1 2018-01-26 Jim Wilson <jimw@sifive.com>
3 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
6 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
8 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
9 and CMP + SUB-immediate -> SUBS.
11 2018-01-26 Martin Sebor <msebor@redhat.com>
13 PR tree-optimization/83896
14 * tree-ssa-strlen.c (get_string_len): Rename...
15 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
16 Avoid assuming length is constant.
17 (handle_char_store): Use HOST_WIDE_INT for string length.
19 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
22 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
23 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
25 2018-01-26 Richard Biener <rguenther@suse.de>
27 PR rtl-optimization/84003
28 * dse.c (record_store): Only record redundant stores when
29 the earlier store aliases at least all accesses the later one does.
31 2018-01-26 Jakub Jelinek <jakub@redhat.com>
33 PR rtl-optimization/83985
34 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
35 REG_CFA_RESTORE insns.
36 (delete_unmarked_insns): Don't ignore separate shrink wrapping
37 REG_CFA_RESTORE insns here.
40 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
41 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
43 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
45 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
46 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
48 (arc_override_options): Likewise.
49 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
51 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
53 * config/arc/arc.h (TARGET_DBNZ): Define.
54 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
55 properly set the tune attribute.
56 (dbnz): Use TARGET_DBNZ guard.
57 * config/arc/arc.opt (mtune): Add core3 option.
59 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
61 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
62 recognize new pic like addresses.
63 (arc_delegitimize_address): Clean up.
65 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
67 * config/arc/arc-arches.def: Option mrf16 valid for all
69 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
70 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
71 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
72 * config/arc/arc-tables.opt: Regenerate.
73 * config/arc/arc.c (arc_conditional_register_usage): Handle
74 reduced register file case.
75 (arc_file_start): Set must have build attributes.
76 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
78 * config/arc/arc.opt (mrf16): Add new option.
79 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
80 * config/arc/genmultilib.awk: Handle new mrf16 option.
81 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
82 * config/arc/t-multilib: Regenerate.
83 * doc/invoke.texi (ARC Options): Document mrf16 option.
85 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
87 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
88 * config/arc/arc.c (arc_handle_secure_attribute): New function.
89 (arc_attribute_table): Add 'secure_call' attribute.
90 (arc_print_operand): Print secure call operand.
91 (arc_function_ok_for_sibcall): Don't optimize tail calls when
93 (arc_is_secure_call_p): New function. * config/arc/arc.md
94 (call_i): Add support for sjli instruction.
95 (call_value_i): Likewise.
96 * config/arc/constraints.md (Csc): New constraint.
98 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
99 John Eric Martin <John.Martin@emmicro-us.com>
101 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
102 * config/arc/arc.c (_arc_jli_section): New struct.
103 (arc_jli_section): New type.
104 (rc_jli_sections): New static variable.
105 (arc_handle_jli_attribute): New function.
106 (arc_attribute_table): Add jli_always and jli_fixed attribute.
107 (arc_file_end): New function.
108 (TARGET_ASM_FILE_END): Define.
109 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
110 (arc_add_jli_section): New function.
111 (jli_call_scan): Likewise.
112 (arc_reorg): Call jli_call_scan.
113 (arc_output_addsi): Remove 'S' from printing asm operand.
114 (arc_is_jli_call_p): New function.
115 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
117 (movhi_insn): Likewise.
118 (movsi_insn): Likewise.
119 (movsi_set_cc_insn): Likewise.
120 (loadqi_update): Likewise.
121 (load_zeroextendqisi_update): Likewise.
122 (load_signextendqisi_update): Likewise.
123 (loadhi_update): Likewise.
124 (load_zeroextendhisi_update): Likewise.
125 (load_signextendhisi_update): Likewise.
126 (loadsi_update): Likewise.
127 (loadsf_update): Likewise.
128 (movsicc_insn): Likewise.
129 (bset_insn): Likewise.
130 (bxor_insn): Likewise.
131 (bclr_insn): Likewise.
132 (bmsk_insn): Likewise.
133 (bicsi3_insn): Likewise.
134 (cmpsi_cc_c_insn): Likewise.
135 (movsi_ne): Likewise.
136 (movsi_cond_exec): Likewise.
137 (clrsbsi2): Likewise.
148 (call_i): Remove 'S' asm letter, add jli instruction.
149 (call_value_i): Likewise.
150 * config/arc/arc.op (mjli-always): New option.
151 * config/arc/constraints.md (Cji): New constraint.
152 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
154 (subsf3_fpx): Likewise.
155 (mulsf3_fpx): Likewise.
156 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
158 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
160 * doc/invoke.texi (ARC): Document mjli-always option.
162 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
164 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
165 and use incw and decw where possible
166 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
168 2018-01-26 Richard Biener <rguenther@suse.de>
170 PR tree-optimization/81082
171 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
172 association if it requires casting to unsigned.
173 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
174 from fold_plusminus_mult_expr to catch important cases late when
175 range info is available.
177 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
179 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
180 * configure.ac (hidden_linkonce): New test.
181 * configure: Regenerate.
182 * config.in: Regenerate.
184 2018-01-26 Julia Koval <julia.koval@intel.com>
186 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
187 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
188 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
189 _mm_mask_bitshuffle_epi64_mask): Fix type.
190 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
191 USI_FTYPE_V4DI_V4DI_USI): Remove.
192 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
193 __builtin_ia32_vpshufbitqmb256_mask,
194 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
195 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
196 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
198 2018-01-26 Alan Modra <amodra@gmail.com>
201 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
202 UNSPEC_VBPERMQ. Sort other unspecs.
204 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
206 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
208 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
211 * predict.c (drop_profile): Do not push/pop cfun; update also
213 (handle_missing_profiles): Fix logic looking for zero profiles.
215 2018-01-25 Jakub Jelinek <jakub@redhat.com>
218 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
219 on functions with #pragma omp declare simd or functions with simd
221 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
222 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
223 Remove trailing \n from warning_at calls.
225 2018-01-25 Tom de Vries <tom@codesourcery.com>
228 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
229 for neutered workers.
231 2018-01-24 Joseph Myers <joseph@codesourcery.com>
234 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
235 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
237 2017-01-08 Jeff Law <law@redhat.com>
240 * i386.c (get_probe_interval): Move to earlier point.
241 (ix86_compute_frame_layout): If -fstack-clash-protection and
242 the frame is larger than the probe interval, then use pushes
243 to save registers rather than reg->mem moves.
244 (ix86_expand_prologue): Remove conditional for int_registers_saved
247 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
250 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
251 min/max for never referenced object.
253 2018-01-24 Jakub Jelinek <jakub@redhat.com>
256 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
258 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
259 attributes from DECL_ATTRIBUTES (decl) without affecting
260 DECL_ATTRIBUTES (current_function_decl).
261 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
262 functions with non-NULL DECL_ABSTRACT_ORIGIN.
264 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
266 PR tree-optimization/83979
267 * fold-const.c (fold_comparison): Use constant_boolean_node
268 instead of boolean_{true,false}_node.
270 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
272 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
275 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
277 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
278 Simplify the clause that sets the length attribute.
279 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
280 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
281 clause that sets the length attribute.
282 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
284 2018-01-24 Tom de Vries <tom@codesourcery.com>
287 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
288 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
289 Add strict parameter.
290 (prevent_branch_around_nothing): Insert dummy insn between branch to
291 label and label with no ptx insn inbetween.
292 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
294 2018-01-24 Tom de Vries <tom@codesourcery.com>
297 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
298 for neutered threads in warp.
299 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
301 2018-01-24 Richard Biener <rguenther@suse.de>
303 PR tree-optimization/83176
304 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
307 2018-01-24 Richard Biener <rguenther@suse.de>
309 PR tree-optimization/82819
310 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
311 code generating pluses that are no-ops in the target precision.
313 2018-01-24 Richard Biener <rguenther@suse.de>
316 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
318 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
320 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
321 to merge probabilities.
322 * predict.c (probably_never_executed): Also mark as cold functions
323 with global 0 profile and guessed local profile.
324 * profile-count.c (profile_probability::combine_with_count): New
326 * profile-count.h (profile_probability::operator*,
327 profile_probability::operator*=, profile_probability::operator/,
328 profile_probability::operator/=): Reduce precision to adjusted
329 and set value to guessed on contradictory divisions.
330 (profile_probability::combine_with_freq): Remove.
331 (profile_probability::combine_wiht_count): Declare.
332 (profile_count::force_nonzero):: Set to adjusted.
333 (profile_count::probability_in):: Set quality to adjusted.
334 * tree-ssa-tail-merge.c (replace_block_by): Use
337 2018-01-23 Andrew Waterman <andrew@sifive.com>
338 Jim Wilson <jimw@sifive.com>
340 * config/riscv/riscv.c (riscv_stack_boundary): New.
341 (riscv_option_override): Set riscv_stack_boundary. Handle
342 riscv_preferred_stack_boundary_arg.
343 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
344 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
345 (STACK_BOUNDARY): Set to riscv_stack_boundary.
346 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
347 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
348 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
350 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
353 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
354 of struct ix86_frame.
355 (ix86_expand_epilogue): Likewise. Add a local variable for
356 the reg_save_offset field in struct ix86_frame.
358 2018-01-23 Bin Cheng <bin.cheng@arm.com>
360 PR tree-optimization/82604
361 * tree-loop-distribution.c (enum partition_kind): New enum item
362 PKIND_PARTIAL_MEMSET.
363 (partition_builtin_p): Support above new enum item.
364 (generate_code_for_partition): Ditto.
365 (compute_access_range): Differentiate cases that equality can be
366 proven at all loops, the innermost loops or no loops.
367 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
368 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
369 (finalize_partitions, distribute_loop): Don't fuse partition of
370 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
371 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
374 2018-01-23 Martin Liska <mliska@suse.cz>
376 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
377 order to ignore the predictor.
378 (PRED_POLYMORPHIC_CALL): Likewise.
379 (PRED_RECURSIVE_CALL): Likewise.
381 2018-01-23 Martin Liska <mliska@suse.cz>
383 * tree-profile.c (tree_profiling): Print function header to
384 aware reader which function we are working on.
385 * value-prof.c (gimple_find_values_to_profile): Do not print
386 not interesting value histograms.
388 2018-01-23 Martin Liska <mliska@suse.cz>
390 * profile-count.h (enum profile_quality): Add
391 profile_uninitialized as the first value. Do not number values
392 as they are zero based.
393 (profile_count::verify): Update sanity check.
394 (profile_probability::verify): Likewise.
396 2018-01-23 Nathan Sidwell <nathan@acm.org>
398 * doc/invoke.texi (ffor-scope): Deprecate.
400 2018-01-23 David Malcolm <dmalcolm@redhat.com>
402 PR tree-optimization/83510
403 * domwalk.c (set_all_edges_as_executable): New function.
404 (dom_walker::dom_walker): Convert bool param
405 "skip_unreachable_blocks" to enum reachability. Move setup of
406 edge flags to set_all_edges_as_executable and only do it when
407 reachability is REACHABLE_BLOCKS.
408 * domwalk.h (enum dom_walker::reachability): New enum.
409 (dom_walker::dom_walker): Convert bool param
410 "skip_unreachable_blocks" to enum reachability.
411 (set_all_edges_as_executable): New decl.
412 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
413 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
415 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
416 but converting true to REACHABLE_BLOCKS.
417 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
419 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
420 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
421 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
423 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
424 if check_all_array_refs will be called.
426 2018-01-23 David Malcolm <dmalcolm@redhat.com>
428 * tree.c (selftest::test_location_wrappers): Add more test
431 2018-01-23 David Malcolm <dmalcolm@redhat.com>
433 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
434 (selftest::test_bit_in_range): Likewise.
436 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
439 * doc/sourcebuild.texi (vect_float): Say that the selector
440 only describes the situation when -funsafe-math-optimizations is on.
441 (vect_float_strict): Document.
443 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
445 PR tree-optimization/83965
446 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
447 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
448 instead of checking only for a reduction.
449 (vect_recog_widen_sum_pattern): Likewise.
451 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
453 * predict.c (probably_never_executed): Only use precise profile info.
454 (compute_function_frequency): Skip after inlining hack since we now
455 have quality checking.
457 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
459 * profile-count.h (profile_probability::very_unlikely,
460 profile_probability::unlikely, profile_probability::even): Set
461 precision to guessed.
463 2018-01-23 Richard Biener <rguenther@suse.de>
465 PR tree-optimization/83963
466 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
467 Properly terminate dominator walk when crossing the exit edge not
468 when visiting its source block.
470 2018-01-23 Jakub Jelinek <jakub@redhat.com>
473 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
474 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
476 2018-01-22 Jakub Jelinek <jakub@redhat.com>
478 PR tree-optimization/83957
479 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
480 semicolon after for body surrounded by braces.
482 PR tree-optimization/83081
483 * profile-count.h (profile_probability::split): New method.
484 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
485 Use profile_probability::split.
486 (do_compare_rtx_and_jump): Fix adjustment of probabilities
487 when splitting a single conditional jump into 2.
489 2018-01-22 David Malcolm <dmalcolm@redhat.com>
491 PR tree-optimization/69452
492 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
495 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
497 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
498 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
499 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
501 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
503 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
504 * config/rl78/rl78.md: New define_expand "movdi"
505 * config/rl78/rl78.c: New function definition rl78_split_movdi
507 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
510 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
512 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
513 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
514 128-bit to produce an UNSPEC move to get the double word with the
515 signbit and then a shift directly to do signbit.
516 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
517 implementation with a new version that just does either a direct
518 move or a regular move. Move memory interface to separate insns.
519 Move insns so they are next to the expander.
520 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
521 with signbit move. Split big and little endian case.
522 (signbit<mode>2_dm_mem_le): Likewise.
523 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
524 (signbit<mode>2_dm2): Likewise.
526 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
528 * config/rl78/rl78.md: New define_expand "anddi3".
530 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
532 * config/rl78/rl78.md: New define_expand "umindi3".
534 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
536 * config/rl78/rl78.md: New define_expand "smindi3".
538 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
540 * config/rl78/rl78.md: New define_expand "smaxdi3".
542 2018-01-22 Carl Love <cel@us.ibm.com>
544 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
545 LVX_V1TI): Add macro expansion.
546 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
547 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
548 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
549 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
550 Change check to determine if the instruction is a byte reversing
551 entry. Fix typo in comment.
552 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
553 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
554 Add def_builtin calls for new builtins.
555 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
556 Add define_insn expansion.
558 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
560 * config/rl78/rl78.md: New define_expand "umaxdi3".
562 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
564 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
565 for non-QImode registers
567 2018-01-22 Richard Biener <rguenther@suse.de>
569 PR tree-optimization/83963
570 * graphite-scop-detection.c (scop_detection::get_sese): Delay
571 including the loop exit block.
572 (scop_detection::merge_sese): Likewise.
573 (scop_detection::add_scop): Do it here instead.
575 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
577 * doc/sourcebuild.texi (arm_softfloat): Document.
579 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
582 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
583 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
584 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
586 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
587 David Edelsohn <dje.gcc@gmail.com>
590 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
591 Change "crset eq" to "crset 2".
592 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
593 (*call_indirect_aix<mode>_nospec): Likewise.
594 (*call_value_indirect_aix<mode>_nospec): Likewise.
595 (*call_indirect_elfv2<mode>_nospec): Likewise.
596 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
597 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
598 change assembly output from . to $.
599 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
600 (indirect_jump<mode>_nospec): Change assembly output from . to $.
601 (*tablejump<mode>_internal1_nospec): Likewise.
603 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
606 * config/sh/sh_optimize_sett_clrt.cc:
607 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
609 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
611 PR tree-optimization/83940
612 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
613 offset_dt to vect_constant_def rather than vect_unknown_def_type.
614 (vect_check_load_store_mask): Add a mask_dt_out parameter and
615 use it to pass back the definition type.
616 (vect_check_store_rhs): Likewise rhs_dt_out.
617 (vect_build_gather_load_calls): Add a mask_dt argument and use
618 it instead of a call to vect_is_simple_use.
619 (vectorizable_store): Update calls to vect_check_load_store_mask
620 and vect_check_store_rhs. Use the dt returned by the latter instead
621 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
622 instead of calls to vect_is_simple_use. Pass the scalar rather
623 than the vector operand to vect_is_simple_use when handling
624 second and subsequent copies of an rhs value.
625 (vectorizable_load): Update calls to vect_check_load_store_mask
626 and vect_build_gather_load_calls. Use the cached mask_dt and
627 gs_info.offset_dt instead of calls to vect_is_simple_use.
629 2018-01-20 Jakub Jelinek <jakub@redhat.com>
632 * tree-emutls.c: Include gimplify.h.
633 (lower_emutls_2): New function.
634 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
635 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
636 it before further processing.
639 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
640 UINTVAL (trueop1) instead of INTVAL (op1).
642 2018-01-19 Jakub Jelinek <jakub@redhat.com>
646 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
647 INCOMING_FRAME_SP_OFFSET if not defined.
648 (scan_trace): Add ENTRY argument. If true and
649 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
650 emit a note to adjust the CFA offset.
651 (create_cfi_notes): Adjust scan_trace callers.
652 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
653 INCOMING_FRAME_SP_OFFSET in the CIE.
654 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
655 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
657 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
658 * doc/tm.texi: Regenerated.
660 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
662 PR rtl-optimization/83147
663 * lra-constraints.c (remove_inheritance_pseudos): Use
664 lra_substitute_pseudo_within_insn.
666 2018-01-19 Tom de Vries <tom@codesourcery.com>
667 Cesar Philippidis <cesar@codesourcery.com>
670 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
672 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
675 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
676 spaces for function labels.
678 2018-01-19 Martin Liska <mliska@suse.cz>
680 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
681 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
682 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
683 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
684 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
685 (PRED_CONST_RETURN): Change from 69 to 65.
686 (PRED_NULL_RETURN): Change from 91 to 71.
687 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
688 (PRED_LOOP_GUARD): Change from 66 to 73.
690 2018-01-19 Martin Liska <mliska@suse.cz>
692 * predict.c (predict_insn_def): Add new assert.
693 (struct branch_predictor): Change type to signed integer.
694 (test_prediction_value_range): Amend test to cover
696 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
697 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
698 (PRED_LOOP_ITERATIONS_MAX): Likewise.
699 (PRED_LOOP_IV_COMPARE): Likewise.
700 * predict.h (PROB_UNINITIALIZED): Define new constant.
702 2018-01-19 Martin Liska <mliska@suse.cz>
704 * predict.c (dump_prediction): Add new format for
705 analyze_brprob.py script which is enabled with -details
707 * profile-count.h (precise_p): New function.
709 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
711 PR tree-optimization/83922
712 * tree-vect-loop.c (vect_verify_full_masking): Return false if
713 there are no statements that need masking.
714 (vect_active_double_reduction_p): New function.
715 (vect_analyze_loop_operations): Use it when handling phis that
716 are not in the loop header.
718 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
720 PR tree-optimization/83914
721 * tree-vect-loop.c (vectorizable_induction): Don't convert
722 init_expr or apply the peeling adjustment for inductions
723 that are nested within the vectorized loop.
725 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
727 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
730 2018-01-18 Jakub Jelinek <jakub@redhat.com>
734 * function.h (gimplify_parameters): Add gimple_seq * argument.
735 * function.c: Include gimple.h and options.h.
736 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
737 for the added local temporaries if needed.
738 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
739 if there are any parameter cleanups, wrap whole body into a
740 try/finally with the cleanups.
742 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
745 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
746 Use GET_MODE_CLASS for scalar floating point.
748 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
752 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
753 Fix call of call_cgraph_insertion_hooks.
755 2018-01-18 Martin Sebor <msebor@redhat.com>
757 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
759 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
762 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
765 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
768 * common.opt: (-ffile-prefix-map): New option.
769 * opts.c (common_handle_option): Defer it.
770 * opts-global.c (handle_common_deferred_options): Handle it.
771 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
772 * file-prefix-map.h: New file.
773 (remap_debug_filename, add_debug_prefix_map): ...here.
774 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
775 * final.c (debug_prefix_map, add_debug_prefix_map
776 remap_debug_filename): Move to...
777 * file-prefix-map.c: New file.
778 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
779 generalize, get rid of alloca(), use strrchr() instead of strchr().
780 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
781 Implement in terms of add_prefix_map().
782 (remap_macro_filename, remap_debug_filename): Implement in term of
784 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
785 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
786 * dbxout.c: Include file-prefix-map.h.
787 * varasm.c: Likewise.
788 * vmsdbgout.c: Likewise.
