1 2018-01-22 Carl Love <cel@us.ibm.com>
3 * doc/extend.tex: Fix typo in second arg in
4 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
6 2018-01-29 Richard Biener <rguenther@suse.de>
8 PR tree-optimization/84086
9 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
10 (flush_ssaname_freelist): When SSA names were released reset
13 2018-01-29 Richard Biener <rguenther@suse.de>
15 PR tree-optimization/84057
16 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
17 removed paths when removing edges.
19 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
21 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
22 -mfunction-return=@var{choice}.
24 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
27 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
29 (layout::print_source_line): Likewise.
30 (test_get_line_width_without_trailing_whitespace): Add test cases.
32 2018-01-27 Jakub Jelinek <jakub@redhat.com>
35 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
38 2018-01-26 Jim Wilson <jimw@sifive.com>
40 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
42 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
45 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
47 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
48 and CMP + SUB-immediate -> SUBS.
50 2018-01-26 Martin Sebor <msebor@redhat.com>
52 PR tree-optimization/83896
53 * tree-ssa-strlen.c (get_string_len): Rename...
54 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
55 Avoid assuming length is constant.
56 (handle_char_store): Use HOST_WIDE_INT for string length.
58 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
61 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
62 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
64 2018-01-26 Richard Biener <rguenther@suse.de>
66 PR rtl-optimization/84003
67 * dse.c (record_store): Only record redundant stores when
68 the earlier store aliases at least all accesses the later one does.
70 2018-01-26 Jakub Jelinek <jakub@redhat.com>
72 PR rtl-optimization/83985
73 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
74 REG_CFA_RESTORE insns.
75 (delete_unmarked_insns): Don't ignore separate shrink wrapping
76 REG_CFA_RESTORE insns here.
79 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
80 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
82 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
84 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
85 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
87 (arc_override_options): Likewise.
88 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
90 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
92 * config/arc/arc.h (TARGET_DBNZ): Define.
93 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
94 properly set the tune attribute.
95 (dbnz): Use TARGET_DBNZ guard.
96 * config/arc/arc.opt (mtune): Add core3 option.
98 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
100 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
101 recognize new pic like addresses.
102 (arc_delegitimize_address): Clean up.
104 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
106 * config/arc/arc-arches.def: Option mrf16 valid for all
108 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
109 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
110 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
111 * config/arc/arc-tables.opt: Regenerate.
112 * config/arc/arc.c (arc_conditional_register_usage): Handle
113 reduced register file case.
114 (arc_file_start): Set must have build attributes.
115 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
117 * config/arc/arc.opt (mrf16): Add new option.
118 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
119 * config/arc/genmultilib.awk: Handle new mrf16 option.
120 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
121 * config/arc/t-multilib: Regenerate.
122 * doc/invoke.texi (ARC Options): Document mrf16 option.
124 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
126 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
127 * config/arc/arc.c (arc_handle_secure_attribute): New function.
128 (arc_attribute_table): Add 'secure_call' attribute.
129 (arc_print_operand): Print secure call operand.
130 (arc_function_ok_for_sibcall): Don't optimize tail calls when
132 (arc_is_secure_call_p): New function. * config/arc/arc.md
133 (call_i): Add support for sjli instruction.
134 (call_value_i): Likewise.
135 * config/arc/constraints.md (Csc): New constraint.
137 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
138 John Eric Martin <John.Martin@emmicro-us.com>
140 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
141 * config/arc/arc.c (_arc_jli_section): New struct.
142 (arc_jli_section): New type.
143 (rc_jli_sections): New static variable.
144 (arc_handle_jli_attribute): New function.
145 (arc_attribute_table): Add jli_always and jli_fixed attribute.
146 (arc_file_end): New function.
147 (TARGET_ASM_FILE_END): Define.
148 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
149 (arc_add_jli_section): New function.
150 (jli_call_scan): Likewise.
151 (arc_reorg): Call jli_call_scan.
152 (arc_output_addsi): Remove 'S' from printing asm operand.
153 (arc_is_jli_call_p): New function.
154 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
156 (movhi_insn): Likewise.
157 (movsi_insn): Likewise.
158 (movsi_set_cc_insn): Likewise.
159 (loadqi_update): Likewise.
160 (load_zeroextendqisi_update): Likewise.
161 (load_signextendqisi_update): Likewise.
162 (loadhi_update): Likewise.
163 (load_zeroextendhisi_update): Likewise.
164 (load_signextendhisi_update): Likewise.
165 (loadsi_update): Likewise.
166 (loadsf_update): Likewise.
167 (movsicc_insn): Likewise.
168 (bset_insn): Likewise.
169 (bxor_insn): Likewise.
170 (bclr_insn): Likewise.
171 (bmsk_insn): Likewise.
172 (bicsi3_insn): Likewise.
173 (cmpsi_cc_c_insn): Likewise.
174 (movsi_ne): Likewise.
175 (movsi_cond_exec): Likewise.
176 (clrsbsi2): Likewise.
187 (call_i): Remove 'S' asm letter, add jli instruction.
188 (call_value_i): Likewise.
189 * config/arc/arc.op (mjli-always): New option.
190 * config/arc/constraints.md (Cji): New constraint.
191 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
193 (subsf3_fpx): Likewise.
194 (mulsf3_fpx): Likewise.
195 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
197 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
199 * doc/invoke.texi (ARC): Document mjli-always option.
201 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
203 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
204 and use incw and decw where possible
205 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
207 2018-01-26 Richard Biener <rguenther@suse.de>
209 PR tree-optimization/81082
210 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
211 association if it requires casting to unsigned.
212 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
213 from fold_plusminus_mult_expr to catch important cases late when
214 range info is available.
216 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
218 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
219 * configure.ac (hidden_linkonce): New test.
220 * configure: Regenerate.
221 * config.in: Regenerate.
223 2018-01-26 Julia Koval <julia.koval@intel.com>
225 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
226 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
227 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
228 _mm_mask_bitshuffle_epi64_mask): Fix type.
229 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
230 USI_FTYPE_V4DI_V4DI_USI): Remove.
231 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
232 __builtin_ia32_vpshufbitqmb256_mask,
233 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
234 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
235 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
237 2018-01-26 Alan Modra <amodra@gmail.com>
240 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
241 UNSPEC_VBPERMQ. Sort other unspecs.
243 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
245 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
247 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
250 * predict.c (drop_profile): Do not push/pop cfun; update also
252 (handle_missing_profiles): Fix logic looking for zero profiles.
254 2018-01-25 Jakub Jelinek <jakub@redhat.com>
257 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
258 on functions with #pragma omp declare simd or functions with simd
260 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
261 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
262 Remove trailing \n from warning_at calls.
264 2018-01-25 Tom de Vries <tom@codesourcery.com>
267 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
268 for neutered workers.
270 2018-01-24 Joseph Myers <joseph@codesourcery.com>
273 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
274 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
276 2017-01-08 Jeff Law <law@redhat.com>
279 * i386.c (get_probe_interval): Move to earlier point.
280 (ix86_compute_frame_layout): If -fstack-clash-protection and
281 the frame is larger than the probe interval, then use pushes
282 to save registers rather than reg->mem moves.
283 (ix86_expand_prologue): Remove conditional for int_registers_saved
286 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
289 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
290 min/max for never referenced object.
292 2018-01-24 Jakub Jelinek <jakub@redhat.com>
295 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
297 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
298 attributes from DECL_ATTRIBUTES (decl) without affecting
299 DECL_ATTRIBUTES (current_function_decl).
300 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
301 functions with non-NULL DECL_ABSTRACT_ORIGIN.
303 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
305 PR tree-optimization/83979
306 * fold-const.c (fold_comparison): Use constant_boolean_node
307 instead of boolean_{true,false}_node.
309 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
311 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
314 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
316 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
317 Simplify the clause that sets the length attribute.
318 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
319 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
320 clause that sets the length attribute.
321 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
323 2018-01-24 Tom de Vries <tom@codesourcery.com>
326 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
327 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
328 Add strict parameter.
329 (prevent_branch_around_nothing): Insert dummy insn between branch to
330 label and label with no ptx insn inbetween.
331 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
333 2018-01-24 Tom de Vries <tom@codesourcery.com>
336 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
337 for neutered threads in warp.
338 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
340 2018-01-24 Richard Biener <rguenther@suse.de>
342 PR tree-optimization/83176
343 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
346 2018-01-24 Richard Biener <rguenther@suse.de>
348 PR tree-optimization/82819
349 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
350 code generating pluses that are no-ops in the target precision.
352 2018-01-24 Richard Biener <rguenther@suse.de>
355 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
357 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
359 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
360 to merge probabilities.
361 * predict.c (probably_never_executed): Also mark as cold functions
362 with global 0 profile and guessed local profile.
363 * profile-count.c (profile_probability::combine_with_count): New
365 * profile-count.h (profile_probability::operator*,
366 profile_probability::operator*=, profile_probability::operator/,
367 profile_probability::operator/=): Reduce precision to adjusted
368 and set value to guessed on contradictory divisions.
369 (profile_probability::combine_with_freq): Remove.
370 (profile_probability::combine_wiht_count): Declare.
371 (profile_count::force_nonzero):: Set to adjusted.
372 (profile_count::probability_in):: Set quality to adjusted.
373 * tree-ssa-tail-merge.c (replace_block_by): Use
376 2018-01-23 Andrew Waterman <andrew@sifive.com>
377 Jim Wilson <jimw@sifive.com>
379 * config/riscv/riscv.c (riscv_stack_boundary): New.
380 (riscv_option_override): Set riscv_stack_boundary. Handle
381 riscv_preferred_stack_boundary_arg.
382 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
383 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
384 (STACK_BOUNDARY): Set to riscv_stack_boundary.
385 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
386 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
387 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
389 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
392 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
393 of struct ix86_frame.
394 (ix86_expand_epilogue): Likewise. Add a local variable for
395 the reg_save_offset field in struct ix86_frame.
397 2018-01-23 Bin Cheng <bin.cheng@arm.com>
399 PR tree-optimization/82604
400 * tree-loop-distribution.c (enum partition_kind): New enum item
401 PKIND_PARTIAL_MEMSET.
402 (partition_builtin_p): Support above new enum item.
403 (generate_code_for_partition): Ditto.
404 (compute_access_range): Differentiate cases that equality can be
405 proven at all loops, the innermost loops or no loops.
406 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
407 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
408 (finalize_partitions, distribute_loop): Don't fuse partition of
409 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
410 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
413 2018-01-23 Martin Liska <mliska@suse.cz>
415 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
416 order to ignore the predictor.
417 (PRED_POLYMORPHIC_CALL): Likewise.
418 (PRED_RECURSIVE_CALL): Likewise.
420 2018-01-23 Martin Liska <mliska@suse.cz>
422 * tree-profile.c (tree_profiling): Print function header to
423 aware reader which function we are working on.
424 * value-prof.c (gimple_find_values_to_profile): Do not print
425 not interesting value histograms.
427 2018-01-23 Martin Liska <mliska@suse.cz>
429 * profile-count.h (enum profile_quality): Add
430 profile_uninitialized as the first value. Do not number values
431 as they are zero based.
432 (profile_count::verify): Update sanity check.
433 (profile_probability::verify): Likewise.
435 2018-01-23 Nathan Sidwell <nathan@acm.org>
437 * doc/invoke.texi (ffor-scope): Deprecate.
439 2018-01-23 David Malcolm <dmalcolm@redhat.com>
441 PR tree-optimization/83510
442 * domwalk.c (set_all_edges_as_executable): New function.
443 (dom_walker::dom_walker): Convert bool param
444 "skip_unreachable_blocks" to enum reachability. Move setup of
445 edge flags to set_all_edges_as_executable and only do it when
446 reachability is REACHABLE_BLOCKS.
447 * domwalk.h (enum dom_walker::reachability): New enum.
448 (dom_walker::dom_walker): Convert bool param
449 "skip_unreachable_blocks" to enum reachability.
450 (set_all_edges_as_executable): New decl.
451 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
452 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
454 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
455 but converting true to REACHABLE_BLOCKS.
456 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
458 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
459 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
460 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
462 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
463 if check_all_array_refs will be called.
465 2018-01-23 David Malcolm <dmalcolm@redhat.com>
467 * tree.c (selftest::test_location_wrappers): Add more test
470 2018-01-23 David Malcolm <dmalcolm@redhat.com>
472 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
473 (selftest::test_bit_in_range): Likewise.
475 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
478 * doc/sourcebuild.texi (vect_float): Say that the selector
479 only describes the situation when -funsafe-math-optimizations is on.
480 (vect_float_strict): Document.
482 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
484 PR tree-optimization/83965
485 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
486 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
487 instead of checking only for a reduction.
488 (vect_recog_widen_sum_pattern): Likewise.
490 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
492 * predict.c (probably_never_executed): Only use precise profile info.
493 (compute_function_frequency): Skip after inlining hack since we now
494 have quality checking.
496 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
498 * profile-count.h (profile_probability::very_unlikely,
499 profile_probability::unlikely, profile_probability::even): Set
500 precision to guessed.
502 2018-01-23 Richard Biener <rguenther@suse.de>
504 PR tree-optimization/83963
505 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
506 Properly terminate dominator walk when crossing the exit edge not
507 when visiting its source block.
509 2018-01-23 Jakub Jelinek <jakub@redhat.com>
512 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
513 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
515 2018-01-22 Jakub Jelinek <jakub@redhat.com>
517 PR tree-optimization/83957
518 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
519 semicolon after for body surrounded by braces.
521 PR tree-optimization/83081
522 * profile-count.h (profile_probability::split): New method.
523 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
524 Use profile_probability::split.
525 (do_compare_rtx_and_jump): Fix adjustment of probabilities
526 when splitting a single conditional jump into 2.
528 2018-01-22 David Malcolm <dmalcolm@redhat.com>
530 PR tree-optimization/69452
531 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
534 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
536 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
537 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
538 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
540 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
542 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
543 * config/rl78/rl78.md: New define_expand "movdi"
544 * config/rl78/rl78.c: New function definition rl78_split_movdi
546 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
549 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
551 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
552 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
553 128-bit to produce an UNSPEC move to get the double word with the
554 signbit and then a shift directly to do signbit.
555 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
556 implementation with a new version that just does either a direct
557 move or a regular move. Move memory interface to separate insns.
558 Move insns so they are next to the expander.
559 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
560 with signbit move. Split big and little endian case.
561 (signbit<mode>2_dm_mem_le): Likewise.
562 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
563 (signbit<mode>2_dm2): Likewise.
565 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
567 * config/rl78/rl78.md: New define_expand "anddi3".
569 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
571 * config/rl78/rl78.md: New define_expand "umindi3".
573 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
575 * config/rl78/rl78.md: New define_expand "smindi3".
577 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
579 * config/rl78/rl78.md: New define_expand "smaxdi3".
581 2018-01-22 Carl Love <cel@us.ibm.com>
583 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
584 LVX_V1TI): Add macro expansion.
585 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
586 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
587 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
588 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
589 Change check to determine if the instruction is a byte reversing
590 entry. Fix typo in comment.
591 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
592 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
593 Add def_builtin calls for new builtins.
594 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
595 Add define_insn expansion.
597 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
599 * config/rl78/rl78.md: New define_expand "umaxdi3".
601 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
603 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
604 for non-QImode registers
606 2018-01-22 Richard Biener <rguenther@suse.de>
608 PR tree-optimization/83963
609 * graphite-scop-detection.c (scop_detection::get_sese): Delay
610 including the loop exit block.
611 (scop_detection::merge_sese): Likewise.
612 (scop_detection::add_scop): Do it here instead.
614 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
616 * doc/sourcebuild.texi (arm_softfloat): Document.
618 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
621 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
622 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
623 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
625 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
626 David Edelsohn <dje.gcc@gmail.com>
629 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
630 Change "crset eq" to "crset 2".
631 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
632 (*call_indirect_aix<mode>_nospec): Likewise.
633 (*call_value_indirect_aix<mode>_nospec): Likewise.
634 (*call_indirect_elfv2<mode>_nospec): Likewise.
635 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
636 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
637 change assembly output from . to $.
638 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
639 (indirect_jump<mode>_nospec): Change assembly output from . to $.
640 (*tablejump<mode>_internal1_nospec): Likewise.
642 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
645 * config/sh/sh_optimize_sett_clrt.cc:
646 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
648 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
650 PR tree-optimization/83940
651 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
652 offset_dt to vect_constant_def rather than vect_unknown_def_type.
653 (vect_check_load_store_mask): Add a mask_dt_out parameter and
654 use it to pass back the definition type.
655 (vect_check_store_rhs): Likewise rhs_dt_out.
656 (vect_build_gather_load_calls): Add a mask_dt argument and use
657 it instead of a call to vect_is_simple_use.
658 (vectorizable_store): Update calls to vect_check_load_store_mask
659 and vect_check_store_rhs. Use the dt returned by the latter instead
660 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
661 instead of calls to vect_is_simple_use. Pass the scalar rather
662 than the vector operand to vect_is_simple_use when handling
663 second and subsequent copies of an rhs value.
664 (vectorizable_load): Update calls to vect_check_load_store_mask
665 and vect_build_gather_load_calls. Use the cached mask_dt and
666 gs_info.offset_dt instead of calls to vect_is_simple_use.
668 2018-01-20 Jakub Jelinek <jakub@redhat.com>
671 * tree-emutls.c: Include gimplify.h.
672 (lower_emutls_2): New function.
673 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
674 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
675 it before further processing.
678 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
679 UINTVAL (trueop1) instead of INTVAL (op1).
681 2018-01-19 Jakub Jelinek <jakub@redhat.com>
685 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
686 INCOMING_FRAME_SP_OFFSET if not defined.
687 (scan_trace): Add ENTRY argument. If true and
688 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
689 emit a note to adjust the CFA offset.
690 (create_cfi_notes): Adjust scan_trace callers.
691 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
692 INCOMING_FRAME_SP_OFFSET in the CIE.
693 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
694 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
696 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
697 * doc/tm.texi: Regenerated.
699 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
701 PR rtl-optimization/83147
702 * lra-constraints.c (remove_inheritance_pseudos): Use
703 lra_substitute_pseudo_within_insn.
705 2018-01-19 Tom de Vries <tom@codesourcery.com>
706 Cesar Philippidis <cesar@codesourcery.com>
709 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
711 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
714 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
715 spaces for function labels.
717 2018-01-19 Martin Liska <mliska@suse.cz>
719 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
720 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
721 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
722 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
723 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
724 (PRED_CONST_RETURN): Change from 69 to 65.
725 (PRED_NULL_RETURN): Change from 91 to 71.
726 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
727 (PRED_LOOP_GUARD): Change from 66 to 73.
729 2018-01-19 Martin Liska <mliska@suse.cz>
731 * predict.c (predict_insn_def): Add new assert.
732 (struct branch_predictor): Change type to signed integer.
733 (test_prediction_value_range): Amend test to cover
735 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
736 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
737 (PRED_LOOP_ITERATIONS_MAX): Likewise.
738 (PRED_LOOP_IV_COMPARE): Likewise.
739 * predict.h (PROB_UNINITIALIZED): Define new constant.
741 2018-01-19 Martin Liska <mliska@suse.cz>
743 * predict.c (dump_prediction): Add new format for
744 analyze_brprob.py script which is enabled with -details
746 * profile-count.h (precise_p): New function.
748 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
750 PR tree-optimization/83922
751 * tree-vect-loop.c (vect_verify_full_masking): Return false if
752 there are no statements that need masking.
753 (vect_active_double_reduction_p): New function.
754 (vect_analyze_loop_operations): Use it when handling phis that
755 are not in the loop header.
757 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
759 PR tree-optimization/83914
760 * tree-vect-loop.c (vectorizable_induction): Don't convert
761 init_expr or apply the peeling adjustment for inductions
762 that are nested within the vectorized loop.
764 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
766 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
769 2018-01-18 Jakub Jelinek <jakub@redhat.com>
773 * function.h (gimplify_parameters): Add gimple_seq * argument.
774 * function.c: Include gimple.h and options.h.
775 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
776 for the added local temporaries if needed.
777 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
778 if there are any parameter cleanups, wrap whole body into a
779 try/finally with the cleanups.
781 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
784 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
785 Use GET_MODE_CLASS for scalar floating point.
787 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
791 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
792 Fix call of call_cgraph_insertion_hooks.
794 2018-01-18 Martin Sebor <msebor@redhat.com>
796 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
798 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
801 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
804 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
807 * common.opt: (-ffile-prefix-map): New option.
808 * opts.c (common_handle_option): Defer it.
809 * opts-global.c (handle_common_deferred_options): Handle it.
810 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
811 * file-prefix-map.h: New file.
812 (remap_debug_filename, add_debug_prefix_map): ...here.
