1 2017-01-30 Jeff Law <law@redhat.com>
4 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
5 INT_REGISTERS_SAVED. Check it prior to calling
6 get_scratch_register_on_entry.
7 (ix86_adjust_stack_and_probe): Similarly.
8 (ix86_emit_probe_stack_range): Similarly.
9 (ix86_expand_prologue): Corresponding changes.
11 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
14 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
15 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
17 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
20 * lra-constraints.c (curr_insn_transform): Process AND in the
23 2018-01-30 Jakub Jelinek <jakub@redhat.com>
25 PR rtl-optimization/83986
26 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
27 dependence against last_pending_memory_flush in addition to
30 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
32 PR tree-optimization/81611
33 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
36 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
38 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
41 2018-01-30 Richard Biener <rguenther@suse.de>
42 Jakub Jelinek <jakub@redhat.com>
44 PR tree-optimization/84111
45 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
46 inner loops added during recursion, as they don't have up-to-date
49 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
52 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
53 (can_inline_edge_by_limits_p): ... here.
54 (can_early_inline_edge_p, check_callers,
55 update_caller_keys, update_callee_keys, recursive_inlining,
56 add_new_edges_to_heap, speculation_useful_p,
57 inline_small_functions,
58 inline_small_functions, flatten_function,
59 inline_to_all_callers_1): Update.
61 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
63 * profile-count.c (profile_count::combine_with_ipa_count): Handle
66 2018-01-30 Richard Biener <rguenther@suse.de>
68 PR tree-optimization/83008
69 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
70 invariant and constant vector uses in stmts when they need
73 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
76 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
77 * configure: Regenerate.
79 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
81 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
83 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
84 Use gen_rtx_REG rather than gen_lowpart.
86 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
88 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
89 rather than 0 when creating partial subregs.
91 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
93 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
96 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
99 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
100 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
101 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
102 flags. This restores the settings used before the 2017-07-24.
103 Turning off pre increment/decrement/modify allows IVOPTS to
104 optimize DF/SF loops where the index is an int.
106 2018-01-29 Richard Biener <rguenther@suse.de>
107 Kelvin Nilsen <kelvin@gcc.gnu.org>
110 * tree-vect-stmts.c (vectorizable_call): Don't call
111 targetm.vectorize_builtin_md_vectorized_function if callee is
114 2018-01-22 Carl Love <cel@us.ibm.com>
116 * doc/extend.tex: Fix typo in second arg in
117 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
119 2018-01-29 Richard Biener <rguenther@suse.de>
121 PR tree-optimization/84086
122 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
123 (flush_ssaname_freelist): When SSA names were released reset
126 2018-01-29 Richard Biener <rguenther@suse.de>
128 PR tree-optimization/84057
129 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
130 removed paths when removing edges.
132 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
134 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
135 -mfunction-return=@var{choice}.
137 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
140 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
142 (layout::print_source_line): Likewise.
143 (test_get_line_width_without_trailing_whitespace): Add test cases.
145 2018-01-27 Jakub Jelinek <jakub@redhat.com>
148 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
151 2018-01-26 Jim Wilson <jimw@sifive.com>
153 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
155 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
158 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
160 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
161 and CMP + SUB-immediate -> SUBS.
163 2018-01-26 Martin Sebor <msebor@redhat.com>
165 PR tree-optimization/83896
166 * tree-ssa-strlen.c (get_string_len): Rename...
167 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
168 Avoid assuming length is constant.
169 (handle_char_store): Use HOST_WIDE_INT for string length.
171 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
174 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
175 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
177 2018-01-26 Richard Biener <rguenther@suse.de>
179 PR rtl-optimization/84003
180 * dse.c (record_store): Only record redundant stores when
181 the earlier store aliases at least all accesses the later one does.
183 2018-01-26 Jakub Jelinek <jakub@redhat.com>
185 PR rtl-optimization/83985
186 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
187 REG_CFA_RESTORE insns.
188 (delete_unmarked_insns): Don't ignore separate shrink wrapping
189 REG_CFA_RESTORE insns here.
192 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
193 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
195 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
197 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
198 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
199 (arc_init): Likewise.
200 (arc_override_options): Likewise.
201 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
203 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
205 * config/arc/arc.h (TARGET_DBNZ): Define.
206 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
207 properly set the tune attribute.
208 (dbnz): Use TARGET_DBNZ guard.
209 * config/arc/arc.opt (mtune): Add core3 option.
211 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
213 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
214 recognize new pic like addresses.
215 (arc_delegitimize_address): Clean up.
217 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
219 * config/arc/arc-arches.def: Option mrf16 valid for all
221 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
222 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
223 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
224 * config/arc/arc-tables.opt: Regenerate.
225 * config/arc/arc.c (arc_conditional_register_usage): Handle
226 reduced register file case.
227 (arc_file_start): Set must have build attributes.
228 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
230 * config/arc/arc.opt (mrf16): Add new option.
231 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
232 * config/arc/genmultilib.awk: Handle new mrf16 option.
233 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
234 * config/arc/t-multilib: Regenerate.
235 * doc/invoke.texi (ARC Options): Document mrf16 option.
237 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
239 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
240 * config/arc/arc.c (arc_handle_secure_attribute): New function.
241 (arc_attribute_table): Add 'secure_call' attribute.
242 (arc_print_operand): Print secure call operand.
243 (arc_function_ok_for_sibcall): Don't optimize tail calls when
245 (arc_is_secure_call_p): New function. * config/arc/arc.md
246 (call_i): Add support for sjli instruction.
247 (call_value_i): Likewise.
248 * config/arc/constraints.md (Csc): New constraint.
250 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
251 John Eric Martin <John.Martin@emmicro-us.com>
253 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
254 * config/arc/arc.c (_arc_jli_section): New struct.
255 (arc_jli_section): New type.
256 (rc_jli_sections): New static variable.
257 (arc_handle_jli_attribute): New function.
258 (arc_attribute_table): Add jli_always and jli_fixed attribute.
259 (arc_file_end): New function.
260 (TARGET_ASM_FILE_END): Define.
261 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
262 (arc_add_jli_section): New function.
263 (jli_call_scan): Likewise.
264 (arc_reorg): Call jli_call_scan.
265 (arc_output_addsi): Remove 'S' from printing asm operand.
266 (arc_is_jli_call_p): New function.
267 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
269 (movhi_insn): Likewise.
270 (movsi_insn): Likewise.
271 (movsi_set_cc_insn): Likewise.
272 (loadqi_update): Likewise.
273 (load_zeroextendqisi_update): Likewise.
274 (load_signextendqisi_update): Likewise.
275 (loadhi_update): Likewise.
276 (load_zeroextendhisi_update): Likewise.
277 (load_signextendhisi_update): Likewise.
278 (loadsi_update): Likewise.
279 (loadsf_update): Likewise.
280 (movsicc_insn): Likewise.
281 (bset_insn): Likewise.
282 (bxor_insn): Likewise.
283 (bclr_insn): Likewise.
284 (bmsk_insn): Likewise.
285 (bicsi3_insn): Likewise.
286 (cmpsi_cc_c_insn): Likewise.
287 (movsi_ne): Likewise.
288 (movsi_cond_exec): Likewise.
289 (clrsbsi2): Likewise.
300 (call_i): Remove 'S' asm letter, add jli instruction.
301 (call_value_i): Likewise.
302 * config/arc/arc.op (mjli-always): New option.
303 * config/arc/constraints.md (Cji): New constraint.
304 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
306 (subsf3_fpx): Likewise.
307 (mulsf3_fpx): Likewise.
308 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
310 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
312 * doc/invoke.texi (ARC): Document mjli-always option.
314 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
316 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
317 and use incw and decw where possible
318 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
320 2018-01-26 Richard Biener <rguenther@suse.de>
322 PR tree-optimization/81082
323 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
324 association if it requires casting to unsigned.
325 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
326 from fold_plusminus_mult_expr to catch important cases late when
327 range info is available.
329 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
331 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
332 * configure.ac (hidden_linkonce): New test.
333 * configure: Regenerate.
334 * config.in: Regenerate.
336 2018-01-26 Julia Koval <julia.koval@intel.com>
338 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
339 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
340 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
341 _mm_mask_bitshuffle_epi64_mask): Fix type.
342 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
343 USI_FTYPE_V4DI_V4DI_USI): Remove.
344 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
345 __builtin_ia32_vpshufbitqmb256_mask,
346 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
347 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
348 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
350 2018-01-26 Alan Modra <amodra@gmail.com>
353 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
354 UNSPEC_VBPERMQ. Sort other unspecs.
356 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
358 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
360 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
363 * predict.c (drop_profile): Do not push/pop cfun; update also
365 (handle_missing_profiles): Fix logic looking for zero profiles.
367 2018-01-25 Jakub Jelinek <jakub@redhat.com>
370 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
371 on functions with #pragma omp declare simd or functions with simd
373 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
374 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
375 Remove trailing \n from warning_at calls.
377 2018-01-25 Tom de Vries <tom@codesourcery.com>
380 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
381 for neutered workers.
383 2018-01-24 Joseph Myers <joseph@codesourcery.com>
386 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
387 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
389 2017-01-24 Jeff Law <law@redhat.com>
392 * i386.c (get_probe_interval): Move to earlier point.
393 (ix86_compute_frame_layout): If -fstack-clash-protection and
394 the frame is larger than the probe interval, then use pushes
395 to save registers rather than reg->mem moves.
396 (ix86_expand_prologue): Remove conditional for int_registers_saved
399 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
402 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
403 min/max for never referenced object.
405 2018-01-24 Jakub Jelinek <jakub@redhat.com>
408 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
410 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
411 attributes from DECL_ATTRIBUTES (decl) without affecting
412 DECL_ATTRIBUTES (current_function_decl).
413 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
414 functions with non-NULL DECL_ABSTRACT_ORIGIN.
416 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
418 PR tree-optimization/83979
419 * fold-const.c (fold_comparison): Use constant_boolean_node
420 instead of boolean_{true,false}_node.
422 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
424 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
427 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
429 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
430 Simplify the clause that sets the length attribute.
431 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
432 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
433 clause that sets the length attribute.
434 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
436 2018-01-24 Tom de Vries <tom@codesourcery.com>
439 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
440 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
441 Add strict parameter.
442 (prevent_branch_around_nothing): Insert dummy insn between branch to
443 label and label with no ptx insn inbetween.
444 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
446 2018-01-24 Tom de Vries <tom@codesourcery.com>
449 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
450 for neutered threads in warp.
451 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
453 2018-01-24 Richard Biener <rguenther@suse.de>
455 PR tree-optimization/83176
456 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
459 2018-01-24 Richard Biener <rguenther@suse.de>
461 PR tree-optimization/82819
462 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
463 code generating pluses that are no-ops in the target precision.
465 2018-01-24 Richard Biener <rguenther@suse.de>
468 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
470 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
472 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
473 to merge probabilities.
474 * predict.c (probably_never_executed): Also mark as cold functions
475 with global 0 profile and guessed local profile.
476 * profile-count.c (profile_probability::combine_with_count): New
478 * profile-count.h (profile_probability::operator*,
479 profile_probability::operator*=, profile_probability::operator/,
480 profile_probability::operator/=): Reduce precision to adjusted
481 and set value to guessed on contradictory divisions.
482 (profile_probability::combine_with_freq): Remove.
483 (profile_probability::combine_wiht_count): Declare.
484 (profile_count::force_nonzero):: Set to adjusted.
485 (profile_count::probability_in):: Set quality to adjusted.
486 * tree-ssa-tail-merge.c (replace_block_by): Use
489 2018-01-23 Andrew Waterman <andrew@sifive.com>
490 Jim Wilson <jimw@sifive.com>
492 * config/riscv/riscv.c (riscv_stack_boundary): New.
493 (riscv_option_override): Set riscv_stack_boundary. Handle
494 riscv_preferred_stack_boundary_arg.
495 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
496 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
497 (STACK_BOUNDARY): Set to riscv_stack_boundary.
498 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
499 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
500 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
502 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
505 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
506 of struct ix86_frame.
507 (ix86_expand_epilogue): Likewise. Add a local variable for
508 the reg_save_offset field in struct ix86_frame.
510 2018-01-23 Bin Cheng <bin.cheng@arm.com>
512 PR tree-optimization/82604
513 * tree-loop-distribution.c (enum partition_kind): New enum item
514 PKIND_PARTIAL_MEMSET.
515 (partition_builtin_p): Support above new enum item.
516 (generate_code_for_partition): Ditto.
517 (compute_access_range): Differentiate cases that equality can be
518 proven at all loops, the innermost loops or no loops.
519 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
520 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
521 (finalize_partitions, distribute_loop): Don't fuse partition of
522 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
523 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
526 2018-01-23 Martin Liska <mliska@suse.cz>
528 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
529 order to ignore the predictor.