789 * xcoffout.c: Likewise.
790 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
791 * doc/cppopts.texi (-fmacro-prefix-map): Document.
792 * doc/invoke.texi (-ffile-prefix-map): Document.
793 (-fdebug-prefix-map): Update description.
795 2018-01-18 Martin Liska <mliska@suse.cz>
797 * config/i386/i386.c (indirect_thunk_name): Document that also
799 (output_indirect_thunk): Document why both instructions
800 (pause and lfence) are generated.
802 2018-01-18 Richard Biener <rguenther@suse.de>
804 PR tree-optimization/83887
805 * graphite-scop-detection.c
806 (scop_detection::get_nearest_dom_with_single_entry): Remove.
807 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
808 (scop_detection::merge_sese): Re-implement with a flood-fill
809 algorithm that properly finds a SESE region if it exists.
811 2018-01-18 Jakub Jelinek <jakub@redhat.com>
814 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
815 pointer_diff optimizations use view_convert instead of convert.
817 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
819 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
820 Generate different code for -mno-speculate-indirect-jumps.
821 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
822 (*call_indirect_aix<mode>): Disable for
823 -mno-speculate-indirect-jumps.
824 (*call_indirect_aix<mode>_nospec): New define_insn.
825 (*call_value_indirect_aix<mode>): Disable for
826 -mno-speculate-indirect-jumps.
827 (*call_value_indirect_aix<mode>_nospec): New define_insn.
828 (*sibcall_nonlocal_sysv<mode>): Generate different code for
829 -mno-speculate-indirect-jumps.
830 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
832 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
834 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
835 long double type, set the flags for noting the default long double
836 type, even if we don't pass or return a long double type.
838 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
841 * ipa-inline.c (flatten_function): Do not overwrite final inlining
844 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
846 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
847 support for merge[hl].
848 (fold_mergehl_helper): New helper function.
849 (tree-vector-builder.h): New #include for tree_vector_builder usage.
850 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
851 (altivec_vmrglw_direct): Add xxmrglw insn.
853 2018-01-17 Andrew Waterman <andrew@sifive.com>
855 * config/riscv/riscv.c (riscv_conditional_register_usage): If
856 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
858 2018-01-17 David Malcolm <dmalcolm@redhat.com>
861 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
862 call the lto_location_cache before reading the
863 DECL_SOURCE_LOCATION of the types.
865 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
866 Richard Sandiford <richard.sandiford@linaro.org>
868 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
869 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
870 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
871 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
872 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
874 * config/aarch64/constraints.md (aarch64_movti_operand):
876 * config/aarch64/predicates.md (Uti): Add new constraint.
878 2018-01-17 Carl Love <cel@us.ibm.com>
879 * config/rs6000/vsx.md (define_expand xl_len_r,
880 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
881 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
883 (define_expand, define_insn): Move the shift left from the
884 define_insn to the define_expand for lxvl and stxvl instructions.
885 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
886 and XL_LEN_R definitions to PURE.
888 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
890 * config/i386/i386.c (indirect_thunk_name): Declare regno
891 as unsigned int. Compare regno with INVALID_REGNUM.
892 (output_indirect_thunk): Ditto.
893 (output_indirect_thunk_function): Ditto.
894 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
895 in the call to output_indirect_thunk_function.
897 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
900 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
901 rather than the size of inner_type to determine the stack slot size
902 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
904 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
907 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
910 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
912 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
913 endian Linux systems to optionally enable multilibs for selecting
914 the long double type if the user configured an explicit type.
915 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
916 have no long double multilibs if not defined.
917 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
918 warn if the user used -mabi={ieee,ibm}longdouble and we built
919 multilibs for long double.
920 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
921 appropriate multilib option.
922 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
924 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
925 for building long double multilibs.
926 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
928 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
930 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
933 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
935 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
938 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
941 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
944 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
946 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
947 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
949 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
951 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
952 different rtl trees depending on TARGET_64BIT.
953 (rs6000_gen_lvx): Likewise.
955 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
957 * config/visium/visium.md (nop): Tweak comment.
958 (hazard_nop): Likewise.
960 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
962 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
963 -mspeculate-indirect-jumps.
964 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
965 for -mno-speculate-indirect-jumps.
966 (*call_indirect_elfv2<mode>_nospec): New define_insn.
967 (*call_value_indirect_elfv2<mode>): Disable for
968 -mno-speculate-indirect-jumps.
969 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
970 (indirect_jump): Emit different RTL for
971 -mno-speculate-indirect-jumps.
972 (*indirect_jump<mode>): Disable for
973 -mno-speculate-indirect-jumps.
974 (*indirect_jump<mode>_nospec): New define_insn.
975 (tablejump): Emit different RTL for
976 -mno-speculate-indirect-jumps.
977 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
978 (tablejumpsi_nospec): New define_expand.
979 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
980 (tablejumpdi_nospec): New define_expand.
981 (*tablejump<mode>_internal1): Disable for
982 -mno-speculate-indirect-jumps.
983 (*tablejump<mode>_internal1_nospec): New define_insn.
984 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
987 2018-01-16 Artyom Skrobov tyomitch@gmail.com
989 * caller-save.c (insert_save): Drop unnecessary parameter. All
992 2018-01-16 Jakub Jelinek <jakub@redhat.com>
993 Richard Biener <rguenth@suse.de>
996 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
997 return early, inline manually is_gimple_sizepos. Make sure if we
998 call gimplify_expr we don't end up with a gimple constant.
999 * tree.c (variably_modified_type_p): Don't return true for
1000 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1001 * gimplify.h (is_gimple_sizepos): Remove.
1003 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1005 PR tree-optimization/83857
1006 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1007 vectorizable_live_operation for pure SLP statements.
1008 (vectorizable_live_operation): Handle PHIs.
1010 2018-01-16 Richard Biener <rguenther@suse.de>
1012 PR tree-optimization/83867
1013 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1014 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1016 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1019 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1020 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1021 If off is not INTEGER_CST, issue a may not be aligned warning
1022 rather than isn't aligned. Use isn%'t rather than isn't.
1023 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1025 <case MULT_EXPR>: Improve the case when bottom and one of the
1026 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1027 operand, in that case check if the other operand is multiple of
1028 bottom divided by the INTEGER_CST operand.
1030 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1033 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1034 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1035 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1036 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1037 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1038 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1039 (pa_function_arg_size): New function.
1041 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1043 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1044 in a separate statement.
1046 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1048 PR tree-optimization/83847
1049 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1050 group gathers and scatters.
1052 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1054 PR rtl-optimization/86620
1055 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1057 PR rtl-optimization/83213
1058 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1059 to last if both are JUMP_INSNs.
1061 PR tree-optimization/83843
1062 * gimple-ssa-store-merging.c
1063 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1064 store_immediate_info for bswap/nop orig_stores.
1066 2018-01-15 Andrew Waterman <andrew@sifive.com>
1068 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1070 <UDIV>: Increase cost if !TARGET_DIV.
1072 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1074 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1075 (define_attr "cr_logical_3op"): New.
1076 (cceq_ior_compare): Adjust.
1077 (cceq_ior_compare_complement): Adjust.
1078 (*cceq_rev_compare): Adjust.
1079 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1080 (is_cracked_insn): Adjust.
1081 (insn_must_be_first_in_group): Adjust.
1082 * config/rs6000/40x.md: Adjust.
1083 * config/rs6000/440.md: Adjust.
1084 * config/rs6000/476.md: Adjust.
1085 * config/rs6000/601.md: Adjust.
1086 * config/rs6000/603.md: Adjust.
1087 * config/rs6000/6xx.md: Adjust.
1088 * config/rs6000/7450.md: Adjust.
1089 * config/rs6000/7xx.md: Adjust.
1090 * config/rs6000/8540.md: Adjust.
1091 * config/rs6000/cell.md: Adjust.
1092 * config/rs6000/e300c2c3.md: Adjust.
1093 * config/rs6000/e500mc.md: Adjust.
1094 * config/rs6000/e500mc64.md: Adjust.
1095 * config/rs6000/e5500.md: Adjust.
1096 * config/rs6000/e6500.md: Adjust.
1097 * config/rs6000/mpc.md: Adjust.
1098 * config/rs6000/power4.md: Adjust.
1099 * config/rs6000/power5.md: Adjust.
1100 * config/rs6000/power6.md: Adjust.
1101 * config/rs6000/power7.md: Adjust.
1102 * config/rs6000/power8.md: Adjust.
1103 * config/rs6000/power9.md: Adjust.
1104 * config/rs6000/rs64.md: Adjust.
1105 * config/rs6000/titan.md: Adjust.
1107 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1109 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1110 ix86_indirect_branch_register logic.
1112 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1114 * config/i386/constraints.md (Bs): Update
1115 ix86_indirect_branch_register check. Don't check
1116 ix86_indirect_branch_register with GOT_memory_operand.
1118 * config/i386/predicates.md (GOT_memory_operand): Don't check
1119 ix86_indirect_branch_register here.
1120 (GOT32_symbol_operand): Likewise.
1122 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1124 * config/i386/predicates.md (constant_call_address_operand):
1125 Rewrite ix86_indirect_branch_register logic.
1126 (sibcall_insn_operand): Likewise.
1128 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1130 * config/i386/constraints.md (Bs): Replace
1131 ix86_indirect_branch_thunk_register with
1132 ix86_indirect_branch_register.
1134 * config/i386/i386.md (indirect_jump): Likewise.
1135 (tablejump): Likewise.
1136 (*sibcall_memory): Likewise.
1137 (*sibcall_value_memory): Likewise.
1138 Peepholes of indirect call and jump via memory: Likewise.
1139 * config/i386/i386.opt: Likewise.
1140 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1141 (GOT_memory_operand): Likewise.
1142 (call_insn_operand): Likewise.
1143 (sibcall_insn_operand): Likewise.
1144 (GOT32_symbol_operand): Likewise.
1146 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1149 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1150 type rather than type addr's type points to.
1151 (expand_omp_atomic_mutex): Likewise.
1152 (expand_omp_atomic): Likewise.
1154 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1157 * config/i386/i386.c (output_indirect_thunk_function): Use
1158 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1159 for __x86_return_thunk.
1161 2018-01-15 Richard Biener <rguenther@suse.de>
1164 * expmed.c (extract_bit_field_1): Fix typo.
1166 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1169 * config/arm/iterators.md (VF): New mode iterator.
1170 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1171 Remove integer-related logic from pattern.
1172 (neon_vabd<mode>_3): Likewise.
1174 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1177 * common.opt (fstrict-overflow): No longer an alias.
1178 (fwrapv-pointer): New option.
1179 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1180 also for pointer types based on flag_wrapv_pointer.
1181 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1182 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1183 opts->x_flag_wrapv got set.
1184 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1185 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1186 POINTER_TYPE_OVERFLOW_UNDEFINED.
1187 * match.pd: Likewise in address comparison pattern.
1188 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1190 2018-01-15 Richard Biener <rguenther@suse.de>
1193 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1194 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1195 Reset type names to their identifier if their TYPE_DECL doesn't
1196 have linkage (and thus is used for ODR and devirt).
1197 (save_debug_info_for_decl): Remove.
1198 (save_debug_info_for_type): Likewise.
1199 (add_tree_to_fld_list): Adjust.
1200 * tree-pretty-print.c (dump_generic_node): Make dumping of
1201 type names more robust.
1203 2018-01-15 Richard Biener <rguenther@suse.de>
1205 * BASE-VER: Bump to 8.0.1.
1207 2018-01-14 Martin Sebor <msebor@redhat.com>
1210 * builtins.c (check_access): Avoid warning when the no-warning bit
1213 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1215 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1216 * ira-color (allocno_hard_regs_compare): Likewise.
1218 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1221 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1222 Use .pushsection/.popsection.
1224 2018-01-14 Martin Sebor <msebor@redhat.com>
1227 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1229 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1231 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1232 entry from extra_headers.
1233 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1234 extra_headers, make the list bitwise identical to the i?86-*-* one.
1236 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1238 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1239 -mcmodel=large with -mindirect-branch=thunk,
1240 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1241 -mfunction-return=thunk-extern.
1242 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1243 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1244 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1246 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1248 * config/i386/i386.c (print_reg): Print the name of the full
1249 integer register without '%'.
1250 (ix86_print_operand): Handle 'V'.
1251 * doc/extend.texi: Document 'V' modifier.
1253 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1255 * config/i386/constraints.md (Bs): Disallow memory operand for
1256 -mindirect-branch-register.
1258 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1259 (GOT_memory_operand): Likewise.
1260 (call_insn_operand): Likewise.
1261 (sibcall_insn_operand): Likewise.
1262 (GOT32_symbol_operand): Likewise.
1263 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1264 for -mindirect-branch-register.
1265 (tablejump): Likewise.
1266 (*sibcall_memory): Likewise.
1267 (*sibcall_value_memory): Likewise.
1268 Disallow peepholes of indirect call and jump via memory for
1269 -mindirect-branch-register.
1270 (*call_pop): Replace m with Bw.
1271 (*call_value_pop): Likewise.
1272 (*sibcall_pop_memory): Replace m with Bs.
1273 * config/i386/i386.opt (mindirect-branch-register): New option.
1274 * doc/invoke.texi: Document -mindirect-branch-register option.
1276 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1278 * config/i386/i386-protos.h (ix86_output_function_return): New.
1279 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1280 set function_return_type.
1281 (indirect_thunk_name): Add ret_p to indicate thunk for function
1283 (output_indirect_thunk_function): Pass false to
1284 indirect_thunk_name.
1285 (ix86_output_indirect_branch_via_reg): Likewise.
1286 (ix86_output_indirect_branch_via_push): Likewise.
1287 (output_indirect_thunk_function): Create alias for function
1288 return thunk if regno < 0.
1289 (ix86_output_function_return): New function.
1290 (ix86_handle_fndecl_attribute): Handle function_return.
1291 (ix86_attribute_table): Add function_return.
1292 * config/i386/i386.h (machine_function): Add
1293 function_return_type.
1294 * config/i386/i386.md (simple_return_internal): Use
1295 ix86_output_function_return.
1296 (simple_return_internal_long): Likewise.
1297 * config/i386/i386.opt (mfunction-return=): New option.
1298 (indirect_branch): Mention -mfunction-return=.
1299 * doc/extend.texi: Document function_return function attribute.
1300 * doc/invoke.texi: Document -mfunction-return= option.
1302 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1304 * config/i386/i386-opts.h (indirect_branch): New.
1305 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1306 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1307 with local indirect jump when converting indirect call and jump.
1308 (ix86_set_indirect_branch_type): New.
1309 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1310 (indirectlabelno): New.
1311 (indirect_thunk_needed): Likewise.
1312 (indirect_thunk_bnd_needed): Likewise.
1313 (indirect_thunks_used): Likewise.
1314 (indirect_thunks_bnd_used): Likewise.
1315 (INDIRECT_LABEL): Likewise.
1316 (indirect_thunk_name): Likewise.
1317 (output_indirect_thunk): Likewise.
1318 (output_indirect_thunk_function): Likewise.
1319 (ix86_output_indirect_branch_via_reg): Likewise.
1320 (ix86_output_indirect_branch_via_push): Likewise.
1321 (ix86_output_indirect_branch): Likewise.
1322 (ix86_output_indirect_jmp): Likewise.
1323 (ix86_code_end): Call output_indirect_thunk_function if needed.
1324 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1326 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1327 (ix86_attribute_table): Add indirect_branch.
1328 * config/i386/i386.h (machine_function): Add indirect_branch_type
1329 and has_local_indirect_jump.
1330 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1332 (tablejump): Likewise.
1333 (*indirect_jump): Use ix86_output_indirect_jmp.
1334 (*tablejump_1): Likewise.
1335 (simple_return_indirect_internal): Likewise.
1336 * config/i386/i386.opt (mindirect-branch=): New option.
1337 (indirect_branch): New.
1340 (thunk-inline): Likewise.
1341 (thunk-extern): Likewise.
1342 * doc/extend.texi: Document indirect_branch function attribute.
1343 * doc/invoke.texi: Document -mindirect-branch= option.
1345 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1348 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1350 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1352 * ipa-inline.c (want_inline_small_function_p): Return false if
1353 inlining has already failed with CIF_FINAL_ERROR.
1354 (update_caller_keys): Call want_inline_small_function_p before
1356 (update_callee_keys): Likewise.
1358 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1360 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1362 (rs6000_quadword_masked_address_p): Likewise.
1363 (quad_aligned_load_p): Likewise.
1364 (quad_aligned_store_p): Likewise.
1365 (const_load_sequence_p): Add comment to describe the outer-most loop.
1366 (mimic_memory_attributes_and_flags): New function.
1367 (rs6000_gen_stvx): Likewise.
1368 (replace_swapped_aligned_store): Likewise.
1369 (rs6000_gen_lvx): Likewise.
1370 (replace_swapped_aligned_load): Likewise.
1371 (replace_swapped_load_constant): Capitalize argument name in
1372 comment describing this function.
1373 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1374 and stores that access quad-word aligned addresses and replace
1375 with stvx or lvx instructions when appropriate.
1376 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1377 New function prototype.
1378 (rs6000_quadword_masked_address_p): Likewise.
1379 (rs6000_gen_lvx): Likewise.
1380 (rs6000_gen_stvx): Likewise.
1381 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1382 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1383 when memory address is aligned.
1384 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1385 this split to select lvx instruction when memory address is aligned.
1386 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1387 instruction when memory address is aligned.
1388 (*vsx_le_perm_load_v16qi): Likewise.
1389 (four unnamed splitters): Modify to select the stvx instruction
1390 when memory is aligned.
1392 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1394 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1395 which appears in the queue multiple times.
1397 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1398 Alan Hayward <alan.hayward@arm.com>
1399 David Sherwood <david.sherwood@arm.com>
1401 * tree-vectorizer.h (vec_lower_bound): New structure.
1402 (_loop_vec_info): Add check_nonzero and lower_bounds.
1403 (LOOP_VINFO_CHECK_NONZERO): New macro.
1404 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1405 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1406 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1407 fields. Make seg_len the distance travelled, not including the
1409 (dr_direction_indicator): Declare.
1410 (dr_zero_step_indicator): Likewise.
1411 (dr_known_forward_stride_p): Likewise.
1412 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1414 (runtime_alias_check_p): Allow runtime alias checks with
1416 (operator ==): Compare access_size and align.
1417 (prune_runtime_alias_test_list): Rework for new distinction between
1418 the access_size and seg_len.
1419 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1421 (get_segment_min_max): New function.
1422 (create_intersect_range_checks): Use it.
1423 (dr_step_indicator): New function.
1424 (dr_direction_indicator): Likewise.
1425 (dr_zero_step_indicator): Likewise.
1426 (dr_known_forward_stride_p): Likewise.
1427 * tree-loop-distribution.c (data_ref_segment_size): Return
1428 DR_STEP * (niters - 1).
1429 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1431 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1432 (vect_preserves_scalar_order_p): New function, split out from...
1433 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1434 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1435 (vect_vfa_access_size): New function.
1436 (vect_vfa_align): Likewise.
1437 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1438 (dump_lower_bound): New function.
1439 (vect_check_lower_bound): Likewise.
1440 (vect_small_gap_p): Likewise.
1441 (vectorizable_with_step_bound_p): Likewise.
1442 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1443 depencies if the vectorization factor is 1. Convert the checks
1444 for nonzero steps into checks on the bounds of DR_STEP. Try using
1445 a bunds check for variable steps if the minimum required step is
1446 relatively small. Update calls to the dr_with_seg_len
1447 constructor and to vect_compile_time_alias.
1448 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1450 (vect_loop_versioning): Call it.
1451 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1453 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1455 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1456 Alan Hayward <alan.hayward@arm.com>
1457 David Sherwood <david.sherwood@arm.com>
1459 * doc/sourcebuild.texi (vect_scatter_store): Document.
1460 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1462 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1464 * genopinit.c (main): Add supports_vec_scatter_store and
1465 supports_vec_scatter_store_cached to target_optabs.
1466 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1467 IFN_MASK_SCATTER_STORE.
1468 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1470 * internal-fn.h (internal_store_fn_p): Declare.
1471 (internal_fn_stored_value_index): Likewise.
1472 * internal-fn.c (scatter_store_direct): New macro.
1473 (expand_scatter_store_optab_fn): New function.
1474 (direct_scatter_store_optab_supported_p): New macro.
1475 (internal_store_fn_p): New function.
1476 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1477 IFN_MASK_SCATTER_STORE.
1478 (internal_fn_mask_index): Likewise.