813 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
814 * final.c (debug_prefix_map, add_debug_prefix_map
815 remap_debug_filename): Move to...
816 * file-prefix-map.c: New file.
817 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
818 generalize, get rid of alloca(), use strrchr() instead of strchr().
819 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
820 Implement in terms of add_prefix_map().
821 (remap_macro_filename, remap_debug_filename): Implement in term of
823 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
824 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
825 * dbxout.c: Include file-prefix-map.h.
826 * varasm.c: Likewise.
827 * vmsdbgout.c: Likewise.
828 * xcoffout.c: Likewise.
829 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
830 * doc/cppopts.texi (-fmacro-prefix-map): Document.
831 * doc/invoke.texi (-ffile-prefix-map): Document.
832 (-fdebug-prefix-map): Update description.
834 2018-01-18 Martin Liska <mliska@suse.cz>
836 * config/i386/i386.c (indirect_thunk_name): Document that also
838 (output_indirect_thunk): Document why both instructions
839 (pause and lfence) are generated.
841 2018-01-18 Richard Biener <rguenther@suse.de>
843 PR tree-optimization/83887
844 * graphite-scop-detection.c
845 (scop_detection::get_nearest_dom_with_single_entry): Remove.
846 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
847 (scop_detection::merge_sese): Re-implement with a flood-fill
848 algorithm that properly finds a SESE region if it exists.
850 2018-01-18 Jakub Jelinek <jakub@redhat.com>
853 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
854 pointer_diff optimizations use view_convert instead of convert.
856 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
858 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
859 Generate different code for -mno-speculate-indirect-jumps.
860 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
861 (*call_indirect_aix<mode>): Disable for
862 -mno-speculate-indirect-jumps.
863 (*call_indirect_aix<mode>_nospec): New define_insn.
864 (*call_value_indirect_aix<mode>): Disable for
865 -mno-speculate-indirect-jumps.
866 (*call_value_indirect_aix<mode>_nospec): New define_insn.
867 (*sibcall_nonlocal_sysv<mode>): Generate different code for
868 -mno-speculate-indirect-jumps.
869 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
871 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
873 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
874 long double type, set the flags for noting the default long double
875 type, even if we don't pass or return a long double type.
877 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
880 * ipa-inline.c (flatten_function): Do not overwrite final inlining
883 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
885 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
886 support for merge[hl].
887 (fold_mergehl_helper): New helper function.
888 (tree-vector-builder.h): New #include for tree_vector_builder usage.
889 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
890 (altivec_vmrglw_direct): Add xxmrglw insn.
892 2018-01-17 Andrew Waterman <andrew@sifive.com>
894 * config/riscv/riscv.c (riscv_conditional_register_usage): If
895 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
897 2018-01-17 David Malcolm <dmalcolm@redhat.com>
900 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
901 call the lto_location_cache before reading the
902 DECL_SOURCE_LOCATION of the types.
904 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
905 Richard Sandiford <richard.sandiford@linaro.org>
907 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
908 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
909 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
910 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
911 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
913 * config/aarch64/constraints.md (aarch64_movti_operand):
915 * config/aarch64/predicates.md (Uti): Add new constraint.
917 2018-01-17 Carl Love <cel@us.ibm.com>
918 * config/rs6000/vsx.md (define_expand xl_len_r,
919 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
920 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
922 (define_expand, define_insn): Move the shift left from the
923 define_insn to the define_expand for lxvl and stxvl instructions.
924 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
925 and XL_LEN_R definitions to PURE.
927 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
929 * config/i386/i386.c (indirect_thunk_name): Declare regno
930 as unsigned int. Compare regno with INVALID_REGNUM.
931 (output_indirect_thunk): Ditto.
932 (output_indirect_thunk_function): Ditto.
933 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
934 in the call to output_indirect_thunk_function.
936 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
939 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
940 rather than the size of inner_type to determine the stack slot size
941 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
943 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
946 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
949 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
951 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
952 endian Linux systems to optionally enable multilibs for selecting
953 the long double type if the user configured an explicit type.
954 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
955 have no long double multilibs if not defined.
956 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
957 warn if the user used -mabi={ieee,ibm}longdouble and we built
958 multilibs for long double.
959 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
960 appropriate multilib option.
961 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
963 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
964 for building long double multilibs.
965 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
967 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
969 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
972 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
974 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
977 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
980 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
983 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
985 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
986 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
988 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
990 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
991 different rtl trees depending on TARGET_64BIT.
992 (rs6000_gen_lvx): Likewise.
994 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
996 * config/visium/visium.md (nop): Tweak comment.
997 (hazard_nop): Likewise.
999 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1001 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1002 -mspeculate-indirect-jumps.
1003 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1004 for -mno-speculate-indirect-jumps.
1005 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1006 (*call_value_indirect_elfv2<mode>): Disable for
1007 -mno-speculate-indirect-jumps.
1008 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1009 (indirect_jump): Emit different RTL for
1010 -mno-speculate-indirect-jumps.
1011 (*indirect_jump<mode>): Disable for
1012 -mno-speculate-indirect-jumps.
1013 (*indirect_jump<mode>_nospec): New define_insn.
1014 (tablejump): Emit different RTL for
1015 -mno-speculate-indirect-jumps.
1016 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1017 (tablejumpsi_nospec): New define_expand.
1018 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1019 (tablejumpdi_nospec): New define_expand.
1020 (*tablejump<mode>_internal1): Disable for
1021 -mno-speculate-indirect-jumps.
1022 (*tablejump<mode>_internal1_nospec): New define_insn.
1023 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1026 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1028 * caller-save.c (insert_save): Drop unnecessary parameter. All
1031 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1032 Richard Biener <rguenth@suse.de>
1035 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1036 return early, inline manually is_gimple_sizepos. Make sure if we
1037 call gimplify_expr we don't end up with a gimple constant.
1038 * tree.c (variably_modified_type_p): Don't return true for
1039 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1040 * gimplify.h (is_gimple_sizepos): Remove.
1042 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1044 PR tree-optimization/83857
1045 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1046 vectorizable_live_operation for pure SLP statements.
1047 (vectorizable_live_operation): Handle PHIs.
1049 2018-01-16 Richard Biener <rguenther@suse.de>
1051 PR tree-optimization/83867
1052 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1053 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1055 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1058 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1059 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1060 If off is not INTEGER_CST, issue a may not be aligned warning
1061 rather than isn't aligned. Use isn%'t rather than isn't.
1062 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1064 <case MULT_EXPR>: Improve the case when bottom and one of the
1065 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1066 operand, in that case check if the other operand is multiple of
1067 bottom divided by the INTEGER_CST operand.
1069 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1072 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1073 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1074 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1075 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1076 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1077 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1078 (pa_function_arg_size): New function.
1080 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1082 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1083 in a separate statement.
1085 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1087 PR tree-optimization/83847
1088 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1089 group gathers and scatters.
1091 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1093 PR rtl-optimization/86620
1094 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1096 PR rtl-optimization/83213
1097 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1098 to last if both are JUMP_INSNs.
1100 PR tree-optimization/83843
1101 * gimple-ssa-store-merging.c
1102 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1103 store_immediate_info for bswap/nop orig_stores.
1105 2018-01-15 Andrew Waterman <andrew@sifive.com>
1107 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1109 <UDIV>: Increase cost if !TARGET_DIV.
1111 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1113 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1114 (define_attr "cr_logical_3op"): New.
1115 (cceq_ior_compare): Adjust.
1116 (cceq_ior_compare_complement): Adjust.
1117 (*cceq_rev_compare): Adjust.
1118 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1119 (is_cracked_insn): Adjust.
1120 (insn_must_be_first_in_group): Adjust.
1121 * config/rs6000/40x.md: Adjust.
1122 * config/rs6000/440.md: Adjust.
1123 * config/rs6000/476.md: Adjust.
1124 * config/rs6000/601.md: Adjust.
1125 * config/rs6000/603.md: Adjust.
1126 * config/rs6000/6xx.md: Adjust.
1127 * config/rs6000/7450.md: Adjust.
1128 * config/rs6000/7xx.md: Adjust.
1129 * config/rs6000/8540.md: Adjust.
1130 * config/rs6000/cell.md: Adjust.
1131 * config/rs6000/e300c2c3.md: Adjust.
1132 * config/rs6000/e500mc.md: Adjust.
1133 * config/rs6000/e500mc64.md: Adjust.
1134 * config/rs6000/e5500.md: Adjust.
1135 * config/rs6000/e6500.md: Adjust.
1136 * config/rs6000/mpc.md: Adjust.
1137 * config/rs6000/power4.md: Adjust.
1138 * config/rs6000/power5.md: Adjust.
1139 * config/rs6000/power6.md: Adjust.
1140 * config/rs6000/power7.md: Adjust.
1141 * config/rs6000/power8.md: Adjust.
1142 * config/rs6000/power9.md: Adjust.
1143 * config/rs6000/rs64.md: Adjust.
1144 * config/rs6000/titan.md: Adjust.
1146 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1148 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1149 ix86_indirect_branch_register logic.
1151 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1153 * config/i386/constraints.md (Bs): Update
1154 ix86_indirect_branch_register check. Don't check
1155 ix86_indirect_branch_register with GOT_memory_operand.
1157 * config/i386/predicates.md (GOT_memory_operand): Don't check
1158 ix86_indirect_branch_register here.
1159 (GOT32_symbol_operand): Likewise.
1161 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1163 * config/i386/predicates.md (constant_call_address_operand):
1164 Rewrite ix86_indirect_branch_register logic.
1165 (sibcall_insn_operand): Likewise.
1167 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1169 * config/i386/constraints.md (Bs): Replace
1170 ix86_indirect_branch_thunk_register with
1171 ix86_indirect_branch_register.
1173 * config/i386/i386.md (indirect_jump): Likewise.
1174 (tablejump): Likewise.
1175 (*sibcall_memory): Likewise.
1176 (*sibcall_value_memory): Likewise.
1177 Peepholes of indirect call and jump via memory: Likewise.
1178 * config/i386/i386.opt: Likewise.
1179 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1180 (GOT_memory_operand): Likewise.
1181 (call_insn_operand): Likewise.
1182 (sibcall_insn_operand): Likewise.
1183 (GOT32_symbol_operand): Likewise.
1185 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1188 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1189 type rather than type addr's type points to.
1190 (expand_omp_atomic_mutex): Likewise.
1191 (expand_omp_atomic): Likewise.
1193 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1196 * config/i386/i386.c (output_indirect_thunk_function): Use
1197 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1198 for __x86_return_thunk.
1200 2018-01-15 Richard Biener <rguenther@suse.de>
1203 * expmed.c (extract_bit_field_1): Fix typo.
1205 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1208 * config/arm/iterators.md (VF): New mode iterator.
1209 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1210 Remove integer-related logic from pattern.
1211 (neon_vabd<mode>_3): Likewise.
1213 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1216 * common.opt (fstrict-overflow): No longer an alias.
1217 (fwrapv-pointer): New option.
1218 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1219 also for pointer types based on flag_wrapv_pointer.
1220 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1221 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1222 opts->x_flag_wrapv got set.
1223 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1224 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1225 POINTER_TYPE_OVERFLOW_UNDEFINED.
1226 * match.pd: Likewise in address comparison pattern.
1227 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1229 2018-01-15 Richard Biener <rguenther@suse.de>
1232 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1233 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1234 Reset type names to their identifier if their TYPE_DECL doesn't
1235 have linkage (and thus is used for ODR and devirt).
1236 (save_debug_info_for_decl): Remove.
1237 (save_debug_info_for_type): Likewise.
1238 (add_tree_to_fld_list): Adjust.
1239 * tree-pretty-print.c (dump_generic_node): Make dumping of
1240 type names more robust.
1242 2018-01-15 Richard Biener <rguenther@suse.de>
1244 * BASE-VER: Bump to 8.0.1.
1246 2018-01-14 Martin Sebor <msebor@redhat.com>
1249 * builtins.c (check_access): Avoid warning when the no-warning bit
1252 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1254 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1255 * ira-color (allocno_hard_regs_compare): Likewise.
1257 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1260 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1261 Use .pushsection/.popsection.
1263 2018-01-14 Martin Sebor <msebor@redhat.com>
1266 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1268 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1270 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1271 entry from extra_headers.
1272 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1273 extra_headers, make the list bitwise identical to the i?86-*-* one.
1275 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1277 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1278 -mcmodel=large with -mindirect-branch=thunk,
1279 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1280 -mfunction-return=thunk-extern.
1281 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1282 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1283 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1285 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1287 * config/i386/i386.c (print_reg): Print the name of the full
1288 integer register without '%'.
1289 (ix86_print_operand): Handle 'V'.
1290 * doc/extend.texi: Document 'V' modifier.
1292 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1294 * config/i386/constraints.md (Bs): Disallow memory operand for
1295 -mindirect-branch-register.
1297 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1298 (GOT_memory_operand): Likewise.
1299 (call_insn_operand): Likewise.
1300 (sibcall_insn_operand): Likewise.
1301 (GOT32_symbol_operand): Likewise.
1302 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1303 for -mindirect-branch-register.
1304 (tablejump): Likewise.
1305 (*sibcall_memory): Likewise.
1306 (*sibcall_value_memory): Likewise.
1307 Disallow peepholes of indirect call and jump via memory for
1308 -mindirect-branch-register.
1309 (*call_pop): Replace m with Bw.
1310 (*call_value_pop): Likewise.
1311 (*sibcall_pop_memory): Replace m with Bs.
1312 * config/i386/i386.opt (mindirect-branch-register): New option.
1313 * doc/invoke.texi: Document -mindirect-branch-register option.
1315 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1317 * config/i386/i386-protos.h (ix86_output_function_return): New.
1318 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1319 set function_return_type.
1320 (indirect_thunk_name): Add ret_p to indicate thunk for function
1322 (output_indirect_thunk_function): Pass false to
1323 indirect_thunk_name.
1324 (ix86_output_indirect_branch_via_reg): Likewise.
1325 (ix86_output_indirect_branch_via_push): Likewise.
1326 (output_indirect_thunk_function): Create alias for function
1327 return thunk if regno < 0.
1328 (ix86_output_function_return): New function.
1329 (ix86_handle_fndecl_attribute): Handle function_return.
1330 (ix86_attribute_table): Add function_return.
1331 * config/i386/i386.h (machine_function): Add
1332 function_return_type.
1333 * config/i386/i386.md (simple_return_internal): Use
1334 ix86_output_function_return.
1335 (simple_return_internal_long): Likewise.
1336 * config/i386/i386.opt (mfunction-return=): New option.
1337 (indirect_branch): Mention -mfunction-return=.
1338 * doc/extend.texi: Document function_return function attribute.
1339 * doc/invoke.texi: Document -mfunction-return= option.
1341 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1343 * config/i386/i386-opts.h (indirect_branch): New.
1344 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1345 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1346 with local indirect jump when converting indirect call and jump.
1347 (ix86_set_indirect_branch_type): New.
1348 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1349 (indirectlabelno): New.
1350 (indirect_thunk_needed): Likewise.
1351 (indirect_thunk_bnd_needed): Likewise.
1352 (indirect_thunks_used): Likewise.
1353 (indirect_thunks_bnd_used): Likewise.
1354 (INDIRECT_LABEL): Likewise.
1355 (indirect_thunk_name): Likewise.
1356 (output_indirect_thunk): Likewise.
1357 (output_indirect_thunk_function): Likewise.
1358 (ix86_output_indirect_branch_via_reg): Likewise.
1359 (ix86_output_indirect_branch_via_push): Likewise.
1360 (ix86_output_indirect_branch): Likewise.
1361 (ix86_output_indirect_jmp): Likewise.
1362 (ix86_code_end): Call output_indirect_thunk_function if needed.
1363 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1365 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1366 (ix86_attribute_table): Add indirect_branch.
1367 * config/i386/i386.h (machine_function): Add indirect_branch_type
1368 and has_local_indirect_jump.
1369 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1371 (tablejump): Likewise.
1372 (*indirect_jump): Use ix86_output_indirect_jmp.
1373 (*tablejump_1): Likewise.
1374 (simple_return_indirect_internal): Likewise.
1375 * config/i386/i386.opt (mindirect-branch=): New option.
1376 (indirect_branch): New.
1379 (thunk-inline): Likewise.
1380 (thunk-extern): Likewise.
1381 * doc/extend.texi: Document indirect_branch function attribute.
1382 * doc/invoke.texi: Document -mindirect-branch= option.
1384 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1387 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1389 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1391 * ipa-inline.c (want_inline_small_function_p): Return false if
1392 inlining has already failed with CIF_FINAL_ERROR.
1393 (update_caller_keys): Call want_inline_small_function_p before
1395 (update_callee_keys): Likewise.
1397 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1399 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1401 (rs6000_quadword_masked_address_p): Likewise.
1402 (quad_aligned_load_p): Likewise.
1403 (quad_aligned_store_p): Likewise.
1404 (const_load_sequence_p): Add comment to describe the outer-most loop.
1405 (mimic_memory_attributes_and_flags): New function.
1406 (rs6000_gen_stvx): Likewise.
1407 (replace_swapped_aligned_store): Likewise.
1408 (rs6000_gen_lvx): Likewise.
1409 (replace_swapped_aligned_load): Likewise.
1410 (replace_swapped_load_constant): Capitalize argument name in
1411 comment describing this function.
1412 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1413 and stores that access quad-word aligned addresses and replace
1414 with stvx or lvx instructions when appropriate.
1415 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1416 New function prototype.
1417 (rs6000_quadword_masked_address_p): Likewise.
1418 (rs6000_gen_lvx): Likewise.
1419 (rs6000_gen_stvx): Likewise.
1420 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1421 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1422 when memory address is aligned.
1423 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1424 this split to select lvx instruction when memory address is aligned.
1425 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1426 instruction when memory address is aligned.
1427 (*vsx_le_perm_load_v16qi): Likewise.
1428 (four unnamed splitters): Modify to select the stvx instruction
1429 when memory is aligned.
1431 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1433 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1434 which appears in the queue multiple times.
1436 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1437 Alan Hayward <alan.hayward@arm.com>
1438 David Sherwood <david.sherwood@arm.com>
1440 * tree-vectorizer.h (vec_lower_bound): New structure.
1441 (_loop_vec_info): Add check_nonzero and lower_bounds.
1442 (LOOP_VINFO_CHECK_NONZERO): New macro.
1443 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1444 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1445 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1446 fields. Make seg_len the distance travelled, not including the
1448 (dr_direction_indicator): Declare.
1449 (dr_zero_step_indicator): Likewise.
1450 (dr_known_forward_stride_p): Likewise.
1451 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1453 (runtime_alias_check_p): Allow runtime alias checks with
1455 (operator ==): Compare access_size and align.
1456 (prune_runtime_alias_test_list): Rework for new distinction between
1457 the access_size and seg_len.
1458 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1460 (get_segment_min_max): New function.
1461 (create_intersect_range_checks): Use it.
1462 (dr_step_indicator): New function.
1463 (dr_direction_indicator): Likewise.
1464 (dr_zero_step_indicator): Likewise.
1465 (dr_known_forward_stride_p): Likewise.
1466 * tree-loop-distribution.c (data_ref_segment_size): Return
1467 DR_STEP * (niters - 1).
1468 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1470 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1471 (vect_preserves_scalar_order_p): New function, split out from...
1472 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1473 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1474 (vect_vfa_access_size): New function.
1475 (vect_vfa_align): Likewise.
1476 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1477 (dump_lower_bound): New function.
1478 (vect_check_lower_bound): Likewise.
1479 (vect_small_gap_p): Likewise.
1480 (vectorizable_with_step_bound_p): Likewise.
1481 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1482 depencies if the vectorization factor is 1. Convert the checks
1483 for nonzero steps into checks on the bounds of DR_STEP. Try using
1484 a bunds check for variable steps if the minimum required step is
1485 relatively small. Update calls to the dr_with_seg_len
1486 constructor and to vect_compile_time_alias.
1487 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1489 (vect_loop_versioning): Call it.
1490 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1492 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1494 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1495 Alan Hayward <alan.hayward@arm.com>
1496 David Sherwood <david.sherwood@arm.com>
1498 * doc/sourcebuild.texi (vect_scatter_store): Document.
1499 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1501 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1503 * genopinit.c (main): Add supports_vec_scatter_store and
1504 supports_vec_scatter_store_cached to target_optabs.
1505 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1506 IFN_MASK_SCATTER_STORE.
1507 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1509 * internal-fn.h (internal_store_fn_p): Declare.
1510 (internal_fn_stored_value_index): Likewise.
1511 * internal-fn.c (scatter_store_direct): New macro.
1512 (expand_scatter_store_optab_fn): New function.