530 (PRED_POLYMORPHIC_CALL): Likewise.
531 (PRED_RECURSIVE_CALL): Likewise.
533 2018-01-23 Martin Liska <mliska@suse.cz>
535 * tree-profile.c (tree_profiling): Print function header to
536 aware reader which function we are working on.
537 * value-prof.c (gimple_find_values_to_profile): Do not print
538 not interesting value histograms.
540 2018-01-23 Martin Liska <mliska@suse.cz>
542 * profile-count.h (enum profile_quality): Add
543 profile_uninitialized as the first value. Do not number values
544 as they are zero based.
545 (profile_count::verify): Update sanity check.
546 (profile_probability::verify): Likewise.
548 2018-01-23 Nathan Sidwell <nathan@acm.org>
550 * doc/invoke.texi (ffor-scope): Deprecate.
552 2018-01-23 David Malcolm <dmalcolm@redhat.com>
554 PR tree-optimization/83510
555 * domwalk.c (set_all_edges_as_executable): New function.
556 (dom_walker::dom_walker): Convert bool param
557 "skip_unreachable_blocks" to enum reachability. Move setup of
558 edge flags to set_all_edges_as_executable and only do it when
559 reachability is REACHABLE_BLOCKS.
560 * domwalk.h (enum dom_walker::reachability): New enum.
561 (dom_walker::dom_walker): Convert bool param
562 "skip_unreachable_blocks" to enum reachability.
563 (set_all_edges_as_executable): New decl.
564 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
565 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
567 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
568 but converting true to REACHABLE_BLOCKS.
569 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
571 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
572 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
573 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
575 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
576 if check_all_array_refs will be called.
578 2018-01-23 David Malcolm <dmalcolm@redhat.com>
580 * tree.c (selftest::test_location_wrappers): Add more test
583 2018-01-23 David Malcolm <dmalcolm@redhat.com>
585 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
586 (selftest::test_bit_in_range): Likewise.
588 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
591 * doc/sourcebuild.texi (vect_float): Say that the selector
592 only describes the situation when -funsafe-math-optimizations is on.
593 (vect_float_strict): Document.
595 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
597 PR tree-optimization/83965
598 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
599 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
600 instead of checking only for a reduction.
601 (vect_recog_widen_sum_pattern): Likewise.
603 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
605 * predict.c (probably_never_executed): Only use precise profile info.
606 (compute_function_frequency): Skip after inlining hack since we now
607 have quality checking.
609 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
611 * profile-count.h (profile_probability::very_unlikely,
612 profile_probability::unlikely, profile_probability::even): Set
613 precision to guessed.
615 2018-01-23 Richard Biener <rguenther@suse.de>
617 PR tree-optimization/83963
618 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
619 Properly terminate dominator walk when crossing the exit edge not
620 when visiting its source block.
622 2018-01-23 Jakub Jelinek <jakub@redhat.com>
625 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
626 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
628 2018-01-22 Jakub Jelinek <jakub@redhat.com>
630 PR tree-optimization/83957
631 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
632 semicolon after for body surrounded by braces.
634 PR tree-optimization/83081
635 * profile-count.h (profile_probability::split): New method.
636 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
637 Use profile_probability::split.
638 (do_compare_rtx_and_jump): Fix adjustment of probabilities
639 when splitting a single conditional jump into 2.
641 2018-01-22 David Malcolm <dmalcolm@redhat.com>
643 PR tree-optimization/69452
644 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
647 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
649 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
650 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
651 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
653 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
655 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
656 * config/rl78/rl78.md: New define_expand "movdi"
657 * config/rl78/rl78.c: New function definition rl78_split_movdi
659 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
662 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
664 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
665 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
666 128-bit to produce an UNSPEC move to get the double word with the
667 signbit and then a shift directly to do signbit.
668 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
669 implementation with a new version that just does either a direct
670 move or a regular move. Move memory interface to separate insns.
671 Move insns so they are next to the expander.
672 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
673 with signbit move. Split big and little endian case.
674 (signbit<mode>2_dm_mem_le): Likewise.
675 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
676 (signbit<mode>2_dm2): Likewise.
678 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
680 * config/rl78/rl78.md: New define_expand "anddi3".
682 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
684 * config/rl78/rl78.md: New define_expand "umindi3".
686 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
688 * config/rl78/rl78.md: New define_expand "smindi3".
690 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
692 * config/rl78/rl78.md: New define_expand "smaxdi3".
694 2018-01-22 Carl Love <cel@us.ibm.com>
696 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
697 LVX_V1TI): Add macro expansion.
698 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
699 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
700 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
701 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
702 Change check to determine if the instruction is a byte reversing
703 entry. Fix typo in comment.
704 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
705 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
706 Add def_builtin calls for new builtins.
707 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
708 Add define_insn expansion.
710 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
712 * config/rl78/rl78.md: New define_expand "umaxdi3".
714 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
716 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
717 for non-QImode registers
719 2018-01-22 Richard Biener <rguenther@suse.de>
721 PR tree-optimization/83963
722 * graphite-scop-detection.c (scop_detection::get_sese): Delay
723 including the loop exit block.
724 (scop_detection::merge_sese): Likewise.
725 (scop_detection::add_scop): Do it here instead.
727 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
729 * doc/sourcebuild.texi (arm_softfloat): Document.
731 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
734 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
735 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
736 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
738 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
739 David Edelsohn <dje.gcc@gmail.com>
742 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
743 Change "crset eq" to "crset 2".
744 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
745 (*call_indirect_aix<mode>_nospec): Likewise.
746 (*call_value_indirect_aix<mode>_nospec): Likewise.
747 (*call_indirect_elfv2<mode>_nospec): Likewise.
748 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
749 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
750 change assembly output from . to $.
751 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
752 (indirect_jump<mode>_nospec): Change assembly output from . to $.
753 (*tablejump<mode>_internal1_nospec): Likewise.
755 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
758 * config/sh/sh_optimize_sett_clrt.cc:
759 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
761 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
763 PR tree-optimization/83940
764 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
765 offset_dt to vect_constant_def rather than vect_unknown_def_type.
766 (vect_check_load_store_mask): Add a mask_dt_out parameter and
767 use it to pass back the definition type.
768 (vect_check_store_rhs): Likewise rhs_dt_out.
769 (vect_build_gather_load_calls): Add a mask_dt argument and use
770 it instead of a call to vect_is_simple_use.
771 (vectorizable_store): Update calls to vect_check_load_store_mask
772 and vect_check_store_rhs. Use the dt returned by the latter instead
773 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
774 instead of calls to vect_is_simple_use. Pass the scalar rather
775 than the vector operand to vect_is_simple_use when handling
776 second and subsequent copies of an rhs value.
777 (vectorizable_load): Update calls to vect_check_load_store_mask
778 and vect_build_gather_load_calls. Use the cached mask_dt and
779 gs_info.offset_dt instead of calls to vect_is_simple_use.
781 2018-01-20 Jakub Jelinek <jakub@redhat.com>
784 * tree-emutls.c: Include gimplify.h.
785 (lower_emutls_2): New function.
786 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
787 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
788 it before further processing.
791 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
792 UINTVAL (trueop1) instead of INTVAL (op1).
794 2018-01-19 Jakub Jelinek <jakub@redhat.com>
798 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
799 INCOMING_FRAME_SP_OFFSET if not defined.
800 (scan_trace): Add ENTRY argument. If true and
801 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
802 emit a note to adjust the CFA offset.
803 (create_cfi_notes): Adjust scan_trace callers.
804 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
805 INCOMING_FRAME_SP_OFFSET in the CIE.
806 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
807 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
809 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
810 * doc/tm.texi: Regenerated.
812 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
814 PR rtl-optimization/83147
815 * lra-constraints.c (remove_inheritance_pseudos): Use
816 lra_substitute_pseudo_within_insn.
818 2018-01-19 Tom de Vries <tom@codesourcery.com>
819 Cesar Philippidis <cesar@codesourcery.com>
822 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
824 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
827 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
828 spaces for function labels.
830 2018-01-19 Martin Liska <mliska@suse.cz>
832 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
833 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
834 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
835 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
836 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
837 (PRED_CONST_RETURN): Change from 69 to 65.
838 (PRED_NULL_RETURN): Change from 91 to 71.
839 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
840 (PRED_LOOP_GUARD): Change from 66 to 73.
842 2018-01-19 Martin Liska <mliska@suse.cz>
844 * predict.c (predict_insn_def): Add new assert.
845 (struct branch_predictor): Change type to signed integer.
846 (test_prediction_value_range): Amend test to cover
848 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
849 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
850 (PRED_LOOP_ITERATIONS_MAX): Likewise.
851 (PRED_LOOP_IV_COMPARE): Likewise.
852 * predict.h (PROB_UNINITIALIZED): Define new constant.
854 2018-01-19 Martin Liska <mliska@suse.cz>
856 * predict.c (dump_prediction): Add new format for
857 analyze_brprob.py script which is enabled with -details
859 * profile-count.h (precise_p): New function.
861 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
863 PR tree-optimization/83922
864 * tree-vect-loop.c (vect_verify_full_masking): Return false if
865 there are no statements that need masking.
866 (vect_active_double_reduction_p): New function.
867 (vect_analyze_loop_operations): Use it when handling phis that
868 are not in the loop header.
870 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
872 PR tree-optimization/83914
873 * tree-vect-loop.c (vectorizable_induction): Don't convert
874 init_expr or apply the peeling adjustment for inductions
875 that are nested within the vectorized loop.
877 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
879 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
882 2018-01-18 Jakub Jelinek <jakub@redhat.com>
886 * function.h (gimplify_parameters): Add gimple_seq * argument.
887 * function.c: Include gimple.h and options.h.
888 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
889 for the added local temporaries if needed.
890 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
891 if there are any parameter cleanups, wrap whole body into a
892 try/finally with the cleanups.
894 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
897 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
898 Use GET_MODE_CLASS for scalar floating point.
900 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
904 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
905 Fix call of call_cgraph_insertion_hooks.
907 2018-01-18 Martin Sebor <msebor@redhat.com>
909 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
911 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
914 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
917 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
920 * common.opt: (-ffile-prefix-map): New option.
921 * opts.c (common_handle_option): Defer it.
922 * opts-global.c (handle_common_deferred_options): Handle it.
923 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
924 * file-prefix-map.h: New file.
925 (remap_debug_filename, add_debug_prefix_map): ...here.
926 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
927 * final.c (debug_prefix_map, add_debug_prefix_map
928 remap_debug_filename): Move to...
929 * file-prefix-map.c: New file.
930 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
931 generalize, get rid of alloca(), use strrchr() instead of strchr().
932 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
933 Implement in terms of add_prefix_map().
934 (remap_macro_filename, remap_debug_filename): Implement in term of
936 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
937 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
938 * dbxout.c: Include file-prefix-map.h.
939 * varasm.c: Likewise.
940 * vmsdbgout.c: Likewise.
941 * xcoffout.c: Likewise.
942 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
943 * doc/cppopts.texi (-fmacro-prefix-map): Document.
944 * doc/invoke.texi (-ffile-prefix-map): Document.
945 (-fdebug-prefix-map): Update description.
947 2018-01-18 Martin Liska <mliska@suse.cz>
949 * config/i386/i386.c (indirect_thunk_name): Document that also
951 (output_indirect_thunk): Document why both instructions
952 (pause and lfence) are generated.
954 2018-01-18 Richard Biener <rguenther@suse.de>
956 PR tree-optimization/83887
957 * graphite-scop-detection.c
958 (scop_detection::get_nearest_dom_with_single_entry): Remove.
959 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
960 (scop_detection::merge_sese): Re-implement with a flood-fill
961 algorithm that properly finds a SESE region if it exists.
963 2018-01-18 Jakub Jelinek <jakub@redhat.com>
966 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
967 pointer_diff optimizations use view_convert instead of convert.
969 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
971 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
972 Generate different code for -mno-speculate-indirect-jumps.
973 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
974 (*call_indirect_aix<mode>): Disable for
975 -mno-speculate-indirect-jumps.
976 (*call_indirect_aix<mode>_nospec): New define_insn.
977 (*call_value_indirect_aix<mode>): Disable for
978 -mno-speculate-indirect-jumps.
979 (*call_value_indirect_aix<mode>_nospec): New define_insn.
980 (*sibcall_nonlocal_sysv<mode>): Generate different code for
981 -mno-speculate-indirect-jumps.
982 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
984 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
986 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
987 long double type, set the flags for noting the default long double
988 type, even if we don't pass or return a long double type.
990 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
993 * ipa-inline.c (flatten_function): Do not overwrite final inlining
996 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
998 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
999 support for merge[hl].
1000 (fold_mergehl_helper): New helper function.
1001 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1002 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1003 (altivec_vmrglw_direct): Add xxmrglw insn.
