new vector rather than modifying a CONST_VECTOR in-place.
5234 (handle_special_swappables): Update call accordingly.
5236 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5238 * simplify-rtx.c (simplify_const_binary_operation): Use
5239 CONST_VECTOR_ELT instead of XVECEXP.
5241 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5243 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5244 the selector elements to be different from the data elements
5245 if the selector is a VECTOR_CST.
5246 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5247 ssizetype for the selector.
5249 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5251 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5252 before testing each element individually.
5253 * tree-vect-generic.c (lower_vec_perm): Likewise.
5255 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5257 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5258 * selftest-run-tests.c (selftest::run_tests): Call it.
5259 * vector-builder.h (vector_builder::operator ==): New function.
5260 (vector_builder::operator !=): Likewise.
5261 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5262 (vec_perm_indices::all_from_input_p): New function.
5263 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5264 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5265 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5266 instead of reading the VECTOR_CST directly. Detect whether both
5267 vector inputs are the same before constructing the vec_perm_indices,
5268 and update the number of inputs argument accordingly. Use the
5269 utility functions added above. Only construct sel2 if we need to.
5271 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5273 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5274 the broadcast of the low byte.
5275 (expand_mult_highpart): Use an explicit encoding for the permutes.
5276 * optabs-query.c (can_mult_highpart_p): Likewise.
5277 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5278 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5279 (vectorizable_bswap): Likewise.
5280 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5281 explicit encoding for the power-of-2 permutes.
5282 (vect_permute_store_chain): Likewise.
5283 (vect_grouped_load_supported): Likewise.
5284 (vect_permute_load_chain): Likewise.
5286 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5288 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5289 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5290 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5291 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5292 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5293 (vect_gen_perm_mask_any): Likewise.
5295 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5297 * int-vector-builder.h: New file.
5298 * vec-perm-indices.h: Include int-vector-builder.h.
5299 (vec_perm_indices): Redefine as an int_vector_builder.
5300 (auto_vec_perm_indices): Delete.
5301 (vec_perm_builder): Redefine as a stand-alone class.
5302 (vec_perm_indices::vec_perm_indices): New function.
5303 (vec_perm_indices::clamp): Likewise.
5304 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5305 (vec_perm_indices::new_vector): New function.
5306 (vec_perm_indices::new_expanded_vector): Update for new
5307 vec_perm_indices class.
5308 (vec_perm_indices::rotate_inputs): New function.
5309 (vec_perm_indices::all_in_range_p): Operate directly on the
5310 encoded form, without computing elided elements.
5311 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5312 encoding. Update for new vec_perm_indices class.
5313 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5314 the given vec_perm_builder.
5315 (expand_vec_perm_var): Update vec_perm_builder constructor.
5316 (expand_mult_highpart): Use vec_perm_builder instead of
5317 auto_vec_perm_indices.
5318 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5319 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5320 or double series encoding as appropriate.
5321 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5322 vec_perm_indices instead of auto_vec_perm_indices.
5323 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5324 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5325 (vect_permute_store_chain): Likewise.
5326 (vect_grouped_load_supported): Likewise.
5327 (vect_permute_load_chain): Likewise.
5328 (vect_shift_permute_load_chain): Likewise.
5329 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5330 (vect_transform_slp_perm_load): Likewise.
5331 (vect_schedule_slp_instance): Likewise.
5332 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5333 (vectorizable_mask_load_store): Likewise.
5334 (vectorizable_bswap): Likewise.
5335 (vectorizable_store): Likewise.
5336 (vectorizable_load): Likewise.
5337 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5338 vec_perm_indices instead of auto_vec_perm_indices. Use
5339 tree_to_vec_perm_builder to read the vector from a tree.
5340 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5341 vec_perm_builder instead of a vec_perm_indices.
5342 (have_whole_vector_shift): Use vec_perm_builder and
5343 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5344 truncation to calc_vec_perm_mask_for_shift.