1513 (direct_scatter_store_optab_supported_p): New macro.
1514 (internal_store_fn_p): New function.
1515 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1516 IFN_MASK_SCATTER_STORE.
1517 (internal_fn_mask_index): Likewise.
1518 (internal_fn_stored_value_index): New function.
1519 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1521 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1522 * optabs-query.c (supports_vec_scatter_store_p): New function.
1523 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1524 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1525 true for scatter stores.
1526 (vect_gather_scatter_fn_p): Handle scatter stores too.
1527 (vect_check_gather_scatter): Consider using scatter stores if
1528 supports_vec_scatter_store_p.
1529 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1531 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1532 internal_fn_stored_value_index.
1533 (check_load_store_masking): Handle scatter stores too.
1534 (vect_get_store_rhs): Make public.
1535 (vectorizable_call): Use internal_store_fn_p.
1536 (vectorizable_store): Handle scatter store internal functions.
1537 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1538 when deciding whether the end of the group has been reached.
1539 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1540 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1541 (mask_scatter_store<mode>): New insns.
1543 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1544 Alan Hayward <alan.hayward@arm.com>
1545 David Sherwood <david.sherwood@arm.com>
1547 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1548 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1549 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1551 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1552 Use vect_truncate_gather_scatter_offset if we can't treat the
1553 operation as a normal gather load or scatter store.
1554 (get_group_load_store_type): Take the gather_scatter_info
1555 as argument. Try using a gather load or scatter store for
1556 single-element groups.
1557 (get_load_store_type): Update calls to get_group_load_store_type
1558 and vect_use_strided_gather_scatters_p.
1560 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1561 Alan Hayward <alan.hayward@arm.com>
1562 David Sherwood <david.sherwood@arm.com>
1564 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1565 optional tree argument.
1566 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1568 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1569 but continue to use the current value as a fallback.
1570 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1571 to compare the updates.
1572 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1573 (get_load_store_type): Use it when handling a strided access.
1574 (vect_get_strided_load_store_ops): New function.
1575 (vect_get_data_ptr_increment): Likewise.
1576 (vectorizable_load): Handle strided gather loads. Always pass
1577 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1579 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1580 Alan Hayward <alan.hayward@arm.com>
1581 David Sherwood <david.sherwood@arm.com>
1583 * doc/md.texi (gather_load@var{m}): Document.
1584 (mask_gather_load@var{m}): Likewise.
1585 * genopinit.c (main): Add supports_vec_gather_load and
1586 supports_vec_gather_load_cached to target_optabs.
1587 * optabs-tree.c (init_tree_optimization_optabs): Use
1588 ggc_cleared_alloc to allocate target_optabs.
1589 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1590 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1592 * internal-fn.h (internal_load_fn_p): Declare.
1593 (internal_gather_scatter_fn_p): Likewise.
1594 (internal_fn_mask_index): Likewise.
1595 (internal_gather_scatter_fn_supported_p): Likewise.
1596 * internal-fn.c (gather_load_direct): New macro.
1597 (expand_gather_load_optab_fn): New function.
1598 (direct_gather_load_optab_supported_p): New macro.
1599 (direct_internal_fn_optab): New function.
1600 (internal_load_fn_p): Likewise.
1601 (internal_gather_scatter_fn_p): Likewise.
1602 (internal_fn_mask_index): Likewise.
1603 (internal_gather_scatter_fn_supported_p): Likewise.
1604 * optabs-query.c (supports_at_least_one_mode_p): New function.
1605 (supports_vec_gather_load_p): Likewise.
1606 * optabs-query.h (supports_vec_gather_load_p): Declare.
1607 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1608 and memory_type field.
1609 (NUM_PATTERNS): Bump to 15.
1610 * tree-vect-data-refs.c: Include internal-fn.h.
1611 (vect_gather_scatter_fn_p): New function.
1612 (vect_describe_gather_scatter_call): Likewise.
1613 (vect_check_gather_scatter): Try using internal functions for
1614 gather loads. Recognize existing calls to a gather load function.
1615 (vect_analyze_data_refs): Consider using gather loads if
1616 supports_vec_gather_load_p.
1617 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1618 (vect_get_gather_scatter_offset_type): Likewise.
1619 (vect_convert_mask_for_vectype): Likewise.
1620 (vect_add_conversion_to_patterm): Likewise.
1621 (vect_try_gather_scatter_pattern): Likewise.
1622 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1623 (vect_vect_recog_func_ptrs): Add it.
1624 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1625 internal_fn_mask_index and internal_gather_scatter_fn_p.
1626 (check_load_store_masking): Take the gather_scatter_info as an
1627 argument and handle gather loads.
1628 (vect_get_gather_scatter_ops): New function.
1629 (vectorizable_call): Check internal_load_fn_p.
1630 (vectorizable_load): Likewise. Handle gather load internal
1632 (vectorizable_store): Update call to check_load_store_masking.
1633 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1634 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1635 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1636 (aarch64_gather_scale_operand_d): New predicates.
1637 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1638 (mask_gather_load<mode>): New insns.
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 * optabs.def (fold_left_plus_optab): New optab.
1645 * doc/md.texi (fold_left_plus_@var{m}): Document.
1646 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1647 * internal-fn.c (fold_left_direct): Define.
1648 (expand_fold_left_optab_fn): Likewise.
1649 (direct_fold_left_optab_supported_p): Likewise.
1650 * fold-const-call.c (fold_const_fold_left): New function.
1651 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1652 * tree-parloops.c (valid_reduction_p): New function.
1653 (gather_scalar_reductions): Use it.
1654 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1655 (vect_finish_replace_stmt): Declare.
1656 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1657 (needs_fold_left_reduction_p): New function, split out from...
1658 (vect_is_simple_reduction): ...here. Accept reductions that
1659 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1660 (vect_force_simple_reduction): Also store the reduction type in
1661 the assignment's STMT_VINFO_REDUC_TYPE.
1662 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1663 (merge_with_identity): New function.
1664 (vect_expand_fold_left): Likewise.
1665 (vectorize_fold_left_reduction): Likewise.
1666 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1667 scalar phi in place for it. Check for target support and reject
1668 cases that would reassociate the operation. Defer the transform
1669 phase to vectorize_fold_left_reduction.
1670 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1671 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1672 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1674 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1676 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1677 call to ifc_temp_var.
1679 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1680 Alan Hayward <alan.hayward@arm.com>
1681 David Sherwood <david.sherwood@arm.com>
1683 * target.def (legitimize_address_displacement): Take the original
1684 offset as a poly_int.
1685 * targhooks.h (default_legitimize_address_displacement): Update
1687 * targhooks.c (default_legitimize_address_displacement): Likewise.
1688 * doc/tm.texi: Regenerate.
1689 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1690 as an argument, moving assert of ad->disp == ad->disp_term to...
1691 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1692 Try calling targetm.legitimize_address_displacement before expanding
1693 the address rather than afterwards, and adjust for the new interface.
1694 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1695 Match the new hook interface. Handle SVE addresses.
1696 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1699 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1701 * Makefile.in (OBJS): Add early-remat.o.
1702 * target.def (select_early_remat_modes): New hook.
1703 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1704 * doc/tm.texi: Regenerate.
1705 * targhooks.h (default_select_early_remat_modes): Declare.
1706 * targhooks.c (default_select_early_remat_modes): New function.
1707 * timevar.def (TV_EARLY_REMAT): New timevar.
1708 * passes.def (pass_early_remat): New pass.
1709 * tree-pass.h (make_pass_early_remat): Declare.
1710 * early-remat.c: New file.
1711 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1713 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1715 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1716 Alan Hayward <alan.hayward@arm.com>
1717 David Sherwood <david.sherwood@arm.com>
1719 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1720 vfm1 with a bound_epilog parameter.
1721 (vect_do_peeling): Update calls accordingly, and move the prologue
1722 call earlier in the function. Treat the base bound_epilog as 0 for
1723 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1724 this base when peeling for gaps.
1725 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1726 with fully-masked loops.
1727 (vect_estimate_min_profitable_iters): Handle the single peeled
1728 iteration in that case.
1730 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1731 Alan Hayward <alan.hayward@arm.com>
1732 David Sherwood <david.sherwood@arm.com>
1734 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1735 single-element interleaving even if the size is not a power of 2.
1736 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1737 accesses for single-element interleaving if the group size is
1740 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1741 Alan Hayward <alan.hayward@arm.com>
1742 David Sherwood <david.sherwood@arm.com>
1744 * doc/md.texi (fold_extract_last_@var{m}): Document.
1745 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1746 * optabs.def (fold_extract_last_optab): New optab.
1747 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1748 * internal-fn.c (fold_extract_direct): New macro.
1749 (expand_fold_extract_optab_fn): Likewise.
1750 (direct_fold_extract_optab_supported_p): Likewise.
1751 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1752 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1753 EXTRACT_LAST_REDUCTION.
1754 (get_initial_def_for_reduction): Do not create an initial vector
1755 for EXTRACT_LAST_REDUCTION reductions.
1756 (vectorizable_reduction): Leave the scalar phi in place for
1757 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1758 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1759 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1760 transform phase to vectorizable_condition.
1761 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1763 (vect_finish_stmt_generation): ...here.
1764 (vect_finish_replace_stmt): New function.
1765 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1766 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1768 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1770 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1771 Alan Hayward <alan.hayward@arm.com>
1772 David Sherwood <david.sherwood@arm.com>
1774 * doc/md.texi (extract_last_@var{m}): Document.
1775 * optabs.def (extract_last_optab): New optab.
1776 * internal-fn.def (EXTRACT_LAST): New internal function.
1777 * internal-fn.c (cond_unary_direct): New macro.
1778 (expand_cond_unary_optab_fn): Likewise.
1779 (direct_cond_unary_optab_supported_p): Likewise.
1780 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1781 loops using EXTRACT_LAST.
1782 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1783 (extract_last_<mode>): ...this optab.
1784 (vec_extract<mode><Vel>): Update accordingly.
1786 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1787 Alan Hayward <alan.hayward@arm.com>
1788 David Sherwood <david.sherwood@arm.com>
1790 * target.def (empty_mask_is_expensive): New hook.
1791 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
1792 * doc/tm.texi: Regenerate.
1793 * targhooks.h (default_empty_mask_is_expensive): Declare.
1794 * targhooks.c (default_empty_mask_is_expensive): New function.
1795 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
1796 if the target says that empty masks are expensive.
1797 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
1799 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
1801 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1802 Alan Hayward <alan.hayward@arm.com>
1803 David Sherwood <david.sherwood@arm.com>
1805 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
1806 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
1807 (vect_use_loop_mask_for_alignment_p): New function.
1808 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
1809 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
1810 niters_skip argument. Make sure that the first niters_skip elements
1811 of the first iteration are inactive.
1812 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
1813 Update call to vect_set_loop_masks_directly.
1814 (get_misalign_in_elems): New function, split out from...
1815 (vect_gen_prolog_loop_niters): ...here.
1816 (vect_update_init_of_dr): Take a code argument that specifies whether
1817 the adjustment should be added or subtracted.
1818 (vect_update_init_of_drs): Likewise.
1819 (vect_prepare_for_masked_peels): New function.
1820 (vect_do_peeling): Skip prologue peeling if we're using a mask
1821 instead. Update call to vect_update_inits_of_drs.
1822 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1824 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
1825 alignment. Do not include the number of peeled iterations in
1826 the minimum threshold in that case.
1827 (vectorizable_induction): Adjust the start value down by
1828 LOOP_VINFO_MASK_SKIP_NITERS iterations.
1829 (vect_transform_loop): Call vect_prepare_for_masked_peels.
1830 Take the number of skipped iterations into account when calculating
1832 * tree-vect-stmts.c (vect_gen_while_not): New function.
1834 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1835 Alan Hayward <alan.hayward@arm.com>
1836 David Sherwood <david.sherwood@arm.com>
1838 * doc/sourcebuild.texi (vect_fully_masked): Document.
1839 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
1841 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
1843 (vect_analyze_loop_2): ...here. Don't check the vectorization
1844 factor against the number of loop iterations if the loop is
1847 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1848 Alan Hayward <alan.hayward@arm.com>
1849 David Sherwood <david.sherwood@arm.com>
1851 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
1852 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
1853 (dump_groups): Update accordingly.
1854 (iv_use::mem_type): New member variable.
1855 (address_p): New function.
1856 (record_use): Add a mem_type argument and initialize the new
1858 (record_group_use): Add a mem_type argument. Use address_p.
1859 Remove obsolete null checks of base_object. Update call to record_use.
1860 (find_interesting_uses_op): Update call to record_group_use.
1861 (find_interesting_uses_cond): Likewise.
1862 (find_interesting_uses_address): Likewise.
1863 (get_mem_type_for_internal_fn): New function.
1864 (find_address_like_use): Likewise.
1865 (find_interesting_uses_stmt): Try find_address_like_use before
1866 calling find_interesting_uses_op.
1867 (addr_offset_valid_p): Use the iv mem_type field as the type
1868 of the addressed memory.
1869 (add_autoinc_candidates): Likewise.
1870 (get_address_cost): Likewise.
1871 (split_small_address_groups_p): Use address_p.
1872 (split_address_groups): Likewise.
1873 (add_iv_candidate_for_use): Likewise.
1874 (autoinc_possible_for_pair): Likewise.
1875 (rewrite_groups): Likewise.
1876 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
1877 (determine_group_iv_cost): Update after split of USE_ADDRESS.
1878 (get_alias_ptr_type_for_ptr_address): New function.
1879 (rewrite_use_address): Rewrite address uses in calls that were
1880 identified by find_address_like_use.
1882 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1883 Alan Hayward <alan.hayward@arm.com>
1884 David Sherwood <david.sherwood@arm.com>
1886 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
1888 * gimple-expr.h (is_gimple_addressable: Likewise.
1889 * gimple-expr.c (is_gimple_address): Likewise.
1890 * internal-fn.c (expand_call_mem_ref): New function.
1891 (expand_mask_load_optab_fn): Use it.
1892 (expand_mask_store_optab_fn): Likewise.
1894 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1895 Alan Hayward <alan.hayward@arm.com>
1896 David Sherwood <david.sherwood@arm.com>
1898 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
1899 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
1900 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
1901 (cond_umax@var{mode}): Document.
1902 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
1903 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
1904 (cond_umin_optab, cond_umax_optab): New optabs.
1905 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
1906 (COND_IOR, COND_XOR): New internal functions.
1907 * internal-fn.h (get_conditional_internal_fn): Declare.
1908 * internal-fn.c (cond_binary_direct): New macro.
1909 (expand_cond_binary_optab_fn): Likewise.
1910 (direct_cond_binary_optab_supported_p): Likewise.
1911 (get_conditional_internal_fn): New function.
1912 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
1913 Cope with reduction statements that are vectorized as calls rather
1915 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
1916 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
1917 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
1918 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
1919 (UNSPEC_COND_EOR): New unspecs.
1920 (optab): Add mappings for them.
1921 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
1922 (sve_int_op, sve_fp_op): New int attributes.
1924 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1925 Alan Hayward <alan.hayward@arm.com>
1926 David Sherwood <david.sherwood@arm.com>
1928 * optabs.def (while_ult_optab): New optab.
1929 * doc/md.texi (while_ult@var{m}@var{n}): Document.
1930 * internal-fn.def (WHILE_ULT): New internal function.
1931 * internal-fn.h (direct_internal_fn_supported_p): New override
1932 that takes two types as argument.
1933 * internal-fn.c (while_direct): New macro.
1934 (expand_while_optab_fn): New function.
1935 (convert_optab_supported_p): Likewise.
1936 (direct_while_optab_supported_p): New macro.
1937 * wide-int.h (wi::udiv_ceil): New function.
1938 * tree-vectorizer.h (rgroup_masks): New structure.
1939 (vec_loop_masks): New typedef.
1940 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
1942 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
1943 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
1944 (vect_max_vf): New function.
1945 (slpeel_make_loop_iterate_ntimes): Delete.
1946 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
1947 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
1948 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
1949 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
1950 internal-fn.h, stor-layout.h and optabs-query.h.
1951 (vect_set_loop_mask): New function.
1952 (add_preheader_seq): Likewise.
1953 (add_header_seq): Likewise.
1954 (interleave_supported_p): Likewise.
1955 (vect_maybe_permute_loop_masks): Likewise.
1956 (vect_set_loop_masks_directly): Likewise.
1957 (vect_set_loop_condition_masked): Likewise.
1958 (vect_set_loop_condition_unmasked): New function, split out from
1959 slpeel_make_loop_iterate_ntimes.
1960 (slpeel_make_loop_iterate_ntimes): Rename to..
1961 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
1962 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
1963 (vect_do_peeling): Update call accordingly.
1964 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
1966 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1967 mask_compare_type, can_fully_mask_p and fully_masked_p.
1968 (release_vec_loop_masks): New function.
1969 (_loop_vec_info): Use it to free the loop masks.
1970 (can_produce_all_loop_masks_p): New function.
1971 (vect_get_max_nscalars_per_iter): Likewise.
1972 (vect_verify_full_masking): Likewise.
1973 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
1974 retries, and free the mask rgroups before retrying. Check loop-wide
1975 reasons for disallowing fully-masked loops. Make the final decision
1976 about whether use a fully-masked loop or not.
1977 (vect_estimate_min_profitable_iters): Do not assume that peeling
1978 for the number of iterations will be needed for fully-masked loops.
1979 (vectorizable_reduction): Disable fully-masked loops.
1980 (vectorizable_live_operation): Likewise.
1981 (vect_halve_mask_nunits): New function.
1982 (vect_double_mask_nunits): Likewise.
1983 (vect_record_loop_mask): Likewise.
1984 (vect_get_loop_mask): Likewise.
1985 (vect_transform_loop): Handle the case in which the final loop
1986 iteration might handle a partial vector. Call vect_set_loop_condition
1987 instead of slpeel_make_loop_iterate_ntimes.
1988 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
1989 (check_load_store_masking): New function.
1990 (prepare_load_store_mask): Likewise.
1991 (vectorizable_store): Handle fully-masked loops.
1992 (vectorizable_load): Likewise.
1993 (supportable_widening_operation): Use vect_halve_mask_nunits for
1995 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
1996 (vect_gen_while): New function.
1997 * config/aarch64/aarch64.md (umax<mode>3): New expander.
1998 (aarch64_uqdec<mode>): New insn.
2000 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2001 Alan Hayward <alan.hayward@arm.com>
2002 David Sherwood <david.sherwood@arm.com>
2004 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2005 (reduc_xor_scal_optab): New optabs.
2006 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2007 (reduc_xor_scal_@var{m}): Document.
2008 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2009 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2011 * fold-const-call.c (fold_const_call): Handle them.
2012 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2013 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2014 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2015 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2016 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2017 (UNSPEC_XORV): New unspecs.
2018 (optab): Add entries for them.
2019 (BITWISEV): New int iterator.
2020 (bit_reduc_op): New int attributes.
2022 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2023 Alan Hayward <alan.hayward@arm.com>
2024 David Sherwood <david.sherwood@arm.com>
2026 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2027 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2028 * optabs.def (vec_shl_insert_optab): New optab.
2029 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2030 (duplicate_and_interleave): Likewise.
2031 * tree-vect-loop.c: Include internal-fn.h.
2032 (neutral_op_for_slp_reduction): New function, split out from
2033 get_initial_defs_for_reduction.
2034 (get_initial_def_for_reduction): Handle option 2 for variable-length
2035 vectors by loading the neutral value into a vector and then shifting
2036 the initial value into element 0.
2037 (get_initial_defs_for_reduction): Replace the code argument with
2038 the neutral value calculated by neutral_op_for_slp_reduction.
2039 Use gimple_build_vector for constant-length vectors.
2040 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2041 but the first group_size elements have a neutral value.
2042 Use duplicate_and_interleave otherwise.
2043 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2044 Update call to get_initial_defs_for_reduction. Handle SLP
2045 reductions for variable-length vectors by creating one vector
2046 result for each scalar result, with the elements associated
2047 with other scalar results stubbed out with the neutral value.
2048 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2049 Require IFN_VEC_SHL_INSERT for double reductions on
2050 variable-length vectors, or SLP reductions that have
2051 a neutral value. Require can_duplicate_and_interleave_p
2052 support for variable-length unchained SLP reductions if there
2053 is no neutral value, such as for MIN/MAX reductions. Also require
2054 the number of vector elements to be a multiple of the number of
2055 SLP statements when doing variable-length unchained SLP reductions.
2056 Update call to vect_create_epilog_for_reduction.
2057 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2058 and remove initial values.
2059 (duplicate_and_interleave): Make public.
2060 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2061 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2063 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2064 Alan Hayward <alan.hayward@arm.com>
2065 David Sherwood <david.sherwood@arm.com>
2067 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2068 (can_duplicate_and_interleave_p): New function.