1005 2018-01-17 Andrew Waterman <andrew@sifive.com>
1007 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1008 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1010 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1013 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1014 call the lto_location_cache before reading the
1015 DECL_SOURCE_LOCATION of the types.
1017 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1018 Richard Sandiford <richard.sandiford@linaro.org>
1020 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1021 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1022 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1023 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1024 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1026 * config/aarch64/constraints.md (aarch64_movti_operand):
1028 * config/aarch64/predicates.md (Uti): Add new constraint.
1030 2018-01-17 Carl Love <cel@us.ibm.com>
1031 * config/rs6000/vsx.md (define_expand xl_len_r,
1032 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1033 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1035 (define_expand, define_insn): Move the shift left from the
1036 define_insn to the define_expand for lxvl and stxvl instructions.
1037 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1038 and XL_LEN_R definitions to PURE.
1040 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1042 * config/i386/i386.c (indirect_thunk_name): Declare regno
1043 as unsigned int. Compare regno with INVALID_REGNUM.
1044 (output_indirect_thunk): Ditto.
1045 (output_indirect_thunk_function): Ditto.
1046 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1047 in the call to output_indirect_thunk_function.
1049 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1052 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1053 rather than the size of inner_type to determine the stack slot size
1054 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1056 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1059 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1062 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1064 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1065 endian Linux systems to optionally enable multilibs for selecting
1066 the long double type if the user configured an explicit type.
1067 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1068 have no long double multilibs if not defined.
1069 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1070 warn if the user used -mabi={ieee,ibm}longdouble and we built
1071 multilibs for long double.
1072 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1073 appropriate multilib option.
1074 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1076 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1077 for building long double multilibs.
1078 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1080 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1082 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1085 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1087 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1090 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1093 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1096 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1098 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1099 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1101 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1103 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1104 different rtl trees depending on TARGET_64BIT.
1105 (rs6000_gen_lvx): Likewise.
1107 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1109 * config/visium/visium.md (nop): Tweak comment.
1110 (hazard_nop): Likewise.
1112 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1114 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1115 -mspeculate-indirect-jumps.
1116 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1117 for -mno-speculate-indirect-jumps.
1118 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1119 (*call_value_indirect_elfv2<mode>): Disable for
1120 -mno-speculate-indirect-jumps.
1121 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1122 (indirect_jump): Emit different RTL for
1123 -mno-speculate-indirect-jumps.
1124 (*indirect_jump<mode>): Disable for
1125 -mno-speculate-indirect-jumps.
1126 (*indirect_jump<mode>_nospec): New define_insn.
1127 (tablejump): Emit different RTL for
1128 -mno-speculate-indirect-jumps.
1129 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1130 (tablejumpsi_nospec): New define_expand.
1131 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1132 (tablejumpdi_nospec): New define_expand.
1133 (*tablejump<mode>_internal1): Disable for
1134 -mno-speculate-indirect-jumps.
1135 (*tablejump<mode>_internal1_nospec): New define_insn.
1136 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1139 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1141 * caller-save.c (insert_save): Drop unnecessary parameter. All
1144 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1145 Richard Biener <rguenth@suse.de>
1148 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1149 return early, inline manually is_gimple_sizepos. Make sure if we
1150 call gimplify_expr we don't end up with a gimple constant.
1151 * tree.c (variably_modified_type_p): Don't return true for
1152 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1153 * gimplify.h (is_gimple_sizepos): Remove.
1155 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1157 PR tree-optimization/83857
1158 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1159 vectorizable_live_operation for pure SLP statements.
1160 (vectorizable_live_operation): Handle PHIs.
1162 2018-01-16 Richard Biener <rguenther@suse.de>
1164 PR tree-optimization/83867
1165 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1166 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1168 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1171 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1172 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1173 If off is not INTEGER_CST, issue a may not be aligned warning
1174 rather than isn't aligned. Use isn%'t rather than isn't.
1175 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1177 <case MULT_EXPR>: Improve the case when bottom and one of the
1178 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1179 operand, in that case check if the other operand is multiple of
1180 bottom divided by the INTEGER_CST operand.
1182 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1185 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1186 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1187 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1188 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1189 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1190 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1191 (pa_function_arg_size): New function.
1193 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1195 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1196 in a separate statement.
1198 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1200 PR tree-optimization/83847
1201 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1202 group gathers and scatters.
1204 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1206 PR rtl-optimization/86620
1207 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1209 PR rtl-optimization/83213
1210 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1211 to last if both are JUMP_INSNs.
1213 PR tree-optimization/83843
1214 * gimple-ssa-store-merging.c
1215 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1216 store_immediate_info for bswap/nop orig_stores.
1218 2018-01-15 Andrew Waterman <andrew@sifive.com>
1220 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1222 <UDIV>: Increase cost if !TARGET_DIV.
1224 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1226 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1227 (define_attr "cr_logical_3op"): New.
1228 (cceq_ior_compare): Adjust.
1229 (cceq_ior_compare_complement): Adjust.
1230 (*cceq_rev_compare): Adjust.
1231 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1232 (is_cracked_insn): Adjust.
1233 (insn_must_be_first_in_group): Adjust.
1234 * config/rs6000/40x.md: Adjust.
1235 * config/rs6000/440.md: Adjust.
1236 * config/rs6000/476.md: Adjust.
1237 * config/rs6000/601.md: Adjust.
1238 * config/rs6000/603.md: Adjust.
1239 * config/rs6000/6xx.md: Adjust.
1240 * config/rs6000/7450.md: Adjust.
1241 * config/rs6000/7xx.md: Adjust.
1242 * config/rs6000/8540.md: Adjust.
1243 * config/rs6000/cell.md: Adjust.
1244 * config/rs6000/e300c2c3.md: Adjust.
1245 * config/rs6000/e500mc.md: Adjust.
1246 * config/rs6000/e500mc64.md: Adjust.
1247 * config/rs6000/e5500.md: Adjust.
1248 * config/rs6000/e6500.md: Adjust.
1249 * config/rs6000/mpc.md: Adjust.
1250 * config/rs6000/power4.md: Adjust.
1251 * config/rs6000/power5.md: Adjust.
1252 * config/rs6000/power6.md: Adjust.
1253 * config/rs6000/power7.md: Adjust.
1254 * config/rs6000/power8.md: Adjust.
1255 * config/rs6000/power9.md: Adjust.
1256 * config/rs6000/rs64.md: Adjust.
1257 * config/rs6000/titan.md: Adjust.
1259 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1261 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1262 ix86_indirect_branch_register logic.
1264 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1266 * config/i386/constraints.md (Bs): Update
1267 ix86_indirect_branch_register check. Don't check
1268 ix86_indirect_branch_register with GOT_memory_operand.
1270 * config/i386/predicates.md (GOT_memory_operand): Don't check
1271 ix86_indirect_branch_register here.
1272 (GOT32_symbol_operand): Likewise.
1274 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1276 * config/i386/predicates.md (constant_call_address_operand):
1277 Rewrite ix86_indirect_branch_register logic.
1278 (sibcall_insn_operand): Likewise.
1280 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1282 * config/i386/constraints.md (Bs): Replace
1283 ix86_indirect_branch_thunk_register with
1284 ix86_indirect_branch_register.
1286 * config/i386/i386.md (indirect_jump): Likewise.
1287 (tablejump): Likewise.
1288 (*sibcall_memory): Likewise.
1289 (*sibcall_value_memory): Likewise.
1290 Peepholes of indirect call and jump via memory: Likewise.
1291 * config/i386/i386.opt: Likewise.
1292 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1293 (GOT_memory_operand): Likewise.
1294 (call_insn_operand): Likewise.
1295 (sibcall_insn_operand): Likewise.
1296 (GOT32_symbol_operand): Likewise.
1298 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1301 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1302 type rather than type addr's type points to.
1303 (expand_omp_atomic_mutex): Likewise.
1304 (expand_omp_atomic): Likewise.
1306 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1309 * config/i386/i386.c (output_indirect_thunk_function): Use
1310 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1311 for __x86_return_thunk.
1313 2018-01-15 Richard Biener <rguenther@suse.de>
1316 * expmed.c (extract_bit_field_1): Fix typo.
1318 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1321 * config/arm/iterators.md (VF): New mode iterator.
1322 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1323 Remove integer-related logic from pattern.
1324 (neon_vabd<mode>_3): Likewise.
1326 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1329 * common.opt (fstrict-overflow): No longer an alias.
1330 (fwrapv-pointer): New option.
1331 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1332 also for pointer types based on flag_wrapv_pointer.
1333 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1334 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1335 opts->x_flag_wrapv got set.
1336 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1337 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1338 POINTER_TYPE_OVERFLOW_UNDEFINED.
1339 * match.pd: Likewise in address comparison pattern.
1340 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1342 2018-01-15 Richard Biener <rguenther@suse.de>
1345 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1346 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1347 Reset type names to their identifier if their TYPE_DECL doesn't
1348 have linkage (and thus is used for ODR and devirt).
1349 (save_debug_info_for_decl): Remove.
1350 (save_debug_info_for_type): Likewise.
1351 (add_tree_to_fld_list): Adjust.
1352 * tree-pretty-print.c (dump_generic_node): Make dumping of
1353 type names more robust.
1355 2018-01-15 Richard Biener <rguenther@suse.de>
1357 * BASE-VER: Bump to 8.0.1.
1359 2018-01-14 Martin Sebor <msebor@redhat.com>
1362 * builtins.c (check_access): Avoid warning when the no-warning bit
1365 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1367 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1368 * ira-color (allocno_hard_regs_compare): Likewise.
1370 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1373 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1374 Use .pushsection/.popsection.
1376 2018-01-14 Martin Sebor <msebor@redhat.com>
1379 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1381 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1383 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1384 entry from extra_headers.
1385 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1386 extra_headers, make the list bitwise identical to the i?86-*-* one.
1388 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1390 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1391 -mcmodel=large with -mindirect-branch=thunk,
1392 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1393 -mfunction-return=thunk-extern.
1394 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1395 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1396 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1398 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1400 * config/i386/i386.c (print_reg): Print the name of the full
1401 integer register without '%'.
1402 (ix86_print_operand): Handle 'V'.
1403 * doc/extend.texi: Document 'V' modifier.
1405 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1407 * config/i386/constraints.md (Bs): Disallow memory operand for
1408 -mindirect-branch-register.
1410 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1411 (GOT_memory_operand): Likewise.
1412 (call_insn_operand): Likewise.
1413 (sibcall_insn_operand): Likewise.
1414 (GOT32_symbol_operand): Likewise.
1415 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1416 for -mindirect-branch-register.
1417 (tablejump): Likewise.
1418 (*sibcall_memory): Likewise.
1419 (*sibcall_value_memory): Likewise.
1420 Disallow peepholes of indirect call and jump via memory for
1421 -mindirect-branch-register.
1422 (*call_pop): Replace m with Bw.
1423 (*call_value_pop): Likewise.
1424 (*sibcall_pop_memory): Replace m with Bs.
1425 * config/i386/i386.opt (mindirect-branch-register): New option.
1426 * doc/invoke.texi: Document -mindirect-branch-register option.
1428 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1430 * config/i386/i386-protos.h (ix86_output_function_return): New.
1431 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1432 set function_return_type.
1433 (indirect_thunk_name): Add ret_p to indicate thunk for function
1435 (output_indirect_thunk_function): Pass false to
1436 indirect_thunk_name.
1437 (ix86_output_indirect_branch_via_reg): Likewise.
1438 (ix86_output_indirect_branch_via_push): Likewise.
1439 (output_indirect_thunk_function): Create alias for function
1440 return thunk if regno < 0.
1441 (ix86_output_function_return): New function.
1442 (ix86_handle_fndecl_attribute): Handle function_return.
1443 (ix86_attribute_table): Add function_return.
1444 * config/i386/i386.h (machine_function): Add
1445 function_return_type.
1446 * config/i386/i386.md (simple_return_internal): Use
1447 ix86_output_function_return.
1448 (simple_return_internal_long): Likewise.
1449 * config/i386/i386.opt (mfunction-return=): New option.
1450 (indirect_branch): Mention -mfunction-return=.
1451 * doc/extend.texi: Document function_return function attribute.
1452 * doc/invoke.texi: Document -mfunction-return= option.
1454 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1456 * config/i386/i386-opts.h (indirect_branch): New.
1457 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1458 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1459 with local indirect jump when converting indirect call and jump.
1460 (ix86_set_indirect_branch_type): New.
1461 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1462 (indirectlabelno): New.
1463 (indirect_thunk_needed): Likewise.
1464 (indirect_thunk_bnd_needed): Likewise.
1465 (indirect_thunks_used): Likewise.
1466 (indirect_thunks_bnd_used): Likewise.
1467 (INDIRECT_LABEL): Likewise.
1468 (indirect_thunk_name): Likewise.
1469 (output_indirect_thunk): Likewise.