1479 (internal_fn_stored_value_index): New function.
1480 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1482 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1483 * optabs-query.c (supports_vec_scatter_store_p): New function.
1484 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1485 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1486 true for scatter stores.
1487 (vect_gather_scatter_fn_p): Handle scatter stores too.
1488 (vect_check_gather_scatter): Consider using scatter stores if
1489 supports_vec_scatter_store_p.
1490 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1492 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1493 internal_fn_stored_value_index.
1494 (check_load_store_masking): Handle scatter stores too.
1495 (vect_get_store_rhs): Make public.
1496 (vectorizable_call): Use internal_store_fn_p.
1497 (vectorizable_store): Handle scatter store internal functions.
1498 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1499 when deciding whether the end of the group has been reached.
1500 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1501 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1502 (mask_scatter_store<mode>): New insns.
1504 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1505 Alan Hayward <alan.hayward@arm.com>
1506 David Sherwood <david.sherwood@arm.com>
1508 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1509 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1510 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1512 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1513 Use vect_truncate_gather_scatter_offset if we can't treat the
1514 operation as a normal gather load or scatter store.
1515 (get_group_load_store_type): Take the gather_scatter_info
1516 as argument. Try using a gather load or scatter store for
1517 single-element groups.
1518 (get_load_store_type): Update calls to get_group_load_store_type
1519 and vect_use_strided_gather_scatters_p.
1521 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1522 Alan Hayward <alan.hayward@arm.com>
1523 David Sherwood <david.sherwood@arm.com>
1525 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1526 optional tree argument.
1527 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1529 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1530 but continue to use the current value as a fallback.
1531 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1532 to compare the updates.
1533 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1534 (get_load_store_type): Use it when handling a strided access.
1535 (vect_get_strided_load_store_ops): New function.
1536 (vect_get_data_ptr_increment): Likewise.
1537 (vectorizable_load): Handle strided gather loads. Always pass
1538 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1540 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1541 Alan Hayward <alan.hayward@arm.com>
1542 David Sherwood <david.sherwood@arm.com>
1544 * doc/md.texi (gather_load@var{m}): Document.
1545 (mask_gather_load@var{m}): Likewise.
1546 * genopinit.c (main): Add supports_vec_gather_load and
1547 supports_vec_gather_load_cached to target_optabs.
1548 * optabs-tree.c (init_tree_optimization_optabs): Use
1549 ggc_cleared_alloc to allocate target_optabs.
1550 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1551 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1553 * internal-fn.h (internal_load_fn_p): Declare.
1554 (internal_gather_scatter_fn_p): Likewise.
1555 (internal_fn_mask_index): Likewise.
1556 (internal_gather_scatter_fn_supported_p): Likewise.
1557 * internal-fn.c (gather_load_direct): New macro.
1558 (expand_gather_load_optab_fn): New function.
1559 (direct_gather_load_optab_supported_p): New macro.
1560 (direct_internal_fn_optab): New function.
1561 (internal_load_fn_p): Likewise.
1562 (internal_gather_scatter_fn_p): Likewise.
1563 (internal_fn_mask_index): Likewise.
1564 (internal_gather_scatter_fn_supported_p): Likewise.
1565 * optabs-query.c (supports_at_least_one_mode_p): New function.
1566 (supports_vec_gather_load_p): Likewise.
1567 * optabs-query.h (supports_vec_gather_load_p): Declare.
1568 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1569 and memory_type field.
1570 (NUM_PATTERNS): Bump to 15.
1571 * tree-vect-data-refs.c: Include internal-fn.h.
1572 (vect_gather_scatter_fn_p): New function.
1573 (vect_describe_gather_scatter_call): Likewise.
1574 (vect_check_gather_scatter): Try using internal functions for
1575 gather loads. Recognize existing calls to a gather load function.
1576 (vect_analyze_data_refs): Consider using gather loads if
1577 supports_vec_gather_load_p.
1578 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1579 (vect_get_gather_scatter_offset_type): Likewise.
1580 (vect_convert_mask_for_vectype): Likewise.
1581 (vect_add_conversion_to_patterm): Likewise.
1582 (vect_try_gather_scatter_pattern): Likewise.
1583 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1584 (vect_vect_recog_func_ptrs): Add it.
1585 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1586 internal_fn_mask_index and internal_gather_scatter_fn_p.
1587 (check_load_store_masking): Take the gather_scatter_info as an
1588 argument and handle gather loads.
1589 (vect_get_gather_scatter_ops): New function.
1590 (vectorizable_call): Check internal_load_fn_p.
1591 (vectorizable_load): Likewise. Handle gather load internal
1593 (vectorizable_store): Update call to check_load_store_masking.
1594 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1595 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1596 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1597 (aarch64_gather_scale_operand_d): New predicates.
1598 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1599 (mask_gather_load<mode>): New insns.
1601 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1602 Alan Hayward <alan.hayward@arm.com>
1603 David Sherwood <david.sherwood@arm.com>
1605 * optabs.def (fold_left_plus_optab): New optab.
1606 * doc/md.texi (fold_left_plus_@var{m}): Document.
1607 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1608 * internal-fn.c (fold_left_direct): Define.
1609 (expand_fold_left_optab_fn): Likewise.
1610 (direct_fold_left_optab_supported_p): Likewise.
1611 * fold-const-call.c (fold_const_fold_left): New function.
1612 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1613 * tree-parloops.c (valid_reduction_p): New function.
1614 (gather_scalar_reductions): Use it.
1615 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1616 (vect_finish_replace_stmt): Declare.
1617 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1618 (needs_fold_left_reduction_p): New function, split out from...
1619 (vect_is_simple_reduction): ...here. Accept reductions that
1620 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1621 (vect_force_simple_reduction): Also store the reduction type in
1622 the assignment's STMT_VINFO_REDUC_TYPE.
1623 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1624 (merge_with_identity): New function.
1625 (vect_expand_fold_left): Likewise.
1626 (vectorize_fold_left_reduction): Likewise.
1627 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1628 scalar phi in place for it. Check for target support and reject
1629 cases that would reassociate the operation. Defer the transform
1630 phase to vectorize_fold_left_reduction.
1631 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1632 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1633 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1635 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1637 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1638 call to ifc_temp_var.
1640 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1641 Alan Hayward <alan.hayward@arm.com>
1642 David Sherwood <david.sherwood@arm.com>
1644 * target.def (legitimize_address_displacement): Take the original
1645 offset as a poly_int.
1646 * targhooks.h (default_legitimize_address_displacement): Update
1648 * targhooks.c (default_legitimize_address_displacement): Likewise.
1649 * doc/tm.texi: Regenerate.
1650 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1651 as an argument, moving assert of ad->disp == ad->disp_term to...
1652 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1653 Try calling targetm.legitimize_address_displacement before expanding
1654 the address rather than afterwards, and adjust for the new interface.
1655 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1656 Match the new hook interface. Handle SVE addresses.
1657 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1660 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1662 * Makefile.in (OBJS): Add early-remat.o.
1663 * target.def (select_early_remat_modes): New hook.
1664 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1665 * doc/tm.texi: Regenerate.
1666 * targhooks.h (default_select_early_remat_modes): Declare.
1667 * targhooks.c (default_select_early_remat_modes): New function.
1668 * timevar.def (TV_EARLY_REMAT): New timevar.
1669 * passes.def (pass_early_remat): New pass.
1670 * tree-pass.h (make_pass_early_remat): Declare.
1671 * early-remat.c: New file.
1672 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1674 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1676 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1677 Alan Hayward <alan.hayward@arm.com>
1678 David Sherwood <david.sherwood@arm.com>
1680 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1681 vfm1 with a bound_epilog parameter.
1682 (vect_do_peeling): Update calls accordingly, and move the prologue
1683 call earlier in the function. Treat the base bound_epilog as 0 for
1684 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1685 this base when peeling for gaps.
1686 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1687 with fully-masked loops.
1688 (vect_estimate_min_profitable_iters): Handle the single peeled
1689 iteration in that case.
1691 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1692 Alan Hayward <alan.hayward@arm.com>
1693 David Sherwood <david.sherwood@arm.com>
1695 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1696 single-element interleaving even if the size is not a power of 2.
1697 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1698 accesses for single-element interleaving if the group size is
1701 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1702 Alan Hayward <alan.hayward@arm.com>
1703 David Sherwood <david.sherwood@arm.com>
1705 * doc/md.texi (fold_extract_last_@var{m}): Document.
1706 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1707 * optabs.def (fold_extract_last_optab): New optab.
1708 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1709 * internal-fn.c (fold_extract_direct): New macro.
1710 (expand_fold_extract_optab_fn): Likewise.
1711 (direct_fold_extract_optab_supported_p): Likewise.
1712 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1713 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1714 EXTRACT_LAST_REDUCTION.
1715 (get_initial_def_for_reduction): Do not create an initial vector
1716 for EXTRACT_LAST_REDUCTION reductions.
1717 (vectorizable_reduction): Leave the scalar phi in place for
1718 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1719 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1720 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1721 transform phase to vectorizable_condition.
1722 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1724 (vect_finish_stmt_generation): ...here.
1725 (vect_finish_replace_stmt): New function.
1726 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1727 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1729 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1731 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1732 Alan Hayward <alan.hayward@arm.com>
1733 David Sherwood <david.sherwood@arm.com>
1735 * doc/md.texi (extract_last_@var{m}): Document.
1736 * optabs.def (extract_last_optab): New optab.
1737 * internal-fn.def (EXTRACT_LAST): New internal function.
1738 * internal-fn.c (cond_unary_direct): New macro.
1739 (expand_cond_unary_optab_fn): Likewise.
1740 (direct_cond_unary_optab_supported_p): Likewise.
1741 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1742 loops using EXTRACT_LAST.
1743 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1744 (extract_last_<mode>): ...this optab.
1745 (vec_extract<mode><Vel>): Update accordingly.
1747 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1748 Alan Hayward <alan.hayward@arm.com>
1749 David Sherwood <david.sherwood@arm.com>
1751 * target.def (empty_mask_is_expensive): New hook.
1752 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
1753 * doc/tm.texi: Regenerate.
1754 * targhooks.h (default_empty_mask_is_expensive): Declare.
1755 * targhooks.c (default_empty_mask_is_expensive): New function.
1756 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
1757 if the target says that empty masks are expensive.
1758 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
1760 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
1762 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1763 Alan Hayward <alan.hayward@arm.com>
1764 David Sherwood <david.sherwood@arm.com>
1766 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
1767 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
1768 (vect_use_loop_mask_for_alignment_p): New function.
1769 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
1770 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
1771 niters_skip argument. Make sure that the first niters_skip elements
1772 of the first iteration are inactive.
1773 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
1774 Update call to vect_set_loop_masks_directly.
1775 (get_misalign_in_elems): New function, split out from...
1776 (vect_gen_prolog_loop_niters): ...here.
1777 (vect_update_init_of_dr): Take a code argument that specifies whether
1778 the adjustment should be added or subtracted.
1779 (vect_update_init_of_drs): Likewise.
1780 (vect_prepare_for_masked_peels): New function.
1781 (vect_do_peeling): Skip prologue peeling if we're using a mask
1782 instead. Update call to vect_update_inits_of_drs.
1783 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1785 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
1786 alignment. Do not include the number of peeled iterations in
1787 the minimum threshold in that case.
1788 (vectorizable_induction): Adjust the start value down by
1789 LOOP_VINFO_MASK_SKIP_NITERS iterations.
1790 (vect_transform_loop): Call vect_prepare_for_masked_peels.
1791 Take the number of skipped iterations into account when calculating
1793 * tree-vect-stmts.c (vect_gen_while_not): New function.
1795 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1796 Alan Hayward <alan.hayward@arm.com>
1797 David Sherwood <david.sherwood@arm.com>
1799 * doc/sourcebuild.texi (vect_fully_masked): Document.
1800 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
1802 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
1804 (vect_analyze_loop_2): ...here. Don't check the vectorization
1805 factor against the number of loop iterations if the loop is
1808 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1809 Alan Hayward <alan.hayward@arm.com>
1810 David Sherwood <david.sherwood@arm.com>
1812 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
1813 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
1814 (dump_groups): Update accordingly.
1815 (iv_use::mem_type): New member variable.
1816 (address_p): New function.
1817 (record_use): Add a mem_type argument and initialize the new
1819 (record_group_use): Add a mem_type argument. Use address_p.
1820 Remove obsolete null checks of base_object. Update call to record_use.
1821 (find_interesting_uses_op): Update call to record_group_use.
1822 (find_interesting_uses_cond): Likewise.
1823 (find_interesting_uses_address): Likewise.
1824 (get_mem_type_for_internal_fn): New function.
1825 (find_address_like_use): Likewise.
1826 (find_interesting_uses_stmt): Try find_address_like_use before
1827 calling find_interesting_uses_op.
1828 (addr_offset_valid_p): Use the iv mem_type field as the type
1829 of the addressed memory.
1830 (add_autoinc_candidates): Likewise.
1831 (get_address_cost): Likewise.
1832 (split_small_address_groups_p): Use address_p.
1833 (split_address_groups): Likewise.
1834 (add_iv_candidate_for_use): Likewise.
1835 (autoinc_possible_for_pair): Likewise.
1836 (rewrite_groups): Likewise.
1837 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
1838 (determine_group_iv_cost): Update after split of USE_ADDRESS.
1839 (get_alias_ptr_type_for_ptr_address): New function.
1840 (rewrite_use_address): Rewrite address uses in calls that were
1841 identified by find_address_like_use.
1843 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1844 Alan Hayward <alan.hayward@arm.com>
1845 David Sherwood <david.sherwood@arm.com>
1847 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
1849 * gimple-expr.h (is_gimple_addressable: Likewise.
1850 * gimple-expr.c (is_gimple_address): Likewise.
1851 * internal-fn.c (expand_call_mem_ref): New function.
1852 (expand_mask_load_optab_fn): Use it.
1853 (expand_mask_store_optab_fn): Likewise.
1855 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1856 Alan Hayward <alan.hayward@arm.com>
1857 David Sherwood <david.sherwood@arm.com>
1859 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
1860 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
1861 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
1862 (cond_umax@var{mode}): Document.
1863 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
1864 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
1865 (cond_umin_optab, cond_umax_optab): New optabs.
1866 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
1867 (COND_IOR, COND_XOR): New internal functions.
1868 * internal-fn.h (get_conditional_internal_fn): Declare.
1869 * internal-fn.c (cond_binary_direct): New macro.
1870 (expand_cond_binary_optab_fn): Likewise.
1871 (direct_cond_binary_optab_supported_p): Likewise.
1872 (get_conditional_internal_fn): New function.
1873 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
1874 Cope with reduction statements that are vectorized as calls rather
1876 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
1877 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
1878 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
1879 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
1880 (UNSPEC_COND_EOR): New unspecs.
1881 (optab): Add mappings for them.
1882 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
1883 (sve_int_op, sve_fp_op): New int attributes.
1885 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1886 Alan Hayward <alan.hayward@arm.com>
1887 David Sherwood <david.sherwood@arm.com>
1889 * optabs.def (while_ult_optab): New optab.
1890 * doc/md.texi (while_ult@var{m}@var{n}): Document.
1891 * internal-fn.def (WHILE_ULT): New internal function.
1892 * internal-fn.h (direct_internal_fn_supported_p): New override
1893 that takes two types as argument.
1894 * internal-fn.c (while_direct): New macro.
1895 (expand_while_optab_fn): New function.
1896 (convert_optab_supported_p): Likewise.
1897 (direct_while_optab_supported_p): New macro.
1898 * wide-int.h (wi::udiv_ceil): New function.
1899 * tree-vectorizer.h (rgroup_masks): New structure.
1900 (vec_loop_masks): New typedef.
1901 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
1903 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
1904 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
1905 (vect_max_vf): New function.
1906 (slpeel_make_loop_iterate_ntimes): Delete.
1907 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
1908 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
1909 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
1910 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
1911 internal-fn.h, stor-layout.h and optabs-query.h.
1912 (vect_set_loop_mask): New function.
1913 (add_preheader_seq): Likewise.
1914 (add_header_seq): Likewise.
1915 (interleave_supported_p): Likewise.
1916 (vect_maybe_permute_loop_masks): Likewise.
1917 (vect_set_loop_masks_directly): Likewise.
1918 (vect_set_loop_condition_masked): Likewise.
1919 (vect_set_loop_condition_unmasked): New function, split out from
1920 slpeel_make_loop_iterate_ntimes.
1921 (slpeel_make_loop_iterate_ntimes): Rename to..
1922 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
1923 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
1924 (vect_do_peeling): Update call accordingly.
1925 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
1927 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1928 mask_compare_type, can_fully_mask_p and fully_masked_p.
1929 (release_vec_loop_masks): New function.
1930 (_loop_vec_info): Use it to free the loop masks.
1931 (can_produce_all_loop_masks_p): New function.
1932 (vect_get_max_nscalars_per_iter): Likewise.
1933 (vect_verify_full_masking): Likewise.
1934 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
1935 retries, and free the mask rgroups before retrying. Check loop-wide
1936 reasons for disallowing fully-masked loops. Make the final decision
1937 about whether use a fully-masked loop or not.
1938 (vect_estimate_min_profitable_iters): Do not assume that peeling
1939 for the number of iterations will be needed for fully-masked loops.
1940 (vectorizable_reduction): Disable fully-masked loops.
1941 (vectorizable_live_operation): Likewise.
1942 (vect_halve_mask_nunits): New function.
1943 (vect_double_mask_nunits): Likewise.
1944 (vect_record_loop_mask): Likewise.
1945 (vect_get_loop_mask): Likewise.
1946 (vect_transform_loop): Handle the case in which the final loop
1947 iteration might handle a partial vector. Call vect_set_loop_condition
1948 instead of slpeel_make_loop_iterate_ntimes.
1949 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
1950 (check_load_store_masking): New function.
1951 (prepare_load_store_mask): Likewise.
1952 (vectorizable_store): Handle fully-masked loops.
1953 (vectorizable_load): Likewise.
1954 (supportable_widening_operation): Use vect_halve_mask_nunits for
1956 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
1957 (vect_gen_while): New function.
1958 * config/aarch64/aarch64.md (umax<mode>3): New expander.
1959 (aarch64_uqdec<mode>): New insn.
1961 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1962 Alan Hayward <alan.hayward@arm.com>
1963 David Sherwood <david.sherwood@arm.com>
1965 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
1966 (reduc_xor_scal_optab): New optabs.
1967 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
1968 (reduc_xor_scal_@var{m}): Document.
1969 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
1970 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
1972 * fold-const-call.c (fold_const_call): Handle them.
1973 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
1974 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
1975 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
1976 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
1977 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
1978 (UNSPEC_XORV): New unspecs.
1979 (optab): Add entries for them.
1980 (BITWISEV): New int iterator.
1981 (bit_reduc_op): New int attributes.
1983 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1984 Alan Hayward <alan.hayward@arm.com>
1985 David Sherwood <david.sherwood@arm.com>
1987 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
1988 * internal-fn.def (VEC_SHL_INSERT): New internal function.
1989 * optabs.def (vec_shl_insert_optab): New optab.
1990 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
1991 (duplicate_and_interleave): Likewise.
1992 * tree-vect-loop.c: Include internal-fn.h.
1993 (neutral_op_for_slp_reduction): New function, split out from
1994 get_initial_defs_for_reduction.
1995 (get_initial_def_for_reduction): Handle option 2 for variable-length
1996 vectors by loading the neutral value into a vector and then shifting
1997 the initial value into element 0.
1998 (get_initial_defs_for_reduction): Replace the code argument with
1999 the neutral value calculated by neutral_op_for_slp_reduction.
2000 Use gimple_build_vector for constant-length vectors.
2001 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2002 but the first group_size elements have a neutral value.
2003 Use duplicate_and_interleave otherwise.
2004 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2005 Update call to get_initial_defs_for_reduction. Handle SLP
2006 reductions for variable-length vectors by creating one vector
2007 result for each scalar result, with the elements associated
2008 with other scalar results stubbed out with the neutral value.
2009 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2010 Require IFN_VEC_SHL_INSERT for double reductions on
2011 variable-length vectors, or SLP reductions that have
2012 a neutral value. Require can_duplicate_and_interleave_p
2013 support for variable-length unchained SLP reductions if there
2014 is no neutral value, such as for MIN/MAX reductions. Also require
2015 the number of vector elements to be a multiple of the number of
2016 SLP statements when doing variable-length unchained SLP reductions.
2017 Update call to vect_create_epilog_for_reduction.
2018 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2019 and remove initial values.