5345 (vect_create_epilog_for_reduction): Likewise.
5346 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5347 from auto_vec_perm_indices to vec_perm_indices.
5348 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5349 instead of changing individual elements.
5350 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5351 the vector in d.perm.
5352 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5353 from auto_vec_perm_indices to vec_perm_indices.
5354 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5355 instead of changing individual elements.
5356 (arm_vectorize_vec_perm_const): Use new_vector to install
5357 the vector in d.perm.
5358 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5359 Update vec_perm_builder constructor.
5360 (rs6000_expand_interleave): Likewise.
5361 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5362 (rs6000_expand_interleave): Likewise.
5364 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5366 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5367 to qimode could truncate the indices.
5368 * optabs.c (expand_vec_perm_var): Likewise.
5370 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5372 * Makefile.in (OBJS): Add vec-perm-indices.o.
5373 * vec-perm-indices.h: New file.
5374 * vec-perm-indices.c: Likewise.
5375 * target.h (vec_perm_indices): Replace with a forward class
5377 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5378 * optabs.h: Include vec-perm-indices.h.
5379 (expand_vec_perm): Delete.
5380 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5381 (expand_vec_perm_const): Declare.
5382 * target.def (vec_perm_const_ok): Replace with...
5383 (vec_perm_const): ...this new hook.
5384 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5385 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5386 * doc/tm.texi: Regenerate.
5387 * optabs.def (vec_perm_const): Delete.
5388 * doc/md.texi (vec_perm_const): Likewise.
5389 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5390 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5391 expand_vec_perm for constant permutation vectors. Assert that
5392 the mode of variable permutation vectors is the integer equivalent
5393 of the mode that is being permuted.
5394 * optabs-query.h (selector_fits_mode_p): Declare.
5395 * optabs-query.c: Include vec-perm-indices.h.
5396 (selector_fits_mode_p): New function.
5397 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5398 is defined, instead of checking whether the vec_perm_const_optab
5399 exists. Use targetm.vectorize.vec_perm_const instead of
5400 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5401 fit in the vector mode before using a variable permute.
5402 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5403 vec_perm_indices instead of an rtx.
5404 (expand_vec_perm): Replace with...
5405 (expand_vec_perm_const): ...this new function. Take the selector
5406 as a vec_perm_indices rather than an rtx. Also take the mode of
5407 the selector. Update call to shift_amt_for_vec_perm_mask.
5408 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5409 Use vec_perm_indices::new_expanded_vector to expand the original
5410 selector into bytes. Check whether the indices fit in the vector
5411 mode before using a variable permute.
5412 (expand_vec_perm_var): Make global.
5413 (expand_mult_highpart): Use expand_vec_perm_const.
5414 * fold-const.c: Includes vec-perm-indices.h.
5415 * tree-ssa-forwprop.c: Likewise.
5416 * tree-vect-data-refs.c: Likewise.
5417 * tree-vect-generic.c: Likewise.
5418 * tree-vect-loop.c: Likewise.
5419 * tree-vect-slp.c: Likewise.
5420 * tree-vect-stmts.c: Likewise.
5421 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5423 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5424 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5425 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5426 (aarch64_vectorize_vec_perm_const): ...this new function.
5427 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5428 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5429 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5430 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5431 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5432 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5433 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5435 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5436 check for NEON modes.
5437 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5438 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5439 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5440 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5442 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5443 the old VEC_PERM_CONST conditions.
5444 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5445 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5446 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5447 (ia64_vectorize_vec_perm_const_ok): Merge into...
5448 (ia64_vectorize_vec_perm_const): ...this new function.
5449 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5450 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5451 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5452 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5453 * config/mips/mips.c (mips_expand_vec_perm_const)
5454 (mips_vectorize_vec_perm_const_ok): Merge into...
5455 (mips_vectorize_vec_perm_const): ...this new function.