1470 (output_indirect_thunk_function): Likewise.
1471 (ix86_output_indirect_branch_via_reg): Likewise.
1472 (ix86_output_indirect_branch_via_push): Likewise.
1473 (ix86_output_indirect_branch): Likewise.
1474 (ix86_output_indirect_jmp): Likewise.
1475 (ix86_code_end): Call output_indirect_thunk_function if needed.
1476 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1478 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1479 (ix86_attribute_table): Add indirect_branch.
1480 * config/i386/i386.h (machine_function): Add indirect_branch_type
1481 and has_local_indirect_jump.
1482 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1484 (tablejump): Likewise.
1485 (*indirect_jump): Use ix86_output_indirect_jmp.
1486 (*tablejump_1): Likewise.
1487 (simple_return_indirect_internal): Likewise.
1488 * config/i386/i386.opt (mindirect-branch=): New option.
1489 (indirect_branch): New.
1492 (thunk-inline): Likewise.
1493 (thunk-extern): Likewise.
1494 * doc/extend.texi: Document indirect_branch function attribute.
1495 * doc/invoke.texi: Document -mindirect-branch= option.
1497 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1500 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1502 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1504 * ipa-inline.c (want_inline_small_function_p): Return false if
1505 inlining has already failed with CIF_FINAL_ERROR.
1506 (update_caller_keys): Call want_inline_small_function_p before
1508 (update_callee_keys): Likewise.
1510 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1512 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1514 (rs6000_quadword_masked_address_p): Likewise.
1515 (quad_aligned_load_p): Likewise.
1516 (quad_aligned_store_p): Likewise.
1517 (const_load_sequence_p): Add comment to describe the outer-most loop.
1518 (mimic_memory_attributes_and_flags): New function.
1519 (rs6000_gen_stvx): Likewise.
1520 (replace_swapped_aligned_store): Likewise.
1521 (rs6000_gen_lvx): Likewise.
1522 (replace_swapped_aligned_load): Likewise.
1523 (replace_swapped_load_constant): Capitalize argument name in
1524 comment describing this function.
1525 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1526 and stores that access quad-word aligned addresses and replace
1527 with stvx or lvx instructions when appropriate.
1528 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1529 New function prototype.
1530 (rs6000_quadword_masked_address_p): Likewise.
1531 (rs6000_gen_lvx): Likewise.
1532 (rs6000_gen_stvx): Likewise.
1533 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1534 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1535 when memory address is aligned.
1536 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1537 this split to select lvx instruction when memory address is aligned.
1538 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1539 instruction when memory address is aligned.
1540 (*vsx_le_perm_load_v16qi): Likewise.
1541 (four unnamed splitters): Modify to select the stvx instruction
1542 when memory is aligned.
1544 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1546 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1547 which appears in the queue multiple times.
1549 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1550 Alan Hayward <alan.hayward@arm.com>
1551 David Sherwood <david.sherwood@arm.com>
1553 * tree-vectorizer.h (vec_lower_bound): New structure.
1554 (_loop_vec_info): Add check_nonzero and lower_bounds.
1555 (LOOP_VINFO_CHECK_NONZERO): New macro.
1556 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1557 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1558 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1559 fields. Make seg_len the distance travelled, not including the
1561 (dr_direction_indicator): Declare.
1562 (dr_zero_step_indicator): Likewise.
1563 (dr_known_forward_stride_p): Likewise.
1564 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1566 (runtime_alias_check_p): Allow runtime alias checks with
1568 (operator ==): Compare access_size and align.
1569 (prune_runtime_alias_test_list): Rework for new distinction between
1570 the access_size and seg_len.
1571 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1573 (get_segment_min_max): New function.
1574 (create_intersect_range_checks): Use it.
1575 (dr_step_indicator): New function.
1576 (dr_direction_indicator): Likewise.
1577 (dr_zero_step_indicator): Likewise.
1578 (dr_known_forward_stride_p): Likewise.
1579 * tree-loop-distribution.c (data_ref_segment_size): Return
1580 DR_STEP * (niters - 1).
1581 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1583 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1584 (vect_preserves_scalar_order_p): New function, split out from...
1585 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1586 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1587 (vect_vfa_access_size): New function.
1588 (vect_vfa_align): Likewise.
1589 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1590 (dump_lower_bound): New function.
1591 (vect_check_lower_bound): Likewise.
1592 (vect_small_gap_p): Likewise.
1593 (vectorizable_with_step_bound_p): Likewise.
1594 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1595 depencies if the vectorization factor is 1. Convert the checks
1596 for nonzero steps into checks on the bounds of DR_STEP. Try using
1597 a bunds check for variable steps if the minimum required step is
1598 relatively small. Update calls to the dr_with_seg_len
1599 constructor and to vect_compile_time_alias.
1600 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1602 (vect_loop_versioning): Call it.
1603 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1605 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1607 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1608 Alan Hayward <alan.hayward@arm.com>
1609 David Sherwood <david.sherwood@arm.com>
1611 * doc/sourcebuild.texi (vect_scatter_store): Document.
1612 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1614 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1616 * genopinit.c (main): Add supports_vec_scatter_store and
1617 supports_vec_scatter_store_cached to target_optabs.
1618 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1619 IFN_MASK_SCATTER_STORE.
1620 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1622 * internal-fn.h (internal_store_fn_p): Declare.
1623 (internal_fn_stored_value_index): Likewise.
1624 * internal-fn.c (scatter_store_direct): New macro.
1625 (expand_scatter_store_optab_fn): New function.
1626 (direct_scatter_store_optab_supported_p): New macro.
1627 (internal_store_fn_p): New function.
1628 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1629 IFN_MASK_SCATTER_STORE.
1630 (internal_fn_mask_index): Likewise.
1631 (internal_fn_stored_value_index): New function.
1632 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1634 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1635 * optabs-query.c (supports_vec_scatter_store_p): New function.
1636 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1637 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1638 true for scatter stores.
1639 (vect_gather_scatter_fn_p): Handle scatter stores too.
1640 (vect_check_gather_scatter): Consider using scatter stores if
1641 supports_vec_scatter_store_p.
1642 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1644 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1645 internal_fn_stored_value_index.
1646 (check_load_store_masking): Handle scatter stores too.
1647 (vect_get_store_rhs): Make public.
1648 (vectorizable_call): Use internal_store_fn_p.
1649 (vectorizable_store): Handle scatter store internal functions.
1650 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1651 when deciding whether the end of the group has been reached.
1652 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1653 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1654 (mask_scatter_store<mode>): New insns.
1656 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1657 Alan Hayward <alan.hayward@arm.com>
1658 David Sherwood <david.sherwood@arm.com>
1660 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1661 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1662 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1664 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1665 Use vect_truncate_gather_scatter_offset if we can't treat the
1666 operation as a normal gather load or scatter store.
1667 (get_group_load_store_type): Take the gather_scatter_info
1668 as argument. Try using a gather load or scatter store for
1669 single-element groups.
1670 (get_load_store_type): Update calls to get_group_load_store_type
1671 and vect_use_strided_gather_scatters_p.
1673 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1674 Alan Hayward <alan.hayward@arm.com>
1675 David Sherwood <david.sherwood@arm.com>
1677 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1678 optional tree argument.
1679 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1681 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1682 but continue to use the current value as a fallback.
1683 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1684 to compare the updates.
1685 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1686 (get_load_store_type): Use it when handling a strided access.
1687 (vect_get_strided_load_store_ops): New function.
1688 (vect_get_data_ptr_increment): Likewise.
1689 (vectorizable_load): Handle strided gather loads. Always pass
1690 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1692 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1693 Alan Hayward <alan.hayward@arm.com>
1694 David Sherwood <david.sherwood@arm.com>
1696 * doc/md.texi (gather_load@var{m}): Document.
1697 (mask_gather_load@var{m}): Likewise.
1698 * genopinit.c (main): Add supports_vec_gather_load and
1699 supports_vec_gather_load_cached to target_optabs.
1700 * optabs-tree.c (init_tree_optimization_optabs): Use
1701 ggc_cleared_alloc to allocate target_optabs.
1702 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1703 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1705 * internal-fn.h (internal_load_fn_p): Declare.
1706 (internal_gather_scatter_fn_p): Likewise.
1707 (internal_fn_mask_index): Likewise.
1708 (internal_gather_scatter_fn_supported_p): Likewise.
1709 * internal-fn.c (gather_load_direct): New macro.
1710 (expand_gather_load_optab_fn): New function.
1711 (direct_gather_load_optab_supported_p): New macro.
1712 (direct_internal_fn_optab): New function.
1713 (internal_load_fn_p): Likewise.
1714 (internal_gather_scatter_fn_p): Likewise.
1715 (internal_fn_mask_index): Likewise.
1716 (internal_gather_scatter_fn_supported_p): Likewise.
1717 * optabs-query.c (supports_at_least_one_mode_p): New function.
1718 (supports_vec_gather_load_p): Likewise.
1719 * optabs-query.h (supports_vec_gather_load_p): Declare.
1720 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1721 and memory_type field.
1722 (NUM_PATTERNS): Bump to 15.
1723 * tree-vect-data-refs.c: Include internal-fn.h.
1724 (vect_gather_scatter_fn_p): New function.
1725 (vect_describe_gather_scatter_call): Likewise.
1726 (vect_check_gather_scatter): Try using internal functions for
1727 gather loads. Recognize existing calls to a gather load function.
1728 (vect_analyze_data_refs): Consider using gather loads if
1729 supports_vec_gather_load_p.
1730 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1731 (vect_get_gather_scatter_offset_type): Likewise.
1732 (vect_convert_mask_for_vectype): Likewise.
1733 (vect_add_conversion_to_patterm): Likewise.
1734 (vect_try_gather_scatter_pattern): Likewise.
1735 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1736 (vect_vect_recog_func_ptrs): Add it.
1737 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1738 internal_fn_mask_index and internal_gather_scatter_fn_p.
1739 (check_load_store_masking): Take the gather_scatter_info as an
1740 argument and handle gather loads.
1741 (vect_get_gather_scatter_ops): New function.
1742 (vectorizable_call): Check internal_load_fn_p.
1743 (vectorizable_load): Likewise. Handle gather load internal
1745 (vectorizable_store): Update call to check_load_store_masking.
1746 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1747 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1748 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1749 (aarch64_gather_scale_operand_d): New predicates.
1750 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1751 (mask_gather_load<mode>): New insns.
1753 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1754 Alan Hayward <alan.hayward@arm.com>
1755 David Sherwood <david.sherwood@arm.com>
1757 * optabs.def (fold_left_plus_optab): New optab.
1758 * doc/md.texi (fold_left_plus_@var{m}): Document.
1759 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1760 * internal-fn.c (fold_left_direct): Define.
1761 (expand_fold_left_optab_fn): Likewise.
1762 (direct_fold_left_optab_supported_p): Likewise.
1763 * fold-const-call.c (fold_const_fold_left): New function.
1764 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1765 * tree-parloops.c (valid_reduction_p): New function.
1766 (gather_scalar_reductions): Use it.
1767 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1768 (vect_finish_replace_stmt): Declare.
1769 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1770 (needs_fold_left_reduction_p): New function, split out from...
1771 (vect_is_simple_reduction): ...here. Accept reductions that
1772 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1773 (vect_force_simple_reduction): Also store the reduction type in
1774 the assignment's STMT_VINFO_REDUC_TYPE.
1775 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1776 (merge_with_identity): New function.
1777 (vect_expand_fold_left): Likewise.
1778 (vectorize_fold_left_reduction): Likewise.
1779 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1780 scalar phi in place for it. Check for target support and reject
1781 cases that would reassociate the operation. Defer the transform
1782 phase to vectorize_fold_left_reduction.
1783 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1784 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1785 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1787 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1789 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1790 call to ifc_temp_var.
1792 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1793 Alan Hayward <alan.hayward@arm.com>
1794 David Sherwood <david.sherwood@arm.com>
1796 * target.def (legitimize_address_displacement): Take the original
1797 offset as a poly_int.
1798 * targhooks.h (default_legitimize_address_displacement): Update
1800 * targhooks.c (default_legitimize_address_displacement): Likewise.
1801 * doc/tm.texi: Regenerate.
1802 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1803 as an argument, moving assert of ad->disp == ad->disp_term to...
1804 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1805 Try calling targetm.legitimize_address_displacement before expanding
1806 the address rather than afterwards, and adjust for the new interface.
1807 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1808 Match the new hook interface. Handle SVE addresses.
1809 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1812 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1814 * Makefile.in (OBJS): Add early-remat.o.
1815 * target.def (select_early_remat_modes): New hook.
1816 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1817 * doc/tm.texi: Regenerate.
1818 * targhooks.h (default_select_early_remat_modes): Declare.
1819 * targhooks.c (default_select_early_remat_modes): New function.
1820 * timevar.def (TV_EARLY_REMAT): New timevar.