2069 (vect_get_and_check_slp_defs): Take the vector of statements
2070 rather than just the current one. Remove excess parentheses.
2071 Restriction rejectinon of vect_constant_def and vect_external_def
2072 for variable-length vectors to boolean types, or types for which
2073 can_duplicate_and_interleave_p is false.
2074 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2075 (duplicate_and_interleave): New function.
2076 (vect_get_constant_vectors): Use gimple_build_vector for
2077 constant-length vectors and suitable variable-length constant
2078 vectors. Use duplicate_and_interleave for other variable-length
2079 vectors. Don't defer the update when inserting new statements.
2081 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2082 Alan Hayward <alan.hayward@arm.com>
2083 David Sherwood <david.sherwood@arm.com>
2085 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2086 min_profitable_iters doesn't go negative.
2088 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2089 Alan Hayward <alan.hayward@arm.com>
2090 David Sherwood <david.sherwood@arm.com>
2092 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2093 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2094 * optabs.def (vec_mask_load_lanes_optab): New optab.
2095 (vec_mask_store_lanes_optab): Likewise.
2096 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2097 (MASK_STORE_LANES): Likewise.
2098 * internal-fn.c (mask_load_lanes_direct): New macro.
2099 (mask_store_lanes_direct): Likewise.
2100 (expand_mask_load_optab_fn): Handle masked operations.
2101 (expand_mask_load_lanes_optab_fn): New macro.
2102 (expand_mask_store_optab_fn): Handle masked operations.
2103 (expand_mask_store_lanes_optab_fn): New macro.
2104 (direct_mask_load_lanes_optab_supported_p): Likewise.
2105 (direct_mask_store_lanes_optab_supported_p): Likewise.
2106 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2108 (vect_load_lanes_supported): Likewise.
2109 * tree-vect-data-refs.c (strip_conversion): New function.
2110 (can_group_stmts_p): Likewise.
2111 (vect_analyze_data_ref_accesses): Use it instead of checking
2112 for a pair of assignments.
2113 (vect_store_lanes_supported): Take a masked_p parameter.
2114 (vect_load_lanes_supported): Likewise.
2115 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2116 vect_store_lanes_supported and vect_load_lanes_supported.
2117 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2118 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2119 parameter. Don't allow gaps for masked accesses.
2120 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2121 and vect_load_lanes_supported.
2122 (get_load_store_type): Take a masked_p parameter and update
2123 call to get_group_load_store_type.
2124 (vectorizable_store): Update call to get_load_store_type.
2125 Handle IFN_MASK_STORE_LANES.
2126 (vectorizable_load): Update call to get_load_store_type.
2127 Handle IFN_MASK_LOAD_LANES.
2129 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2130 Alan Hayward <alan.hayward@arm.com>
2131 David Sherwood <david.sherwood@arm.com>
2133 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2135 * config/aarch64/aarch64-protos.h
2136 (aarch64_sve_struct_memory_operand_p): Declare.
2137 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2138 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2139 (VPRED, vpred): Handle SVE structure modes.
2140 * config/aarch64/constraints.md (Utx): New constraint.
2141 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2142 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2143 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2144 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2145 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2146 structure modes. Split into pieces after RA.
2147 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2148 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2150 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2151 SVE structure modes.
2152 (aarch64_classify_address): Likewise.
2153 (sizetochar): Move earlier in file.
2154 (aarch64_print_operand): Handle SVE register lists.
2155 (aarch64_array_mode): New function.
2156 (aarch64_sve_struct_memory_operand_p): Likewise.
2157 (TARGET_ARRAY_MODE): Redefine.
2159 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2160 Alan Hayward <alan.hayward@arm.com>
2161 David Sherwood <david.sherwood@arm.com>
2163 * target.def (array_mode): New target hook.
2164 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2165 * doc/tm.texi: Regenerate.
2166 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2167 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2168 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2170 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2173 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2174 Alan Hayward <alan.hayward@arm.com>
2175 David Sherwood <david.sherwood@arm.com>
2177 * fold-const.c (fold_binary_loc): Check the argument types
2178 rather than the result type when testing for a vector operation.
2180 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2182 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2183 * doc/tm.texi: Regenerate.
2185 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2186 Alan Hayward <alan.hayward@arm.com>
2187 David Sherwood <david.sherwood@arm.com>
2189 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2190 (sve): Document new AArch64 extension.
2191 * doc/md.texi (w): Extend the description of the AArch64
2192 constraint to include SVE vectors.
2193 (Upl, Upa): Document new AArch64 predicate constraints.
2194 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2196 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2197 (msve-vector-bits=): New option.
2198 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2199 SVE when these are disabled.
2200 (sve): New extension.
2201 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2202 modes. Adjust their number of units based on aarch64_sve_vg.
2203 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2204 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2205 aarch64_addr_query_type.
2206 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2207 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2208 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2209 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2210 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2211 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2212 (aarch64_simd_imm_zero_p): Delete.
2213 (aarch64_check_zero_based_sve_index_immediate): Declare.
2214 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2215 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2216 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2217 (aarch64_sve_float_mul_immediate_p): Likewise.
2218 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2220 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2221 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2222 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2223 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2224 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2225 (aarch64_regmode_natural_size): Likewise.
2226 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2227 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2229 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2230 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2231 for VG and the SVE predicate registers.
2232 (V_ALIASES): Add a "z"-prefixed alias.
2233 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2234 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2235 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2236 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2237 (REG_CLASS_NAMES): Add entries for them.
2238 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2239 and the predicate registers.
2240 (aarch64_sve_vg): Declare.
2241 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2242 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2243 (REGMODE_NATURAL_SIZE): Define.
2244 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2246 * config/aarch64/aarch64.c: Include cfgrtl.h.
2247 (simd_immediate_info): Add a constructor for series vectors,
2248 and an associated step field.
2249 (aarch64_sve_vg): New variable.
2250 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2251 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2252 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2253 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2254 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2255 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2256 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2257 (aarch64_get_mask_mode): New functions.
2258 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2259 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2260 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2261 predicate modes and predicate registers. Explicitly restrict
2262 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2263 to store a vector mode if it is recognized by
2264 aarch64_classify_vector_mode.
2265 (aarch64_regmode_natural_size): New function.
2266 (aarch64_hard_regno_caller_save_mode): Return the original mode
2268 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2269 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2270 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2271 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2273 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2274 does not overlap dest if the function is frame-related. Handle
2276 (aarch64_split_add_offset): New function.
2277 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2278 them aarch64_add_offset.
2279 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2280 and update call to aarch64_sub_sp.
2281 (aarch64_add_cfa_expression): New function.
2282 (aarch64_expand_prologue): Pass extra temporary registers to the
2283 functions above. Handle the case in which we need to emit new
2284 DW_CFA_expressions for registers that were originally saved
2285 relative to the stack pointer, but now have to be expressed
2286 relative to the frame pointer.
2287 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2289 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2290 IP0 and IP1 values for SVE frames.
2291 (aarch64_expand_vec_series): New function.
2292 (aarch64_expand_sve_widened_duplicate): Likewise.
2293 (aarch64_expand_sve_const_vector): Likewise.
2294 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2295 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2296 into the register, rather than emitting a SET directly.
2297 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2298 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2299 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2300 (offset_9bit_signed_scaled_p): New functions.
2301 (aarch64_replicate_bitmask_imm): New function.
2302 (aarch64_bitmask_imm): Use it.
2303 (aarch64_cannot_force_const_mem): Reject expressions involving
2304 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2305 (aarch64_classify_index): Handle SVE indices, by requiring
2306 a plain register index with a scale that matches the element size.
2307 (aarch64_classify_address): Handle SVE addresses. Assert that
2308 the mode of the address is VOIDmode or an integer mode.
2309 Update call to aarch64_classify_symbol.
2310 (aarch64_classify_symbolic_expression): Update call to
2311 aarch64_classify_symbol.
2312 (aarch64_const_vec_all_in_range_p): New function.
2313 (aarch64_print_vector_float_operand): Likewise.
2314 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2315 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2316 and the FP immediates 1.0 and 0.5.
2317 (aarch64_print_address_internal): Handle SVE addresses.
2318 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2319 (aarch64_regno_regclass): Handle predicate registers.
2320 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2322 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2323 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2324 (aarch64_convert_sve_vector_bits): New function.
2325 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2326 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2328 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2329 Handle SVE vector and predicate modes. Accept VL-based constants
2330 that need only one temporary register, and VL offsets that require
2331 no temporary registers.
2332 (aarch64_conditional_register_usage): Mark the predicate registers
2333 as fixed if SVE isn't available.
2334 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2335 Return true for SVE vector and predicate modes.
2336 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2337 rather than an unsigned int. Handle SVE modes.
2338 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2340 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2342 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2343 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2344 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2345 (aarch64_sve_float_mul_immediate_p): New functions.
2346 (aarch64_sve_valid_immediate): New function.
2347 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2348 Explicitly reject structure modes. Check for INDEX constants.
2349 Handle PTRUE and PFALSE constants.
2350 (aarch64_check_zero_based_sve_index_immediate): New function.
2351 (aarch64_simd_imm_zero_p): Delete.
2352 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2353 vector modes. Accept constants in the range of CNT[BHWD].
2354 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2355 ask for an Advanced SIMD mode.
2356 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2357 (aarch64_simd_vector_alignment): Handle SVE predicates.
2358 (aarch64_vectorize_preferred_vector_alignment): New function.
2359 (aarch64_simd_vector_alignment_reachable): Use it instead of
2361 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2362 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2364 (MAX_VECT_LEN): Delete.
2365 (expand_vec_perm_d): Add a vec_flags field.
2366 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2367 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2368 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2370 (aarch64_evpc_rev): Rename to...
2371 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2372 (aarch64_evpc_rev_global): New function.
2373 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2374 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2376 (aarch64_evpc_sve_tbl): New function.
2377 (aarch64_expand_vec_perm_const_1): Update after rename of
2378 aarch64_evpc_rev. Handle SVE permutes too, trying
2379 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2380 than aarch64_evpc_tbl.
2381 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2382 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2383 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2384 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2385 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2386 (aarch64_expand_sve_vcond): New functions.
2387 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2388 of aarch64_vector_mode_p.
2389 (aarch64_dwarf_poly_indeterminate_value): New function.
2390 (aarch64_compute_pressure_classes): Likewise.
2391 (aarch64_can_change_mode_class): Likewise.
2392 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2393 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2394 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2395 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2396 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2397 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2398 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2399 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2401 (Dn, Dl, Dr): Accept const as well as const_vector.
2402 (Dz): Likewise. Compare against CONST0_RTX.
2403 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2404 of "vector" where appropriate.
2405 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2406 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2407 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2408 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2409 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2410 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2411 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2412 (v_int_equiv): Extend to SVE modes.
2413 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2415 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2416 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2417 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2418 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2419 (SVE_COND_FP_CMP): New int iterators.
2420 (perm_hilo): Handle the new unpack unspecs.
2421 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2423 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2424 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2425 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2426 (aarch64_equality_operator, aarch64_constant_vector_operand)
2427 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2428 (aarch64_sve_nonimmediate_operand): Likewise.
2429 (aarch64_sve_general_operand): Likewise.
2430 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2431 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2432 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2433 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2434 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2435 (aarch64_sve_float_arith_immediate): Likewise.
2436 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2437 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2438 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2439 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2440 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2441 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2442 (aarch64_sve_float_arith_operand): Likewise.
2443 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2444 (aarch64_sve_float_mul_operand): Likewise.
2445 (aarch64_sve_vec_perm_operand): Likewise.
2446 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2447 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2448 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2449 as well as const_vector.
2450 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2451 in file. Use CONST0_RTX and CONSTM1_RTX.
2452 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2453 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2454 Use aarch64_simd_imm_zero.
2455 * config/aarch64/aarch64-sve.md: New file.
2456 * config/aarch64/aarch64.md: Include it.
2457 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2458 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2459 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2460 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2461 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2462 (sve): New attribute.
2463 (enabled): Disable instructions with the sve attribute unless
2465 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2466 aarch64_expand_mov_immediate.
2467 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2468 CNT[BHSD] immediates.
2469 (movti): Split CONST_POLY_INT moves into two halves.
2470 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2471 Split additions that need a temporary here if the destination
2472 is the stack pointer.
2473 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2474 (*add<mode>3_poly_1): New instruction.
2475 (set_clobber_cc): New expander.
2477 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2479 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2480 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2481 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2482 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2483 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2484 Change innermode from fixed_mode_size to machine_mode.
2485 (simplify_subreg): Update call accordingly. Handle a constant-sized
2486 subreg of a variable-length CONST_VECTOR.
2488 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2489 Alan Hayward <alan.hayward@arm.com>
2490 David Sherwood <david.sherwood@arm.com>
2492 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2493 (add_offset_to_base): New function, split out from...
2494 (create_mem_ref): ...here. When handling a scale other than 1,
2495 check first whether the address is valid without the offset.
2496 Add it into the base if so, leaving the index and scale as-is.
2498 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2501 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2502 fold_for_warn before checking if arg2 is INTEGER_CST.
2504 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2506 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2507 (store_multiple_operation): Delete.
2508 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2509 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2510 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2511 guarded by TARGET_STRING.
2512 (rs6000_output_load_multiple): Delete.
2513 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2514 OPTION_MASK_STRING / TARGET_STRING handling.
2515 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2516 (const rs6000_opt_masks) <"string">: Change mask to 0.
2517 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2518 (MASK_STRING): Delete.
2519 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2521 (load_multiple): Delete.
2528 (store_multiple): Delete.
2535 (movmemsi_8reg): Delete.
2536 (corresponding unnamed define_insn): Delete.
2537 (movmemsi_6reg): Delete.
2538 (corresponding unnamed define_insn): Delete.
2539 (movmemsi_4reg): Delete.
2540 (corresponding unnamed define_insn): Delete.
2541 (movmemsi_2reg): Delete.
2542 (corresponding unnamed define_insn): Delete.
2543 (movmemsi_1reg): Delete.
2544 (corresponding unnamed define_insn): Delete.
2545 * config/rs6000/rs6000.opt (mno-string): New.
2546 (mstring): Replace by deprecation warning stub.
2547 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2549 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2551 * regrename.c (regrename_do_replace): If replacing the same
2552 reg multiple times, try to reuse last created gen_raw_REG.
2555 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2556 main to workaround a bug in GDB.
2558 2018-01-12 Tom de Vries <tom@codesourcery.com>
2561 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2563 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2565 PR rtl-optimization/80481
2566 * ira-color.c (get_cap_member): New function.
2567 (allocnos_conflict_by_live_ranges_p): Use it.
2568 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2569 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2571 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2574 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2575 (*saddl_se_1): Ditto.
2577 (*ssubl_se_1): Ditto.
2579 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2581 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2582 rather than wi::to_widest for DR_INITs.
2583 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2584 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2585 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2587 (vect_analyze_group_access_1): Note that here.
2589 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2591 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2592 polynomial type sizes.
2594 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2596 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2597 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2598 (gimple_add_tmp_var): Likewise.
2600 2018-01-12 Martin Liska <mliska@suse.cz>
2602 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2603 (gimple_alloc_sizes): Likewise.
2604 (dump_gimple_statistics): Use PRIu64 in printf format.
2605 * gimple.h: Change uint64_t to int.
2607 2018-01-12 Martin Liska <mliska@suse.cz>
2609 * tree-core.h: Use uint64_t instead of int.
2610 * tree.c (tree_node_counts): Likewise.
2611 (tree_node_sizes): Likewise.
2612 (dump_tree_statistics): Use PRIu64 in printf format.
2614 2018-01-12 Martin Liska <mliska@suse.cz>
2616 * Makefile.in: As qsort_chk is implemented in vec.c, add
2617 vec.o to linkage of gencfn-macros.
2618 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2619 passing the info to record_node_allocation_statistics.
2620 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2622 * ggc-common.c (struct ggc_usage): Add operator== and use
2623 it in operator< and compare function.
2624 * mem-stats.h (struct mem_usage): Likewise.
2625 * vec.c (struct vec_usage): Remove operator< and compare
2626 function. Can be simply inherited.
2628 2018-01-12 Martin Jambor <mjambor@suse.cz>
2631 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2632 * tree-ssa-math-opts.c: Include domwalk.h.
2633 (convert_mult_to_fma_1): New function.
2634 (fma_transformation_info): New type.
2635 (fma_deferring_state): Likewise.
2636 (cancel_fma_deferring): New function.
2637 (result_of_phi): Likewise.
2638 (last_fma_candidate_feeds_initial_phi): Likewise.
2639 (convert_mult_to_fma): Added deferring logic, split actual
2640 transformation to convert_mult_to_fma_1.
2641 (math_opts_dom_walker): New type.
2642 (math_opts_dom_walker::after_dom_children): New method, body moved
2643 here from pass_optimize_widening_mul::execute, added deferring logic
2645 (pass_optimize_widening_mul::execute): Moved most of code to
2646 math_opts_dom_walker::after_dom_children.
2647 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2648 * config/i386/i386.c (ix86_option_override_internal): Added
2649 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2651 2018-01-12 Richard Biener <rguenther@suse.de>
2654 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2655 inline instance vars.
2657 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2660 * config/rx/rx.c (rx_is_restricted_memory_address):
2663 2018-01-12 Richard Biener <rguenther@suse.de>
2665 PR tree-optimization/80846
2666 * target.def (split_reduction): New target hook.
2667 * targhooks.c (default_split_reduction): New function.
2668 * targhooks.h (default_split_reduction): Declare.
2669 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2670 target requests first reduce vectors by combining low and high
2672 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2673 (get_vectype_for_scalar_type_and_size): Export.
2674 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2675 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2676 * doc/tm.texi: Regenerate.
2677 * config/i386/i386.c (ix86_split_reduction): Implement
2678 TARGET_VECTORIZE_SPLIT_REDUCTION.
2680 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2683 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2684 in PIC mode except for TARGET_VXWORKS_RTP.
2685 * config/sparc/sparc.c: Include cfgrtl.h.
2686 (TARGET_INIT_PIC_REG): Define.
2687 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2688 (sparc_pic_register_p): New predicate.
2689 (sparc_legitimate_address_p): Use it.
2690 (sparc_legitimize_pic_address): Likewise.
2691 (sparc_delegitimize_address): Likewise.
2692 (sparc_mode_dependent_address_p): Likewise.
2693 (gen_load_pcrel_sym): Remove 4th parameter.
2694 (load_got_register): Adjust call to above. Remove obsolete stuff.
2695 (sparc_expand_prologue): Do not call load_got_register here.
2696 (sparc_flat_expand_prologue): Likewise.
2697 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2698 (sparc_use_pseudo_pic_reg): New function.
2699 (sparc_init_pic_reg): Likewise.
2700 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2701 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2703 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2705 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2706 Add item for branch_cost.
2708 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2710 PR rtl-optimization/83565
2711 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2712 not extend the result to a larger mode for rotate operations.
2713 (num_sign_bit_copies1): Likewise.
2715 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2718 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2720 Use values-Xc.o for -pedantic.
2721 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2723 2018-01-12 Martin Liska <mliska@suse.cz>
2726 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2728 (possible_polymorphic_call_targets): Use it.
2729 (ipa_devirt): Likewise.
2731 2018-01-12 Martin Liska <mliska@suse.cz>
2733 * profile-count.h (enum profile_quality): Use 0 as invalid
2734 enum value of profile_quality.
2736 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2738 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2739 -mext-string options.
2741 2018-01-12 Richard Biener <rguenther@suse.de>
2743 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2744 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2745 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2747 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2749 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2751 * configure.ac (--with-long-double-format): Add support for the
2752 configuration option to change the default long double format on
2754 * config.gcc (powerpc*-linux*-*): Likewise.
2755 * configure: Regenerate.
2756 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2757 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2758 used without modification.
2760 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2762 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2763 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2764 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2765 MISC_BUILTIN_SPEC_BARRIER.
2766 (rs6000_init_builtins): Likewise.
2767 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2769 (speculation_barrier): New define_insn.
2770 * doc/extend.texi: Document __builtin_speculation_barrier.
2772 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2775 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2776 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2777 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2779 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2780 integral modes instead of "ss" and "sd".
2781 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2782 vectors with 32-bit and 64-bit elements.
2783 (vecdupssescalarmodesuffix): New mode attribute.
2784 (vec_dup<mode>): Use it.
2786 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2789 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2790 frame if argument is passed on stack.
2792 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2795 * ree.c (combine_reaching_defs): Optimize also
2796 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
2797 reg2=any_extend(exp); reg1=reg2;, formatting fix.
2799 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2802 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
2804 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2807 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
2808 after they are computed.