2020 (duplicate_and_interleave): Make public.
2021 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2022 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2024 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2025 Alan Hayward <alan.hayward@arm.com>
2026 David Sherwood <david.sherwood@arm.com>
2028 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2029 (can_duplicate_and_interleave_p): New function.
2030 (vect_get_and_check_slp_defs): Take the vector of statements
2031 rather than just the current one. Remove excess parentheses.
2032 Restriction rejectinon of vect_constant_def and vect_external_def
2033 for variable-length vectors to boolean types, or types for which
2034 can_duplicate_and_interleave_p is false.
2035 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2036 (duplicate_and_interleave): New function.
2037 (vect_get_constant_vectors): Use gimple_build_vector for
2038 constant-length vectors and suitable variable-length constant
2039 vectors. Use duplicate_and_interleave for other variable-length
2040 vectors. Don't defer the update when inserting new statements.
2042 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2043 Alan Hayward <alan.hayward@arm.com>
2044 David Sherwood <david.sherwood@arm.com>
2046 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2047 min_profitable_iters doesn't go negative.
2049 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2050 Alan Hayward <alan.hayward@arm.com>
2051 David Sherwood <david.sherwood@arm.com>
2053 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2054 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2055 * optabs.def (vec_mask_load_lanes_optab): New optab.
2056 (vec_mask_store_lanes_optab): Likewise.
2057 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2058 (MASK_STORE_LANES): Likewise.
2059 * internal-fn.c (mask_load_lanes_direct): New macro.
2060 (mask_store_lanes_direct): Likewise.
2061 (expand_mask_load_optab_fn): Handle masked operations.
2062 (expand_mask_load_lanes_optab_fn): New macro.
2063 (expand_mask_store_optab_fn): Handle masked operations.
2064 (expand_mask_store_lanes_optab_fn): New macro.
2065 (direct_mask_load_lanes_optab_supported_p): Likewise.
2066 (direct_mask_store_lanes_optab_supported_p): Likewise.
2067 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2069 (vect_load_lanes_supported): Likewise.
2070 * tree-vect-data-refs.c (strip_conversion): New function.
2071 (can_group_stmts_p): Likewise.
2072 (vect_analyze_data_ref_accesses): Use it instead of checking
2073 for a pair of assignments.
2074 (vect_store_lanes_supported): Take a masked_p parameter.
2075 (vect_load_lanes_supported): Likewise.
2076 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2077 vect_store_lanes_supported and vect_load_lanes_supported.
2078 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2079 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2080 parameter. Don't allow gaps for masked accesses.
2081 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2082 and vect_load_lanes_supported.
2083 (get_load_store_type): Take a masked_p parameter and update
2084 call to get_group_load_store_type.
2085 (vectorizable_store): Update call to get_load_store_type.
2086 Handle IFN_MASK_STORE_LANES.
2087 (vectorizable_load): Update call to get_load_store_type.
2088 Handle IFN_MASK_LOAD_LANES.
2090 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2091 Alan Hayward <alan.hayward@arm.com>
2092 David Sherwood <david.sherwood@arm.com>
2094 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2096 * config/aarch64/aarch64-protos.h
2097 (aarch64_sve_struct_memory_operand_p): Declare.
2098 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2099 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2100 (VPRED, vpred): Handle SVE structure modes.
2101 * config/aarch64/constraints.md (Utx): New constraint.
2102 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2103 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2104 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2105 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2106 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2107 structure modes. Split into pieces after RA.
2108 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2109 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2111 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2112 SVE structure modes.
2113 (aarch64_classify_address): Likewise.
2114 (sizetochar): Move earlier in file.
2115 (aarch64_print_operand): Handle SVE register lists.
2116 (aarch64_array_mode): New function.
2117 (aarch64_sve_struct_memory_operand_p): Likewise.
2118 (TARGET_ARRAY_MODE): Redefine.
2120 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2121 Alan Hayward <alan.hayward@arm.com>
2122 David Sherwood <david.sherwood@arm.com>
2124 * target.def (array_mode): New target hook.
2125 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2126 * doc/tm.texi: Regenerate.
2127 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2128 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2129 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2131 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2134 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2135 Alan Hayward <alan.hayward@arm.com>
2136 David Sherwood <david.sherwood@arm.com>
2138 * fold-const.c (fold_binary_loc): Check the argument types
2139 rather than the result type when testing for a vector operation.
2141 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2143 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2144 * doc/tm.texi: Regenerate.
2146 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2147 Alan Hayward <alan.hayward@arm.com>
2148 David Sherwood <david.sherwood@arm.com>
2150 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2151 (sve): Document new AArch64 extension.
2152 * doc/md.texi (w): Extend the description of the AArch64
2153 constraint to include SVE vectors.
2154 (Upl, Upa): Document new AArch64 predicate constraints.
2155 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2157 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2158 (msve-vector-bits=): New option.
2159 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2160 SVE when these are disabled.
2161 (sve): New extension.
2162 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2163 modes. Adjust their number of units based on aarch64_sve_vg.
2164 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2165 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2166 aarch64_addr_query_type.
2167 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2168 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2169 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2170 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2171 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2172 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2173 (aarch64_simd_imm_zero_p): Delete.
2174 (aarch64_check_zero_based_sve_index_immediate): Declare.
2175 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2176 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2177 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2178 (aarch64_sve_float_mul_immediate_p): Likewise.
2179 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2181 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2182 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2183 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2184 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2185 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2186 (aarch64_regmode_natural_size): Likewise.
2187 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2188 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2190 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2191 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2192 for VG and the SVE predicate registers.
2193 (V_ALIASES): Add a "z"-prefixed alias.
2194 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2195 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2196 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2197 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2198 (REG_CLASS_NAMES): Add entries for them.
2199 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2200 and the predicate registers.
2201 (aarch64_sve_vg): Declare.
2202 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2203 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2204 (REGMODE_NATURAL_SIZE): Define.
2205 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2207 * config/aarch64/aarch64.c: Include cfgrtl.h.
2208 (simd_immediate_info): Add a constructor for series vectors,
2209 and an associated step field.
2210 (aarch64_sve_vg): New variable.
2211 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2212 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2213 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2214 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2215 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2216 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2217 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2218 (aarch64_get_mask_mode): New functions.
2219 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2220 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2221 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2222 predicate modes and predicate registers. Explicitly restrict
2223 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2224 to store a vector mode if it is recognized by
2225 aarch64_classify_vector_mode.
2226 (aarch64_regmode_natural_size): New function.
2227 (aarch64_hard_regno_caller_save_mode): Return the original mode
2229 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2230 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2231 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2232 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2234 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2235 does not overlap dest if the function is frame-related. Handle
2237 (aarch64_split_add_offset): New function.
2238 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2239 them aarch64_add_offset.
2240 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2241 and update call to aarch64_sub_sp.
2242 (aarch64_add_cfa_expression): New function.
2243 (aarch64_expand_prologue): Pass extra temporary registers to the
2244 functions above. Handle the case in which we need to emit new
2245 DW_CFA_expressions for registers that were originally saved
2246 relative to the stack pointer, but now have to be expressed
2247 relative to the frame pointer.
2248 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2250 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2251 IP0 and IP1 values for SVE frames.
2252 (aarch64_expand_vec_series): New function.
2253 (aarch64_expand_sve_widened_duplicate): Likewise.
2254 (aarch64_expand_sve_const_vector): Likewise.
2255 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2256 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2257 into the register, rather than emitting a SET directly.
2258 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2259 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2260 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2261 (offset_9bit_signed_scaled_p): New functions.
2262 (aarch64_replicate_bitmask_imm): New function.
2263 (aarch64_bitmask_imm): Use it.
2264 (aarch64_cannot_force_const_mem): Reject expressions involving
2265 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2266 (aarch64_classify_index): Handle SVE indices, by requiring
2267 a plain register index with a scale that matches the element size.
2268 (aarch64_classify_address): Handle SVE addresses. Assert that
2269 the mode of the address is VOIDmode or an integer mode.
2270 Update call to aarch64_classify_symbol.
2271 (aarch64_classify_symbolic_expression): Update call to
2272 aarch64_classify_symbol.
2273 (aarch64_const_vec_all_in_range_p): New function.
2274 (aarch64_print_vector_float_operand): Likewise.
2275 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2276 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2277 and the FP immediates 1.0 and 0.5.
2278 (aarch64_print_address_internal): Handle SVE addresses.
2279 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2280 (aarch64_regno_regclass): Handle predicate registers.
2281 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2283 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2284 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2285 (aarch64_convert_sve_vector_bits): New function.
2286 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2287 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2289 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2290 Handle SVE vector and predicate modes. Accept VL-based constants
2291 that need only one temporary register, and VL offsets that require
2292 no temporary registers.
2293 (aarch64_conditional_register_usage): Mark the predicate registers
2294 as fixed if SVE isn't available.
2295 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2296 Return true for SVE vector and predicate modes.
2297 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2298 rather than an unsigned int. Handle SVE modes.
2299 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2301 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2303 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2304 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2305 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2306 (aarch64_sve_float_mul_immediate_p): New functions.
2307 (aarch64_sve_valid_immediate): New function.
2308 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2309 Explicitly reject structure modes. Check for INDEX constants.
2310 Handle PTRUE and PFALSE constants.
2311 (aarch64_check_zero_based_sve_index_immediate): New function.
2312 (aarch64_simd_imm_zero_p): Delete.
2313 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2314 vector modes. Accept constants in the range of CNT[BHWD].
2315 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2316 ask for an Advanced SIMD mode.
2317 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2318 (aarch64_simd_vector_alignment): Handle SVE predicates.
2319 (aarch64_vectorize_preferred_vector_alignment): New function.
2320 (aarch64_simd_vector_alignment_reachable): Use it instead of
2322 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2323 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2325 (MAX_VECT_LEN): Delete.
2326 (expand_vec_perm_d): Add a vec_flags field.
2327 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2328 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2329 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2331 (aarch64_evpc_rev): Rename to...
2332 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2333 (aarch64_evpc_rev_global): New function.
2334 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2335 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2337 (aarch64_evpc_sve_tbl): New function.
2338 (aarch64_expand_vec_perm_const_1): Update after rename of
2339 aarch64_evpc_rev. Handle SVE permutes too, trying
2340 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2341 than aarch64_evpc_tbl.
2342 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2343 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2344 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2345 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2346 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2347 (aarch64_expand_sve_vcond): New functions.
2348 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2349 of aarch64_vector_mode_p.
2350 (aarch64_dwarf_poly_indeterminate_value): New function.
2351 (aarch64_compute_pressure_classes): Likewise.
2352 (aarch64_can_change_mode_class): Likewise.
2353 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2354 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2355 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2356 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2357 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2358 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2359 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2360 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2362 (Dn, Dl, Dr): Accept const as well as const_vector.
2363 (Dz): Likewise. Compare against CONST0_RTX.
2364 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2365 of "vector" where appropriate.
2366 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2367 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2368 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2369 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2370 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2371 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2372 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2373 (v_int_equiv): Extend to SVE modes.
2374 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2376 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2377 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2378 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2379 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2380 (SVE_COND_FP_CMP): New int iterators.
2381 (perm_hilo): Handle the new unpack unspecs.
2382 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2384 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2385 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2386 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2387 (aarch64_equality_operator, aarch64_constant_vector_operand)
2388 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2389 (aarch64_sve_nonimmediate_operand): Likewise.
2390 (aarch64_sve_general_operand): Likewise.
2391 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2392 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2393 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2394 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2395 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2396 (aarch64_sve_float_arith_immediate): Likewise.
2397 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2398 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2399 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2400 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2401 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2402 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2403 (aarch64_sve_float_arith_operand): Likewise.
2404 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2405 (aarch64_sve_float_mul_operand): Likewise.
2406 (aarch64_sve_vec_perm_operand): Likewise.
2407 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2408 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2409 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2410 as well as const_vector.
2411 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2412 in file. Use CONST0_RTX and CONSTM1_RTX.
2413 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2414 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2415 Use aarch64_simd_imm_zero.
2416 * config/aarch64/aarch64-sve.md: New file.
2417 * config/aarch64/aarch64.md: Include it.
2418 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2419 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2420 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2421 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2422 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2423 (sve): New attribute.
2424 (enabled): Disable instructions with the sve attribute unless
2426 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2427 aarch64_expand_mov_immediate.
2428 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2429 CNT[BHSD] immediates.
2430 (movti): Split CONST_POLY_INT moves into two halves.
2431 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2432 Split additions that need a temporary here if the destination
2433 is the stack pointer.
2434 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2435 (*add<mode>3_poly_1): New instruction.
2436 (set_clobber_cc): New expander.
2438 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2440 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2441 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2442 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2443 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2444 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2445 Change innermode from fixed_mode_size to machine_mode.
2446 (simplify_subreg): Update call accordingly. Handle a constant-sized
2447 subreg of a variable-length CONST_VECTOR.
2449 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2450 Alan Hayward <alan.hayward@arm.com>
2451 David Sherwood <david.sherwood@arm.com>
2453 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2454 (add_offset_to_base): New function, split out from...
2455 (create_mem_ref): ...here. When handling a scale other than 1,
2456 check first whether the address is valid without the offset.
2457 Add it into the base if so, leaving the index and scale as-is.
2459 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2462 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2463 fold_for_warn before checking if arg2 is INTEGER_CST.
2465 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2467 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2468 (store_multiple_operation): Delete.
2469 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2470 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2471 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2472 guarded by TARGET_STRING.
2473 (rs6000_output_load_multiple): Delete.
2474 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2475 OPTION_MASK_STRING / TARGET_STRING handling.
2476 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2477 (const rs6000_opt_masks) <"string">: Change mask to 0.
2478 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2479 (MASK_STRING): Delete.
2480 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2482 (load_multiple): Delete.
2489 (store_multiple): Delete.
2496 (movmemsi_8reg): Delete.
2497 (corresponding unnamed define_insn): Delete.
2498 (movmemsi_6reg): Delete.
2499 (corresponding unnamed define_insn): Delete.
2500 (movmemsi_4reg): Delete.
2501 (corresponding unnamed define_insn): Delete.
2502 (movmemsi_2reg): Delete.
2503 (corresponding unnamed define_insn): Delete.
2504 (movmemsi_1reg): Delete.
2505 (corresponding unnamed define_insn): Delete.
2506 * config/rs6000/rs6000.opt (mno-string): New.
2507 (mstring): Replace by deprecation warning stub.
2508 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2510 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2512 * regrename.c (regrename_do_replace): If replacing the same
2513 reg multiple times, try to reuse last created gen_raw_REG.
2516 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2517 main to workaround a bug in GDB.
2519 2018-01-12 Tom de Vries <tom@codesourcery.com>
2522 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2524 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2526 PR rtl-optimization/80481
2527 * ira-color.c (get_cap_member): New function.
2528 (allocnos_conflict_by_live_ranges_p): Use it.
2529 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2530 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2532 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2535 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2536 (*saddl_se_1): Ditto.
2538 (*ssubl_se_1): Ditto.
2540 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2542 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2543 rather than wi::to_widest for DR_INITs.
2544 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2545 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2546 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2548 (vect_analyze_group_access_1): Note that here.
2550 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2552 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2553 polynomial type sizes.
2555 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2557 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2558 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2559 (gimple_add_tmp_var): Likewise.
2561 2018-01-12 Martin Liska <mliska@suse.cz>
2563 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2564 (gimple_alloc_sizes): Likewise.
2565 (dump_gimple_statistics): Use PRIu64 in printf format.
2566 * gimple.h: Change uint64_t to int.
2568 2018-01-12 Martin Liska <mliska@suse.cz>
2570 * tree-core.h: Use uint64_t instead of int.
2571 * tree.c (tree_node_counts): Likewise.
2572 (tree_node_sizes): Likewise.
2573 (dump_tree_statistics): Use PRIu64 in printf format.
2575 2018-01-12 Martin Liska <mliska@suse.cz>
2577 * Makefile.in: As qsort_chk is implemented in vec.c, add
2578 vec.o to linkage of gencfn-macros.
2579 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2580 passing the info to record_node_allocation_statistics.
2581 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2583 * ggc-common.c (struct ggc_usage): Add operator== and use
2584 it in operator< and compare function.
2585 * mem-stats.h (struct mem_usage): Likewise.
2586 * vec.c (struct vec_usage): Remove operator< and compare
2587 function. Can be simply inherited.
2589 2018-01-12 Martin Jambor <mjambor@suse.cz>
2592 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2593 * tree-ssa-math-opts.c: Include domwalk.h.
2594 (convert_mult_to_fma_1): New function.
2595 (fma_transformation_info): New type.
2596 (fma_deferring_state): Likewise.
2597 (cancel_fma_deferring): New function.
2598 (result_of_phi): Likewise.
2599 (last_fma_candidate_feeds_initial_phi): Likewise.
2600 (convert_mult_to_fma): Added deferring logic, split actual
2601 transformation to convert_mult_to_fma_1.
2602 (math_opts_dom_walker): New type.
2603 (math_opts_dom_walker::after_dom_children): New method, body moved
2604 here from pass_optimize_widening_mul::execute, added deferring logic
2606 (pass_optimize_widening_mul::execute): Moved most of code to
2607 math_opts_dom_walker::after_dom_children.
2608 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2609 * config/i386/i386.c (ix86_option_override_internal): Added
2610 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2612 2018-01-12 Richard Biener <rguenther@suse.de>
2615 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2616 inline instance vars.
2618 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2621 * config/rx/rx.c (rx_is_restricted_memory_address):
2624 2018-01-12 Richard Biener <rguenther@suse.de>
2626 PR tree-optimization/80846
2627 * target.def (split_reduction): New target hook.
2628 * targhooks.c (default_split_reduction): New function.
2629 * targhooks.h (default_split_reduction): Declare.
2630 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2631 target requests first reduce vectors by combining low and high
2633 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2634 (get_vectype_for_scalar_type_and_size): Export.
2635 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2636 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2637 * doc/tm.texi: Regenerate.
2638 * config/i386/i386.c (ix86_split_reduction): Implement
2639 TARGET_VECTORIZE_SPLIT_REDUCTION.
2641 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2644 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2645 in PIC mode except for TARGET_VXWORKS_RTP.
2646 * config/sparc/sparc.c: Include cfgrtl.h.
2647 (TARGET_INIT_PIC_REG): Define.
2648 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2649 (sparc_pic_register_p): New predicate.
2650 (sparc_legitimate_address_p): Use it.
2651 (sparc_legitimize_pic_address): Likewise.
2652 (sparc_delegitimize_address): Likewise.
2653 (sparc_mode_dependent_address_p): Likewise.
2654 (gen_load_pcrel_sym): Remove 4th parameter.
2655 (load_got_register): Adjust call to above. Remove obsolete stuff.
2656 (sparc_expand_prologue): Do not call load_got_register here.
2657 (sparc_flat_expand_prologue): Likewise.
2658 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2659 (sparc_use_pseudo_pic_reg): New function.
2660 (sparc_init_pic_reg): Likewise.
2661 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2662 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2664 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2666 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2667 Add item for branch_cost.
2669 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2671 PR rtl-optimization/83565
2672 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2673 not extend the result to a larger mode for rotate operations.
2674 (num_sign_bit_copies1): Likewise.
2676 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2679 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2681 Use values-Xc.o for -pedantic.
2682 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2684 2018-01-12 Martin Liska <mliska@suse.cz>
2687 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2689 (possible_polymorphic_call_targets): Use it.
2690 (ipa_devirt): Likewise.
2692 2018-01-12 Martin Liska <mliska@suse.cz>
2694 * profile-count.h (enum profile_quality): Use 0 as invalid
2695 enum value of profile_quality.
2697 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2699 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2700 -mext-string options.
2702 2018-01-12 Richard Biener <rguenther@suse.de>
2704 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2705 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2706 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2708 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2710 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2712 * configure.ac (--with-long-double-format): Add support for the
2713 configuration option to change the default long double format on
2715 * config.gcc (powerpc*-linux*-*): Likewise.
2716 * configure: Regenerate.
2717 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2718 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2719 used without modification.
2721 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2723 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2724 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2725 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2726 MISC_BUILTIN_SPEC_BARRIER.
2727 (rs6000_init_builtins): Likewise.
2728 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2730 (speculation_barrier): New define_insn.
2731 * doc/extend.texi: Document __builtin_speculation_barrier.
2733 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2736 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2737 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2738 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2740 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2741 integral modes instead of "ss" and "sd".
2742 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2743 vectors with 32-bit and 64-bit elements.