1821 * passes.def (pass_early_remat): New pass.
1822 * tree-pass.h (make_pass_early_remat): Declare.
1823 * early-remat.c: New file.
1824 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1826 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1828 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1829 Alan Hayward <alan.hayward@arm.com>
1830 David Sherwood <david.sherwood@arm.com>
1832 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1833 vfm1 with a bound_epilog parameter.
1834 (vect_do_peeling): Update calls accordingly, and move the prologue
1835 call earlier in the function. Treat the base bound_epilog as 0 for
1836 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1837 this base when peeling for gaps.
1838 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1839 with fully-masked loops.
1840 (vect_estimate_min_profitable_iters): Handle the single peeled
1841 iteration in that case.
1843 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1844 Alan Hayward <alan.hayward@arm.com>
1845 David Sherwood <david.sherwood@arm.com>
1847 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1848 single-element interleaving even if the size is not a power of 2.
1849 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1850 accesses for single-element interleaving if the group size is
1853 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1854 Alan Hayward <alan.hayward@arm.com>
1855 David Sherwood <david.sherwood@arm.com>
1857 * doc/md.texi (fold_extract_last_@var{m}): Document.
1858 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1859 * optabs.def (fold_extract_last_optab): New optab.
1860 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1861 * internal-fn.c (fold_extract_direct): New macro.
1862 (expand_fold_extract_optab_fn): Likewise.
1863 (direct_fold_extract_optab_supported_p): Likewise.
1864 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1865 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1866 EXTRACT_LAST_REDUCTION.
1867 (get_initial_def_for_reduction): Do not create an initial vector
1868 for EXTRACT_LAST_REDUCTION reductions.
1869 (vectorizable_reduction): Leave the scalar phi in place for
1870 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1871 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1872 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1873 transform phase to vectorizable_condition.
1874 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1876 (vect_finish_stmt_generation): ...here.
1877 (vect_finish_replace_stmt): New function.
1878 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1879 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1881 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1883 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1884 Alan Hayward <alan.hayward@arm.com>
1885 David Sherwood <david.sherwood@arm.com>
1887 * doc/md.texi (extract_last_@var{m}): Document.
1888 * optabs.def (extract_last_optab): New optab.
1889 * internal-fn.def (EXTRACT_LAST): New internal function.
1890 * internal-fn.c (cond_unary_direct): New macro.
1891 (expand_cond_unary_optab_fn): Likewise.
1892 (direct_cond_unary_optab_supported_p): Likewise.
1893 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1894 loops using EXTRACT_LAST.
1895 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1896 (extract_last_<mode>): ...this optab.
1897 (vec_extract<mode><Vel>): Update accordingly.
1899 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1900 Alan Hayward <alan.hayward@arm.com>
1901 David Sherwood <david.sherwood@arm.com>
1903 * target.def (empty_mask_is_expensive): New hook.
1904 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
1905 * doc/tm.texi: Regenerate.
1906 * targhooks.h (default_empty_mask_is_expensive): Declare.
1907 * targhooks.c (default_empty_mask_is_expensive): New function.
1908 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
1909 if the target says that empty masks are expensive.
1910 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
1912 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
1914 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1915 Alan Hayward <alan.hayward@arm.com>
1916 David Sherwood <david.sherwood@arm.com>
1918 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
1919 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
1920 (vect_use_loop_mask_for_alignment_p): New function.
1921 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
1922 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
1923 niters_skip argument. Make sure that the first niters_skip elements
1924 of the first iteration are inactive.
1925 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
1926 Update call to vect_set_loop_masks_directly.
1927 (get_misalign_in_elems): New function, split out from...
1928 (vect_gen_prolog_loop_niters): ...here.
1929 (vect_update_init_of_dr): Take a code argument that specifies whether
1930 the adjustment should be added or subtracted.
1931 (vect_update_init_of_drs): Likewise.
1932 (vect_prepare_for_masked_peels): New function.
1933 (vect_do_peeling): Skip prologue peeling if we're using a mask
1934 instead. Update call to vect_update_inits_of_drs.
1935 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1937 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
1938 alignment. Do not include the number of peeled iterations in
1939 the minimum threshold in that case.
1940 (vectorizable_induction): Adjust the start value down by
1941 LOOP_VINFO_MASK_SKIP_NITERS iterations.
1942 (vect_transform_loop): Call vect_prepare_for_masked_peels.
1943 Take the number of skipped iterations into account when calculating
1945 * tree-vect-stmts.c (vect_gen_while_not): New function.
1947 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1948 Alan Hayward <alan.hayward@arm.com>
1949 David Sherwood <david.sherwood@arm.com>
1951 * doc/sourcebuild.texi (vect_fully_masked): Document.
1952 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
1954 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
1956 (vect_analyze_loop_2): ...here. Don't check the vectorization
1957 factor against the number of loop iterations if the loop is
1960 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1961 Alan Hayward <alan.hayward@arm.com>
1962 David Sherwood <david.sherwood@arm.com>
1964 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
1965 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
1966 (dump_groups): Update accordingly.
1967 (iv_use::mem_type): New member variable.
1968 (address_p): New function.
1969 (record_use): Add a mem_type argument and initialize the new
1971 (record_group_use): Add a mem_type argument. Use address_p.
1972 Remove obsolete null checks of base_object. Update call to record_use.
1973 (find_interesting_uses_op): Update call to record_group_use.
1974 (find_interesting_uses_cond): Likewise.
1975 (find_interesting_uses_address): Likewise.
1976 (get_mem_type_for_internal_fn): New function.
1977 (find_address_like_use): Likewise.
1978 (find_interesting_uses_stmt): Try find_address_like_use before
1979 calling find_interesting_uses_op.
1980 (addr_offset_valid_p): Use the iv mem_type field as the type
1981 of the addressed memory.
1982 (add_autoinc_candidates): Likewise.
1983 (get_address_cost): Likewise.
1984 (split_small_address_groups_p): Use address_p.
1985 (split_address_groups): Likewise.
1986 (add_iv_candidate_for_use): Likewise.
1987 (autoinc_possible_for_pair): Likewise.
1988 (rewrite_groups): Likewise.
1989 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
1990 (determine_group_iv_cost): Update after split of USE_ADDRESS.
1991 (get_alias_ptr_type_for_ptr_address): New function.
1992 (rewrite_use_address): Rewrite address uses in calls that were
1993 identified by find_address_like_use.
1995 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1996 Alan Hayward <alan.hayward@arm.com>
1997 David Sherwood <david.sherwood@arm.com>
1999 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2001 * gimple-expr.h (is_gimple_addressable: Likewise.
2002 * gimple-expr.c (is_gimple_address): Likewise.
2003 * internal-fn.c (expand_call_mem_ref): New function.
2004 (expand_mask_load_optab_fn): Use it.
2005 (expand_mask_store_optab_fn): Likewise.
2007 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2008 Alan Hayward <alan.hayward@arm.com>
2009 David Sherwood <david.sherwood@arm.com>
2011 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2012 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2013 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2014 (cond_umax@var{mode}): Document.
2015 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2016 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2017 (cond_umin_optab, cond_umax_optab): New optabs.
2018 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2019 (COND_IOR, COND_XOR): New internal functions.
2020 * internal-fn.h (get_conditional_internal_fn): Declare.
2021 * internal-fn.c (cond_binary_direct): New macro.
2022 (expand_cond_binary_optab_fn): Likewise.
2023 (direct_cond_binary_optab_supported_p): Likewise.
2024 (get_conditional_internal_fn): New function.
2025 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2026 Cope with reduction statements that are vectorized as calls rather
2028 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2029 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2030 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2031 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2032 (UNSPEC_COND_EOR): New unspecs.
2033 (optab): Add mappings for them.
2034 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2035 (sve_int_op, sve_fp_op): New int attributes.
2037 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2038 Alan Hayward <alan.hayward@arm.com>
2039 David Sherwood <david.sherwood@arm.com>
2041 * optabs.def (while_ult_optab): New optab.
2042 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2043 * internal-fn.def (WHILE_ULT): New internal function.
2044 * internal-fn.h (direct_internal_fn_supported_p): New override
2045 that takes two types as argument.
2046 * internal-fn.c (while_direct): New macro.
2047 (expand_while_optab_fn): New function.
2048 (convert_optab_supported_p): Likewise.
2049 (direct_while_optab_supported_p): New macro.
2050 * wide-int.h (wi::udiv_ceil): New function.
2051 * tree-vectorizer.h (rgroup_masks): New structure.
2052 (vec_loop_masks): New typedef.
2053 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2055 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2056 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2057 (vect_max_vf): New function.
2058 (slpeel_make_loop_iterate_ntimes): Delete.
2059 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2060 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2061 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2062 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2063 internal-fn.h, stor-layout.h and optabs-query.h.
2064 (vect_set_loop_mask): New function.
2065 (add_preheader_seq): Likewise.
2066 (add_header_seq): Likewise.
2067 (interleave_supported_p): Likewise.
2068 (vect_maybe_permute_loop_masks): Likewise.
2069 (vect_set_loop_masks_directly): Likewise.
2070 (vect_set_loop_condition_masked): Likewise.
2071 (vect_set_loop_condition_unmasked): New function, split out from
2072 slpeel_make_loop_iterate_ntimes.
2073 (slpeel_make_loop_iterate_ntimes): Rename to..
2074 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2075 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2076 (vect_do_peeling): Update call accordingly.
2077 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2079 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2080 mask_compare_type, can_fully_mask_p and fully_masked_p.
2081 (release_vec_loop_masks): New function.
2082 (_loop_vec_info): Use it to free the loop masks.
2083 (can_produce_all_loop_masks_p): New function.
2084 (vect_get_max_nscalars_per_iter): Likewise.
2085 (vect_verify_full_masking): Likewise.
2086 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2087 retries, and free the mask rgroups before retrying. Check loop-wide
2088 reasons for disallowing fully-masked loops. Make the final decision
2089 about whether use a fully-masked loop or not.
2090 (vect_estimate_min_profitable_iters): Do not assume that peeling
2091 for the number of iterations will be needed for fully-masked loops.
2092 (vectorizable_reduction): Disable fully-masked loops.
2093 (vectorizable_live_operation): Likewise.
2094 (vect_halve_mask_nunits): New function.
2095 (vect_double_mask_nunits): Likewise.
2096 (vect_record_loop_mask): Likewise.
2097 (vect_get_loop_mask): Likewise.
2098 (vect_transform_loop): Handle the case in which the final loop
2099 iteration might handle a partial vector. Call vect_set_loop_condition
2100 instead of slpeel_make_loop_iterate_ntimes.
2101 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2102 (check_load_store_masking): New function.
2103 (prepare_load_store_mask): Likewise.
2104 (vectorizable_store): Handle fully-masked loops.
2105 (vectorizable_load): Likewise.
2106 (supportable_widening_operation): Use vect_halve_mask_nunits for
2108 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2109 (vect_gen_while): New function.
2110 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2111 (aarch64_uqdec<mode>): New insn.
2113 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2114 Alan Hayward <alan.hayward@arm.com>
2115 David Sherwood <david.sherwood@arm.com>
2117 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2118 (reduc_xor_scal_optab): New optabs.
2119 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2120 (reduc_xor_scal_@var{m}): Document.
2121 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2122 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2124 * fold-const-call.c (fold_const_call): Handle them.
2125 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2126 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2127 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2128 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2129 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2130 (UNSPEC_XORV): New unspecs.
2131 (optab): Add entries for them.
2132 (BITWISEV): New int iterator.
2133 (bit_reduc_op): New int attributes.
2135 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2136 Alan Hayward <alan.hayward@arm.com>
2137 David Sherwood <david.sherwood@arm.com>
2139 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2140 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2141 * optabs.def (vec_shl_insert_optab): New optab.
2142 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2143 (duplicate_and_interleave): Likewise.
2144 * tree-vect-loop.c: Include internal-fn.h.
2145 (neutral_op_for_slp_reduction): New function, split out from
2146 get_initial_defs_for_reduction.
2147 (get_initial_def_for_reduction): Handle option 2 for variable-length
2148 vectors by loading the neutral value into a vector and then shifting
2149 the initial value into element 0.
2150 (get_initial_defs_for_reduction): Replace the code argument with
2151 the neutral value calculated by neutral_op_for_slp_reduction.
2152 Use gimple_build_vector for constant-length vectors.
2153 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2154 but the first group_size elements have a neutral value.
2155 Use duplicate_and_interleave otherwise.
2156 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2157 Update call to get_initial_defs_for_reduction. Handle SLP
2158 reductions for variable-length vectors by creating one vector
2159 result for each scalar result, with the elements associated
2160 with other scalar results stubbed out with the neutral value.