2810 2018-01-11 Bin Cheng <bin.cheng@arm.com>
2812 PR tree-optimization/83695
2813 * gimple-loop-linterchange.cc
2814 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
2815 reset cached scev information after interchange.
2816 (pass_linterchange::execute): Remove call to scev_reset_htab.
2818 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2820 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
2821 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
2822 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
2823 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
2824 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
2825 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
2826 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
2827 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
2828 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
2829 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
2830 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
2831 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
2832 (V_lane_reg): Likewise.
2833 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
2835 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
2836 (vfmal_lane_low<mode>_intrinsic,
2837 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
2838 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
2839 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
2840 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
2841 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
2842 vfmsl_lane_high<mode>_intrinsic): New define_insns.
2844 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2846 * config/arm/arm-cpus.in (fp16fml): New feature.
2847 (ALL_SIMD): Add fp16fml.
2848 (armv8.2-a): Add fp16fml as an option.
2849 (armv8.3-a): Likewise.
2850 (armv8.4-a): Add fp16fml as part of fp16.
2851 * config/arm/arm.h (TARGET_FP16FML): Define.
2852 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
2854 * config/arm/arm-modes.def (V2HF): Define.
2855 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
2856 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
2857 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
2858 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
2859 vfmsl_low, vfmsl_high): New set of builtins.
2860 * config/arm/iterators.md (PLUSMINUS): New code iterator.
2861 (vfml_op): New code attribute.
2862 (VFMLHALVES): New int iterator.
2863 (VFML, VFMLSEL): New mode attributes.
2864 (V_reg): Define mapping for V2HF.
2865 (V_hi, V_lo): New mode attributes.
2866 (VF_constraint): Likewise.
2867 (vfml_half, vfml_half_selector): New int attributes.
2868 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
2870 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
2871 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
2873 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
2874 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
2875 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
2876 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
2878 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
2879 Document new effective target and option set.
2881 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2883 * config/arm/arm-cpus.in (armv8_4): New feature.
2884 (ARMv8_4a): New fgroup.
2885 (armv8.4-a): New arch.
2886 * config/arm/arm-tables.opt: Regenerate.
2887 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
2888 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
2889 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
2890 Add matching rules for -march=armv8.4-a and extensions.
2891 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
2893 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
2896 * config/rx/rx.md (BW): New mode attribute.
2897 (sync_lock_test_and_setsi): Add mode suffix to insn output.
2899 2018-01-11 Richard Biener <rguenther@suse.de>
2901 PR tree-optimization/83435
2902 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
2903 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
2904 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
2906 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2907 Alan Hayward <alan.hayward@arm.com>
2908 David Sherwood <david.sherwood@arm.com>
2910 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
2912 (aarch64_classify_address): Initialize it. Track polynomial offsets.
2913 (aarch64_print_address_internal): Use it to check for a zero offset.
2915 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2916 Alan Hayward <alan.hayward@arm.com>
2917 David Sherwood <david.sherwood@arm.com>
2919 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
2920 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
2921 Return a poly_int64 rather than a HOST_WIDE_INT.
2922 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
2923 rather than a HOST_WIDE_INT.
2924 * config/aarch64/aarch64.h (aarch64_frame): Protect with
2925 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
2926 hard_fp_offset, frame_size, initial_adjust, callee_offset and
2927 final_offset from HOST_WIDE_INT to poly_int64.
2928 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
2929 to_constant when getting the number of units in an Advanced SIMD
2931 (aarch64_builtin_vectorized_function): Check for a constant number
2933 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
2935 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
2936 attribute instead of GET_MODE_NUNITS.
2937 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
2938 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
2939 GET_MODE_SIZE for fixed-size registers.
2940 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
2941 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
2942 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
2943 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
2944 (aarch64_print_operand, aarch64_print_address_internal)
2945 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
2946 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
2947 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
2948 Handle polynomial GET_MODE_SIZE.
2949 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
2950 wider than SImode without modification.
2951 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
2952 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
2953 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
2954 passing and returning SVE modes.
2955 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
2956 rather than GEN_INT.
2957 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
2958 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
2959 (aarch64_allocate_and_probe_stack_space): Likewise.
2960 (aarch64_layout_frame): Cope with polynomial offsets.
2961 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
2962 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
2964 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
2965 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
2966 poly_int64 rather than a HOST_WIDE_INT.
2967 (aarch64_get_separate_components, aarch64_process_components)
2968 (aarch64_expand_prologue, aarch64_expand_epilogue)
2969 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
2970 (aarch64_anchor_offset): New function, split out from...
2971 (aarch64_legitimize_address): ...here.
2972 (aarch64_builtin_vectorization_cost): Handle polynomial
2973 TYPE_VECTOR_SUBPARTS.
2974 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
2976 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
2977 number of elements from the PARALLEL rather than the mode.
2978 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
2979 rather than GET_MODE_BITSIZE.
2980 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
2981 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
2982 (aarch64_expand_vec_perm_const_1): Handle polynomial
2983 d->perm.length () and d->perm elements.
2984 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
2985 Apply to_constant to d->perm elements.
2986 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
2987 polynomial CONST_VECTOR_NUNITS.
2988 (aarch64_move_pointer): Take amount as a poly_int64 rather
2990 (aarch64_progress_pointer): Avoid temporary variable.
2991 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
2992 the mode attribute instead of GET_MODE.
2994 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2995 Alan Hayward <alan.hayward@arm.com>
2996 David Sherwood <david.sherwood@arm.com>
2998 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
2999 x exists before using it.
3000 (aarch64_add_constant_internal): Rename to...
3001 (aarch64_add_offset_1): ...this. Replace regnum with separate
3002 src and dest rtxes. Handle the case in which they're different,
3003 including when the offset is zero. Replace scratchreg with an rtx.
3004 Use 2 additions if there is no spare register into which we can
3005 move a 16-bit constant.
3006 (aarch64_add_constant): Delete.
3007 (aarch64_add_offset): Replace reg with separate src and dest
3008 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3009 Use aarch64_add_offset_1.
3010 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3011 an rtx rather than an int. Take the delta as a poly_int64
3012 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3013 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3014 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3015 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3016 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3018 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3019 aarch64_add_constant.
3021 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3023 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3024 Use scalar_float_mode.
3026 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3028 * config/aarch64/aarch64-simd.md
3029 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3030 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3031 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3032 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3033 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3034 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3035 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3036 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3037 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3038 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3040 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3043 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3044 targ_options->x_arm_arch_string is non NULL.
3046 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3048 * config/aarch64/aarch64.h
3049 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3051 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3054 * expmed.c (emit_store_flag_force): Swap if const op0
3055 and change VOIDmode to mode of op0.
3057 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3059 PR rtl-optimization/83761
3060 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3061 than bytes to mode_for_size.
3063 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3066 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3067 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3070 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3073 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3074 when in layout mode.
3075 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3076 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3079 2018-01-10 Michael Collison <michael.collison@arm.com>
3081 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3082 * config/aarch64/aarch64-option-extension.def: Add
3083 AARCH64_OPT_EXTENSION of 'fp16fml'.
3084 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3085 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3086 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3087 * config/aarch64/constraints.md (Ui7): New constraint.
3088 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3089 (VFMLA_SEL_W): Ditto.
3092 (VFMLA16_LOW): New int iterator.
3093 (VFMLA16_HIGH): Ditto.
3094 (UNSPEC_FMLAL): New unspec.
3095 (UNSPEC_FMLSL): Ditto.
3096 (UNSPEC_FMLAL2): Ditto.
3097 (UNSPEC_FMLSL2): Ditto.
3098 (f16mac): New code attribute.
3099 * config/aarch64/aarch64-simd-builtins.def
3100 (aarch64_fmlal_lowv2sf): Ditto.
3101 (aarch64_fmlsl_lowv2sf): Ditto.
3102 (aarch64_fmlalq_lowv4sf): Ditto.
3103 (aarch64_fmlslq_lowv4sf): Ditto.
3104 (aarch64_fmlal_highv2sf): Ditto.
3105 (aarch64_fmlsl_highv2sf): Ditto.
3106 (aarch64_fmlalq_highv4sf): Ditto.
3107 (aarch64_fmlslq_highv4sf): Ditto.
3108 (aarch64_fmlal_lane_lowv2sf): Ditto.
3109 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3110 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3111 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3112 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3113 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3114 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3115 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3116 (aarch64_fmlal_lane_highv2sf): Ditto.
3117 (aarch64_fmlsl_lane_highv2sf): Ditto.
3118 (aarch64_fmlal_laneq_highv2sf): Ditto.
3119 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3120 (aarch64_fmlalq_lane_highv4sf): Ditto.
3121 (aarch64_fmlsl_lane_highv4sf): Ditto.
3122 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3123 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3124 * config/aarch64/aarch64-simd.md:
3125 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3126 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3127 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3128 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3129 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3130 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3131 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3132 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3133 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3134 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3135 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3136 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3137 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3138 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3139 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3140 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3141 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3142 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3143 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3144 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3145 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3146 (vfmlsl_low_u32): Ditto.
3147 (vfmlalq_low_u32): Ditto.
3148 (vfmlslq_low_u32): Ditto.
3149 (vfmlal_high_u32): Ditto.
3150 (vfmlsl_high_u32): Ditto.
3151 (vfmlalq_high_u32): Ditto.
3152 (vfmlslq_high_u32): Ditto.
3153 (vfmlal_lane_low_u32): Ditto.
3154 (vfmlsl_lane_low_u32): Ditto.
3155 (vfmlal_laneq_low_u32): Ditto.
3156 (vfmlsl_laneq_low_u32): Ditto.
3157 (vfmlalq_lane_low_u32): Ditto.
3158 (vfmlslq_lane_low_u32): Ditto.
3159 (vfmlalq_laneq_low_u32): Ditto.
3160 (vfmlslq_laneq_low_u32): Ditto.
3161 (vfmlal_lane_high_u32): Ditto.
3162 (vfmlsl_lane_high_u32): Ditto.
3163 (vfmlal_laneq_high_u32): Ditto.
3164 (vfmlsl_laneq_high_u32): Ditto.
3165 (vfmlalq_lane_high_u32): Ditto.
3166 (vfmlslq_lane_high_u32): Ditto.
3167 (vfmlalq_laneq_high_u32): Ditto.
3168 (vfmlslq_laneq_high_u32): Ditto.
3169 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3170 (AARCH64_FL_FOR_ARCH8_4): New.
3171 (AARCH64_ISA_F16FML): New ISA flag.
3172 (TARGET_F16FML): New feature flag for fp16fml.
3173 (doc/invoke.texi): Document new fp16fml option.
3175 2018-01-10 Michael Collison <michael.collison@arm.com>
3177 * config/aarch64/aarch64-builtins.c:
3178 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3179 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3180 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3181 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3182 (AARCH64_ISA_SHA3): New ISA flag.
3183 (TARGET_SHA3): New feature flag for sha3.
3184 * config/aarch64/iterators.md (sha512_op): New int attribute.
3185 (CRYPTO_SHA512): New int iterator.
3186 (UNSPEC_SHA512H): New unspec.
3187 (UNSPEC_SHA512H2): Ditto.
3188 (UNSPEC_SHA512SU0): Ditto.
3189 (UNSPEC_SHA512SU1): Ditto.
3190 * config/aarch64/aarch64-simd-builtins.def
3191 (aarch64_crypto_sha512hqv2di): New builtin.
3192 (aarch64_crypto_sha512h2qv2di): Ditto.
3193 (aarch64_crypto_sha512su0qv2di): Ditto.
3194 (aarch64_crypto_sha512su1qv2di): Ditto.
3195 (aarch64_eor3qv8hi): Ditto.
3196 (aarch64_rax1qv2di): Ditto.
3197 (aarch64_xarqv2di): Ditto.
3198 (aarch64_bcaxqv8hi): Ditto.
3199 * config/aarch64/aarch64-simd.md:
3200 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3201 (aarch64_crypto_sha512su0qv2di): Ditto.
3202 (aarch64_crypto_sha512su1qv2di): Ditto.
3203 (aarch64_eor3qv8hi): Ditto.
3204 (aarch64_rax1qv2di): Ditto.
3205 (aarch64_xarqv2di): Ditto.
3206 (aarch64_bcaxqv8hi): Ditto.
3207 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3208 (vsha512h2q_u64): Ditto.
3209 (vsha512su0q_u64): Ditto.
3210 (vsha512su1q_u64): Ditto.
3211 (veor3q_u16): Ditto.
3212 (vrax1q_u64): Ditto.
3214 (vbcaxq_u16): Ditto.
3215 * config/arm/types.md (crypto_sha512): New type attribute.
3216 (crypto_sha3): Ditto.
3217 (doc/invoke.texi): Document new sha3 option.
3219 2018-01-10 Michael Collison <michael.collison@arm.com>
3221 * config/aarch64/aarch64-builtins.c:
3222 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3223 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3224 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3225 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3226 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3227 (AARCH64_ISA_SM4): New ISA flag.
3228 (TARGET_SM4): New feature flag for sm4.
3229 * config/aarch64/aarch64-simd-builtins.def
3230 (aarch64_sm3ss1qv4si): Ditto.
3231 (aarch64_sm3tt1aq4si): Ditto.
3232 (aarch64_sm3tt1bq4si): Ditto.
3233 (aarch64_sm3tt2aq4si): Ditto.
3234 (aarch64_sm3tt2bq4si): Ditto.
3235 (aarch64_sm3partw1qv4si): Ditto.
3236 (aarch64_sm3partw2qv4si): Ditto.
3237 (aarch64_sm4eqv4si): Ditto.
3238 (aarch64_sm4ekeyqv4si): Ditto.
3239 * config/aarch64/aarch64-simd.md:
3240 (aarch64_sm3ss1qv4si): Ditto.
3241 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3242 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3243 (aarch64_sm4eqv4si): Ditto.
3244 (aarch64_sm4ekeyqv4si): Ditto.
3245 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3246 (sm3part_op): Ditto.
3247 (CRYPTO_SM3TT): Ditto.
3248 (CRYPTO_SM3PART): Ditto.
3249 (UNSPEC_SM3SS1): New unspec.
3250 (UNSPEC_SM3TT1A): Ditto.
3251 (UNSPEC_SM3TT1B): Ditto.
3252 (UNSPEC_SM3TT2A): Ditto.
3253 (UNSPEC_SM3TT2B): Ditto.
3254 (UNSPEC_SM3PARTW1): Ditto.
3255 (UNSPEC_SM3PARTW2): Ditto.
3256 (UNSPEC_SM4E): Ditto.
3257 (UNSPEC_SM4EKEY): Ditto.
3258 * config/aarch64/constraints.md (Ui2): New constraint.
3259 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3260 * config/arm/types.md (crypto_sm3): New type attribute.
3261 (crypto_sm4): Ditto.
3262 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3263 (vsm3tt1aq_u32): Ditto.
3264 (vsm3tt1bq_u32): Ditto.
3265 (vsm3tt2aq_u32): Ditto.
3266 (vsm3tt2bq_u32): Ditto.
3267 (vsm3partw1q_u32): Ditto.
3268 (vsm3partw2q_u32): Ditto.
3269 (vsm4eq_u32): Ditto.
3270 (vsm4ekeyq_u32): Ditto.
3271 (doc/invoke.texi): Document new sm4 option.
3273 2018-01-10 Michael Collison <michael.collison@arm.com>
3275 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3276 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3277 (AARCH64_FL_FOR_ARCH8_4): New.
3278 (AARCH64_FL_V8_4): New flag.
3279 (doc/invoke.texi): Document new armv8.4-a option.
3281 2018-01-10 Michael Collison <michael.collison@arm.com>
3283 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3284 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3285 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3286 * config/aarch64/aarch64-option-extension.def: Add
3287 AARCH64_OPT_EXTENSION of 'sha2'.
3288 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3289 (crypto): Disable sha2 and aes if crypto disabled.
3290 (crypto): Enable aes and sha2 if enabled.
3291 (simd): Disable sha2 and aes if simd disabled.
3292 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3294 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3295 (TARGET_SHA2): New feature flag for sha2.
3296 (TARGET_AES): New feature flag for aes.
3297 * config/aarch64/aarch64-simd.md:
3298 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3299 conditional on TARGET_AES.
3300 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3301 (aarch64_crypto_sha1hsi): Make pattern conditional
3303 (aarch64_crypto_sha1hv4si): Ditto.
3304 (aarch64_be_crypto_sha1hv4si): Ditto.
3305 (aarch64_crypto_sha1su1v4si): Ditto.
3306 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3307 (aarch64_crypto_sha1su0v4si): Ditto.
3308 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3309 (aarch64_crypto_sha256su0v4si): Ditto.
3310 (aarch64_crypto_sha256su1v4si): Ditto.
3311 (doc/invoke.texi): Document new aes and sha2 options.
3313 2018-01-10 Martin Sebor <msebor@redhat.com>
3315 PR tree-optimization/83781
3316 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3319 2018-01-11 Martin Sebor <msebor@gmail.com>
3320 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3322 PR tree-optimization/83501
3323 PR tree-optimization/81703
3325 * tree-ssa-strlen.c (get_string_cst): Rename...
3326 (get_string_len): ...to this. Handle global constants.
3327 (handle_char_store): Adjust.
3329 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3330 Jim Wilson <jimw@sifive.com>
3332 * config/riscv/riscv-protos.h (riscv_output_return): New.
3333 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3334 (riscv_attribute_table, riscv_output_return),
3335 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3336 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3337 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3339 (riscv_expand_prologue): Add early return for naked function.
3340 (riscv_expand_epilogue): Likewise.
3341 (riscv_function_ok_for_sibcall): Return false for naked function.
3342 (riscv_set_current_function): New.
3343 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3344 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3345 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3346 * doc/extend.texi (RISC-V Function Attributes): New.
3348 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3350 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3351 check for 128-bit long double before checking TCmode.
3352 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3353 128-bit long doubles before checking TFmode or TCmode.
3354 (FLOAT128_IBM_P): Likewise.
3356 2018-01-10 Martin Sebor <msebor@redhat.com>
3358 PR tree-optimization/83671
3359 * builtins.c (c_strlen): Unconditionally return zero for the empty
3361 Use -Warray-bounds for warnings.
3362 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3363 for non-constant array indices with COMPONENT_REF, arrays of
3364 arrays, and pointers to arrays.
3365 (gimple_fold_builtin_strlen): Determine and set length range for
3366 non-constant character arrays.
3368 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3371 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3374 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3376 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3378 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3381 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3382 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3383 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3384 indexed_or_indirect_operand predicate.
3385 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3386 (*vsx_le_perm_load_v8hi): Likewise.
3387 (*vsx_le_perm_load_v16qi): Likewise.
3388 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3389 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3390 (*vsx_le_perm_store_v8hi): Likewise.
3391 (*vsx_le_perm_store_v16qi): Likewise.
3392 (eight unnamed splitters): Likewise.
3394 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3396 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3397 * config/rs6000/emmintrin.h: Likewise.
3398 * config/rs6000/mmintrin.h: Likewise.
3399 * config/rs6000/xmmintrin.h: Likewise.
3401 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3404 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3406 * tree.c (tree_nop_conversion): Return true for location wrapper
3408 (maybe_wrap_with_location): New function.
3409 (selftest::check_strip_nops): New function.
3410 (selftest::test_location_wrappers): New function.
3411 (selftest::tree_c_tests): Call it.
3412 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3413 (maybe_wrap_with_location): New decl.
3414 (EXPR_LOCATION_WRAPPER_P): New macro.
3415 (location_wrapper_p): New inline function.
3416 (tree_strip_any_location_wrapper): New inline function.
3418 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3421 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3422 stack_realign_offset for the largest alignment of stack slot
3424 (ix86_find_max_used_stack_alignment): New function.
3425 (ix86_finalize_stack_frame_flags): Use it. Set
3426 max_used_stack_alignment if we don't realign stack.
3427 * config/i386/i386.h (machine_function): Add
3428 max_used_stack_alignment.
3430 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3432 * config/arm/arm.opt (-mbranch-cost): New option.
3433 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3436 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3439 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3440 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3442 2018-01-10 Richard Biener <rguenther@suse.de>
3445 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3446 early out so it also covers the case where we have a non-NULL
3449 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3451 PR tree-optimization/83753
3452 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3453 for non-strided grouped accesses if the number of elements is 1.
3455 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3458 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3459 * i386.h (TARGET_USE_GATHER): Define.
3460 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3462 2018-01-10 Martin Liska <mliska@suse.cz>
3465 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3466 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3468 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3469 CLEANUP_NO_PARTITIONING is not set.
3471 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3473 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3474 for vectors, as a partial revert of r254296.
3475 * rtl.h (const_vec_p): Delete.