2744 (vecdupssescalarmodesuffix): New mode attribute.
2745 (vec_dup<mode>): Use it.
2747 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2750 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2751 frame if argument is passed on stack.
2753 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2756 * ree.c (combine_reaching_defs): Optimize also
2757 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
2758 reg2=any_extend(exp); reg1=reg2;, formatting fix.
2760 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2763 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
2765 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2768 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
2769 after they are computed.
2771 2018-01-11 Bin Cheng <bin.cheng@arm.com>
2773 PR tree-optimization/83695
2774 * gimple-loop-linterchange.cc
2775 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
2776 reset cached scev information after interchange.
2777 (pass_linterchange::execute): Remove call to scev_reset_htab.
2779 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2781 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
2782 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
2783 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
2784 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
2785 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
2786 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
2787 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
2788 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
2789 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
2790 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
2791 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
2792 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
2793 (V_lane_reg): Likewise.
2794 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
2796 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
2797 (vfmal_lane_low<mode>_intrinsic,
2798 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
2799 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
2800 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
2801 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
2802 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
2803 vfmsl_lane_high<mode>_intrinsic): New define_insns.
2805 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2807 * config/arm/arm-cpus.in (fp16fml): New feature.
2808 (ALL_SIMD): Add fp16fml.
2809 (armv8.2-a): Add fp16fml as an option.
2810 (armv8.3-a): Likewise.
2811 (armv8.4-a): Add fp16fml as part of fp16.
2812 * config/arm/arm.h (TARGET_FP16FML): Define.
2813 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
2815 * config/arm/arm-modes.def (V2HF): Define.
2816 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
2817 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
2818 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
2819 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
2820 vfmsl_low, vfmsl_high): New set of builtins.
2821 * config/arm/iterators.md (PLUSMINUS): New code iterator.
2822 (vfml_op): New code attribute.
2823 (VFMLHALVES): New int iterator.
2824 (VFML, VFMLSEL): New mode attributes.
2825 (V_reg): Define mapping for V2HF.
2826 (V_hi, V_lo): New mode attributes.
2827 (VF_constraint): Likewise.
2828 (vfml_half, vfml_half_selector): New int attributes.
2829 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
2831 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
2832 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
2834 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
2835 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
2836 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
2837 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
2839 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
2840 Document new effective target and option set.
2842 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2844 * config/arm/arm-cpus.in (armv8_4): New feature.
2845 (ARMv8_4a): New fgroup.
2846 (armv8.4-a): New arch.
2847 * config/arm/arm-tables.opt: Regenerate.
2848 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
2849 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
2850 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
2851 Add matching rules for -march=armv8.4-a and extensions.
2852 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
2854 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
2857 * config/rx/rx.md (BW): New mode attribute.
2858 (sync_lock_test_and_setsi): Add mode suffix to insn output.
2860 2018-01-11 Richard Biener <rguenther@suse.de>
2862 PR tree-optimization/83435
2863 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
2864 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
2865 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
2867 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2868 Alan Hayward <alan.hayward@arm.com>
2869 David Sherwood <david.sherwood@arm.com>
2871 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
2873 (aarch64_classify_address): Initialize it. Track polynomial offsets.
2874 (aarch64_print_address_internal): Use it to check for a zero offset.
2876 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2877 Alan Hayward <alan.hayward@arm.com>
2878 David Sherwood <david.sherwood@arm.com>
2880 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
2881 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
2882 Return a poly_int64 rather than a HOST_WIDE_INT.
2883 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
2884 rather than a HOST_WIDE_INT.
2885 * config/aarch64/aarch64.h (aarch64_frame): Protect with
2886 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
2887 hard_fp_offset, frame_size, initial_adjust, callee_offset and
2888 final_offset from HOST_WIDE_INT to poly_int64.
2889 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
2890 to_constant when getting the number of units in an Advanced SIMD
2892 (aarch64_builtin_vectorized_function): Check for a constant number
2894 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
2896 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
2897 attribute instead of GET_MODE_NUNITS.
2898 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
2899 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
2900 GET_MODE_SIZE for fixed-size registers.
2901 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
2902 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
2903 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
2904 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
2905 (aarch64_print_operand, aarch64_print_address_internal)
2906 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
2907 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
2908 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
2909 Handle polynomial GET_MODE_SIZE.
2910 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
2911 wider than SImode without modification.
2912 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
2913 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
2914 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
2915 passing and returning SVE modes.
2916 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
2917 rather than GEN_INT.
2918 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
2919 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
2920 (aarch64_allocate_and_probe_stack_space): Likewise.
2921 (aarch64_layout_frame): Cope with polynomial offsets.
2922 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
2923 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
2925 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
2926 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
2927 poly_int64 rather than a HOST_WIDE_INT.
2928 (aarch64_get_separate_components, aarch64_process_components)
2929 (aarch64_expand_prologue, aarch64_expand_epilogue)
2930 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
2931 (aarch64_anchor_offset): New function, split out from...
2932 (aarch64_legitimize_address): ...here.
2933 (aarch64_builtin_vectorization_cost): Handle polynomial
2934 TYPE_VECTOR_SUBPARTS.
2935 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
2937 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
2938 number of elements from the PARALLEL rather than the mode.
2939 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
2940 rather than GET_MODE_BITSIZE.
2941 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
2942 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
2943 (aarch64_expand_vec_perm_const_1): Handle polynomial
2944 d->perm.length () and d->perm elements.
2945 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
2946 Apply to_constant to d->perm elements.
2947 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
2948 polynomial CONST_VECTOR_NUNITS.
2949 (aarch64_move_pointer): Take amount as a poly_int64 rather
2951 (aarch64_progress_pointer): Avoid temporary variable.
2952 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
2953 the mode attribute instead of GET_MODE.
2955 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2956 Alan Hayward <alan.hayward@arm.com>
2957 David Sherwood <david.sherwood@arm.com>
2959 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
2960 x exists before using it.
2961 (aarch64_add_constant_internal): Rename to...
2962 (aarch64_add_offset_1): ...this. Replace regnum with separate
2963 src and dest rtxes. Handle the case in which they're different,
2964 including when the offset is zero. Replace scratchreg with an rtx.
2965 Use 2 additions if there is no spare register into which we can
2966 move a 16-bit constant.
2967 (aarch64_add_constant): Delete.
2968 (aarch64_add_offset): Replace reg with separate src and dest
2969 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
2970 Use aarch64_add_offset_1.
2971 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
2972 an rtx rather than an int. Take the delta as a poly_int64
2973 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
2974 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
2975 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
2976 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
2977 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
2979 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
2980 aarch64_add_constant.
2982 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2984 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
2985 Use scalar_float_mode.
2987 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2989 * config/aarch64/aarch64-simd.md
2990 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
2991 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
2992 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
2993 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
2994 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
2995 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
2996 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
2997 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
2998 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
2999 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3001 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3004 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3005 targ_options->x_arm_arch_string is non NULL.
3007 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3009 * config/aarch64/aarch64.h
3010 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3012 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3015 * expmed.c (emit_store_flag_force): Swap if const op0
3016 and change VOIDmode to mode of op0.
3018 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3020 PR rtl-optimization/83761
3021 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3022 than bytes to mode_for_size.
3024 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3027 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3028 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3031 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3034 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3035 when in layout mode.
3036 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3037 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3040 2018-01-10 Michael Collison <michael.collison@arm.com>
3042 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3043 * config/aarch64/aarch64-option-extension.def: Add
3044 AARCH64_OPT_EXTENSION of 'fp16fml'.
3045 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3046 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3047 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3048 * config/aarch64/constraints.md (Ui7): New constraint.
3049 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3050 (VFMLA_SEL_W): Ditto.
3053 (VFMLA16_LOW): New int iterator.
3054 (VFMLA16_HIGH): Ditto.
3055 (UNSPEC_FMLAL): New unspec.
3056 (UNSPEC_FMLSL): Ditto.
3057 (UNSPEC_FMLAL2): Ditto.
3058 (UNSPEC_FMLSL2): Ditto.
3059 (f16mac): New code attribute.
3060 * config/aarch64/aarch64-simd-builtins.def
3061 (aarch64_fmlal_lowv2sf): Ditto.
3062 (aarch64_fmlsl_lowv2sf): Ditto.
3063 (aarch64_fmlalq_lowv4sf): Ditto.
3064 (aarch64_fmlslq_lowv4sf): Ditto.
3065 (aarch64_fmlal_highv2sf): Ditto.
3066 (aarch64_fmlsl_highv2sf): Ditto.
3067 (aarch64_fmlalq_highv4sf): Ditto.
3068 (aarch64_fmlslq_highv4sf): Ditto.
3069 (aarch64_fmlal_lane_lowv2sf): Ditto.
3070 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3071 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3072 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3073 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3074 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3075 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3076 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3077 (aarch64_fmlal_lane_highv2sf): Ditto.
3078 (aarch64_fmlsl_lane_highv2sf): Ditto.
3079 (aarch64_fmlal_laneq_highv2sf): Ditto.
3080 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3081 (aarch64_fmlalq_lane_highv4sf): Ditto.
3082 (aarch64_fmlsl_lane_highv4sf): Ditto.
3083 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3084 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3085 * config/aarch64/aarch64-simd.md:
3086 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3087 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3088 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3089 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3090 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3091 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3092 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3093 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3094 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3095 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3096 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3097 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3098 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3099 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3100 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3101 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3102 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3103 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3104 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3105 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3106 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3107 (vfmlsl_low_u32): Ditto.
3108 (vfmlalq_low_u32): Ditto.
3109 (vfmlslq_low_u32): Ditto.
3110 (vfmlal_high_u32): Ditto.
3111 (vfmlsl_high_u32): Ditto.
3112 (vfmlalq_high_u32): Ditto.
3113 (vfmlslq_high_u32): Ditto.
3114 (vfmlal_lane_low_u32): Ditto.
3115 (vfmlsl_lane_low_u32): Ditto.
3116 (vfmlal_laneq_low_u32): Ditto.
3117 (vfmlsl_laneq_low_u32): Ditto.
3118 (vfmlalq_lane_low_u32): Ditto.
3119 (vfmlslq_lane_low_u32): Ditto.
3120 (vfmlalq_laneq_low_u32): Ditto.
3121 (vfmlslq_laneq_low_u32): Ditto.
3122 (vfmlal_lane_high_u32): Ditto.
3123 (vfmlsl_lane_high_u32): Ditto.
3124 (vfmlal_laneq_high_u32): Ditto.
3125 (vfmlsl_laneq_high_u32): Ditto.
3126 (vfmlalq_lane_high_u32): Ditto.
3127 (vfmlslq_lane_high_u32): Ditto.
3128 (vfmlalq_laneq_high_u32): Ditto.
3129 (vfmlslq_laneq_high_u32): Ditto.
3130 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3131 (AARCH64_FL_FOR_ARCH8_4): New.
3132 (AARCH64_ISA_F16FML): New ISA flag.
3133 (TARGET_F16FML): New feature flag for fp16fml.
3134 (doc/invoke.texi): Document new fp16fml option.
3136 2018-01-10 Michael Collison <michael.collison@arm.com>
3138 * config/aarch64/aarch64-builtins.c:
3139 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3140 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3141 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3142 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3143 (AARCH64_ISA_SHA3): New ISA flag.
3144 (TARGET_SHA3): New feature flag for sha3.
3145 * config/aarch64/iterators.md (sha512_op): New int attribute.
3146 (CRYPTO_SHA512): New int iterator.
3147 (UNSPEC_SHA512H): New unspec.
3148 (UNSPEC_SHA512H2): Ditto.
3149 (UNSPEC_SHA512SU0): Ditto.
3150 (UNSPEC_SHA512SU1): Ditto.
3151 * config/aarch64/aarch64-simd-builtins.def
3152 (aarch64_crypto_sha512hqv2di): New builtin.
3153 (aarch64_crypto_sha512h2qv2di): Ditto.
3154 (aarch64_crypto_sha512su0qv2di): Ditto.
3155 (aarch64_crypto_sha512su1qv2di): Ditto.
3156 (aarch64_eor3qv8hi): Ditto.
3157 (aarch64_rax1qv2di): Ditto.
3158 (aarch64_xarqv2di): Ditto.
3159 (aarch64_bcaxqv8hi): Ditto.
3160 * config/aarch64/aarch64-simd.md:
3161 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
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/arm_neon.h (vsha512hq_u64): New intrinsic.
3169 (vsha512h2q_u64): Ditto.
3170 (vsha512su0q_u64): Ditto.
3171 (vsha512su1q_u64): Ditto.
3172 (veor3q_u16): Ditto.
3173 (vrax1q_u64): Ditto.
3175 (vbcaxq_u16): Ditto.
3176 * config/arm/types.md (crypto_sha512): New type attribute.
3177 (crypto_sha3): Ditto.
3178 (doc/invoke.texi): Document new sha3 option.
3180 2018-01-10 Michael Collison <michael.collison@arm.com>
3182 * config/aarch64/aarch64-builtins.c:
3183 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3184 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3185 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3186 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3187 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3188 (AARCH64_ISA_SM4): New ISA flag.
3189 (TARGET_SM4): New feature flag for sm4.
3190 * config/aarch64/aarch64-simd-builtins.def
3191 (aarch64_sm3ss1qv4si): Ditto.
3192 (aarch64_sm3tt1aq4si): Ditto.
3193 (aarch64_sm3tt1bq4si): Ditto.
3194 (aarch64_sm3tt2aq4si): Ditto.
3195 (aarch64_sm3tt2bq4si): Ditto.
3196 (aarch64_sm3partw1qv4si): Ditto.
3197 (aarch64_sm3partw2qv4si): Ditto.
3198 (aarch64_sm4eqv4si): Ditto.
3199 (aarch64_sm4ekeyqv4si): Ditto.
3200 * config/aarch64/aarch64-simd.md:
3201 (aarch64_sm3ss1qv4si): Ditto.
3202 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3203 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3204 (aarch64_sm4eqv4si): Ditto.
3205 (aarch64_sm4ekeyqv4si): Ditto.
3206 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3207 (sm3part_op): Ditto.
3208 (CRYPTO_SM3TT): Ditto.
3209 (CRYPTO_SM3PART): Ditto.
3210 (UNSPEC_SM3SS1): New unspec.
3211 (UNSPEC_SM3TT1A): Ditto.
3212 (UNSPEC_SM3TT1B): Ditto.
3213 (UNSPEC_SM3TT2A): Ditto.
3214 (UNSPEC_SM3TT2B): Ditto.
3215 (UNSPEC_SM3PARTW1): Ditto.
3216 (UNSPEC_SM3PARTW2): Ditto.
3217 (UNSPEC_SM4E): Ditto.
3218 (UNSPEC_SM4EKEY): Ditto.
3219 * config/aarch64/constraints.md (Ui2): New constraint.
3220 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3221 * config/arm/types.md (crypto_sm3): New type attribute.
3222 (crypto_sm4): Ditto.
3223 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3224 (vsm3tt1aq_u32): Ditto.
3225 (vsm3tt1bq_u32): Ditto.
3226 (vsm3tt2aq_u32): Ditto.
3227 (vsm3tt2bq_u32): Ditto.
3228 (vsm3partw1q_u32): Ditto.
3229 (vsm3partw2q_u32): Ditto.
3230 (vsm4eq_u32): Ditto.
3231 (vsm4ekeyq_u32): Ditto.
3232 (doc/invoke.texi): Document new sm4 option.
3234 2018-01-10 Michael Collison <michael.collison@arm.com>
3236 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3237 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3238 (AARCH64_FL_FOR_ARCH8_4): New.
3239 (AARCH64_FL_V8_4): New flag.
3240 (doc/invoke.texi): Document new armv8.4-a option.
3242 2018-01-10 Michael Collison <michael.collison@arm.com>
3244 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3245 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3246 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3247 * config/aarch64/aarch64-option-extension.def: Add
3248 AARCH64_OPT_EXTENSION of 'sha2'.
3249 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3250 (crypto): Disable sha2 and aes if crypto disabled.
3251 (crypto): Enable aes and sha2 if enabled.
3252 (simd): Disable sha2 and aes if simd disabled.
3253 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3255 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3256 (TARGET_SHA2): New feature flag for sha2.
3257 (TARGET_AES): New feature flag for aes.
3258 * config/aarch64/aarch64-simd.md:
3259 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3260 conditional on TARGET_AES.
3261 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3262 (aarch64_crypto_sha1hsi): Make pattern conditional
3264 (aarch64_crypto_sha1hv4si): Ditto.
3265 (aarch64_be_crypto_sha1hv4si): Ditto.
3266 (aarch64_crypto_sha1su1v4si): Ditto.
3267 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3268 (aarch64_crypto_sha1su0v4si): Ditto.
3269 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3270 (aarch64_crypto_sha256su0v4si): Ditto.
3271 (aarch64_crypto_sha256su1v4si): Ditto.
3272 (doc/invoke.texi): Document new aes and sha2 options.
3274 2018-01-10 Martin Sebor <msebor@redhat.com>
3276 PR tree-optimization/83781
3277 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3280 2018-01-11 Martin Sebor <msebor@gmail.com>
3281 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3283 PR tree-optimization/83501
3284 PR tree-optimization/81703
3286 * tree-ssa-strlen.c (get_string_cst): Rename...
3287 (get_string_len): ...to this. Handle global constants.
3288 (handle_char_store): Adjust.
3290 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3291 Jim Wilson <jimw@sifive.com>
3293 * config/riscv/riscv-protos.h (riscv_output_return): New.
3294 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3295 (riscv_attribute_table, riscv_output_return),
3296 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3297 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3298 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3300 (riscv_expand_prologue): Add early return for naked function.
3301 (riscv_expand_epilogue): Likewise.
3302 (riscv_function_ok_for_sibcall): Return false for naked function.
3303 (riscv_set_current_function): New.
3304 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3305 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3306 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3307 * doc/extend.texi (RISC-V Function Attributes): New.
3309 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3311 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3312 check for 128-bit long double before checking TCmode.
3313 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3314 128-bit long doubles before checking TFmode or TCmode.
3315 (FLOAT128_IBM_P): Likewise.
3317 2018-01-10 Martin Sebor <msebor@redhat.com>
3319 PR tree-optimization/83671
3320 * builtins.c (c_strlen): Unconditionally return zero for the empty
3322 Use -Warray-bounds for warnings.
3323 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3324 for non-constant array indices with COMPONENT_REF, arrays of
3325 arrays, and pointers to arrays.
3326 (gimple_fold_builtin_strlen): Determine and set length range for
3327 non-constant character arrays.
3329 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3332 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3335 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3337 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3339 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3342 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3343 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3344 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3345 indexed_or_indirect_operand predicate.
3346 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3347 (*vsx_le_perm_load_v8hi): Likewise.
3348 (*vsx_le_perm_load_v16qi): Likewise.
3349 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3350 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3351 (*vsx_le_perm_store_v8hi): Likewise.
3352 (*vsx_le_perm_store_v16qi): Likewise.
3353 (eight unnamed splitters): Likewise.
3355 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3357 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3358 * config/rs6000/emmintrin.h: Likewise.
3359 * config/rs6000/mmintrin.h: Likewise.
3360 * config/rs6000/xmmintrin.h: Likewise.
3362 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3365 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3367 * tree.c (tree_nop_conversion): Return true for location wrapper
3369 (maybe_wrap_with_location): New function.
3370 (selftest::check_strip_nops): New function.
3371 (selftest::test_location_wrappers): New function.
3372 (selftest::tree_c_tests): Call it.
3373 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3374 (maybe_wrap_with_location): New decl.
3375 (EXPR_LOCATION_WRAPPER_P): New macro.
3376 (location_wrapper_p): New inline function.
3377 (tree_strip_any_location_wrapper): New inline function.
3379 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3382 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3383 stack_realign_offset for the largest alignment of stack slot
3385 (ix86_find_max_used_stack_alignment): New function.
3386 (ix86_finalize_stack_frame_flags): Use it. Set
3387 max_used_stack_alignment if we don't realign stack.
3388 * config/i386/i386.h (machine_function): Add
3389 max_used_stack_alignment.
3391 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3393 * config/arm/arm.opt (-mbranch-cost): New option.
3394 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3397 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3400 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3401 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3403 2018-01-10 Richard Biener <rguenther@suse.de>
3406 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3407 early out so it also covers the case where we have a non-NULL
3410 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3412 PR tree-optimization/83753
3413 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3414 for non-strided grouped accesses if the number of elements is 1.
3416 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3419 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3420 * i386.h (TARGET_USE_GATHER): Define.