2161 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2162 Require IFN_VEC_SHL_INSERT for double reductions on
2163 variable-length vectors, or SLP reductions that have
2164 a neutral value. Require can_duplicate_and_interleave_p
2165 support for variable-length unchained SLP reductions if there
2166 is no neutral value, such as for MIN/MAX reductions. Also require
2167 the number of vector elements to be a multiple of the number of
2168 SLP statements when doing variable-length unchained SLP reductions.
2169 Update call to vect_create_epilog_for_reduction.
2170 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2171 and remove initial values.
2172 (duplicate_and_interleave): Make public.
2173 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2174 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2176 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2177 Alan Hayward <alan.hayward@arm.com>
2178 David Sherwood <david.sherwood@arm.com>
2180 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2181 (can_duplicate_and_interleave_p): New function.
2182 (vect_get_and_check_slp_defs): Take the vector of statements
2183 rather than just the current one. Remove excess parentheses.
2184 Restriction rejectinon of vect_constant_def and vect_external_def
2185 for variable-length vectors to boolean types, or types for which
2186 can_duplicate_and_interleave_p is false.
2187 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2188 (duplicate_and_interleave): New function.
2189 (vect_get_constant_vectors): Use gimple_build_vector for
2190 constant-length vectors and suitable variable-length constant
2191 vectors. Use duplicate_and_interleave for other variable-length
2192 vectors. Don't defer the update when inserting new statements.
2194 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2195 Alan Hayward <alan.hayward@arm.com>
2196 David Sherwood <david.sherwood@arm.com>
2198 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2199 min_profitable_iters doesn't go negative.
2201 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2202 Alan Hayward <alan.hayward@arm.com>
2203 David Sherwood <david.sherwood@arm.com>
2205 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2206 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2207 * optabs.def (vec_mask_load_lanes_optab): New optab.
2208 (vec_mask_store_lanes_optab): Likewise.
2209 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2210 (MASK_STORE_LANES): Likewise.
2211 * internal-fn.c (mask_load_lanes_direct): New macro.
2212 (mask_store_lanes_direct): Likewise.
2213 (expand_mask_load_optab_fn): Handle masked operations.
2214 (expand_mask_load_lanes_optab_fn): New macro.
2215 (expand_mask_store_optab_fn): Handle masked operations.
2216 (expand_mask_store_lanes_optab_fn): New macro.
2217 (direct_mask_load_lanes_optab_supported_p): Likewise.
2218 (direct_mask_store_lanes_optab_supported_p): Likewise.
2219 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2221 (vect_load_lanes_supported): Likewise.
2222 * tree-vect-data-refs.c (strip_conversion): New function.
2223 (can_group_stmts_p): Likewise.
2224 (vect_analyze_data_ref_accesses): Use it instead of checking
2225 for a pair of assignments.
2226 (vect_store_lanes_supported): Take a masked_p parameter.
2227 (vect_load_lanes_supported): Likewise.
2228 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2229 vect_store_lanes_supported and vect_load_lanes_supported.
2230 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2231 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2232 parameter. Don't allow gaps for masked accesses.
2233 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2234 and vect_load_lanes_supported.
2235 (get_load_store_type): Take a masked_p parameter and update
2236 call to get_group_load_store_type.
2237 (vectorizable_store): Update call to get_load_store_type.
2238 Handle IFN_MASK_STORE_LANES.
2239 (vectorizable_load): Update call to get_load_store_type.
2240 Handle IFN_MASK_LOAD_LANES.
2242 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2243 Alan Hayward <alan.hayward@arm.com>
2244 David Sherwood <david.sherwood@arm.com>
2246 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2248 * config/aarch64/aarch64-protos.h
2249 (aarch64_sve_struct_memory_operand_p): Declare.
2250 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2251 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2252 (VPRED, vpred): Handle SVE structure modes.
2253 * config/aarch64/constraints.md (Utx): New constraint.
2254 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2255 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2256 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2257 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2258 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2259 structure modes. Split into pieces after RA.
2260 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2261 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2263 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2264 SVE structure modes.
2265 (aarch64_classify_address): Likewise.
2266 (sizetochar): Move earlier in file.
2267 (aarch64_print_operand): Handle SVE register lists.
2268 (aarch64_array_mode): New function.
2269 (aarch64_sve_struct_memory_operand_p): Likewise.
2270 (TARGET_ARRAY_MODE): Redefine.
2272 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2273 Alan Hayward <alan.hayward@arm.com>
2274 David Sherwood <david.sherwood@arm.com>
2276 * target.def (array_mode): New target hook.
2277 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2278 * doc/tm.texi: Regenerate.
2279 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2280 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2281 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2283 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2286 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2287 Alan Hayward <alan.hayward@arm.com>
2288 David Sherwood <david.sherwood@arm.com>
2290 * fold-const.c (fold_binary_loc): Check the argument types
2291 rather than the result type when testing for a vector operation.
2293 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2295 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2296 * doc/tm.texi: Regenerate.
2298 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2299 Alan Hayward <alan.hayward@arm.com>
2300 David Sherwood <david.sherwood@arm.com>
2302 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2303 (sve): Document new AArch64 extension.
2304 * doc/md.texi (w): Extend the description of the AArch64
2305 constraint to include SVE vectors.
2306 (Upl, Upa): Document new AArch64 predicate constraints.
2307 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2309 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2310 (msve-vector-bits=): New option.
2311 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2312 SVE when these are disabled.
2313 (sve): New extension.
2314 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2315 modes. Adjust their number of units based on aarch64_sve_vg.
2316 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2317 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2318 aarch64_addr_query_type.
2319 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2320 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2321 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2322 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2323 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2324 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2325 (aarch64_simd_imm_zero_p): Delete.
2326 (aarch64_check_zero_based_sve_index_immediate): Declare.
2327 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2328 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2329 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2330 (aarch64_sve_float_mul_immediate_p): Likewise.
2331 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2333 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2334 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2335 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2336 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2337 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2338 (aarch64_regmode_natural_size): Likewise.
2339 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2340 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2342 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2343 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2344 for VG and the SVE predicate registers.
2345 (V_ALIASES): Add a "z"-prefixed alias.
2346 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2347 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2348 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2349 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2350 (REG_CLASS_NAMES): Add entries for them.
2351 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2352 and the predicate registers.
2353 (aarch64_sve_vg): Declare.
2354 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2355 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2356 (REGMODE_NATURAL_SIZE): Define.
2357 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2359 * config/aarch64/aarch64.c: Include cfgrtl.h.
2360 (simd_immediate_info): Add a constructor for series vectors,
2361 and an associated step field.
2362 (aarch64_sve_vg): New variable.
2363 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2364 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2365 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2366 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2367 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2368 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2369 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2370 (aarch64_get_mask_mode): New functions.
2371 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2372 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2373 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2374 predicate modes and predicate registers. Explicitly restrict
2375 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2376 to store a vector mode if it is recognized by
2377 aarch64_classify_vector_mode.
2378 (aarch64_regmode_natural_size): New function.
2379 (aarch64_hard_regno_caller_save_mode): Return the original mode
2381 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2382 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2383 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2384 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2386 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2387 does not overlap dest if the function is frame-related. Handle
2389 (aarch64_split_add_offset): New function.
2390 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2391 them aarch64_add_offset.
2392 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2393 and update call to aarch64_sub_sp.
2394 (aarch64_add_cfa_expression): New function.
2395 (aarch64_expand_prologue): Pass extra temporary registers to the
2396 functions above. Handle the case in which we need to emit new
2397 DW_CFA_expressions for registers that were originally saved
2398 relative to the stack pointer, but now have to be expressed
2399 relative to the frame pointer.
2400 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2402 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2403 IP0 and IP1 values for SVE frames.
2404 (aarch64_expand_vec_series): New function.
2405 (aarch64_expand_sve_widened_duplicate): Likewise.
2406 (aarch64_expand_sve_const_vector): Likewise.
2407 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2408 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2409 into the register, rather than emitting a SET directly.
2410 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2411 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2412 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2413 (offset_9bit_signed_scaled_p): New functions.
2414 (aarch64_replicate_bitmask_imm): New function.
2415 (aarch64_bitmask_imm): Use it.
2416 (aarch64_cannot_force_const_mem): Reject expressions involving
2417 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2418 (aarch64_classify_index): Handle SVE indices, by requiring
2419 a plain register index with a scale that matches the element size.
2420 (aarch64_classify_address): Handle SVE addresses. Assert that
2421 the mode of the address is VOIDmode or an integer mode.
2422 Update call to aarch64_classify_symbol.
2423 (aarch64_classify_symbolic_expression): Update call to
2424 aarch64_classify_symbol.
2425 (aarch64_const_vec_all_in_range_p): New function.
2426 (aarch64_print_vector_float_operand): Likewise.
2427 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2428 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2429 and the FP immediates 1.0 and 0.5.
2430 (aarch64_print_address_internal): Handle SVE addresses.
2431 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2432 (aarch64_regno_regclass): Handle predicate registers.
2433 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2435 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2436 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2437 (aarch64_convert_sve_vector_bits): New function.
2438 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2439 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2441 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2442 Handle SVE vector and predicate modes. Accept VL-based constants
2443 that need only one temporary register, and VL offsets that require
2444 no temporary registers.
2445 (aarch64_conditional_register_usage): Mark the predicate registers
2446 as fixed if SVE isn't available.
2447 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2448 Return true for SVE vector and predicate modes.
2449 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2450 rather than an unsigned int. Handle SVE modes.
2451 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2453 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2455 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2456 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2457 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2458 (aarch64_sve_float_mul_immediate_p): New functions.
2459 (aarch64_sve_valid_immediate): New function.
2460 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2461 Explicitly reject structure modes. Check for INDEX constants.
2462 Handle PTRUE and PFALSE constants.
2463 (aarch64_check_zero_based_sve_index_immediate): New function.
2464 (aarch64_simd_imm_zero_p): Delete.
2465 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2466 vector modes. Accept constants in the range of CNT[BHWD].
2467 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2468 ask for an Advanced SIMD mode.
2469 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2470 (aarch64_simd_vector_alignment): Handle SVE predicates.
2471 (aarch64_vectorize_preferred_vector_alignment): New function.
2472 (aarch64_simd_vector_alignment_reachable): Use it instead of
2474 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2475 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2477 (MAX_VECT_LEN): Delete.
2478 (expand_vec_perm_d): Add a vec_flags field.
2479 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2480 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2481 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2483 (aarch64_evpc_rev): Rename to...
2484 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2485 (aarch64_evpc_rev_global): New function.
2486 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2487 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2489 (aarch64_evpc_sve_tbl): New function.
2490 (aarch64_expand_vec_perm_const_1): Update after rename of
2491 aarch64_evpc_rev. Handle SVE permutes too, trying
2492 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2493 than aarch64_evpc_tbl.
2494 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2495 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2496 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2497 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2498 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2499 (aarch64_expand_sve_vcond): New functions.
2500 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2501 of aarch64_vector_mode_p.
2502 (aarch64_dwarf_poly_indeterminate_value): New function.
2503 (aarch64_compute_pressure_classes): Likewise.
2504 (aarch64_can_change_mode_class): Likewise.
2505 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2506 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2507 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2508 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2509 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2510 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2511 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2512 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2514 (Dn, Dl, Dr): Accept const as well as const_vector.
2515 (Dz): Likewise. Compare against CONST0_RTX.
2516 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2517 of "vector" where appropriate.
2518 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2519 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2520 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2521 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2522 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2523 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2524 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2525 (v_int_equiv): Extend to SVE modes.
2526 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2528 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2529 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2530 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2531 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2532 (SVE_COND_FP_CMP): New int iterators.
2533 (perm_hilo): Handle the new unpack unspecs.
2534 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2536 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2537 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2538 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2539 (aarch64_equality_operator, aarch64_constant_vector_operand)
2540 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2541 (aarch64_sve_nonimmediate_operand): Likewise.
2542 (aarch64_sve_general_operand): Likewise.
2543 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2544 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2545 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2546 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2547 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2548 (aarch64_sve_float_arith_immediate): Likewise.
2549 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2550 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2551 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2552 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2553 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2554 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2555 (aarch64_sve_float_arith_operand): Likewise.
2556 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2557 (aarch64_sve_float_mul_operand): Likewise.
2558 (aarch64_sve_vec_perm_operand): Likewise.
2559 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2560 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2561 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2562 as well as const_vector.
2563 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2564 in file. Use CONST0_RTX and CONSTM1_RTX.
2565 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2566 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2567 Use aarch64_simd_imm_zero.
2568 * config/aarch64/aarch64-sve.md: New file.
2569 * config/aarch64/aarch64.md: Include it.
2570 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2571 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2572 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2573 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2574 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2575 (sve): New attribute.
2576 (enabled): Disable instructions with the sve attribute unless
2578 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2579 aarch64_expand_mov_immediate.
2580 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2581 CNT[BHSD] immediates.
2582 (movti): Split CONST_POLY_INT moves into two halves.