3476 (const_vec_duplicate_p): Don't test for vector CONSTs.
3477 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3478 * expmed.c (make_tree): Likewise.
3481 * common.md (E, F): Use CONSTANT_P instead of checking for
3483 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3484 checking for CONST_VECTOR.
3486 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3489 * predict.c (force_edge_cold): Handle in more sane way edges
3492 2018-01-09 Carl Love <cel@us.ibm.com>
3494 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3496 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3497 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3498 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3499 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3500 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3501 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3502 * config/rs6000/rs6000-protos.h: Add extern defition for
3503 rs6000_generate_float2_double_code.
3504 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3506 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3507 (float2_v2df): Add define_expand.
3509 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3512 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3513 op_mode in the force_to_mode call.
3515 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3517 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3518 instead of checking each element individually.
3519 (aarch64_evpc_uzp): Likewise.
3520 (aarch64_evpc_zip): Likewise.
3521 (aarch64_evpc_ext): Likewise.
3522 (aarch64_evpc_rev): Likewise.
3523 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3524 instead of checking each element individually. Return true without
3526 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3527 whether all selected elements come from the same input, instead of
3528 checking each element individually. Remove calls to gen_rtx_REG,
3529 start_sequence and end_sequence and instead assert that no rtl is
3532 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3534 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3535 order of HIGH and CONST checks.
3537 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3539 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3540 if the destination isn't an SSA_NAME.
3542 2018-01-09 Richard Biener <rguenther@suse.de>
3544 PR tree-optimization/83668
3545 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3547 (canonicalize_loop_form): ... here, renamed from ...
3548 (canonicalize_loop_closed_ssa_form): ... this and amended to
3549 swap successor edges for loop exit blocks to make us use
3550 the RPO order we need for initial schedule generation.
3552 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3554 PR tree-optimization/64811
3555 * match.pd: When optimizing comparisons with Inf, avoid
3556 introducing or losing exceptions from comparisons with NaN.
3558 2018-01-09 Martin Liska <mliska@suse.cz>
3561 * asan.c (shadow_mem_size): Add gcc_assert.
3563 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3565 Don't save registers in main().
3568 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3569 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3570 * config/avr/avr.c (avr_set_current_function): Don't error if
3571 naked, OS_task or OS_main are specified at the same time.
3572 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3574 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3576 * common/config/avr/avr-common.c (avr_option_optimization_table):
3577 Switch on -mmain-is-OS_task for optimizing compilations.
3579 2018-01-09 Richard Biener <rguenther@suse.de>
3581 PR tree-optimization/83572
3582 * graphite.c: Include cfganal.h.
3583 (graphite_transform_loops): Connect infinite loops to exit
3584 and remove fake edges at the end.
3586 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3588 * ipa-inline.c (edge_badness): Revert accidental checkin.
3590 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3593 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3594 symbols; not inline clones.
3596 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3599 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3600 hard registers. Formatting fixes.
3602 PR preprocessor/83722
3603 * gcc.c (try_generate_repro): Pass
3604 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3605 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3608 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3609 Kito Cheng <kito.cheng@gmail.com>
3611 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3612 (riscv_leaf_function_p): Delete.
3613 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3615 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3617 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3619 (do_ifelse): New function.
3620 (do_isel): New function.
3621 (do_sub3): New function.
3622 (do_add3): New function.
3623 (do_load_mask_compare): New function.
3624 (do_overlap_load_compare): New function.
3625 (expand_compare_loop): New function.
3626 (expand_block_compare): Call expand_compare_loop() when appropriate.
3627 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3629 (-mblock-compare-inline-loop-limit): New option.
3631 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3634 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3635 Reverse order of second and third operands in first alternative.
3636 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3637 of first and second elements in UNSPEC_VPERMR vector.
3638 (altivec_expand_vec_perm_le): Likewise.
3640 2017-01-08 Jeff Law <law@redhat.com>
3642 PR rtl-optimizatin/81308
3643 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3644 (process_switch): If group_case_labels makes a change, then set
3646 (pass_convert_switch::execute): If a switch is converted, then
3647 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3649 PR rtl-optimization/81308
3650 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3653 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3655 PR target/83663 - Revert r255946
3656 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3657 generation for cases where splatting a value is not useful.
3658 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3659 across a vec_duplicate and a paradoxical subreg forming a vector
3660 mode to a vec_concat.
3662 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3664 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3665 -march=armv8.3-a variants.
3666 * config/arm/t-multilib: Likewise.
3667 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3669 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3671 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3673 (cceq_ior_compare_complement): Give it a name so I can use it, and
3674 change boolean_or_operator predicate to boolean_operator so it can
3675 be used to generate a crand.
3676 (eqne): New code iterator.
3677 (bd/bd_neg): New code_attrs.
3678 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3679 a single define_insn.
3680 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3681 decrement (bdnzt/bdnzf/bdzt/bdzf).
3682 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3683 with the new names of the branch decrement patterns, and added the
3684 names of the branch decrement conditional patterns.
3686 2018-01-08 Richard Biener <rguenther@suse.de>
3688 PR tree-optimization/83563
3689 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3692 2018-01-08 Richard Biener <rguenther@suse.de>
3695 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3697 2018-01-08 Richard Biener <rguenther@suse.de>
3699 PR tree-optimization/83685
3700 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3701 references to abnormals.
3703 2018-01-08 Richard Biener <rguenther@suse.de>
3706 * dwarf2out.c (output_indirect_strings): Handle empty
3707 skeleton_debug_str_hash.
3708 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3710 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3712 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3713 (emit_store_direct): Likewise.
3714 (arc_trampoline_adjust_address): Likewise.
3715 (arc_asm_trampoline_template): New function.
3716 (arc_initialize_trampoline): Use asm_trampoline_template.
3717 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3718 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3719 * config/arc/arc.md (flush_icache): Delete pattern.
3721 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3723 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3724 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3727 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3730 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3731 by not USED_FOR_TARGET.
3732 (make_pass_resolve_sw_modes): Likewise.
3734 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3736 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3739 2018-01-08 Richard Biener <rguenther@suse.de>
3742 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3744 2018-01-08 Richard Biener <rguenther@suse.de>
3747 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3749 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3752 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3753 basic blocks with a small number of successors.
3754 (convert_control_dep_chain_into_preds): Improve handling of
3756 (dump_predicates): Split apart into...
3757 (dump_pred_chain): ...here...
3758 (dump_pred_info): ...and here.
3759 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3760 (can_chain_union_be_invalidated_p): Improve check for invalidation
3762 (uninit_uses_cannot_happen): Avoid unnecessary if
3763 convert_control_dep_chain_into_preds yielded nothing.
3765 2018-01-06 Martin Sebor <msebor@redhat.com>
3767 PR tree-optimization/83640
3768 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3769 subtracting negative offset from size.
3770 (builtin_access::overlap): Adjust offset bounds of the access to fall
3771 within the size of the object if possible.
3773 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3775 PR rtl-optimization/83699
3776 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3777 extract_bit_field_as_subreg to cases in which the extracted
3778 value is also a vector.
3780 * lra-constraints.c (process_alt_operands): Test for the equivalence
3781 substitutions when detecting a possible reload cycle.
3783 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3786 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3787 by default if flag_selective_schedling{,2}. Formatting fixes.
3789 PR rtl-optimization/83682
3790 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
3791 if it has non-VECTOR_MODE element mode.
3792 (vec_duplicate_p): Likewise.
3795 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
3796 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
3798 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3801 * config/i386/i386-builtin.def
3802 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
3803 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
3804 Require also OPTION_MASK_ISA_AVX512F in addition to
3805 OPTION_MASK_ISA_GFNI.
3806 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
3807 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
3808 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
3809 to OPTION_MASK_ISA_GFNI.
3810 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
3811 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
3812 OPTION_MASK_ISA_AVX512BW.
3813 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
3814 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
3815 addition to OPTION_MASK_ISA_GFNI.
3816 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
3817 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
3818 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
3819 to OPTION_MASK_ISA_GFNI.
3820 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
3821 a requirement for all ISAs rather than any of them with a few
3823 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
3825 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
3826 bitmasks to be enabled with 3 exceptions, instead of requiring any
3827 enabled ISA with lots of exceptions.
3828 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
3829 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
3830 Change avx512bw in isa attribute to avx512f.
3831 * config/i386/sgxintrin.h: Add license boilerplate.
3832 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
3833 to __AVX512F__ and __AVX512VL to __AVX512VL__.
3834 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
3835 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
3837 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
3838 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
3839 temporarily sse2 rather than sse if not enabled already.
3842 * config/i386/sse.md (VI248_VLBW): Rename to ...
3843 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
3844 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
3845 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
3846 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
3847 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
3848 mode iterator instead of VI248_VLBW.
3850 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
3852 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
3853 (record_modified): Skip clobbers; add debug output.
3854 (param_change_prob): Use sreal frequencies.
3856 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3858 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
3859 punt for user-aligned variables.
3861 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3863 * tree-chrec.c (chrec_contains_symbols): Return true for
3866 2018-01-05 Sudakshina Das <sudi.das@arm.com>
3869 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
3870 of (x|y) == x for BICS pattern.
3872 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3874 PR tree-optimization/83605
3875 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
3876 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
3879 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
3881 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
3882 * config/epiphany/rtems.h: New file.
3884 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3885 Uros Bizjak <ubizjak@gmail.com>
3888 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
3889 QIreg_operand instead of register_operand predicate.
3890 * config/i386/i386.c (ix86_rop_should_change_byte_p,
3891 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
3892 comments instead of -fmitigate[-_]rop.
3894 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3897 * cgraphunit.c (symbol_table::compile): Switch to text_section
3898 before calling assembly_start debug hook.
3899 * run-rtl-passes.c (run_rtl_passes): Likewise.
3902 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3904 * tree-vrp.c (extract_range_from_binary_expr_1): Check
3905 range_int_cst_p rather than !symbolic_range_p before calling
3906 extract_range_from_multiplicative_op_1.
3908 2017-01-04 Jeff Law <law@redhat.com>
3910 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
3911 redundant test in assertion.
3913 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3915 * doc/rtl.texi: Document machine_mode wrapper classes.
3917 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3919 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
3922 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3924 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
3925 the VEC_PERM_EXPR fold to fail.
3927 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3930 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
3931 to switched_sections.
3933 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3936 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
3939 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
3942 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
3943 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
3945 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3948 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
3949 is BLKmode and bitpos not zero or mode change is needed.
3951 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3954 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
3957 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
3960 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
3961 instead of MULT rtx. Update all corresponding splitters.
3963 (*ssub<modesuffix>): Ditto.
3965 (*cmp_sadd_di): Update split patterns.
3966 (*cmp_sadd_si): Ditto.
3967 (*cmp_sadd_sidi): Ditto.
3968 (*cmp_ssub_di): Ditto.
3969 (*cmp_ssub_si): Ditto.
3970 (*cmp_ssub_sidi): Ditto.
3971 * config/alpha/predicates.md (const23_operand): New predicate.
3972 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
3973 Look for ASHIFT, not MULT inner operand.
3974 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
3976 2018-01-04 Martin Liska <mliska@suse.cz>
3978 PR gcov-profile/83669
3979 * gcov.c (output_intermediate_file): Add version to intermediate
3981 * doc/gcov.texi: Document new field 'version' in intermediate
3982 file format. Fix location of '-k' option of gcov command.
3984 2018-01-04 Martin Liska <mliska@suse.cz>
3987 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
3989 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3991 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
3993 2018-01-03 Martin Sebor <msebor@redhat.com>
3995 PR tree-optimization/83655
3996 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
3997 checking calls with invalid arguments.
3999 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4001 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4002 (vectorizable_mask_load_store): Delete.
4003 (vectorizable_call): Return false for masked loads and stores.
4004 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4005 instead of gimple_assign_rhs1.
4006 (vectorizable_load): Handle IFN_MASK_LOAD.
4007 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4009 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4011 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4013 (vectorizable_mask_load_store): ...here.
4014 (vectorizable_load): ...and here.
4016 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4018 * tree-vect-stmts.c (vect_build_all_ones_mask)
4019 (vect_build_zero_merge_argument): New functions, split out from...
4020 (vectorizable_load): ...here.
4022 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4024 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4026 (vectorizable_mask_load_store): ...here.
4027 (vectorizable_store): ...and here.
4029 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4031 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4033 (vectorizable_mask_load_store): ...here.
4035 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4037 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4038 (vect_model_store_cost): Take a vec_load_store_type instead of a
4040 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4041 (vect_model_store_cost): Take a vec_load_store_type instead of a
4043 (vectorizable_mask_load_store): Update accordingly.
4044 (vectorizable_store): Likewise.
4045 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4047 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4049 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4050 IFN_MASK_LOAD calls here rather than...
4051 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4053 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4054 Alan Hayward <alan.hayward@arm.com>
4055 David Sherwood <david.sherwood@arm.com>
4057 * expmed.c (extract_bit_field_1): For vector extracts,
4058 fall back to extract_bit_field_as_subreg if vec_extract
4061 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4062 Alan Hayward <alan.hayward@arm.com>
4063 David Sherwood <david.sherwood@arm.com>
4065 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4066 they are variable or constant sized.
4067 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4068 slots for constant-sized data.
4070 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4071 Alan Hayward <alan.hayward@arm.com>
4072 David Sherwood <david.sherwood@arm.com>
4074 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4075 handling COND_EXPRs with boolean comparisons, try to find a better
4076 basis for the mask type than the boolean itself.
4078 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4080 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4081 is calculated and how it can be overridden.
4082 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4083 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4085 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4088 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4089 Alan Hayward <alan.hayward@arm.com>
4090 David Sherwood <david.sherwood@arm.com>
4092 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4093 Remove the mode argument.
4094 (aarch64_simd_valid_immediate): Remove the mode and inverse
4096 * config/aarch64/iterators.md (bitsize): New iterator.
4097 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4098 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4099 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4100 aarch64_simd_valid_immediate.
4101 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4102 (aarch64_reg_or_bic_imm): Likewise.
4103 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4104 with an insn_type enum and msl with a modifier_type enum.
4105 Replace element_width with a scalar_mode. Change the shift
4106 to unsigned int. Add constructors for scalar_float_mode and
4107 scalar_int_mode elements.
4108 (aarch64_vect_float_const_representable_p): Delete.
4109 (aarch64_can_const_movi_rtx_p)
4110 (aarch64_simd_scalar_immediate_valid_for_move)
4111 (aarch64_simd_make_constant): Update call to
4112 aarch64_simd_valid_immediate.
4113 (aarch64_advsimd_valid_immediate_hs): New function.
4114 (aarch64_advsimd_valid_immediate): Likewise.
4115 (aarch64_simd_valid_immediate): Remove mode and inverse
4116 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4117 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4118 and aarch64_float_const_representable_p on the result.
4119 (aarch64_output_simd_mov_immediate): Remove mode argument.
4120 Update call to aarch64_simd_valid_immediate and use of
4121 simd_immediate_info.
4122 (aarch64_output_scalar_simd_mov_immediate): Update call
4125 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4126 Alan Hayward <alan.hayward@arm.com>
4127 David Sherwood <david.sherwood@arm.com>
4129 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4130 (mode_nunits): Likewise CONST_MODE_NUNITS.
4131 * machmode.def (ADJUST_NUNITS): Document.
4132 * genmodes.c (mode_data::need_nunits_adj): New field.
4133 (blank_mode): Update accordingly.
4134 (adj_nunits): New variable.
4135 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4137 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4138 listed in adj_nunits.
4139 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4140 listed in adj_nunits. Don't emit case statements for such modes.
4141 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4142 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4143 nothing if adj_nunits is nonnull.
4144 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4145 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4146 (emit_mode_fbit): Update use of print_maybe_const_decl.
4147 (emit_move_size): Likewise. Treat the array as non-const
4149 (emit_mode_adjustments): Handle adj_nunits.
4151 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4153 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4154 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4155 (VECTOR_MODES): Use it.
4156 (make_vector_modes): Take the prefix as an argument.
4158 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4159 Alan Hayward <alan.hayward@arm.com>
4160 David Sherwood <david.sherwood@arm.com>
4162 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4163 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4164 for MODE_VECTOR_BOOL.
4165 * machmode.def (VECTOR_BOOL_MODE): Document.
4166 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4167 (make_vector_bool_mode): New function.
4168 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4170 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4171 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4173 * stor-layout.c (int_mode_for_mode): Likewise.
4174 * tree.c (build_vector_type_for_mode): Likewise.
4175 * varasm.c (output_constant_pool_2): Likewise.
4176 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4177 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4178 for MODE_VECTOR_BOOL.
4179 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4180 of mode class checks.
4181 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4182 instead of a list of mode class checks.
4183 (expand_vector_scalar_condition): Likewise.
4184 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4186 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4187 Alan Hayward <alan.hayward@arm.com>
4188 David Sherwood <david.sherwood@arm.com>
4190 * machmode.h (mode_size): Change from unsigned short to
4192 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4193 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4194 or if measurement_type is not polynomial.
4195 (fixed_size_mode::includes_p): Check for constant-sized modes.
4196 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4197 return a poly_uint16 rather than an unsigned short.
4198 (emit_mode_size): Change the type of mode_size from unsigned short
4199 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4200 (emit_mode_adjustments): Cope with polynomial vector sizes.
4201 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4203 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4205 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4206 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4207 * caller-save.c (setup_save_areas): Likewise.
4208 (replace_reg_with_saved_mem): Likewise.
4209 * calls.c (emit_library_call_value_1): Likewise.
4210 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4211 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4212 (gen_lowpart_for_combine): Likewise.
4213 * convert.c (convert_to_integer_1): Likewise.
4214 * cse.c (equiv_constant, cse_insn): Likewise.
4215 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4216 (cselib_subst_to_values): Likewise.
4217 * dce.c (word_dce_process_block): Likewise.
4218 * df-problems.c (df_word_lr_mark_ref): Likewise.
4219 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4220 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4221 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4222 (rtl_for_decl_location): Likewise.
4223 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4224 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4225 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4226 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4227 (expand_expr_real_1): Likewise.
4228 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4229 (pad_below): Likewise.
4230 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4231 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4232 * ira.c (get_subreg_tracking_sizes): Likewise.
4233 * ira-build.c (ira_create_allocno_objects): Likewise.
4234 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4235 (ira_sort_regnos_for_alter_reg): Likewise.
4236 * ira-costs.c (record_operand_costs): Likewise.
4237 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4238 (resolve_simple_move): Likewise.
4239 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4240 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4241 (lra_constraints): Likewise.
4242 (CONST_POOL_OK_P): Reject variable-sized modes.
4243 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4244 (add_pseudo_to_slot, lra_spill): Likewise.
4245 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4246 * optabs-query.c (get_best_extraction_insn): Likewise.
4247 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4248 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4249 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4250 * recog.c (offsettable_address_addr_space_p): Likewise.
4251 * regcprop.c (maybe_mode_change): Likewise.
4252 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4253 * regrename.c (build_def_use): Likewise.
4254 * regstat.c (dump_reg_info): Likewise.
4255 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4256 (find_reloads, find_reloads_subreg_address): Likewise.
4257 * reload1.c (eliminate_regs_1): Likewise.
4258 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4259 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4260 (simplify_binary_operation_1, simplify_subreg): Likewise.
4261 * targhooks.c (default_function_arg_padding): Likewise.
4262 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4263 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4264 (verify_gimple_assign_ternary): Likewise.
4265 * tree-inline.c (estimate_move_cost): Likewise.
4266 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4267 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4268 (get_address_cost_ainc): Likewise.
4269 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4270 (vect_supportable_dr_alignment): Likewise.
4271 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4272 (vectorizable_reduction): Likewise.
4273 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4274 (vectorizable_operation, vectorizable_load): Likewise.
4275 * tree.c (build_same_sized_truth_vector_type): Likewise.
4276 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4277 * var-tracking.c (emit_note_insn_var_location): Likewise.
4278 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4279 (ADDR_VEC_ALIGN): Likewise.
4281 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4282 Alan Hayward <alan.hayward@arm.com>
4283 David Sherwood <david.sherwood@arm.com>
4285 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4287 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4288 or if measurement_type is polynomial.
4289 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4290 * combine.c (make_extraction): Likewise.
4291 * dse.c (find_shift_sequence): Likewise.
4292 * dwarf2out.c (mem_loc_descriptor): Likewise.
4293 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4294 (extract_bit_field, extract_low_bits): Likewise.
4295 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4296 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4297 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4298 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4299 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4300 * reload.c (find_reloads): Likewise.
4301 * reload1.c (alter_reg): Likewise.
4302 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4303 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4304 * tree-if-conv.c (predicate_mem_writes): Likewise.
4305 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4306 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4307 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4308 * valtrack.c (dead_debug_insert_temp): Likewise.