3421 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3423 2018-01-10 Martin Liska <mliska@suse.cz>
3426 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3427 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3429 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3430 CLEANUP_NO_PARTITIONING is not set.
3432 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3434 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3435 for vectors, as a partial revert of r254296.
3436 * rtl.h (const_vec_p): Delete.
3437 (const_vec_duplicate_p): Don't test for vector CONSTs.
3438 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3439 * expmed.c (make_tree): Likewise.
3442 * common.md (E, F): Use CONSTANT_P instead of checking for
3444 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3445 checking for CONST_VECTOR.
3447 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3450 * predict.c (force_edge_cold): Handle in more sane way edges
3453 2018-01-09 Carl Love <cel@us.ibm.com>
3455 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3457 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3458 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3459 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3460 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3461 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3462 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3463 * config/rs6000/rs6000-protos.h: Add extern defition for
3464 rs6000_generate_float2_double_code.
3465 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3467 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3468 (float2_v2df): Add define_expand.
3470 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3473 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3474 op_mode in the force_to_mode call.
3476 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3478 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3479 instead of checking each element individually.
3480 (aarch64_evpc_uzp): Likewise.
3481 (aarch64_evpc_zip): Likewise.
3482 (aarch64_evpc_ext): Likewise.
3483 (aarch64_evpc_rev): Likewise.
3484 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3485 instead of checking each element individually. Return true without
3487 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3488 whether all selected elements come from the same input, instead of
3489 checking each element individually. Remove calls to gen_rtx_REG,
3490 start_sequence and end_sequence and instead assert that no rtl is
3493 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3495 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3496 order of HIGH and CONST checks.
3498 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3500 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3501 if the destination isn't an SSA_NAME.
3503 2018-01-09 Richard Biener <rguenther@suse.de>
3505 PR tree-optimization/83668
3506 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3508 (canonicalize_loop_form): ... here, renamed from ...
3509 (canonicalize_loop_closed_ssa_form): ... this and amended to
3510 swap successor edges for loop exit blocks to make us use
3511 the RPO order we need for initial schedule generation.
3513 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3515 PR tree-optimization/64811
3516 * match.pd: When optimizing comparisons with Inf, avoid
3517 introducing or losing exceptions from comparisons with NaN.
3519 2018-01-09 Martin Liska <mliska@suse.cz>
3522 * asan.c (shadow_mem_size): Add gcc_assert.
3524 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3526 Don't save registers in main().
3529 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3530 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3531 * config/avr/avr.c (avr_set_current_function): Don't error if
3532 naked, OS_task or OS_main are specified at the same time.
3533 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3535 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3537 * common/config/avr/avr-common.c (avr_option_optimization_table):
3538 Switch on -mmain-is-OS_task for optimizing compilations.
3540 2018-01-09 Richard Biener <rguenther@suse.de>
3542 PR tree-optimization/83572
3543 * graphite.c: Include cfganal.h.
3544 (graphite_transform_loops): Connect infinite loops to exit
3545 and remove fake edges at the end.
3547 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3549 * ipa-inline.c (edge_badness): Revert accidental checkin.
3551 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3554 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3555 symbols; not inline clones.
3557 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3560 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3561 hard registers. Formatting fixes.
3563 PR preprocessor/83722
3564 * gcc.c (try_generate_repro): Pass
3565 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3566 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3569 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3570 Kito Cheng <kito.cheng@gmail.com>
3572 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3573 (riscv_leaf_function_p): Delete.
3574 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3576 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3578 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3580 (do_ifelse): New function.
3581 (do_isel): New function.
3582 (do_sub3): New function.
3583 (do_add3): New function.
3584 (do_load_mask_compare): New function.
3585 (do_overlap_load_compare): New function.
3586 (expand_compare_loop): New function.
3587 (expand_block_compare): Call expand_compare_loop() when appropriate.
3588 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3590 (-mblock-compare-inline-loop-limit): New option.
3592 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3595 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3596 Reverse order of second and third operands in first alternative.
3597 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3598 of first and second elements in UNSPEC_VPERMR vector.
3599 (altivec_expand_vec_perm_le): Likewise.
3601 2017-01-08 Jeff Law <law@redhat.com>
3603 PR rtl-optimizatin/81308
3604 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3605 (process_switch): If group_case_labels makes a change, then set
3607 (pass_convert_switch::execute): If a switch is converted, then
3608 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3610 PR rtl-optimization/81308
3611 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3614 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3616 PR target/83663 - Revert r255946
3617 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3618 generation for cases where splatting a value is not useful.
3619 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3620 across a vec_duplicate and a paradoxical subreg forming a vector
3621 mode to a vec_concat.
3623 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3625 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3626 -march=armv8.3-a variants.
3627 * config/arm/t-multilib: Likewise.
3628 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3630 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3632 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3634 (cceq_ior_compare_complement): Give it a name so I can use it, and
3635 change boolean_or_operator predicate to boolean_operator so it can
3636 be used to generate a crand.
3637 (eqne): New code iterator.
3638 (bd/bd_neg): New code_attrs.
3639 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3640 a single define_insn.
3641 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3642 decrement (bdnzt/bdnzf/bdzt/bdzf).
3643 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3644 with the new names of the branch decrement patterns, and added the
3645 names of the branch decrement conditional patterns.
3647 2018-01-08 Richard Biener <rguenther@suse.de>
3649 PR tree-optimization/83563
3650 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3653 2018-01-08 Richard Biener <rguenther@suse.de>
3656 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3658 2018-01-08 Richard Biener <rguenther@suse.de>
3660 PR tree-optimization/83685
3661 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3662 references to abnormals.
3664 2018-01-08 Richard Biener <rguenther@suse.de>
3667 * dwarf2out.c (output_indirect_strings): Handle empty
3668 skeleton_debug_str_hash.
3669 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3671 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3673 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3674 (emit_store_direct): Likewise.
3675 (arc_trampoline_adjust_address): Likewise.
3676 (arc_asm_trampoline_template): New function.
3677 (arc_initialize_trampoline): Use asm_trampoline_template.
3678 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3679 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3680 * config/arc/arc.md (flush_icache): Delete pattern.
3682 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3684 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3685 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3688 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3691 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3692 by not USED_FOR_TARGET.
3693 (make_pass_resolve_sw_modes): Likewise.
3695 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3697 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3700 2018-01-08 Richard Biener <rguenther@suse.de>
3703 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3705 2018-01-08 Richard Biener <rguenther@suse.de>
3708 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3710 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3713 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3714 basic blocks with a small number of successors.
3715 (convert_control_dep_chain_into_preds): Improve handling of
3717 (dump_predicates): Split apart into...
3718 (dump_pred_chain): ...here...
3719 (dump_pred_info): ...and here.
3720 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3721 (can_chain_union_be_invalidated_p): Improve check for invalidation
3723 (uninit_uses_cannot_happen): Avoid unnecessary if
3724 convert_control_dep_chain_into_preds yielded nothing.
3726 2018-01-06 Martin Sebor <msebor@redhat.com>
3728 PR tree-optimization/83640
3729 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3730 subtracting negative offset from size.
3731 (builtin_access::overlap): Adjust offset bounds of the access to fall
3732 within the size of the object if possible.
3734 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3736 PR rtl-optimization/83699
3737 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3738 extract_bit_field_as_subreg to cases in which the extracted
3739 value is also a vector.
3741 * lra-constraints.c (process_alt_operands): Test for the equivalence
3742 substitutions when detecting a possible reload cycle.
3744 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3747 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3748 by default if flag_selective_schedling{,2}. Formatting fixes.
3750 PR rtl-optimization/83682
3751 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
3752 if it has non-VECTOR_MODE element mode.
3753 (vec_duplicate_p): Likewise.
3756 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
3757 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
3759 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3762 * config/i386/i386-builtin.def
3763 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
3764 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
3765 Require also OPTION_MASK_ISA_AVX512F in addition to
3766 OPTION_MASK_ISA_GFNI.
3767 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
3768 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
3769 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
3770 to OPTION_MASK_ISA_GFNI.
3771 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
3772 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
3773 OPTION_MASK_ISA_AVX512BW.
3774 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
3775 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
3776 addition to OPTION_MASK_ISA_GFNI.
3777 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
3778 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
3779 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
3780 to OPTION_MASK_ISA_GFNI.
3781 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
3782 a requirement for all ISAs rather than any of them with a few
3784 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
3786 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
3787 bitmasks to be enabled with 3 exceptions, instead of requiring any
3788 enabled ISA with lots of exceptions.
3789 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
3790 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
3791 Change avx512bw in isa attribute to avx512f.
3792 * config/i386/sgxintrin.h: Add license boilerplate.
3793 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
3794 to __AVX512F__ and __AVX512VL to __AVX512VL__.
3795 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
3796 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
3798 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
3799 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
3800 temporarily sse2 rather than sse if not enabled already.
3803 * config/i386/sse.md (VI248_VLBW): Rename to ...
3804 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
3805 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
3806 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
3807 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
3808 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
3809 mode iterator instead of VI248_VLBW.
3811 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
3813 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
3814 (record_modified): Skip clobbers; add debug output.
3815 (param_change_prob): Use sreal frequencies.
3817 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3819 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
3820 punt for user-aligned variables.
3822 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3824 * tree-chrec.c (chrec_contains_symbols): Return true for
3827 2018-01-05 Sudakshina Das <sudi.das@arm.com>
3830 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
3831 of (x|y) == x for BICS pattern.
3833 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3835 PR tree-optimization/83605
3836 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
3837 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
3840 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
3842 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
3843 * config/epiphany/rtems.h: New file.
3845 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3846 Uros Bizjak <ubizjak@gmail.com>
3849 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
3850 QIreg_operand instead of register_operand predicate.
3851 * config/i386/i386.c (ix86_rop_should_change_byte_p,
3852 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
3853 comments instead of -fmitigate[-_]rop.
3855 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3858 * cgraphunit.c (symbol_table::compile): Switch to text_section
3859 before calling assembly_start debug hook.
3860 * run-rtl-passes.c (run_rtl_passes): Likewise.
3863 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3865 * tree-vrp.c (extract_range_from_binary_expr_1): Check
3866 range_int_cst_p rather than !symbolic_range_p before calling
3867 extract_range_from_multiplicative_op_1.
3869 2017-01-04 Jeff Law <law@redhat.com>
3871 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
3872 redundant test in assertion.
3874 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3876 * doc/rtl.texi: Document machine_mode wrapper classes.
3878 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3880 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
3883 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3885 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
3886 the VEC_PERM_EXPR fold to fail.
3888 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3891 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
3892 to switched_sections.
3894 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3897 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
3900 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
3903 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
3904 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
3906 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3909 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
3910 is BLKmode and bitpos not zero or mode change is needed.
3912 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3915 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
3918 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
3921 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
3922 instead of MULT rtx. Update all corresponding splitters.
3924 (*ssub<modesuffix>): Ditto.
3926 (*cmp_sadd_di): Update split patterns.
3927 (*cmp_sadd_si): Ditto.
3928 (*cmp_sadd_sidi): Ditto.
3929 (*cmp_ssub_di): Ditto.
3930 (*cmp_ssub_si): Ditto.
3931 (*cmp_ssub_sidi): Ditto.
3932 * config/alpha/predicates.md (const23_operand): New predicate.
3933 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
3934 Look for ASHIFT, not MULT inner operand.
3935 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
3937 2018-01-04 Martin Liska <mliska@suse.cz>
3939 PR gcov-profile/83669
3940 * gcov.c (output_intermediate_file): Add version to intermediate
3942 * doc/gcov.texi: Document new field 'version' in intermediate
3943 file format. Fix location of '-k' option of gcov command.
3945 2018-01-04 Martin Liska <mliska@suse.cz>
3948 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
3950 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3952 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
3954 2018-01-03 Martin Sebor <msebor@redhat.com>
3956 PR tree-optimization/83655
3957 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
3958 checking calls with invalid arguments.
3960 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3962 * tree-vect-stmts.c (vect_get_store_rhs): New function.
3963 (vectorizable_mask_load_store): Delete.
3964 (vectorizable_call): Return false for masked loads and stores.
3965 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
3966 instead of gimple_assign_rhs1.
3967 (vectorizable_load): Handle IFN_MASK_LOAD.
3968 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
3970 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3972 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
3974 (vectorizable_mask_load_store): ...here.
3975 (vectorizable_load): ...and here.
3977 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3979 * tree-vect-stmts.c (vect_build_all_ones_mask)
3980 (vect_build_zero_merge_argument): New functions, split out from...
3981 (vectorizable_load): ...here.
3983 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3985 * tree-vect-stmts.c (vect_check_store_rhs): New function,
3987 (vectorizable_mask_load_store): ...here.
3988 (vectorizable_store): ...and here.
3990 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3992 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
3994 (vectorizable_mask_load_store): ...here.
3996 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3998 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
3999 (vect_model_store_cost): Take a vec_load_store_type instead of a
4001 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4002 (vect_model_store_cost): Take a vec_load_store_type instead of a
4004 (vectorizable_mask_load_store): Update accordingly.
4005 (vectorizable_store): Likewise.
4006 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4008 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4010 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4011 IFN_MASK_LOAD calls here rather than...
4012 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4014 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4015 Alan Hayward <alan.hayward@arm.com>
4016 David Sherwood <david.sherwood@arm.com>
4018 * expmed.c (extract_bit_field_1): For vector extracts,
4019 fall back to extract_bit_field_as_subreg if vec_extract
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 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4027 they are variable or constant sized.
4028 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4029 slots for constant-sized data.
4031 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4032 Alan Hayward <alan.hayward@arm.com>
4033 David Sherwood <david.sherwood@arm.com>
4035 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4036 handling COND_EXPRs with boolean comparisons, try to find a better
4037 basis for the mask type than the boolean itself.
4039 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4041 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4042 is calculated and how it can be overridden.
4043 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4044 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4046 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4049 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4050 Alan Hayward <alan.hayward@arm.com>
4051 David Sherwood <david.sherwood@arm.com>
4053 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4054 Remove the mode argument.
4055 (aarch64_simd_valid_immediate): Remove the mode and inverse
4057 * config/aarch64/iterators.md (bitsize): New iterator.
4058 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4059 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4060 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4061 aarch64_simd_valid_immediate.
4062 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4063 (aarch64_reg_or_bic_imm): Likewise.
4064 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4065 with an insn_type enum and msl with a modifier_type enum.
4066 Replace element_width with a scalar_mode. Change the shift
4067 to unsigned int. Add constructors for scalar_float_mode and
4068 scalar_int_mode elements.
4069 (aarch64_vect_float_const_representable_p): Delete.
4070 (aarch64_can_const_movi_rtx_p)
4071 (aarch64_simd_scalar_immediate_valid_for_move)
4072 (aarch64_simd_make_constant): Update call to
4073 aarch64_simd_valid_immediate.
4074 (aarch64_advsimd_valid_immediate_hs): New function.
4075 (aarch64_advsimd_valid_immediate): Likewise.
4076 (aarch64_simd_valid_immediate): Remove mode and inverse
4077 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4078 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4079 and aarch64_float_const_representable_p on the result.
4080 (aarch64_output_simd_mov_immediate): Remove mode argument.
4081 Update call to aarch64_simd_valid_immediate and use of
4082 simd_immediate_info.
4083 (aarch64_output_scalar_simd_mov_immediate): Update call
4086 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4087 Alan Hayward <alan.hayward@arm.com>
4088 David Sherwood <david.sherwood@arm.com>
4090 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4091 (mode_nunits): Likewise CONST_MODE_NUNITS.
4092 * machmode.def (ADJUST_NUNITS): Document.
4093 * genmodes.c (mode_data::need_nunits_adj): New field.
4094 (blank_mode): Update accordingly.
4095 (adj_nunits): New variable.
4096 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4098 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4099 listed in adj_nunits.
4100 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4101 listed in adj_nunits. Don't emit case statements for such modes.
4102 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4103 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4104 nothing if adj_nunits is nonnull.
4105 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4106 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4107 (emit_mode_fbit): Update use of print_maybe_const_decl.
4108 (emit_move_size): Likewise. Treat the array as non-const
4110 (emit_mode_adjustments): Handle adj_nunits.
4112 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4114 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4115 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4116 (VECTOR_MODES): Use it.
4117 (make_vector_modes): Take the prefix as an argument.
4119 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4120 Alan Hayward <alan.hayward@arm.com>
4121 David Sherwood <david.sherwood@arm.com>
4123 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4124 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4125 for MODE_VECTOR_BOOL.
4126 * machmode.def (VECTOR_BOOL_MODE): Document.
4127 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4128 (make_vector_bool_mode): New function.
4129 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4131 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4132 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4134 * stor-layout.c (int_mode_for_mode): Likewise.
4135 * tree.c (build_vector_type_for_mode): Likewise.
4136 * varasm.c (output_constant_pool_2): Likewise.
4137 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4138 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4139 for MODE_VECTOR_BOOL.
4140 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4141 of mode class checks.
4142 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4143 instead of a list of mode class checks.
4144 (expand_vector_scalar_condition): Likewise.
4145 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4147 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4148 Alan Hayward <alan.hayward@arm.com>
4149 David Sherwood <david.sherwood@arm.com>
4151 * machmode.h (mode_size): Change from unsigned short to
4153 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4154 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4155 or if measurement_type is not polynomial.
4156 (fixed_size_mode::includes_p): Check for constant-sized modes.
4157 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4158 return a poly_uint16 rather than an unsigned short.
4159 (emit_mode_size): Change the type of mode_size from unsigned short
4160 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4161 (emit_mode_adjustments): Cope with polynomial vector sizes.
4162 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4164 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4166 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4167 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4168 * caller-save.c (setup_save_areas): Likewise.
4169 (replace_reg_with_saved_mem): Likewise.
4170 * calls.c (emit_library_call_value_1): Likewise.
4171 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4172 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4173 (gen_lowpart_for_combine): Likewise.
4174 * convert.c (convert_to_integer_1): Likewise.
4175 * cse.c (equiv_constant, cse_insn): Likewise.
4176 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4177 (cselib_subst_to_values): Likewise.
4178 * dce.c (word_dce_process_block): Likewise.
4179 * df-problems.c (df_word_lr_mark_ref): Likewise.
4180 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4181 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4182 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4183 (rtl_for_decl_location): Likewise.
4184 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4185 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4186 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4187 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4188 (expand_expr_real_1): Likewise.
4189 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4190 (pad_below): Likewise.
4191 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4192 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4193 * ira.c (get_subreg_tracking_sizes): Likewise.
4194 * ira-build.c (ira_create_allocno_objects): Likewise.
4195 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4196 (ira_sort_regnos_for_alter_reg): Likewise.
4197 * ira-costs.c (record_operand_costs): Likewise.
4198 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4199 (resolve_simple_move): Likewise.
4200 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4201 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4202 (lra_constraints): Likewise.
4203 (CONST_POOL_OK_P): Reject variable-sized modes.
4204 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4205 (add_pseudo_to_slot, lra_spill): Likewise.
4206 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4207 * optabs-query.c (get_best_extraction_insn): Likewise.
4208 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4209 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4210 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4211 * recog.c (offsettable_address_addr_space_p): Likewise.
4212 * regcprop.c (maybe_mode_change): Likewise.
4213 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4214 * regrename.c (build_def_use): Likewise.
4215 * regstat.c (dump_reg_info): Likewise.
4216 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4217 (find_reloads, find_reloads_subreg_address): Likewise.
4218 * reload1.c (eliminate_regs_1): Likewise.
4219 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4220 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4221 (simplify_binary_operation_1, simplify_subreg): Likewise.
4222 * targhooks.c (default_function_arg_padding): Likewise.
4223 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4224 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4225 (verify_gimple_assign_ternary): Likewise.
4226 * tree-inline.c (estimate_move_cost): Likewise.
4227 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4228 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4229 (get_address_cost_ainc): Likewise.
4230 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4231 (vect_supportable_dr_alignment): Likewise.
4232 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4233 (vectorizable_reduction): Likewise.
4234 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4235 (vectorizable_operation, vectorizable_load): Likewise.
4236 * tree.c (build_same_sized_truth_vector_type): Likewise.
4237 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4238 * var-tracking.c (emit_note_insn_var_location): Likewise.
4239 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4240 (ADDR_VEC_ALIGN): Likewise.
4242 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4243 Alan Hayward <alan.hayward@arm.com>
4244 David Sherwood <david.sherwood@arm.com>
4246 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4248 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4249 or if measurement_type is polynomial.
4250 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4251 * combine.c (make_extraction): Likewise.