2583 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2584 Split additions that need a temporary here if the destination
2585 is the stack pointer.
2586 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2587 (*add<mode>3_poly_1): New instruction.
2588 (set_clobber_cc): New expander.
2590 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2592 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2593 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2594 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2595 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2596 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2597 Change innermode from fixed_mode_size to machine_mode.
2598 (simplify_subreg): Update call accordingly. Handle a constant-sized
2599 subreg of a variable-length CONST_VECTOR.
2601 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2602 Alan Hayward <alan.hayward@arm.com>
2603 David Sherwood <david.sherwood@arm.com>
2605 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2606 (add_offset_to_base): New function, split out from...
2607 (create_mem_ref): ...here. When handling a scale other than 1,
2608 check first whether the address is valid without the offset.
2609 Add it into the base if so, leaving the index and scale as-is.
2611 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2614 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2615 fold_for_warn before checking if arg2 is INTEGER_CST.
2617 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2619 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2620 (store_multiple_operation): Delete.
2621 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2622 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2623 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2624 guarded by TARGET_STRING.
2625 (rs6000_output_load_multiple): Delete.
2626 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2627 OPTION_MASK_STRING / TARGET_STRING handling.
2628 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2629 (const rs6000_opt_masks) <"string">: Change mask to 0.
2630 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2631 (MASK_STRING): Delete.
2632 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2634 (load_multiple): Delete.
2641 (store_multiple): Delete.
2648 (movmemsi_8reg): Delete.
2649 (corresponding unnamed define_insn): Delete.
2650 (movmemsi_6reg): Delete.
2651 (corresponding unnamed define_insn): Delete.
2652 (movmemsi_4reg): Delete.
2653 (corresponding unnamed define_insn): Delete.
2654 (movmemsi_2reg): Delete.
2655 (corresponding unnamed define_insn): Delete.
2656 (movmemsi_1reg): Delete.
2657 (corresponding unnamed define_insn): Delete.
2658 * config/rs6000/rs6000.opt (mno-string): New.
2659 (mstring): Replace by deprecation warning stub.
2660 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2662 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2664 * regrename.c (regrename_do_replace): If replacing the same
2665 reg multiple times, try to reuse last created gen_raw_REG.
2668 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2669 main to workaround a bug in GDB.
2671 2018-01-12 Tom de Vries <tom@codesourcery.com>
2674 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2676 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2678 PR rtl-optimization/80481
2679 * ira-color.c (get_cap_member): New function.
2680 (allocnos_conflict_by_live_ranges_p): Use it.
2681 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2682 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2684 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2687 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2688 (*saddl_se_1): Ditto.
2690 (*ssubl_se_1): Ditto.
2692 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2694 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2695 rather than wi::to_widest for DR_INITs.
2696 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2697 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2698 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2700 (vect_analyze_group_access_1): Note that here.
2702 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2704 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2705 polynomial type sizes.
2707 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2709 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2710 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2711 (gimple_add_tmp_var): Likewise.
2713 2018-01-12 Martin Liska <mliska@suse.cz>
2715 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2716 (gimple_alloc_sizes): Likewise.
2717 (dump_gimple_statistics): Use PRIu64 in printf format.
2718 * gimple.h: Change uint64_t to int.
2720 2018-01-12 Martin Liska <mliska@suse.cz>
2722 * tree-core.h: Use uint64_t instead of int.
2723 * tree.c (tree_node_counts): Likewise.
2724 (tree_node_sizes): Likewise.
2725 (dump_tree_statistics): Use PRIu64 in printf format.
2727 2018-01-12 Martin Liska <mliska@suse.cz>
2729 * Makefile.in: As qsort_chk is implemented in vec.c, add
2730 vec.o to linkage of gencfn-macros.
2731 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2732 passing the info to record_node_allocation_statistics.
2733 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2735 * ggc-common.c (struct ggc_usage): Add operator== and use
2736 it in operator< and compare function.
2737 * mem-stats.h (struct mem_usage): Likewise.
2738 * vec.c (struct vec_usage): Remove operator< and compare
2739 function. Can be simply inherited.
2741 2018-01-12 Martin Jambor <mjambor@suse.cz>
2744 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2745 * tree-ssa-math-opts.c: Include domwalk.h.
2746 (convert_mult_to_fma_1): New function.
2747 (fma_transformation_info): New type.
2748 (fma_deferring_state): Likewise.
2749 (cancel_fma_deferring): New function.
2750 (result_of_phi): Likewise.
2751 (last_fma_candidate_feeds_initial_phi): Likewise.
2752 (convert_mult_to_fma): Added deferring logic, split actual
2753 transformation to convert_mult_to_fma_1.
2754 (math_opts_dom_walker): New type.
2755 (math_opts_dom_walker::after_dom_children): New method, body moved
2756 here from pass_optimize_widening_mul::execute, added deferring logic
2758 (pass_optimize_widening_mul::execute): Moved most of code to
2759 math_opts_dom_walker::after_dom_children.
2760 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2761 * config/i386/i386.c (ix86_option_override_internal): Added
2762 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2764 2018-01-12 Richard Biener <rguenther@suse.de>
2767 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2768 inline instance vars.
2770 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2773 * config/rx/rx.c (rx_is_restricted_memory_address):
2776 2018-01-12 Richard Biener <rguenther@suse.de>
2778 PR tree-optimization/80846
2779 * target.def (split_reduction): New target hook.
2780 * targhooks.c (default_split_reduction): New function.
2781 * targhooks.h (default_split_reduction): Declare.
2782 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2783 target requests first reduce vectors by combining low and high
2785 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2786 (get_vectype_for_scalar_type_and_size): Export.
2787 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2788 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2789 * doc/tm.texi: Regenerate.
2790 * config/i386/i386.c (ix86_split_reduction): Implement
2791 TARGET_VECTORIZE_SPLIT_REDUCTION.
2793 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2796 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2797 in PIC mode except for TARGET_VXWORKS_RTP.
2798 * config/sparc/sparc.c: Include cfgrtl.h.
2799 (TARGET_INIT_PIC_REG): Define.
2800 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2801 (sparc_pic_register_p): New predicate.
2802 (sparc_legitimate_address_p): Use it.
2803 (sparc_legitimize_pic_address): Likewise.
2804 (sparc_delegitimize_address): Likewise.
2805 (sparc_mode_dependent_address_p): Likewise.
2806 (gen_load_pcrel_sym): Remove 4th parameter.
2807 (load_got_register): Adjust call to above. Remove obsolete stuff.
2808 (sparc_expand_prologue): Do not call load_got_register here.
2809 (sparc_flat_expand_prologue): Likewise.
2810 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2811 (sparc_use_pseudo_pic_reg): New function.
2812 (sparc_init_pic_reg): Likewise.
2813 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2814 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2816 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2818 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2819 Add item for branch_cost.
2821 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2823 PR rtl-optimization/83565
2824 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2825 not extend the result to a larger mode for rotate operations.
2826 (num_sign_bit_copies1): Likewise.
2828 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2831 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2833 Use values-Xc.o for -pedantic.
2834 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2836 2018-01-12 Martin Liska <mliska@suse.cz>
2839 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2841 (possible_polymorphic_call_targets): Use it.
2842 (ipa_devirt): Likewise.
2844 2018-01-12 Martin Liska <mliska@suse.cz>
2846 * profile-count.h (enum profile_quality): Use 0 as invalid
2847 enum value of profile_quality.
2849 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2851 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2852 -mext-string options.
2854 2018-01-12 Richard Biener <rguenther@suse.de>
2856 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2857 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2858 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2860 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2862 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2864 * configure.ac (--with-long-double-format): Add support for the
2865 configuration option to change the default long double format on
2867 * config.gcc (powerpc*-linux*-*): Likewise.
2868 * configure: Regenerate.
2869 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2870 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2871 used without modification.
2873 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2875 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2876 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2877 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2878 MISC_BUILTIN_SPEC_BARRIER.
2879 (rs6000_init_builtins): Likewise.
2880 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2882 (speculation_barrier): New define_insn.
2883 * doc/extend.texi: Document __builtin_speculation_barrier.
2885 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2888 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2889 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2890 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2892 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2893 integral modes instead of "ss" and "sd".
2894 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2895 vectors with 32-bit and 64-bit elements.
2896 (vecdupssescalarmodesuffix): New mode attribute.
2897 (vec_dup<mode>): Use it.
2899 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2902 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2903 frame if argument is passed on stack.
2905 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2908 * ree.c (combine_reaching_defs): Optimize also
2909 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
2910 reg2=any_extend(exp); reg1=reg2;, formatting fix.
2912 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2915 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
2917 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2920 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
2921 after they are computed.
2923 2018-01-11 Bin Cheng <bin.cheng@arm.com>
2925 PR tree-optimization/83695
2926 * gimple-loop-linterchange.cc
2927 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
2928 reset cached scev information after interchange.
2929 (pass_linterchange::execute): Remove call to scev_reset_htab.
2931 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2933 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
2934 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
2935 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
2936 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
2937 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
2938 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
2939 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
2940 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
2941 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
2942 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
2943 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
2944 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
2945 (V_lane_reg): Likewise.
2946 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
2948 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
2949 (vfmal_lane_low<mode>_intrinsic,
2950 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
2951 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
2952 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
2953 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
2954 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
2955 vfmsl_lane_high<mode>_intrinsic): New define_insns.
2957 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2959 * config/arm/arm-cpus.in (fp16fml): New feature.
2960 (ALL_SIMD): Add fp16fml.
2961 (armv8.2-a): Add fp16fml as an option.
2962 (armv8.3-a): Likewise.
2963 (armv8.4-a): Add fp16fml as part of fp16.
2964 * config/arm/arm.h (TARGET_FP16FML): Define.
2965 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
2967 * config/arm/arm-modes.def (V2HF): Define.
2968 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
2969 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
2970 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
2971 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
2972 vfmsl_low, vfmsl_high): New set of builtins.
2973 * config/arm/iterators.md (PLUSMINUS): New code iterator.
2974 (vfml_op): New code attribute.
2975 (VFMLHALVES): New int iterator.
2976 (VFML, VFMLSEL): New mode attributes.
2977 (V_reg): Define mapping for V2HF.
2978 (V_hi, V_lo): New mode attributes.
2979 (VF_constraint): Likewise.
2980 (vfml_half, vfml_half_selector): New int attributes.
2981 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
2983 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
2984 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
2986 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
2987 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
2988 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
2989 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
2991 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
2992 Document new effective target and option set.
2994 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2996 * config/arm/arm-cpus.in (armv8_4): New feature.
2997 (ARMv8_4a): New fgroup.
2998 (armv8.4-a): New arch.
2999 * config/arm/arm-tables.opt: Regenerate.
3000 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3001 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3002 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3003 Add matching rules for -march=armv8.4-a and extensions.
3004 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3006 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3009 * config/rx/rx.md (BW): New mode attribute.
3010 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3012 2018-01-11 Richard Biener <rguenther@suse.de>
3014 PR tree-optimization/83435
3015 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3016 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3017 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3019 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3020 Alan Hayward <alan.hayward@arm.com>
3021 David Sherwood <david.sherwood@arm.com>
3023 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3025 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3026 (aarch64_print_address_internal): Use it to check for a zero offset.
3028 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3029 Alan Hayward <alan.hayward@arm.com>
3030 David Sherwood <david.sherwood@arm.com>
3032 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3033 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3034 Return a poly_int64 rather than a HOST_WIDE_INT.
3035 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3036 rather than a HOST_WIDE_INT.
3037 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3038 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3039 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3040 final_offset from HOST_WIDE_INT to poly_int64.
3041 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3042 to_constant when getting the number of units in an Advanced SIMD
3044 (aarch64_builtin_vectorized_function): Check for a constant number
3046 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3048 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3049 attribute instead of GET_MODE_NUNITS.
3050 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3051 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3052 GET_MODE_SIZE for fixed-size registers.
3053 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3054 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3055 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3056 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3057 (aarch64_print_operand, aarch64_print_address_internal)
3058 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3059 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3060 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3061 Handle polynomial GET_MODE_SIZE.
3062 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3063 wider than SImode without modification.
3064 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3065 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3066 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3067 passing and returning SVE modes.
3068 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3069 rather than GEN_INT.
3070 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3071 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3072 (aarch64_allocate_and_probe_stack_space): Likewise.
3073 (aarch64_layout_frame): Cope with polynomial offsets.
3074 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3075 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3077 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3078 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3079 poly_int64 rather than a HOST_WIDE_INT.
3080 (aarch64_get_separate_components, aarch64_process_components)
3081 (aarch64_expand_prologue, aarch64_expand_epilogue)
3082 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3083 (aarch64_anchor_offset): New function, split out from...
3084 (aarch64_legitimize_address): ...here.
3085 (aarch64_builtin_vectorization_cost): Handle polynomial
3086 TYPE_VECTOR_SUBPARTS.
3087 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3089 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3090 number of elements from the PARALLEL rather than the mode.