5456 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5457 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5458 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5459 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5460 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5461 (rs6000_expand_vec_perm_const): Delete.
5462 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5464 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5465 (altivec_expand_vec_perm_const_le): Take each operand individually.
5466 Operate on constant selectors rather than rtxes.
5467 (altivec_expand_vec_perm_const): Likewise. Update call to
5468 altivec_expand_vec_perm_const_le.
5469 (rs6000_expand_vec_perm_const): Delete.
5470 (rs6000_vectorize_vec_perm_const_ok): Delete.
5471 (rs6000_vectorize_vec_perm_const): New function.
5472 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5473 an element count and rtx array.
5474 (rs6000_expand_extract_even): Update call accordingly.
5475 (rs6000_expand_interleave): Likewise.
5476 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5477 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5478 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5479 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5480 (rs6000_expand_vec_perm_const): Delete.
5481 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5482 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5483 (altivec_expand_vec_perm_const_le): Take each operand individually.
5484 Operate on constant selectors rather than rtxes.
5485 (altivec_expand_vec_perm_const): Likewise. Update call to
5486 altivec_expand_vec_perm_const_le.
5487 (rs6000_expand_vec_perm_const): Delete.
5488 (rs6000_vectorize_vec_perm_const_ok): Delete.
5489 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5490 reference to the SPE evmerge intructions.
5491 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5492 an element count and rtx array.
5493 (rs6000_expand_extract_even): Update call accordingly.
5494 (rs6000_expand_interleave): Likewise.
5495 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5496 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5498 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5500 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5502 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5503 vector mode and that that mode matches the mode of the data
5505 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5506 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5507 directly using expand_vec_perm_1 when forcing selectors into
5509 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5511 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5513 * optabs-query.h (can_vec_perm_p): Delete.
5514 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5515 * optabs-query.c (can_vec_perm_p): Split into...
5516 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5517 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5518 particular selector is valid.
5519 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5520 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5521 (vect_grouped_load_supported): Likewise.
5522 (vect_shift_permute_load_chain): Likewise.
5523 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5524 (vect_transform_slp_perm_load): Likewise.
5525 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5526 (vectorizable_bswap): Likewise.
5527 (vect_gen_perm_mask_checked): Likewise.
5528 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5529 implementations of variable permutation vectors into account
5530 when deciding which selector to use.
5531 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5532 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5533 with a false third argument.
5534 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5535 to test whether the constant selector is valid and can_vec_perm_var_p
5536 to test whether a variable selector is valid.
5538 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5540 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5541 * optabs-query.c (can_vec_perm_p): Likewise.
5542 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5543 instead of vec_perm_indices.
5544 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5545 (vect_gen_perm_mask_checked): Likewise,
5546 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5547 (vect_gen_perm_mask_checked): Likewise,
5549 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5551 * optabs-query.h (qimode_for_vec_perm): Declare.
5552 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5553 (qimode_for_vec_perm): ...this new function.
5554 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5556 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5558 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5559 does not have a conditional at the top.
5561 2018-01-02 Richard Biener <rguenther@suse.de>
5563 * ipa-inline.c (big_speedup_p): Fix expression.
5565 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5568 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5571 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5575 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5576 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5577 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5578 cond_taken_branch_cost 3->4.
5580 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5582 PR tree-optimization/83581
5583 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5584 TODO_cleanup_cfg if any changes have been made.
5587 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5588 convert_modes if target mode has the right side, but different mode
5592 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5593 last argument when extracting from CONCAT. If either from_real or
5594 from_imag is NULL, use expansion through memory. If result is not
5595 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5596 the parts directly to inner mode, if even that fails, use expansion
5600 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5601 check for bswap in mode rather than HImode and use that in expand_unop
5604 Copyright (C) 2018 Free Software Foundation, Inc.
5606 Copying and distribution of this file, with or without modification,
5607 are permitted in any medium without royalty provided the copyright
5608 notice and this notice are preserved.