4252 * dse.c (find_shift_sequence): Likewise.
4253 * dwarf2out.c (mem_loc_descriptor): Likewise.
4254 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4255 (extract_bit_field, extract_low_bits): Likewise.
4256 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4257 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4258 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4259 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4260 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4261 * reload.c (find_reloads): Likewise.
4262 * reload1.c (alter_reg): Likewise.
4263 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4264 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4265 * tree-if-conv.c (predicate_mem_writes): Likewise.
4266 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4267 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4268 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4269 * valtrack.c (dead_debug_insert_temp): Likewise.
4270 * varasm.c (mergeable_constant_section): Likewise.
4271 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4273 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4274 Alan Hayward <alan.hayward@arm.com>
4275 David Sherwood <david.sherwood@arm.com>
4277 * expr.c (expand_assignment): Cope with polynomial mode sizes
4278 when assigning to a CONCAT.
4280 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4281 Alan Hayward <alan.hayward@arm.com>
4282 David Sherwood <david.sherwood@arm.com>
4284 * machmode.h (mode_precision): Change from unsigned short to
4286 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4288 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4289 or if measurement_type is not polynomial.
4290 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4291 in which the mode is already known to be a scalar_int_mode.
4292 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4293 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4295 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4296 for GET_MODE_PRECISION.
4297 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4298 for GET_MODE_PRECISION.
4299 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4301 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4302 (expand_field_assignment, make_extraction): Likewise.
4303 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4304 (get_last_value): Likewise.
4305 * convert.c (convert_to_integer_1): Likewise.
4306 * cse.c (cse_insn): Likewise.
4307 * expr.c (expand_expr_real_1): Likewise.
4308 * lra-constraints.c (simplify_operand_subreg): Likewise.
4309 * optabs-query.c (can_atomic_load_p): Likewise.
4310 * optabs.c (expand_atomic_load): Likewise.
4311 (expand_atomic_store): Likewise.
4312 * ree.c (combine_reaching_defs): Likewise.
4313 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4314 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4315 * tree.h (type_has_mode_precision_p): Likewise.
4316 * ubsan.c (instrument_si_overflow): Likewise.
4318 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4319 Alan Hayward <alan.hayward@arm.com>
4320 David Sherwood <david.sherwood@arm.com>
4322 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4323 polynomial numbers of units.
4324 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4325 (valid_vector_subparts_p): New function.
4326 (build_vector_type): Remove temporary shim and take the number
4327 of units as a poly_uint64 rather than an int.
4328 (build_opaque_vector_type): Take the number of units as a
4329 poly_uint64 rather than an int.
4330 * tree.c (build_vector_from_ctor): Handle polynomial
4331 TYPE_VECTOR_SUBPARTS.
4332 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4333 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4334 (build_vector_from_val): If the number of units is variable,
4335 use build_vec_duplicate_cst for constant operands and
4336 VEC_DUPLICATE_EXPR otherwise.
4337 (make_vector_type): Remove temporary is_constant ().
4338 (build_vector_type, build_opaque_vector_type): Take the number of
4339 units as a poly_uint64 rather than an int.
4340 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4342 * cfgexpand.c (expand_debug_expr): Likewise.
4343 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4344 (store_constructor, expand_expr_real_1): Likewise.
4345 (const_scalar_mask_from_tree): Likewise.
4346 * fold-const-call.c (fold_const_reduction): Likewise.
4347 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4348 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4349 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4350 (fold_relational_const): Likewise.
4351 (native_interpret_vector): Likewise. Change the size from an
4352 int to an unsigned int.
4353 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4354 TYPE_VECTOR_SUBPARTS.
4355 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4356 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4357 duplicating a non-constant operand into a variable-length vector.
4358 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4359 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4360 * ipa-icf.c (sem_variable::equals): Likewise.
4361 * match.pd: Likewise.
4362 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4363 * print-tree.c (print_node): Likewise.
4364 * stor-layout.c (layout_type): Likewise.
4365 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4366 * tree-cfg.c (verify_gimple_comparison): Likewise.
4367 (verify_gimple_assign_binary): Likewise.
4368 (verify_gimple_assign_ternary): Likewise.
4369 (verify_gimple_assign_single): Likewise.
4370 * tree-pretty-print.c (dump_generic_node): Likewise.
4371 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4372 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4373 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4374 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4375 (vect_shift_permute_load_chain): Likewise.
4376 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4377 (expand_vector_condition, optimize_vector_constructor): Likewise.
4378 (lower_vec_perm, get_compute_type): Likewise.
4379 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4380 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4381 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4382 (vect_recog_mask_conversion_pattern): Likewise.
4383 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4384 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4385 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4386 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4387 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4388 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4389 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4390 (supportable_widening_operation): Likewise.
4391 (supportable_narrowing_operation): Likewise.
4392 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4394 * varasm.c (output_constant): Likewise.
4396 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4397 Alan Hayward <alan.hayward@arm.com>
4398 David Sherwood <david.sherwood@arm.com>
4400 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4401 so that both the length == 3 and length != 3 cases set up their
4402 own permute vectors. Add comments explaining why we know the
4403 number of elements is constant.
4404 (vect_permute_load_chain): Likewise.
4406 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4407 Alan Hayward <alan.hayward@arm.com>
4408 David Sherwood <david.sherwood@arm.com>
4410 * machmode.h (mode_nunits): Change from unsigned char to
4412 (ONLY_FIXED_SIZE_MODES): New macro.
4413 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4414 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4415 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4417 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4418 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4419 or if measurement_type is not polynomial.
4420 * genmodes.c (ZERO_COEFFS): New macro.
4421 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4423 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4424 Use ZERO_COEFFS when emitting initializers.
4425 * data-streamer.h (bp_pack_poly_value): New function.
4426 (bp_unpack_poly_value): Likewise.
4427 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4428 for GET_MODE_NUNITS.
4429 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4430 for GET_MODE_NUNITS.
4431 * tree.c (make_vector_type): Remove temporary shim and make
4432 the real function take the number of units as a poly_uint64
4434 (build_vector_type_for_mode): Handle polynomial nunits.
4435 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4436 * emit-rtl.c (const_vec_series_p_1): Likewise.
4437 (gen_rtx_CONST_VECTOR): Likewise.
4438 * fold-const.c (test_vec_duplicate_folding): Likewise.
4439 * genrecog.c (validate_pattern): Likewise.
4440 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4441 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4442 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4443 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4444 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4445 * rtlanal.c (subreg_get_info): Likewise.
4446 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4447 (vect_grouped_load_supported): Likewise.
4448 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4449 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4450 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4451 (simplify_const_unary_operation, simplify_binary_operation_1)
4452 (simplify_const_binary_operation, simplify_ternary_operation)
4453 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4454 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4455 instead of CONST_VECTOR_NUNITS.
4456 * varasm.c (output_constant_pool_2): Likewise.
4457 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4458 explicit-encoded elements in the XVEC for variable-length vectors.
4460 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4462 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4464 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4465 Alan Hayward <alan.hayward@arm.com>
4466 David Sherwood <david.sherwood@arm.com>
4468 * coretypes.h (fixed_size_mode): Declare.
4469 (fixed_size_mode_pod): New typedef.
4470 * builtins.h (target_builtins::x_apply_args_mode)
4471 (target_builtins::x_apply_result_mode): Change type to
4472 fixed_size_mode_pod.
4473 * builtins.c (apply_args_size, apply_result_size, result_vector)
4474 (expand_builtin_apply_args_1, expand_builtin_apply)
4475 (expand_builtin_return): Update accordingly.
4477 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4479 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4480 * cselib.c (cselib_hash_rtx): Likewise.
4481 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4482 CONST_VECTOR encoding.
4484 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4485 Jeff Law <law@redhat.com>
4488 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4489 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4490 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4491 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4494 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4495 explicitly probe *sp in a noreturn function if there were any callee
4496 register saves or frame pointer is needed.
4498 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4501 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4502 BLKmode for ternary, binary or unary expressions.
4505 * var-tracking.c (delete_vta_debug_insn): New inline function.
4506 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4507 insns from get_insns () to NULL instead of each bb separately.
4508 Use delete_vta_debug_insn. No longer static.
4509 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4510 delete_vta_debug_insns callers.
4511 * rtl.h (delete_vta_debug_insns): Declare.
4512 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4513 instead of variable_tracking_main.
4515 2018-01-03 Martin Sebor <msebor@redhat.com>
4517 PR tree-optimization/83603
4518 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4519 arguments past the endof the argument list in functions declared
4520 without a prototype.
4521 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4522 Avoid checking when arguments are null.
4524 2018-01-03 Martin Sebor <msebor@redhat.com>
4527 * doc/extend.texi (attribute const): Fix a typo.
4528 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4529 issuing -Wsuggest-attribute for void functions.
4531 2018-01-03 Martin Sebor <msebor@redhat.com>
4533 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4534 offset_int::from instead of wide_int::to_shwi.
4535 (maybe_diag_overlap): Remove assertion.
4536 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4537 * gimple-ssa-sprintf.c (format_directive): Same.
4538 (parse_directive): Same.
4539 (sprintf_dom_walker::compute_format_length): Same.
4540 (try_substitute_return_value): Same.
4542 2017-01-03 Jeff Law <law@redhat.com>
4545 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4546 non-constant residual for zero at runtime and avoid probing in
4547 that case. Reorganize code for trailing problem to mirror handling
4550 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4552 PR tree-optimization/83501
4553 * tree-ssa-strlen.c (get_string_cst): New.
4554 (handle_char_store): Call get_string_cst.
4556 2018-01-03 Martin Liska <mliska@suse.cz>
4558 PR tree-optimization/83593
4559 * tree-ssa-strlen.c: Include tree-cfg.h.
4560 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4561 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4562 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4564 (strlen_dom_walker::before_dom_children): Call
4565 gimple_purge_dead_eh_edges. Dump tranformation with details
4567 (strlen_dom_walker::before_dom_children): Update call by adding
4568 new argument cleanup_eh.
4569 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4571 2018-01-03 Martin Liska <mliska@suse.cz>
4574 * cif-code.def (VARIADIC_THUNK): New enum value.
4575 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4578 2018-01-03 Jan Beulich <jbeulich@suse.com>
4580 * sse.md (mov<mode>_internal): Tighten condition for when to use
4581 vmovdqu<ssescalarsize> for TI and OI modes.
4583 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4585 Update copyright years.
4587 2018-01-03 Martin Liska <mliska@suse.cz>
4590 * ipa-visibility.c (function_and_variable_visibility): Skip
4591 functions with noipa attribure.
4593 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4595 * gcc.c (process_command): Update copyright notice dates.
4596 * gcov-dump.c (print_version): Ditto.
4597 * gcov.c (print_version): Ditto.
4598 * gcov-tool.c (print_version): Ditto.
4599 * gengtype.c (create_file): Ditto.
4600 * doc/cpp.texi: Bump @copying's copyright year.
4601 * doc/cppinternals.texi: Ditto.
4602 * doc/gcc.texi: Ditto.
4603 * doc/gccint.texi: Ditto.
4604 * doc/gcov.texi: Ditto.
4605 * doc/install.texi: Ditto.
4606 * doc/invoke.texi: Ditto.
4608 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4610 * vector-builder.h (vector_builder::m_full_nelts): Change from
4611 unsigned int to poly_uint64.
4612 (vector_builder::full_nelts): Update prototype accordingly.
4613 (vector_builder::new_vector): Likewise.
4614 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4615 (vector_builder::operator ==): Likewise.
4616 (vector_builder::finalize): Likewise.
4617 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4618 Take the number of elements as a poly_uint64 rather than an
4620 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4621 from unsigned int to poly_uint64.
4622 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4623 (vec_perm_indices::new_vector): Likewise.
4624 (vec_perm_indices::length): Likewise.
4625 (vec_perm_indices::nelts_per_input): Likewise.
4626 (vec_perm_indices::input_nelts): Likewise.
4627 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4628 number of elements per input as a poly_uint64 rather than an
4629 unsigned int. Use the original encoding for variable-length
4630 vectors, rather than clamping each individual element.
4631 For the second and subsequent elements in each pattern,
4632 clamp the step and base before clamping their sum.
4633 (vec_perm_indices::series_p): Handle polynomial element counts.
4634 (vec_perm_indices::all_in_range_p): Likewise.
4635 (vec_perm_indices_to_tree): Likewise.
4636 (vec_perm_indices_to_rtx): Likewise.
4637 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4638 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4639 (tree_vector_builder::new_binary_operation): Handle polynomial
4640 element counts. Return false if we need to know the number
4641 of elements at compile time.
4642 * fold-const.c (fold_vec_perm): Punt if the number of elements
4643 isn't known at compile time.
4645 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4647 * vec-perm-indices.h (vec_perm_builder): Change element type
4648 from HOST_WIDE_INT to poly_int64.
4649 (vec_perm_indices::element_type): Update accordingly.
4650 (vec_perm_indices::clamp): Handle polynomial element_types.
4651 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4652 (vec_perm_indices::all_in_range_p): Likewise.
4653 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4655 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4656 polynomial vec_perm_indices element types.
4657 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4658 * fold-const.c (fold_vec_perm): Likewise.
4659 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4660 * tree-vect-generic.c (lower_vec_perm): Likewise.
4661 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4662 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4663 element type to HOST_WIDE_INT.
4665 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4666 Alan Hayward <alan.hayward@arm.com>
4667 David Sherwood <david.sherwood@arm.com>
4669 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4670 rather than an int. Use plus_constant.
4671 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4672 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4674 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4675 Alan Hayward <alan.hayward@arm.com>
4676 David Sherwood <david.sherwood@arm.com>
4678 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4679 a HOST_WIDE_INT to a poly_int64.
4681 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4682 Alan Hayward <alan.hayward@arm.com>
4683 David Sherwood <david.sherwood@arm.com>
4685 * calls.c (load_register_parameters): Cope with polynomial
4686 mode sizes. Require a constant size for BLKmode parameters
4687 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4688 forces a parameter to be padded at the lsb end in order to
4689 fill a complete number of words, require the parameter size
4690 to be ordered wrt UNITS_PER_WORD.
4692 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4693 Alan Hayward <alan.hayward@arm.com>
4694 David Sherwood <david.sherwood@arm.com>
4696 * reload1.c (spill_stack_slot_width): Change element type
4697 from unsigned int to poly_uint64_pod.
4698 (alter_reg): Treat mode sizes as polynomial.
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 * reload.c (complex_word_subreg_p): New function.
4705 (reload_inner_reg_of_subreg, push_reload): Use it.
4707 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4708 Alan Hayward <alan.hayward@arm.com>
4709 David Sherwood <david.sherwood@arm.com>
4711 * lra-constraints.c (process_alt_operands): Reject matched
4712 operands whose sizes aren't ordered.
4713 (match_reload): Refer to this check here.
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 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4720 that the mode size is in the set {1, 2, 4, 8, 16}.
4722 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4723 Alan Hayward <alan.hayward@arm.com>
4724 David Sherwood <david.sherwood@arm.com>
4726 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4727 Use plus_constant instead of gen_rtx_PLUS.
4729 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4730 Alan Hayward <alan.hayward@arm.com>
4731 David Sherwood <david.sherwood@arm.com>
4733 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4734 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4735 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4736 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4737 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4738 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4739 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4740 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4741 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4742 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4744 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4745 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4746 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4747 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4748 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4749 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4750 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4751 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
4752 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
4753 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
4755 * expr.c (emit_move_resolve_push): Treat the input and result
4756 of PUSH_ROUNDING as a poly_int64.
4757 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
4758 (emit_push_insn): Likewise.
4759 * lra-eliminations.c (mark_not_eliminable): Likewise.
4760 * recog.c (push_operand): Likewise.
4761 * reload1.c (elimination_effects): Likewise.
4762 * rtlanal.c (nonzero_bits1): Likewise.
4763 * calls.c (store_one_arg): Likewise. Require the padding to be
4764 known at compile time.
4766 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4767 Alan Hayward <alan.hayward@arm.com>
4768 David Sherwood <david.sherwood@arm.com>
4770 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
4771 Use plus_constant instead of gen_rtx_PLUS.
4773 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4774 Alan Hayward <alan.hayward@arm.com>
4775 David Sherwood <david.sherwood@arm.com>
4777 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
4780 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4781 Alan Hayward <alan.hayward@arm.com>
4782 David Sherwood <david.sherwood@arm.com>
4784 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
4785 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
4786 via stack temporaries. Treat the mode size as polynomial too.
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_2): When handling conversions involving
4793 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
4794 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
4795 as a poly_uint64 too.
4797 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4798 Alan Hayward <alan.hayward@arm.com>
4799 David Sherwood <david.sherwood@arm.com>
4801 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
4803 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4804 Alan Hayward <alan.hayward@arm.com>
4805 David Sherwood <david.sherwood@arm.com>
4807 * combine.c (can_change_dest_mode): Handle polynomial
4808 REGMODE_NATURAL_SIZE.
4809 * expmed.c (store_bit_field_1): Likewise.
4810 * expr.c (store_constructor): Likewise.
4811 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
4812 and polynomial REGMODE_NATURAL_SIZE.
4813 (gen_lowpart_common): Likewise.
4814 * reginfo.c (record_subregs_of_mode): Likewise.
4815 * rtlanal.c (read_modify_subreg_p): Likewise.
4817 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4818 Alan Hayward <alan.hayward@arm.com>
4819 David Sherwood <david.sherwood@arm.com>
4821 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
4822 numbers of elements.
4824 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4825 Alan Hayward <alan.hayward@arm.com>
4826 David Sherwood <david.sherwood@arm.com>
4828 * match.pd: Cope with polynomial numbers of vector elements.
4830 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4831 Alan Hayward <alan.hayward@arm.com>
4832 David Sherwood <david.sherwood@arm.com>
4834 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
4835 in a POINTER_PLUS_EXPR.
4837 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4838 Alan Hayward <alan.hayward@arm.com>
4839 David Sherwood <david.sherwood@arm.com>
4841 * omp-simd-clone.c (simd_clone_subparts): New function.
4842 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
4843 (ipa_simd_modify_function_body): Likewise.
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 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
4850 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
4851 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
4852 (expand_vector_condition, vector_element): Likewise.
4853 (subparts_gt): New function.
4854 (get_compute_type): Use subparts_gt.
4855 (count_type_subparts): Delete.
4856 (expand_vector_operations_1): Use subparts_gt instead of
4857 count_type_subparts.
4859 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4860 Alan Hayward <alan.hayward@arm.com>
4861 David Sherwood <david.sherwood@arm.com>
4863 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
4864 (vect_compile_time_alias): ...this new function. Do the calculation
4865 on poly_ints rather than trees.
4866 (vect_prune_runtime_alias_test_list): Update call accordingly.
4868 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4869 Alan Hayward <alan.hayward@arm.com>
4870 David Sherwood <david.sherwood@arm.com>
4872 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
4874 (vect_schedule_slp_instance): Likewise.
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_get_and_check_slp_defs): Reject
4881 constant and extern definitions for variable-length vectors.
4882 (vect_get_constant_vectors): Note that the number of units
4883 is known to be constant.
4885 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4886 Alan Hayward <alan.hayward@arm.com>
4887 David Sherwood <david.sherwood@arm.com>
4889 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
4890 of units as polynomial. Choose between WIDE and NARROW based
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 (simd_clone_subparts): New function.
4898 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
4900 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4901 Alan Hayward <alan.hayward@arm.com>
4902 David Sherwood <david.sherwood@arm.com>
4904 * tree-vect-stmts.c (vectorizable_call): Treat the number of
4905 vectors as polynomial. Use build_index_vector for
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 (get_load_store_type): Treat the number of
4913 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
4914 for variable-length vectors.
4915 (vectorizable_mask_load_store): Treat the number of units as
4916 polynomial, asserting that it is constant if the condition has
4917 already been enforced.
4918 (vectorizable_store, vectorizable_load): Likewise.
4920 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4921 Alan Hayward <alan.hayward@arm.com>
4922 David Sherwood <david.sherwood@arm.com>
4924 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
4925 of units as polynomial. Punt if we can't tell at compile time
4926 which vector contains the final result.
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_induction): Treat the number
4933 of units as polynomial. Punt on SLP inductions. Use an integer
4934 VEC_SERIES_EXPR for variable-length integer reductions. Use a
4935 cast of such a series for variable-length floating-point
4938 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4939 Alan Hayward <alan.hayward@arm.com>
4940 David Sherwood <david.sherwood@arm.com>
4942 * tree.h (build_index_vector): Declare.
4943 * tree.c (build_index_vector): New function.