4309 * varasm.c (mergeable_constant_section): Likewise.
4310 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4312 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4313 Alan Hayward <alan.hayward@arm.com>
4314 David Sherwood <david.sherwood@arm.com>
4316 * expr.c (expand_assignment): Cope with polynomial mode sizes
4317 when assigning to a CONCAT.
4319 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4320 Alan Hayward <alan.hayward@arm.com>
4321 David Sherwood <david.sherwood@arm.com>
4323 * machmode.h (mode_precision): Change from unsigned short to
4325 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4327 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4328 or if measurement_type is not polynomial.
4329 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4330 in which the mode is already known to be a scalar_int_mode.
4331 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4332 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4334 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4335 for GET_MODE_PRECISION.
4336 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4337 for GET_MODE_PRECISION.
4338 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4340 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4341 (expand_field_assignment, make_extraction): Likewise.
4342 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4343 (get_last_value): Likewise.
4344 * convert.c (convert_to_integer_1): Likewise.
4345 * cse.c (cse_insn): Likewise.
4346 * expr.c (expand_expr_real_1): Likewise.
4347 * lra-constraints.c (simplify_operand_subreg): Likewise.
4348 * optabs-query.c (can_atomic_load_p): Likewise.
4349 * optabs.c (expand_atomic_load): Likewise.
4350 (expand_atomic_store): Likewise.
4351 * ree.c (combine_reaching_defs): Likewise.
4352 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4353 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4354 * tree.h (type_has_mode_precision_p): Likewise.
4355 * ubsan.c (instrument_si_overflow): Likewise.
4357 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4358 Alan Hayward <alan.hayward@arm.com>
4359 David Sherwood <david.sherwood@arm.com>
4361 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4362 polynomial numbers of units.
4363 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4364 (valid_vector_subparts_p): New function.
4365 (build_vector_type): Remove temporary shim and take the number
4366 of units as a poly_uint64 rather than an int.
4367 (build_opaque_vector_type): Take the number of units as a
4368 poly_uint64 rather than an int.
4369 * tree.c (build_vector_from_ctor): Handle polynomial
4370 TYPE_VECTOR_SUBPARTS.
4371 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4372 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4373 (build_vector_from_val): If the number of units is variable,
4374 use build_vec_duplicate_cst for constant operands and
4375 VEC_DUPLICATE_EXPR otherwise.
4376 (make_vector_type): Remove temporary is_constant ().
4377 (build_vector_type, build_opaque_vector_type): Take the number of
4378 units as a poly_uint64 rather than an int.
4379 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4381 * cfgexpand.c (expand_debug_expr): Likewise.
4382 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4383 (store_constructor, expand_expr_real_1): Likewise.
4384 (const_scalar_mask_from_tree): Likewise.
4385 * fold-const-call.c (fold_const_reduction): Likewise.
4386 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4387 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4388 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4389 (fold_relational_const): Likewise.
4390 (native_interpret_vector): Likewise. Change the size from an
4391 int to an unsigned int.
4392 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4393 TYPE_VECTOR_SUBPARTS.
4394 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4395 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4396 duplicating a non-constant operand into a variable-length vector.
4397 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4398 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4399 * ipa-icf.c (sem_variable::equals): Likewise.
4400 * match.pd: Likewise.
4401 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4402 * print-tree.c (print_node): Likewise.
4403 * stor-layout.c (layout_type): Likewise.
4404 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4405 * tree-cfg.c (verify_gimple_comparison): Likewise.
4406 (verify_gimple_assign_binary): Likewise.
4407 (verify_gimple_assign_ternary): Likewise.
4408 (verify_gimple_assign_single): Likewise.
4409 * tree-pretty-print.c (dump_generic_node): Likewise.
4410 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4411 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4412 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4413 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4414 (vect_shift_permute_load_chain): Likewise.
4415 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4416 (expand_vector_condition, optimize_vector_constructor): Likewise.
4417 (lower_vec_perm, get_compute_type): Likewise.
4418 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4419 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4420 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4421 (vect_recog_mask_conversion_pattern): Likewise.
4422 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4423 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4424 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4425 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4426 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4427 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4428 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4429 (supportable_widening_operation): Likewise.
4430 (supportable_narrowing_operation): Likewise.
4431 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4433 * varasm.c (output_constant): Likewise.
4435 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4436 Alan Hayward <alan.hayward@arm.com>
4437 David Sherwood <david.sherwood@arm.com>
4439 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4440 so that both the length == 3 and length != 3 cases set up their
4441 own permute vectors. Add comments explaining why we know the
4442 number of elements is constant.
4443 (vect_permute_load_chain): Likewise.
4445 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4446 Alan Hayward <alan.hayward@arm.com>
4447 David Sherwood <david.sherwood@arm.com>
4449 * machmode.h (mode_nunits): Change from unsigned char to
4451 (ONLY_FIXED_SIZE_MODES): New macro.
4452 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4453 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4454 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4456 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4457 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4458 or if measurement_type is not polynomial.
4459 * genmodes.c (ZERO_COEFFS): New macro.
4460 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4462 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4463 Use ZERO_COEFFS when emitting initializers.
4464 * data-streamer.h (bp_pack_poly_value): New function.
4465 (bp_unpack_poly_value): Likewise.
4466 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4467 for GET_MODE_NUNITS.
4468 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4469 for GET_MODE_NUNITS.
4470 * tree.c (make_vector_type): Remove temporary shim and make
4471 the real function take the number of units as a poly_uint64
4473 (build_vector_type_for_mode): Handle polynomial nunits.
4474 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4475 * emit-rtl.c (const_vec_series_p_1): Likewise.
4476 (gen_rtx_CONST_VECTOR): Likewise.
4477 * fold-const.c (test_vec_duplicate_folding): Likewise.
4478 * genrecog.c (validate_pattern): Likewise.
4479 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4480 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4481 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4482 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4483 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4484 * rtlanal.c (subreg_get_info): Likewise.
4485 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4486 (vect_grouped_load_supported): Likewise.
4487 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4488 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4489 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4490 (simplify_const_unary_operation, simplify_binary_operation_1)
4491 (simplify_const_binary_operation, simplify_ternary_operation)
4492 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4493 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4494 instead of CONST_VECTOR_NUNITS.
4495 * varasm.c (output_constant_pool_2): Likewise.
4496 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4497 explicit-encoded elements in the XVEC for variable-length vectors.
4499 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4501 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4503 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4504 Alan Hayward <alan.hayward@arm.com>
4505 David Sherwood <david.sherwood@arm.com>
4507 * coretypes.h (fixed_size_mode): Declare.
4508 (fixed_size_mode_pod): New typedef.
4509 * builtins.h (target_builtins::x_apply_args_mode)
4510 (target_builtins::x_apply_result_mode): Change type to
4511 fixed_size_mode_pod.
4512 * builtins.c (apply_args_size, apply_result_size, result_vector)
4513 (expand_builtin_apply_args_1, expand_builtin_apply)
4514 (expand_builtin_return): Update accordingly.
4516 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4518 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4519 * cselib.c (cselib_hash_rtx): Likewise.
4520 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4521 CONST_VECTOR encoding.
4523 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4524 Jeff Law <law@redhat.com>
4527 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4528 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4529 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4530 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4533 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4534 explicitly probe *sp in a noreturn function if there were any callee
4535 register saves or frame pointer is needed.
4537 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4540 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4541 BLKmode for ternary, binary or unary expressions.
4544 * var-tracking.c (delete_vta_debug_insn): New inline function.
4545 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4546 insns from get_insns () to NULL instead of each bb separately.
4547 Use delete_vta_debug_insn. No longer static.
4548 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4549 delete_vta_debug_insns callers.
4550 * rtl.h (delete_vta_debug_insns): Declare.
4551 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4552 instead of variable_tracking_main.
4554 2018-01-03 Martin Sebor <msebor@redhat.com>
4556 PR tree-optimization/83603
4557 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4558 arguments past the endof the argument list in functions declared
4559 without a prototype.
4560 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4561 Avoid checking when arguments are null.
4563 2018-01-03 Martin Sebor <msebor@redhat.com>
4566 * doc/extend.texi (attribute const): Fix a typo.
4567 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4568 issuing -Wsuggest-attribute for void functions.
4570 2018-01-03 Martin Sebor <msebor@redhat.com>
4572 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4573 offset_int::from instead of wide_int::to_shwi.
4574 (maybe_diag_overlap): Remove assertion.
4575 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4576 * gimple-ssa-sprintf.c (format_directive): Same.
4577 (parse_directive): Same.
4578 (sprintf_dom_walker::compute_format_length): Same.
4579 (try_substitute_return_value): Same.
4581 2017-01-03 Jeff Law <law@redhat.com>
4584 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4585 non-constant residual for zero at runtime and avoid probing in
4586 that case. Reorganize code for trailing problem to mirror handling
4589 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4591 PR tree-optimization/83501
4592 * tree-ssa-strlen.c (get_string_cst): New.
4593 (handle_char_store): Call get_string_cst.
4595 2018-01-03 Martin Liska <mliska@suse.cz>
4597 PR tree-optimization/83593
4598 * tree-ssa-strlen.c: Include tree-cfg.h.
4599 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4600 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4601 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4603 (strlen_dom_walker::before_dom_children): Call
4604 gimple_purge_dead_eh_edges. Dump tranformation with details
4606 (strlen_dom_walker::before_dom_children): Update call by adding
4607 new argument cleanup_eh.
4608 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4610 2018-01-03 Martin Liska <mliska@suse.cz>
4613 * cif-code.def (VARIADIC_THUNK): New enum value.
4614 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4617 2018-01-03 Jan Beulich <jbeulich@suse.com>
4619 * sse.md (mov<mode>_internal): Tighten condition for when to use
4620 vmovdqu<ssescalarsize> for TI and OI modes.
4622 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4624 Update copyright years.
4626 2018-01-03 Martin Liska <mliska@suse.cz>
4629 * ipa-visibility.c (function_and_variable_visibility): Skip
4630 functions with noipa attribure.
4632 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4634 * gcc.c (process_command): Update copyright notice dates.
4635 * gcov-dump.c (print_version): Ditto.
4636 * gcov.c (print_version): Ditto.
4637 * gcov-tool.c (print_version): Ditto.
4638 * gengtype.c (create_file): Ditto.
4639 * doc/cpp.texi: Bump @copying's copyright year.
4640 * doc/cppinternals.texi: Ditto.
4641 * doc/gcc.texi: Ditto.
4642 * doc/gccint.texi: Ditto.
4643 * doc/gcov.texi: Ditto.
4644 * doc/install.texi: Ditto.
4645 * doc/invoke.texi: Ditto.
4647 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4649 * vector-builder.h (vector_builder::m_full_nelts): Change from
4650 unsigned int to poly_uint64.
4651 (vector_builder::full_nelts): Update prototype accordingly.
4652 (vector_builder::new_vector): Likewise.
4653 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4654 (vector_builder::operator ==): Likewise.
4655 (vector_builder::finalize): Likewise.
4656 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4657 Take the number of elements as a poly_uint64 rather than an
4659 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4660 from unsigned int to poly_uint64.
4661 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4662 (vec_perm_indices::new_vector): Likewise.
4663 (vec_perm_indices::length): Likewise.
4664 (vec_perm_indices::nelts_per_input): Likewise.
4665 (vec_perm_indices::input_nelts): Likewise.
4666 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4667 number of elements per input as a poly_uint64 rather than an
4668 unsigned int. Use the original encoding for variable-length
4669 vectors, rather than clamping each individual element.
4670 For the second and subsequent elements in each pattern,
4671 clamp the step and base before clamping their sum.
4672 (vec_perm_indices::series_p): Handle polynomial element counts.
4673 (vec_perm_indices::all_in_range_p): Likewise.
4674 (vec_perm_indices_to_tree): Likewise.
4675 (vec_perm_indices_to_rtx): Likewise.
4676 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4677 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4678 (tree_vector_builder::new_binary_operation): Handle polynomial
4679 element counts. Return false if we need to know the number
4680 of elements at compile time.
4681 * fold-const.c (fold_vec_perm): Punt if the number of elements
4682 isn't known at compile time.
4684 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4686 * vec-perm-indices.h (vec_perm_builder): Change element type
4687 from HOST_WIDE_INT to poly_int64.
4688 (vec_perm_indices::element_type): Update accordingly.
4689 (vec_perm_indices::clamp): Handle polynomial element_types.
4690 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4691 (vec_perm_indices::all_in_range_p): Likewise.
4692 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4694 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4695 polynomial vec_perm_indices element types.
4696 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4697 * fold-const.c (fold_vec_perm): Likewise.
4698 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4699 * tree-vect-generic.c (lower_vec_perm): Likewise.
4700 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4701 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4702 element type to HOST_WIDE_INT.
4704 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4705 Alan Hayward <alan.hayward@arm.com>
4706 David Sherwood <david.sherwood@arm.com>
4708 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4709 rather than an int. Use plus_constant.
4710 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4711 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4713 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4714 Alan Hayward <alan.hayward@arm.com>
4715 David Sherwood <david.sherwood@arm.com>
4717 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4718 a HOST_WIDE_INT to a poly_int64.
4720 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4721 Alan Hayward <alan.hayward@arm.com>
4722 David Sherwood <david.sherwood@arm.com>
4724 * calls.c (load_register_parameters): Cope with polynomial
4725 mode sizes. Require a constant size for BLKmode parameters
4726 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4727 forces a parameter to be padded at the lsb end in order to
4728 fill a complete number of words, require the parameter size
4729 to be ordered wrt UNITS_PER_WORD.
4731 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4732 Alan Hayward <alan.hayward@arm.com>
4733 David Sherwood <david.sherwood@arm.com>
4735 * reload1.c (spill_stack_slot_width): Change element type
4736 from unsigned int to poly_uint64_pod.
4737 (alter_reg): Treat mode sizes as polynomial.
4739 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4740 Alan Hayward <alan.hayward@arm.com>
4741 David Sherwood <david.sherwood@arm.com>
4743 * reload.c (complex_word_subreg_p): New function.
4744 (reload_inner_reg_of_subreg, push_reload): Use it.
4746 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4747 Alan Hayward <alan.hayward@arm.com>
4748 David Sherwood <david.sherwood@arm.com>
4750 * lra-constraints.c (process_alt_operands): Reject matched
4751 operands whose sizes aren't ordered.
4752 (match_reload): Refer to this check here.
4754 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4755 Alan Hayward <alan.hayward@arm.com>
4756 David Sherwood <david.sherwood@arm.com>
4758 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4759 that the mode size is in the set {1, 2, 4, 8, 16}.
4761 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4762 Alan Hayward <alan.hayward@arm.com>
4763 David Sherwood <david.sherwood@arm.com>
4765 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4766 Use plus_constant instead of gen_rtx_PLUS.
4768 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4769 Alan Hayward <alan.hayward@arm.com>
4770 David Sherwood <david.sherwood@arm.com>
4772 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4773 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4774 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4775 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4776 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4777 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4778 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4779 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4780 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4781 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4783 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4784 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4785 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4786 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4787 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4788 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4789 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4790 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
4791 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
4792 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
4794 * expr.c (emit_move_resolve_push): Treat the input and result
4795 of PUSH_ROUNDING as a poly_int64.
4796 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
4797 (emit_push_insn): Likewise.
4798 * lra-eliminations.c (mark_not_eliminable): Likewise.
4799 * recog.c (push_operand): Likewise.
4800 * reload1.c (elimination_effects): Likewise.
4801 * rtlanal.c (nonzero_bits1): Likewise.
4802 * calls.c (store_one_arg): Likewise. Require the padding to be
4803 known at compile time.
4805 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4806 Alan Hayward <alan.hayward@arm.com>
4807 David Sherwood <david.sherwood@arm.com>
4809 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
4810 Use plus_constant instead of gen_rtx_PLUS.
4812 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4813 Alan Hayward <alan.hayward@arm.com>
4814 David Sherwood <david.sherwood@arm.com>
4816 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
4819 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4820 Alan Hayward <alan.hayward@arm.com>
4821 David Sherwood <david.sherwood@arm.com>
4823 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
4824 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
4825 via stack temporaries. Treat the mode size as polynomial too.
4827 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4828 Alan Hayward <alan.hayward@arm.com>
4829 David Sherwood <david.sherwood@arm.com>
4831 * expr.c (expand_expr_real_2): When handling conversions involving
4832 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
4833 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
4834 as a poly_uint64 too.
4836 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4837 Alan Hayward <alan.hayward@arm.com>
4838 David Sherwood <david.sherwood@arm.com>
4840 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
4842 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4843 Alan Hayward <alan.hayward@arm.com>
4844 David Sherwood <david.sherwood@arm.com>
4846 * combine.c (can_change_dest_mode): Handle polynomial
4847 REGMODE_NATURAL_SIZE.
4848 * expmed.c (store_bit_field_1): Likewise.
4849 * expr.c (store_constructor): Likewise.
4850 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
4851 and polynomial REGMODE_NATURAL_SIZE.
4852 (gen_lowpart_common): Likewise.
4853 * reginfo.c (record_subregs_of_mode): Likewise.
4854 * rtlanal.c (read_modify_subreg_p): Likewise.
4856 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4857 Alan Hayward <alan.hayward@arm.com>
4858 David Sherwood <david.sherwood@arm.com>
4860 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
4861 numbers of elements.
4863 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4864 Alan Hayward <alan.hayward@arm.com>
4865 David Sherwood <david.sherwood@arm.com>
4867 * match.pd: Cope with polynomial numbers of vector elements.
4869 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4870 Alan Hayward <alan.hayward@arm.com>
4871 David Sherwood <david.sherwood@arm.com>
4873 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
4874 in a POINTER_PLUS_EXPR.
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 * omp-simd-clone.c (simd_clone_subparts): New function.
4881 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
4882 (ipa_simd_modify_function_body): Likewise.
4884 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4885 Alan Hayward <alan.hayward@arm.com>
4886 David Sherwood <david.sherwood@arm.com>
4888 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
4889 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
4890 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
4891 (expand_vector_condition, vector_element): Likewise.
4892 (subparts_gt): New function.
4893 (get_compute_type): Use subparts_gt.
4894 (count_type_subparts): Delete.
4895 (expand_vector_operations_1): Use subparts_gt instead of
4896 count_type_subparts.
4898 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4899 Alan Hayward <alan.hayward@arm.com>
4900 David Sherwood <david.sherwood@arm.com>
4902 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
4903 (vect_compile_time_alias): ...this new function. Do the calculation
4904 on poly_ints rather than trees.
4905 (vect_prune_runtime_alias_test_list): Update call accordingly.
4907 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4908 Alan Hayward <alan.hayward@arm.com>
4909 David Sherwood <david.sherwood@arm.com>
4911 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
4913 (vect_schedule_slp_instance): Likewise.
4915 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4916 Alan Hayward <alan.hayward@arm.com>
4917 David Sherwood <david.sherwood@arm.com>
4919 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
4920 constant and extern definitions for variable-length vectors.
4921 (vect_get_constant_vectors): Note that the number of units
4922 is known to be constant.
4924 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4925 Alan Hayward <alan.hayward@arm.com>
4926 David Sherwood <david.sherwood@arm.com>
4928 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
4929 of units as polynomial. Choose between WIDE and NARROW based
4932 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4933 Alan Hayward <alan.hayward@arm.com>
4934 David Sherwood <david.sherwood@arm.com>
4936 * tree-vect-stmts.c (simd_clone_subparts): New function.
4937 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
4939 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4940 Alan Hayward <alan.hayward@arm.com>
4941 David Sherwood <david.sherwood@arm.com>
4943 * tree-vect-stmts.c (vectorizable_call): Treat the number of
4944 vectors as polynomial. Use build_index_vector for
4947 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4948 Alan Hayward <alan.hayward@arm.com>
4949 David Sherwood <david.sherwood@arm.com>
4951 * tree-vect-stmts.c (get_load_store_type): Treat the number of
4952 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
4953 for variable-length vectors.
4954 (vectorizable_mask_load_store): Treat the number of units as
4955 polynomial, asserting that it is constant if the condition has
4956 already been enforced.
4957 (vectorizable_store, vectorizable_load): Likewise.
4959 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4960 Alan Hayward <alan.hayward@arm.com>
4961 David Sherwood <david.sherwood@arm.com>
4963 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
4964 of units as polynomial. Punt if we can't tell at compile time
4965 which vector contains the final result.
4967 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4968 Alan Hayward <alan.hayward@arm.com>
4969 David Sherwood <david.sherwood@arm.com>
4971 * tree-vect-loop.c (vectorizable_induction): Treat the number
4972 of units as polynomial. Punt on SLP inductions. Use an integer
4973 VEC_SERIES_EXPR for variable-length integer reductions. Use a
4974 cast of such a series for variable-length floating-point
4977 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4978 Alan Hayward <alan.hayward@arm.com>
4979 David Sherwood <david.sherwood@arm.com>
4981 * tree.h (build_index_vector): Declare.