4944 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
4945 of units as polynomial, forcibly converting it to a constant if
4946 vectorizable_reduction has already enforced the condition.
4947 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
4948 to create a {1,2,3,...} vector.
4949 (vectorizable_reduction): Treat the number of units as polynomial.
4950 Choose vectype_in based on the largest scalar element size rather
4951 than the smallest number of units. Enforce the restrictions
4954 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4955 Alan Hayward <alan.hayward@arm.com>
4956 David Sherwood <david.sherwood@arm.com>
4958 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
4959 number of units as polynomial.
4961 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4962 Alan Hayward <alan.hayward@arm.com>
4963 David Sherwood <david.sherwood@arm.com>
4965 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
4966 * target.def (autovectorize_vector_sizes): Return the vector sizes
4967 by pointer, using vector_sizes rather than a bitmask.
4968 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
4969 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
4970 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
4972 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
4973 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
4974 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
4975 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
4976 * omp-general.c (omp_max_vf): Likewise.
4977 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4978 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
4979 * tree-vect-loop.c (vect_analyze_loop): Likewise.
4980 * tree-vect-slp.c (vect_slp_bb): Likewise.
4981 * doc/tm.texi: Regenerate.
4982 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
4984 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
4985 the vector size as a poly_uint64 rather than an unsigned int.
4986 (current_vector_size): Change from an unsigned int to a poly_uint64.
4987 (get_vectype_for_scalar_type): Update accordingly.
4988 * tree.h (build_truth_vector_type): Take the size and number of
4989 units as a poly_uint64 rather than an unsigned int.
4990 (build_vector_type): Add a temporary overload that takes
4991 the number of units as a poly_uint64 rather than an unsigned int.
4992 * tree.c (make_vector_type): Likewise.
4993 (build_truth_vector_type): Take the number of units as a poly_uint64
4994 rather than an unsigned int.
4996 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4997 Alan Hayward <alan.hayward@arm.com>
4998 David Sherwood <david.sherwood@arm.com>
5000 * target.def (get_mask_mode): Take the number of units and length
5001 as poly_uint64s rather than unsigned ints.
5002 * targhooks.h (default_get_mask_mode): Update accordingly.
5003 * targhooks.c (default_get_mask_mode): Likewise.
5004 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5005 * doc/tm.texi: Regenerate.
5007 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5008 Alan Hayward <alan.hayward@arm.com>
5009 David Sherwood <david.sherwood@arm.com>
5011 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5012 * omp-general.c (omp_max_vf): Likewise.
5013 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5014 (expand_omp_simd): Handle polynomial safelen.
5015 * omp-low.c (omplow_simd_context): Add a default constructor.
5016 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5017 (lower_rec_simd_input_clauses): Update accordingly.
5018 (lower_rec_input_clauses): Likewise.
5020 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5021 Alan Hayward <alan.hayward@arm.com>
5022 David Sherwood <david.sherwood@arm.com>
5024 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5025 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5026 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5027 (vect_analyze_slp_cost): Likewise.
5028 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5029 (vect_model_load_cost): Likewise.
5031 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5032 Alan Hayward <alan.hayward@arm.com>
5033 David Sherwood <david.sherwood@arm.com>
5035 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5036 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5037 from an unsigned int * to a poly_uint64_pod *.
5038 (calculate_unrolling_factor): New function.
5039 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5041 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5042 Alan Hayward <alan.hayward@arm.com>
5043 David Sherwood <david.sherwood@arm.com>
5045 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5046 from an unsigned int to a poly_uint64.
5047 (_loop_vec_info::slp_unrolling_factor): Likewise.
5048 (_loop_vec_info::vectorization_factor): Change from an int
5050 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5051 (vect_get_num_vectors): New function.
5052 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5053 (vect_get_num_copies): Use vect_get_num_vectors.
5054 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5055 to an unsigned int *.
5056 (vect_analyze_data_refs): Change min_vf from an int * to a
5058 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5059 than an unsigned HOST_WIDE_INT.
5060 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5061 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5062 to an unsigned int *.
5063 (vect_analyze_data_ref_dependences): Likewise.
5064 (vect_compute_data_ref_alignment): Handle polynomial vf.
5065 (vect_enhance_data_refs_alignment): Likewise.
5066 (vect_prune_runtime_alias_test_list): Likewise.
5067 (vect_shift_permute_load_chain): Likewise.
5068 (vect_supportable_dr_alignment): Likewise.
5069 (dependence_distance_ge_vf): Take the vectorization factor as a
5070 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5071 (vect_analyze_data_refs): Change min_vf from an int * to a
5073 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5074 vfm1 as a poly_uint64 rather than an int. Make the same change
5075 for the returned bound_scalar.
5076 (vect_gen_vector_loop_niters): Handle polynomial vf.
5077 (vect_do_peeling): Likewise. Update call to
5078 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5079 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5081 * tree-vect-loop.c (vect_determine_vectorization_factor)
5082 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5083 (vect_get_known_peeling_cost): Likewise.
5084 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5085 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5086 (vect_transform_loop): Likewise. Use the lowest possible VF when
5087 updating the upper bounds of the loop.
5088 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5090 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5091 polynomial unroll factors.
5092 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5093 (vect_make_slp_decision): Likewise.
5094 (vect_supported_load_permutation_p): Likewise, and polynomial
5096 (vect_analyze_slp_cost): Handle polynomial vf.
5097 (vect_slp_analyze_node_operations): Likewise.
5098 (vect_slp_analyze_bb_1): Likewise.
5099 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5100 than an unsigned HOST_WIDE_INT.
5101 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5102 (vectorizable_load): Handle polynomial vf.
5103 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5105 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5107 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5108 Alan Hayward <alan.hayward@arm.com>
5109 David Sherwood <david.sherwood@arm.com>
5111 * match.pd: Handle bit operations involving three constants
5112 and try to fold one pair.
5114 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5116 * tree-vect-loop-manip.c: Include gimple-fold.h.
5117 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5118 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5119 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5120 Add a path that uses a step of VF instead of 1, but disable it
5122 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5123 and niters_no_overflow parameters. Update calls to
5124 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5125 Create a new SSA name if the latter choses to use a ste other
5126 than zero, and return it via niters_vector_mult_vf_var.
5127 * tree-vect-loop.c (vect_transform_loop): Update calls to
5128 vect_do_peeling, vect_gen_vector_loop_niters and
5129 slpeel_make_loop_iterate_ntimes.
5130 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5131 (vect_gen_vector_loop_niters): Update declarations after above changes.
5133 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5135 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5136 128-bit round to integer instructions.
5137 (ceil<mode>2): Likewise.
5138 (btrunc<mode>2): Likewise.
5139 (round<mode>2): Likewise.
5141 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5143 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5144 unaligned VSX load/store on P8/P9.
5145 (expand_block_clear): Allow the use of unaligned VSX
5146 load/store on P8/P9.
5148 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5150 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5152 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5153 swap associated with both a load and a store.
5155 2018-01-02 Andrew Waterman <andrew@sifive.com>
5157 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5158 * config/riscv/riscv.md (clear_cache): Use it.
5160 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5162 * web.c: Remove out-of-date comment.
5164 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5166 * expr.c (fixup_args_size_notes): Check that any existing
5167 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5168 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5169 (emit_single_push_insn): ...here.
5171 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5173 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5174 (const_vector_encoded_nelts): New function.
5175 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5176 (const_vector_int_elt, const_vector_elt): Declare.
5177 * emit-rtl.c (const_vector_int_elt_1): New function.
5178 (const_vector_elt): Likewise.
5179 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5180 of CONST_VECTOR_ELT.
5182 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5184 * expr.c: Include rtx-vector-builder.h.
5185 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5186 directly on the tree encoding.
5187 (const_vector_from_tree): Likewise.
5188 * optabs.c: Include rtx-vector-builder.h.
5189 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5190 sequence of "u" values.
5191 * vec-perm-indices.c: Include rtx-vector-builder.h.
5192 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5193 directly on the vec_perm_indices encoding.
5195 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5197 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5198 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5199 * rtx-vector-builder.h: New file.
5200 * rtx-vector-builder.c: Likewise.
5201 * rtl.h (rtx_def::u2): Add a const_vector field.
5202 (CONST_VECTOR_NPATTERNS): New macro.
5203 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5204 (CONST_VECTOR_DUPLICATE_P): Likewise.
5205 (CONST_VECTOR_STEPPED_P): Likewise.
5206 (CONST_VECTOR_ENCODED_ELT): Likewise.
5207 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5208 (unwrap_const_vec_duplicate): Likewise.
5209 (const_vec_series_p): Check for a non-duplicated vector encoding.
5210 Say that the function only returns true for integer vectors.
5211 * emit-rtl.c: Include rtx-vector-builder.h.
5212 (gen_const_vec_duplicate_1): Delete.
5213 (gen_const_vector): Call gen_const_vec_duplicate instead of
5214 gen_const_vec_duplicate_1.
5215 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5216 (gen_const_vec_duplicate): Use rtx_vector_builder.
5217 (gen_const_vec_series): Likewise.
5218 (gen_rtx_CONST_VECTOR): Likewise.
5219 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5220 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5221 Build a new vector rather than modifying a CONST_VECTOR in-place.
5222 (handle_special_swappables): Update call accordingly.
5223 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5224 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5225 Build a new vector rather than modifying a CONST_VECTOR in-place.
5226 (handle_special_swappables): Update call accordingly.
5228 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5230 * simplify-rtx.c (simplify_const_binary_operation): Use
5231 CONST_VECTOR_ELT instead of XVECEXP.
5233 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5235 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5236 the selector elements to be different from the data elements
5237 if the selector is a VECTOR_CST.
5238 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5239 ssizetype for the selector.
5241 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5243 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5244 before testing each element individually.
5245 * tree-vect-generic.c (lower_vec_perm): Likewise.
5247 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5249 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5250 * selftest-run-tests.c (selftest::run_tests): Call it.
5251 * vector-builder.h (vector_builder::operator ==): New function.
5252 (vector_builder::operator !=): Likewise.
5253 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5254 (vec_perm_indices::all_from_input_p): New function.
5255 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5256 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5257 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5258 instead of reading the VECTOR_CST directly. Detect whether both
5259 vector inputs are the same before constructing the vec_perm_indices,
5260 and update the number of inputs argument accordingly. Use the
5261 utility functions added above. Only construct sel2 if we need to.
5263 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5265 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5266 the broadcast of the low byte.
5267 (expand_mult_highpart): Use an explicit encoding for the permutes.
5268 * optabs-query.c (can_mult_highpart_p): Likewise.
5269 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5270 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5271 (vectorizable_bswap): Likewise.
5272 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5273 explicit encoding for the power-of-2 permutes.
5274 (vect_permute_store_chain): Likewise.
5275 (vect_grouped_load_supported): Likewise.
5276 (vect_permute_load_chain): Likewise.
5278 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5280 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5281 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5282 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5283 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5284 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5285 (vect_gen_perm_mask_any): Likewise.
5287 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5289 * int-vector-builder.h: New file.
5290 * vec-perm-indices.h: Include int-vector-builder.h.
5291 (vec_perm_indices): Redefine as an int_vector_builder.
5292 (auto_vec_perm_indices): Delete.
5293 (vec_perm_builder): Redefine as a stand-alone class.
5294 (vec_perm_indices::vec_perm_indices): New function.
5295 (vec_perm_indices::clamp): Likewise.
5296 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5297 (vec_perm_indices::new_vector): New function.
5298 (vec_perm_indices::new_expanded_vector): Update for new
5299 vec_perm_indices class.
5300 (vec_perm_indices::rotate_inputs): New function.
5301 (vec_perm_indices::all_in_range_p): Operate directly on the
5302 encoded form, without computing elided elements.
5303 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5304 encoding. Update for new vec_perm_indices class.
5305 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5306 the given vec_perm_builder.
5307 (expand_vec_perm_var): Update vec_perm_builder constructor.
5308 (expand_mult_highpart): Use vec_perm_builder instead of
5309 auto_vec_perm_indices.
5310 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5311 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5312 or double series encoding as appropriate.
5313 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5314 vec_perm_indices instead of auto_vec_perm_indices.
5315 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5316 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5317 (vect_permute_store_chain): Likewise.
5318 (vect_grouped_load_supported): Likewise.
5319 (vect_permute_load_chain): Likewise.
5320 (vect_shift_permute_load_chain): Likewise.
5321 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5322 (vect_transform_slp_perm_load): Likewise.
5323 (vect_schedule_slp_instance): Likewise.
5324 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5325 (vectorizable_mask_load_store): Likewise.
5326 (vectorizable_bswap): Likewise.
5327 (vectorizable_store): Likewise.
5328 (vectorizable_load): Likewise.
5329 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5330 vec_perm_indices instead of auto_vec_perm_indices. Use
5331 tree_to_vec_perm_builder to read the vector from a tree.
5332 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5333 vec_perm_builder instead of a vec_perm_indices.
5334 (have_whole_vector_shift): Use vec_perm_builder and
5335 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5336 truncation to calc_vec_perm_mask_for_shift.
5337 (vect_create_epilog_for_reduction): Likewise.
5338 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5339 from auto_vec_perm_indices to vec_perm_indices.
5340 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5341 instead of changing individual elements.
5342 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5343 the vector in d.perm.
5344 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5345 from auto_vec_perm_indices to vec_perm_indices.
5346 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5347 instead of changing individual elements.
5348 (arm_vectorize_vec_perm_const): Use new_vector to install
5349 the vector in d.perm.
5350 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5351 Update vec_perm_builder constructor.
5352 (rs6000_expand_interleave): Likewise.
5353 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5354 (rs6000_expand_interleave): Likewise.
5356 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5358 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5359 to qimode could truncate the indices.
5360 * optabs.c (expand_vec_perm_var): Likewise.
5362 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5364 * Makefile.in (OBJS): Add vec-perm-indices.o.
5365 * vec-perm-indices.h: New file.
5366 * vec-perm-indices.c: Likewise.
5367 * target.h (vec_perm_indices): Replace with a forward class
5369 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5370 * optabs.h: Include vec-perm-indices.h.
5371 (expand_vec_perm): Delete.
5372 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5373 (expand_vec_perm_const): Declare.
5374 * target.def (vec_perm_const_ok): Replace with...
5375 (vec_perm_const): ...this new hook.
5376 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5377 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5378 * doc/tm.texi: Regenerate.
5379 * optabs.def (vec_perm_const): Delete.
5380 * doc/md.texi (vec_perm_const): Likewise.
5381 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5382 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5383 expand_vec_perm for constant permutation vectors. Assert that
5384 the mode of variable permutation vectors is the integer equivalent
5385 of the mode that is being permuted.
5386 * optabs-query.h (selector_fits_mode_p): Declare.
5387 * optabs-query.c: Include vec-perm-indices.h.
5388 (selector_fits_mode_p): New function.
5389 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5390 is defined, instead of checking whether the vec_perm_const_optab
5391 exists. Use targetm.vectorize.vec_perm_const instead of
5392 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5393 fit in the vector mode before using a variable permute.
5394 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5395 vec_perm_indices instead of an rtx.
5396 (expand_vec_perm): Replace with...
5397 (expand_vec_perm_const): ...this new function. Take the selector
5398 as a vec_perm_indices rather than an rtx. Also take the mode of
5399 the selector. Update call to shift_amt_for_vec_perm_mask.
5400 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5401 Use vec_perm_indices::new_expanded_vector to expand the original
5402 selector into bytes. Check whether the indices fit in the vector
5403 mode before using a variable permute.
5404 (expand_vec_perm_var): Make global.
5405 (expand_mult_highpart): Use expand_vec_perm_const.
5406 * fold-const.c: Includes vec-perm-indices.h.
5407 * tree-ssa-forwprop.c: Likewise.
5408 * tree-vect-data-refs.c: Likewise.
5409 * tree-vect-generic.c: Likewise.
5410 * tree-vect-loop.c: Likewise.
5411 * tree-vect-slp.c: Likewise.
5412 * tree-vect-stmts.c: Likewise.
5413 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5415 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5416 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5417 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5418 (aarch64_vectorize_vec_perm_const): ...this new function.
5419 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5420 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5421 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5422 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5423 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5424 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5425 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5427 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5428 check for NEON modes.
5429 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5430 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5431 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5432 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5434 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5435 the old VEC_PERM_CONST conditions.
5436 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5437 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5438 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5439 (ia64_vectorize_vec_perm_const_ok): Merge into...
5440 (ia64_vectorize_vec_perm_const): ...this new function.
5441 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5442 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5443 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5444 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5445 * config/mips/mips.c (mips_expand_vec_perm_const)
5446 (mips_vectorize_vec_perm_const_ok): Merge into...
5447 (mips_vectorize_vec_perm_const): ...this new function.
5448 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5449 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5450 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5451 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5452 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5453 (rs6000_expand_vec_perm_const): Delete.
5454 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5456 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5457 (altivec_expand_vec_perm_const_le): Take each operand individually.
5458 Operate on constant selectors rather than rtxes.
5459 (altivec_expand_vec_perm_const): Likewise. Update call to
5460 altivec_expand_vec_perm_const_le.
5461 (rs6000_expand_vec_perm_const): Delete.
5462 (rs6000_vectorize_vec_perm_const_ok): Delete.
5463 (rs6000_vectorize_vec_perm_const): New function.
5464 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5465 an element count and rtx array.
5466 (rs6000_expand_extract_even): Update call accordingly.
5467 (rs6000_expand_interleave): Likewise.
5468 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5469 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5470 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5471 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5472 (rs6000_expand_vec_perm_const): Delete.
5473 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5474 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5475 (altivec_expand_vec_perm_const_le): Take each operand individually.
5476 Operate on constant selectors rather than rtxes.
5477 (altivec_expand_vec_perm_const): Likewise. Update call to
5478 altivec_expand_vec_perm_const_le.
5479 (rs6000_expand_vec_perm_const): Delete.
5480 (rs6000_vectorize_vec_perm_const_ok): Delete.
5481 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5482 reference to the SPE evmerge intructions.
5483 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5484 an element count and rtx array.
5485 (rs6000_expand_extract_even): Update call accordingly.
5486 (rs6000_expand_interleave): Likewise.
5487 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5488 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5490 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5492 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5494 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5495 vector mode and that that mode matches the mode of the data
5497 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5498 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5499 directly using expand_vec_perm_1 when forcing selectors into
5501 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5503 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5505 * optabs-query.h (can_vec_perm_p): Delete.
5506 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5507 * optabs-query.c (can_vec_perm_p): Split into...
5508 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5509 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5510 particular selector is valid.
5511 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5512 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5513 (vect_grouped_load_supported): Likewise.
5514 (vect_shift_permute_load_chain): Likewise.
5515 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5516 (vect_transform_slp_perm_load): Likewise.
5517 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5518 (vectorizable_bswap): Likewise.
5519 (vect_gen_perm_mask_checked): Likewise.
5520 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5521 implementations of variable permutation vectors into account
5522 when deciding which selector to use.
5523 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5524 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5525 with a false third argument.
5526 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5527 to test whether the constant selector is valid and can_vec_perm_var_p
5528 to test whether a variable selector is valid.
5530 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5532 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5533 * optabs-query.c (can_vec_perm_p): Likewise.
5534 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5535 instead of vec_perm_indices.
5536 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5537 (vect_gen_perm_mask_checked): Likewise,
5538 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5539 (vect_gen_perm_mask_checked): Likewise,
5541 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5543 * optabs-query.h (qimode_for_vec_perm): Declare.
5544 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5545 (qimode_for_vec_perm): ...this new function.
5546 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5548 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5550 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5551 does not have a conditional at the top.
5553 2018-01-02 Richard Biener <rguenther@suse.de>
5555 * ipa-inline.c (big_speedup_p): Fix expression.
5557 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5560 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5563 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5567 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5568 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5569 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5570 cond_taken_branch_cost 3->4.
5572 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5574 PR tree-optimization/83581
5575 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5576 TODO_cleanup_cfg if any changes have been made.
5579 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5580 convert_modes if target mode has the right side, but different mode
5584 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5585 last argument when extracting from CONCAT. If either from_real or
5586 from_imag is NULL, use expansion through memory. If result is not
5587 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5588 the parts directly to inner mode, if even that fails, use expansion
5592 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5593 check for bswap in mode rather than HImode and use that in expand_unop
5596 Copyright (C) 2018 Free Software Foundation, Inc.
5598 Copying and distribution of this file, with or without modification,
5599 are permitted in any medium without royalty provided the copyright
5600 notice and this notice are preserved.