4982 * tree.c (build_index_vector): New function.
4983 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
4984 of units as polynomial, forcibly converting it to a constant if
4985 vectorizable_reduction has already enforced the condition.
4986 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
4987 to create a {1,2,3,...} vector.
4988 (vectorizable_reduction): Treat the number of units as polynomial.
4989 Choose vectype_in based on the largest scalar element size rather
4990 than the smallest number of units. Enforce the restrictions
4993 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4994 Alan Hayward <alan.hayward@arm.com>
4995 David Sherwood <david.sherwood@arm.com>
4997 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
4998 number of units as polynomial.
5000 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5001 Alan Hayward <alan.hayward@arm.com>
5002 David Sherwood <david.sherwood@arm.com>
5004 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5005 * target.def (autovectorize_vector_sizes): Return the vector sizes
5006 by pointer, using vector_sizes rather than a bitmask.
5007 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5008 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5009 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5011 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5012 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5013 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5014 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5015 * omp-general.c (omp_max_vf): Likewise.
5016 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5017 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5018 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5019 * tree-vect-slp.c (vect_slp_bb): Likewise.
5020 * doc/tm.texi: Regenerate.
5021 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5023 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5024 the vector size as a poly_uint64 rather than an unsigned int.
5025 (current_vector_size): Change from an unsigned int to a poly_uint64.
5026 (get_vectype_for_scalar_type): Update accordingly.
5027 * tree.h (build_truth_vector_type): Take the size and number of
5028 units as a poly_uint64 rather than an unsigned int.
5029 (build_vector_type): Add a temporary overload that takes
5030 the number of units as a poly_uint64 rather than an unsigned int.
5031 * tree.c (make_vector_type): Likewise.
5032 (build_truth_vector_type): Take the number of units as a poly_uint64
5033 rather than an unsigned int.
5035 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5036 Alan Hayward <alan.hayward@arm.com>
5037 David Sherwood <david.sherwood@arm.com>
5039 * target.def (get_mask_mode): Take the number of units and length
5040 as poly_uint64s rather than unsigned ints.
5041 * targhooks.h (default_get_mask_mode): Update accordingly.
5042 * targhooks.c (default_get_mask_mode): Likewise.
5043 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5044 * doc/tm.texi: Regenerate.
5046 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5047 Alan Hayward <alan.hayward@arm.com>
5048 David Sherwood <david.sherwood@arm.com>
5050 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5051 * omp-general.c (omp_max_vf): Likewise.
5052 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5053 (expand_omp_simd): Handle polynomial safelen.
5054 * omp-low.c (omplow_simd_context): Add a default constructor.
5055 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5056 (lower_rec_simd_input_clauses): Update accordingly.
5057 (lower_rec_input_clauses): Likewise.
5059 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5060 Alan Hayward <alan.hayward@arm.com>
5061 David Sherwood <david.sherwood@arm.com>
5063 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5064 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5065 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5066 (vect_analyze_slp_cost): Likewise.
5067 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5068 (vect_model_load_cost): Likewise.
5070 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5071 Alan Hayward <alan.hayward@arm.com>
5072 David Sherwood <david.sherwood@arm.com>
5074 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5075 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5076 from an unsigned int * to a poly_uint64_pod *.
5077 (calculate_unrolling_factor): New function.
5078 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5080 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5081 Alan Hayward <alan.hayward@arm.com>
5082 David Sherwood <david.sherwood@arm.com>
5084 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5085 from an unsigned int to a poly_uint64.
5086 (_loop_vec_info::slp_unrolling_factor): Likewise.
5087 (_loop_vec_info::vectorization_factor): Change from an int
5089 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5090 (vect_get_num_vectors): New function.
5091 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5092 (vect_get_num_copies): Use vect_get_num_vectors.
5093 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5094 to an unsigned int *.
5095 (vect_analyze_data_refs): Change min_vf from an int * to a
5097 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5098 than an unsigned HOST_WIDE_INT.
5099 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5100 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5101 to an unsigned int *.
5102 (vect_analyze_data_ref_dependences): Likewise.
5103 (vect_compute_data_ref_alignment): Handle polynomial vf.
5104 (vect_enhance_data_refs_alignment): Likewise.
5105 (vect_prune_runtime_alias_test_list): Likewise.
5106 (vect_shift_permute_load_chain): Likewise.
5107 (vect_supportable_dr_alignment): Likewise.
5108 (dependence_distance_ge_vf): Take the vectorization factor as a
5109 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5110 (vect_analyze_data_refs): Change min_vf from an int * to a
5112 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5113 vfm1 as a poly_uint64 rather than an int. Make the same change
5114 for the returned bound_scalar.
5115 (vect_gen_vector_loop_niters): Handle polynomial vf.
5116 (vect_do_peeling): Likewise. Update call to
5117 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5118 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5120 * tree-vect-loop.c (vect_determine_vectorization_factor)
5121 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5122 (vect_get_known_peeling_cost): Likewise.
5123 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5124 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5125 (vect_transform_loop): Likewise. Use the lowest possible VF when
5126 updating the upper bounds of the loop.
5127 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5129 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5130 polynomial unroll factors.
5131 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5132 (vect_make_slp_decision): Likewise.
5133 (vect_supported_load_permutation_p): Likewise, and polynomial
5135 (vect_analyze_slp_cost): Handle polynomial vf.
5136 (vect_slp_analyze_node_operations): Likewise.
5137 (vect_slp_analyze_bb_1): Likewise.
5138 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5139 than an unsigned HOST_WIDE_INT.
5140 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5141 (vectorizable_load): Handle polynomial vf.
5142 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5144 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5146 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5147 Alan Hayward <alan.hayward@arm.com>
5148 David Sherwood <david.sherwood@arm.com>
5150 * match.pd: Handle bit operations involving three constants
5151 and try to fold one pair.
5153 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5155 * tree-vect-loop-manip.c: Include gimple-fold.h.
5156 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5157 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5158 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5159 Add a path that uses a step of VF instead of 1, but disable it
5161 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5162 and niters_no_overflow parameters. Update calls to
5163 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5164 Create a new SSA name if the latter choses to use a ste other
5165 than zero, and return it via niters_vector_mult_vf_var.
5166 * tree-vect-loop.c (vect_transform_loop): Update calls to
5167 vect_do_peeling, vect_gen_vector_loop_niters and
5168 slpeel_make_loop_iterate_ntimes.
5169 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5170 (vect_gen_vector_loop_niters): Update declarations after above changes.
5172 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5174 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5175 128-bit round to integer instructions.
5176 (ceil<mode>2): Likewise.
5177 (btrunc<mode>2): Likewise.
5178 (round<mode>2): Likewise.
5180 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5182 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5183 unaligned VSX load/store on P8/P9.
5184 (expand_block_clear): Allow the use of unaligned VSX
5185 load/store on P8/P9.
5187 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5189 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5191 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5192 swap associated with both a load and a store.
5194 2018-01-02 Andrew Waterman <andrew@sifive.com>
5196 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5197 * config/riscv/riscv.md (clear_cache): Use it.
5199 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5201 * web.c: Remove out-of-date comment.
5203 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5205 * expr.c (fixup_args_size_notes): Check that any existing
5206 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5207 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5208 (emit_single_push_insn): ...here.
5210 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5212 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5213 (const_vector_encoded_nelts): New function.
5214 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5215 (const_vector_int_elt, const_vector_elt): Declare.
5216 * emit-rtl.c (const_vector_int_elt_1): New function.
5217 (const_vector_elt): Likewise.
5218 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5219 of CONST_VECTOR_ELT.
5221 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5223 * expr.c: Include rtx-vector-builder.h.
5224 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5225 directly on the tree encoding.
5226 (const_vector_from_tree): Likewise.
5227 * optabs.c: Include rtx-vector-builder.h.
5228 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5229 sequence of "u" values.
5230 * vec-perm-indices.c: Include rtx-vector-builder.h.
5231 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5232 directly on the vec_perm_indices encoding.
5234 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5236 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5237 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5238 * rtx-vector-builder.h: New file.
5239 * rtx-vector-builder.c: Likewise.
5240 * rtl.h (rtx_def::u2): Add a const_vector field.
5241 (CONST_VECTOR_NPATTERNS): New macro.
5242 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5243 (CONST_VECTOR_DUPLICATE_P): Likewise.
5244 (CONST_VECTOR_STEPPED_P): Likewise.
5245 (CONST_VECTOR_ENCODED_ELT): Likewise.
5246 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5247 (unwrap_const_vec_duplicate): Likewise.
5248 (const_vec_series_p): Check for a non-duplicated vector encoding.
5249 Say that the function only returns true for integer vectors.
5250 * emit-rtl.c: Include rtx-vector-builder.h.
5251 (gen_const_vec_duplicate_1): Delete.
5252 (gen_const_vector): Call gen_const_vec_duplicate instead of
5253 gen_const_vec_duplicate_1.
5254 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5255 (gen_const_vec_duplicate): Use rtx_vector_builder.
5256 (gen_const_vec_series): Likewise.
5257 (gen_rtx_CONST_VECTOR): Likewise.
5258 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5259 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5260 Build a new vector rather than modifying a CONST_VECTOR in-place.
5261 (handle_special_swappables): Update call accordingly.
5262 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5263 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5264 Build a new vector rather than modifying a CONST_VECTOR in-place.
5265 (handle_special_swappables): Update call accordingly.
5267 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5269 * simplify-rtx.c (simplify_const_binary_operation): Use
5270 CONST_VECTOR_ELT instead of XVECEXP.
5272 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5274 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5275 the selector elements to be different from the data elements
5276 if the selector is a VECTOR_CST.
5277 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5278 ssizetype for the selector.
5280 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5282 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5283 before testing each element individually.
5284 * tree-vect-generic.c (lower_vec_perm): Likewise.
5286 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5288 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5289 * selftest-run-tests.c (selftest::run_tests): Call it.
5290 * vector-builder.h (vector_builder::operator ==): New function.
5291 (vector_builder::operator !=): Likewise.
5292 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5293 (vec_perm_indices::all_from_input_p): New function.
5294 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5295 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5296 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5297 instead of reading the VECTOR_CST directly. Detect whether both
5298 vector inputs are the same before constructing the vec_perm_indices,
5299 and update the number of inputs argument accordingly. Use the
5300 utility functions added above. Only construct sel2 if we need to.
5302 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5304 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5305 the broadcast of the low byte.
5306 (expand_mult_highpart): Use an explicit encoding for the permutes.
5307 * optabs-query.c (can_mult_highpart_p): Likewise.
5308 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5309 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5310 (vectorizable_bswap): Likewise.
5311 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5312 explicit encoding for the power-of-2 permutes.
5313 (vect_permute_store_chain): Likewise.
5314 (vect_grouped_load_supported): Likewise.
5315 (vect_permute_load_chain): Likewise.
5317 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5319 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5320 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5321 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5322 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5323 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5324 (vect_gen_perm_mask_any): Likewise.
5326 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5328 * int-vector-builder.h: New file.
5329 * vec-perm-indices.h: Include int-vector-builder.h.
5330 (vec_perm_indices): Redefine as an int_vector_builder.
5331 (auto_vec_perm_indices): Delete.
5332 (vec_perm_builder): Redefine as a stand-alone class.
5333 (vec_perm_indices::vec_perm_indices): New function.
5334 (vec_perm_indices::clamp): Likewise.
5335 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5336 (vec_perm_indices::new_vector): New function.
5337 (vec_perm_indices::new_expanded_vector): Update for new
5338 vec_perm_indices class.
5339 (vec_perm_indices::rotate_inputs): New function.
5340 (vec_perm_indices::all_in_range_p): Operate directly on the
5341 encoded form, without computing elided elements.
5342 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5343 encoding. Update for new vec_perm_indices class.
5344 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5345 the given vec_perm_builder.
5346 (expand_vec_perm_var): Update vec_perm_builder constructor.
5347 (expand_mult_highpart): Use vec_perm_builder instead of
5348 auto_vec_perm_indices.
5349 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5350 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5351 or double series encoding as appropriate.
5352 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5353 vec_perm_indices instead of auto_vec_perm_indices.
5354 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5355 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5356 (vect_permute_store_chain): Likewise.
5357 (vect_grouped_load_supported): Likewise.
5358 (vect_permute_load_chain): Likewise.
5359 (vect_shift_permute_load_chain): Likewise.
5360 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5361 (vect_transform_slp_perm_load): Likewise.
5362 (vect_schedule_slp_instance): Likewise.
5363 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5364 (vectorizable_mask_load_store): Likewise.
5365 (vectorizable_bswap): Likewise.
5366 (vectorizable_store): Likewise.
5367 (vectorizable_load): Likewise.
5368 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5369 vec_perm_indices instead of auto_vec_perm_indices. Use
5370 tree_to_vec_perm_builder to read the vector from a tree.
5371 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5372 vec_perm_builder instead of a vec_perm_indices.
5373 (have_whole_vector_shift): Use vec_perm_builder and
5374 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5375 truncation to calc_vec_perm_mask_for_shift.
5376 (vect_create_epilog_for_reduction): Likewise.
5377 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5378 from auto_vec_perm_indices to vec_perm_indices.
5379 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5380 instead of changing individual elements.
5381 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5382 the vector in d.perm.
5383 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5384 from auto_vec_perm_indices to vec_perm_indices.
5385 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5386 instead of changing individual elements.
5387 (arm_vectorize_vec_perm_const): Use new_vector to install
5388 the vector in d.perm.
5389 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5390 Update vec_perm_builder constructor.
5391 (rs6000_expand_interleave): Likewise.
5392 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5393 (rs6000_expand_interleave): Likewise.
5395 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5397 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5398 to qimode could truncate the indices.
5399 * optabs.c (expand_vec_perm_var): Likewise.
5401 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5403 * Makefile.in (OBJS): Add vec-perm-indices.o.
5404 * vec-perm-indices.h: New file.
5405 * vec-perm-indices.c: Likewise.
5406 * target.h (vec_perm_indices): Replace with a forward class
5408 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5409 * optabs.h: Include vec-perm-indices.h.
5410 (expand_vec_perm): Delete.
5411 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5412 (expand_vec_perm_const): Declare.
5413 * target.def (vec_perm_const_ok): Replace with...
5414 (vec_perm_const): ...this new hook.
5415 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5416 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5417 * doc/tm.texi: Regenerate.
5418 * optabs.def (vec_perm_const): Delete.
5419 * doc/md.texi (vec_perm_const): Likewise.
5420 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5421 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5422 expand_vec_perm for constant permutation vectors. Assert that
5423 the mode of variable permutation vectors is the integer equivalent
5424 of the mode that is being permuted.
5425 * optabs-query.h (selector_fits_mode_p): Declare.
5426 * optabs-query.c: Include vec-perm-indices.h.
5427 (selector_fits_mode_p): New function.
5428 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5429 is defined, instead of checking whether the vec_perm_const_optab
5430 exists. Use targetm.vectorize.vec_perm_const instead of
5431 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5432 fit in the vector mode before using a variable permute.
5433 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5434 vec_perm_indices instead of an rtx.
5435 (expand_vec_perm): Replace with...
5436 (expand_vec_perm_const): ...this new function. Take the selector
5437 as a vec_perm_indices rather than an rtx. Also take the mode of
5438 the selector. Update call to shift_amt_for_vec_perm_mask.
5439 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5440 Use vec_perm_indices::new_expanded_vector to expand the original
5441 selector into bytes. Check whether the indices fit in the vector
5442 mode before using a variable permute.
5443 (expand_vec_perm_var): Make global.
5444 (expand_mult_highpart): Use expand_vec_perm_const.
5445 * fold-const.c: Includes vec-perm-indices.h.
5446 * tree-ssa-forwprop.c: Likewise.
5447 * tree-vect-data-refs.c: Likewise.
5448 * tree-vect-generic.c: Likewise.
5449 * tree-vect-loop.c: Likewise.
5450 * tree-vect-slp.c: Likewise.
5451 * tree-vect-stmts.c: Likewise.
5452 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5454 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5455 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5456 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5457 (aarch64_vectorize_vec_perm_const): ...this new function.
5458 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5459 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5460 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5461 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5462 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5463 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5464 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5466 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5467 check for NEON modes.
5468 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5469 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5470 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5471 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5473 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5474 the old VEC_PERM_CONST conditions.
5475 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5476 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5477 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5478 (ia64_vectorize_vec_perm_const_ok): Merge into...
5479 (ia64_vectorize_vec_perm_const): ...this new function.
5480 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5481 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5482 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5483 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5484 * config/mips/mips.c (mips_expand_vec_perm_const)
5485 (mips_vectorize_vec_perm_const_ok): Merge into...
5486 (mips_vectorize_vec_perm_const): ...this new function.
5487 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5488 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5489 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5490 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5491 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5492 (rs6000_expand_vec_perm_const): Delete.
5493 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5495 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5496 (altivec_expand_vec_perm_const_le): Take each operand individually.
5497 Operate on constant selectors rather than rtxes.
5498 (altivec_expand_vec_perm_const): Likewise. Update call to
5499 altivec_expand_vec_perm_const_le.
5500 (rs6000_expand_vec_perm_const): Delete.
5501 (rs6000_vectorize_vec_perm_const_ok): Delete.
5502 (rs6000_vectorize_vec_perm_const): New function.
5503 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5504 an element count and rtx array.
5505 (rs6000_expand_extract_even): Update call accordingly.
5506 (rs6000_expand_interleave): Likewise.
5507 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5508 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5509 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5510 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5511 (rs6000_expand_vec_perm_const): Delete.
5512 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5513 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5514 (altivec_expand_vec_perm_const_le): Take each operand individually.
5515 Operate on constant selectors rather than rtxes.
5516 (altivec_expand_vec_perm_const): Likewise. Update call to
5517 altivec_expand_vec_perm_const_le.
5518 (rs6000_expand_vec_perm_const): Delete.
5519 (rs6000_vectorize_vec_perm_const_ok): Delete.
5520 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5521 reference to the SPE evmerge intructions.
5522 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5523 an element count and rtx array.
5524 (rs6000_expand_extract_even): Update call accordingly.
5525 (rs6000_expand_interleave): Likewise.
5526 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5527 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5529 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5531 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5533 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5534 vector mode and that that mode matches the mode of the data
5536 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5537 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5538 directly using expand_vec_perm_1 when forcing selectors into
5540 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5542 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5544 * optabs-query.h (can_vec_perm_p): Delete.
5545 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5546 * optabs-query.c (can_vec_perm_p): Split into...
5547 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5548 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5549 particular selector is valid.
5550 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5551 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5552 (vect_grouped_load_supported): Likewise.
5553 (vect_shift_permute_load_chain): Likewise.
5554 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5555 (vect_transform_slp_perm_load): Likewise.
5556 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5557 (vectorizable_bswap): Likewise.
5558 (vect_gen_perm_mask_checked): Likewise.
5559 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5560 implementations of variable permutation vectors into account
5561 when deciding which selector to use.
5562 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5563 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5564 with a false third argument.
5565 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5566 to test whether the constant selector is valid and can_vec_perm_var_p
5567 to test whether a variable selector is valid.
5569 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5571 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5572 * optabs-query.c (can_vec_perm_p): Likewise.
5573 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5574 instead of vec_perm_indices.
5575 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5576 (vect_gen_perm_mask_checked): Likewise,
5577 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5578 (vect_gen_perm_mask_checked): Likewise,
5580 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5582 * optabs-query.h (qimode_for_vec_perm): Declare.
5583 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5584 (qimode_for_vec_perm): ...this new function.
5585 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5587 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5589 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5590 does not have a conditional at the top.
5592 2018-01-02 Richard Biener <rguenther@suse.de>
5594 * ipa-inline.c (big_speedup_p): Fix expression.
5596 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5599 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5602 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5606 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5607 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5608 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5609 cond_taken_branch_cost 3->4.
5611 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5613 PR tree-optimization/83581
5614 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5615 TODO_cleanup_cfg if any changes have been made.
5618 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5619 convert_modes if target mode has the right side, but different mode
5623 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5624 last argument when extracting from CONCAT. If either from_real or
5625 from_imag is NULL, use expansion through memory. If result is not
5626 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5627 the parts directly to inner mode, if even that fails, use expansion
5631 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5632 check for bswap in mode rather than HImode and use that in expand_unop
5635 Copyright (C) 2018 Free Software Foundation, Inc.
5637 Copying and distribution of this file, with or without modification,
5638 are permitted in any medium without royalty provided the copyright
5639 notice and this notice are preserved.