3091 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3092 rather than GET_MODE_BITSIZE.
3093 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3094 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3095 (aarch64_expand_vec_perm_const_1): Handle polynomial
3096 d->perm.length () and d->perm elements.
3097 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3098 Apply to_constant to d->perm elements.
3099 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3100 polynomial CONST_VECTOR_NUNITS.
3101 (aarch64_move_pointer): Take amount as a poly_int64 rather
3103 (aarch64_progress_pointer): Avoid temporary variable.
3104 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3105 the mode attribute instead of GET_MODE.
3107 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3108 Alan Hayward <alan.hayward@arm.com>
3109 David Sherwood <david.sherwood@arm.com>
3111 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3112 x exists before using it.
3113 (aarch64_add_constant_internal): Rename to...
3114 (aarch64_add_offset_1): ...this. Replace regnum with separate
3115 src and dest rtxes. Handle the case in which they're different,
3116 including when the offset is zero. Replace scratchreg with an rtx.
3117 Use 2 additions if there is no spare register into which we can
3118 move a 16-bit constant.
3119 (aarch64_add_constant): Delete.
3120 (aarch64_add_offset): Replace reg with separate src and dest
3121 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3122 Use aarch64_add_offset_1.
3123 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3124 an rtx rather than an int. Take the delta as a poly_int64
3125 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3126 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3127 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3128 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3129 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3131 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3132 aarch64_add_constant.
3134 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3136 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3137 Use scalar_float_mode.
3139 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3141 * config/aarch64/aarch64-simd.md
3142 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3143 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3144 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3145 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3146 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3147 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3148 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3149 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3150 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3151 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3153 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3156 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3157 targ_options->x_arm_arch_string is non NULL.
3159 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3161 * config/aarch64/aarch64.h
3162 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3164 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3167 * expmed.c (emit_store_flag_force): Swap if const op0
3168 and change VOIDmode to mode of op0.
3170 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3172 PR rtl-optimization/83761
3173 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3174 than bytes to mode_for_size.
3176 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3179 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3180 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3183 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3186 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3187 when in layout mode.
3188 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3189 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3192 2018-01-10 Michael Collison <michael.collison@arm.com>
3194 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3195 * config/aarch64/aarch64-option-extension.def: Add
3196 AARCH64_OPT_EXTENSION of 'fp16fml'.
3197 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3198 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3199 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3200 * config/aarch64/constraints.md (Ui7): New constraint.
3201 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3202 (VFMLA_SEL_W): Ditto.
3205 (VFMLA16_LOW): New int iterator.
3206 (VFMLA16_HIGH): Ditto.
3207 (UNSPEC_FMLAL): New unspec.
3208 (UNSPEC_FMLSL): Ditto.
3209 (UNSPEC_FMLAL2): Ditto.
3210 (UNSPEC_FMLSL2): Ditto.
3211 (f16mac): New code attribute.
3212 * config/aarch64/aarch64-simd-builtins.def
3213 (aarch64_fmlal_lowv2sf): Ditto.
3214 (aarch64_fmlsl_lowv2sf): Ditto.
3215 (aarch64_fmlalq_lowv4sf): Ditto.
3216 (aarch64_fmlslq_lowv4sf): Ditto.
3217 (aarch64_fmlal_highv2sf): Ditto.
3218 (aarch64_fmlsl_highv2sf): Ditto.
3219 (aarch64_fmlalq_highv4sf): Ditto.
3220 (aarch64_fmlslq_highv4sf): Ditto.
3221 (aarch64_fmlal_lane_lowv2sf): Ditto.
3222 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3223 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3224 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3225 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3226 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3227 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3228 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3229 (aarch64_fmlal_lane_highv2sf): Ditto.
3230 (aarch64_fmlsl_lane_highv2sf): Ditto.
3231 (aarch64_fmlal_laneq_highv2sf): Ditto.
3232 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3233 (aarch64_fmlalq_lane_highv4sf): Ditto.
3234 (aarch64_fmlsl_lane_highv4sf): Ditto.
3235 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3236 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3237 * config/aarch64/aarch64-simd.md:
3238 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3239 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3240 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3241 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3242 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3243 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3244 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3245 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3246 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3247 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3248 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3249 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3250 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3251 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3252 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3253 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3254 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3255 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3256 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3257 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3258 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3259 (vfmlsl_low_u32): Ditto.
3260 (vfmlalq_low_u32): Ditto.
3261 (vfmlslq_low_u32): Ditto.
3262 (vfmlal_high_u32): Ditto.
3263 (vfmlsl_high_u32): Ditto.
3264 (vfmlalq_high_u32): Ditto.
3265 (vfmlslq_high_u32): Ditto.
3266 (vfmlal_lane_low_u32): Ditto.
3267 (vfmlsl_lane_low_u32): Ditto.
3268 (vfmlal_laneq_low_u32): Ditto.
3269 (vfmlsl_laneq_low_u32): Ditto.
3270 (vfmlalq_lane_low_u32): Ditto.
3271 (vfmlslq_lane_low_u32): Ditto.
3272 (vfmlalq_laneq_low_u32): Ditto.
3273 (vfmlslq_laneq_low_u32): Ditto.
3274 (vfmlal_lane_high_u32): Ditto.
3275 (vfmlsl_lane_high_u32): Ditto.
3276 (vfmlal_laneq_high_u32): Ditto.
3277 (vfmlsl_laneq_high_u32): Ditto.
3278 (vfmlalq_lane_high_u32): Ditto.
3279 (vfmlslq_lane_high_u32): Ditto.
3280 (vfmlalq_laneq_high_u32): Ditto.
3281 (vfmlslq_laneq_high_u32): Ditto.
3282 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3283 (AARCH64_FL_FOR_ARCH8_4): New.
3284 (AARCH64_ISA_F16FML): New ISA flag.
3285 (TARGET_F16FML): New feature flag for fp16fml.
3286 (doc/invoke.texi): Document new fp16fml option.
3288 2018-01-10 Michael Collison <michael.collison@arm.com>
3290 * config/aarch64/aarch64-builtins.c:
3291 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3292 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3293 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3294 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3295 (AARCH64_ISA_SHA3): New ISA flag.
3296 (TARGET_SHA3): New feature flag for sha3.
3297 * config/aarch64/iterators.md (sha512_op): New int attribute.
3298 (CRYPTO_SHA512): New int iterator.
3299 (UNSPEC_SHA512H): New unspec.
3300 (UNSPEC_SHA512H2): Ditto.
3301 (UNSPEC_SHA512SU0): Ditto.
3302 (UNSPEC_SHA512SU1): Ditto.
3303 * config/aarch64/aarch64-simd-builtins.def
3304 (aarch64_crypto_sha512hqv2di): New builtin.
3305 (aarch64_crypto_sha512h2qv2di): Ditto.
3306 (aarch64_crypto_sha512su0qv2di): Ditto.
3307 (aarch64_crypto_sha512su1qv2di): Ditto.
3308 (aarch64_eor3qv8hi): Ditto.
3309 (aarch64_rax1qv2di): Ditto.
3310 (aarch64_xarqv2di): Ditto.
3311 (aarch64_bcaxqv8hi): Ditto.
3312 * config/aarch64/aarch64-simd.md:
3313 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3314 (aarch64_crypto_sha512su0qv2di): Ditto.
3315 (aarch64_crypto_sha512su1qv2di): Ditto.
3316 (aarch64_eor3qv8hi): Ditto.
3317 (aarch64_rax1qv2di): Ditto.
3318 (aarch64_xarqv2di): Ditto.
3319 (aarch64_bcaxqv8hi): Ditto.
3320 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3321 (vsha512h2q_u64): Ditto.
3322 (vsha512su0q_u64): Ditto.
3323 (vsha512su1q_u64): Ditto.
3324 (veor3q_u16): Ditto.
3325 (vrax1q_u64): Ditto.
3327 (vbcaxq_u16): Ditto.
3328 * config/arm/types.md (crypto_sha512): New type attribute.
3329 (crypto_sha3): Ditto.
3330 (doc/invoke.texi): Document new sha3 option.
3332 2018-01-10 Michael Collison <michael.collison@arm.com>
3334 * config/aarch64/aarch64-builtins.c:
3335 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3336 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3337 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3338 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3339 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3340 (AARCH64_ISA_SM4): New ISA flag.
3341 (TARGET_SM4): New feature flag for sm4.
3342 * config/aarch64/aarch64-simd-builtins.def
3343 (aarch64_sm3ss1qv4si): Ditto.
3344 (aarch64_sm3tt1aq4si): Ditto.
3345 (aarch64_sm3tt1bq4si): Ditto.
3346 (aarch64_sm3tt2aq4si): Ditto.
3347 (aarch64_sm3tt2bq4si): Ditto.
3348 (aarch64_sm3partw1qv4si): Ditto.
3349 (aarch64_sm3partw2qv4si): Ditto.
3350 (aarch64_sm4eqv4si): Ditto.
3351 (aarch64_sm4ekeyqv4si): Ditto.
3352 * config/aarch64/aarch64-simd.md:
3353 (aarch64_sm3ss1qv4si): Ditto.
3354 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3355 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3356 (aarch64_sm4eqv4si): Ditto.
3357 (aarch64_sm4ekeyqv4si): Ditto.
3358 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3359 (sm3part_op): Ditto.
3360 (CRYPTO_SM3TT): Ditto.
3361 (CRYPTO_SM3PART): Ditto.
3362 (UNSPEC_SM3SS1): New unspec.
3363 (UNSPEC_SM3TT1A): Ditto.
3364 (UNSPEC_SM3TT1B): Ditto.
3365 (UNSPEC_SM3TT2A): Ditto.
3366 (UNSPEC_SM3TT2B): Ditto.
3367 (UNSPEC_SM3PARTW1): Ditto.
3368 (UNSPEC_SM3PARTW2): Ditto.
3369 (UNSPEC_SM4E): Ditto.
3370 (UNSPEC_SM4EKEY): Ditto.
3371 * config/aarch64/constraints.md (Ui2): New constraint.
3372 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3373 * config/arm/types.md (crypto_sm3): New type attribute.
3374 (crypto_sm4): Ditto.
3375 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3376 (vsm3tt1aq_u32): Ditto.
3377 (vsm3tt1bq_u32): Ditto.
3378 (vsm3tt2aq_u32): Ditto.
3379 (vsm3tt2bq_u32): Ditto.
3380 (vsm3partw1q_u32): Ditto.
3381 (vsm3partw2q_u32): Ditto.
3382 (vsm4eq_u32): Ditto.
3383 (vsm4ekeyq_u32): Ditto.
3384 (doc/invoke.texi): Document new sm4 option.
3386 2018-01-10 Michael Collison <michael.collison@arm.com>
3388 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3389 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3390 (AARCH64_FL_FOR_ARCH8_4): New.
3391 (AARCH64_FL_V8_4): New flag.
3392 (doc/invoke.texi): Document new armv8.4-a option.
3394 2018-01-10 Michael Collison <michael.collison@arm.com>
3396 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3397 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3398 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3399 * config/aarch64/aarch64-option-extension.def: Add
3400 AARCH64_OPT_EXTENSION of 'sha2'.
3401 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3402 (crypto): Disable sha2 and aes if crypto disabled.
3403 (crypto): Enable aes and sha2 if enabled.
3404 (simd): Disable sha2 and aes if simd disabled.
3405 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3407 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3408 (TARGET_SHA2): New feature flag for sha2.
3409 (TARGET_AES): New feature flag for aes.
3410 * config/aarch64/aarch64-simd.md:
3411 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3412 conditional on TARGET_AES.
3413 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3414 (aarch64_crypto_sha1hsi): Make pattern conditional
3416 (aarch64_crypto_sha1hv4si): Ditto.
3417 (aarch64_be_crypto_sha1hv4si): Ditto.
3418 (aarch64_crypto_sha1su1v4si): Ditto.
3419 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3420 (aarch64_crypto_sha1su0v4si): Ditto.
3421 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3422 (aarch64_crypto_sha256su0v4si): Ditto.
3423 (aarch64_crypto_sha256su1v4si): Ditto.
3424 (doc/invoke.texi): Document new aes and sha2 options.
3426 2018-01-10 Martin Sebor <msebor@redhat.com>
3428 PR tree-optimization/83781
3429 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3432 2018-01-11 Martin Sebor <msebor@gmail.com>
3433 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3435 PR tree-optimization/83501
3436 PR tree-optimization/81703
3438 * tree-ssa-strlen.c (get_string_cst): Rename...
3439 (get_string_len): ...to this. Handle global constants.
3440 (handle_char_store): Adjust.
3442 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3443 Jim Wilson <jimw@sifive.com>
3445 * config/riscv/riscv-protos.h (riscv_output_return): New.
3446 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3447 (riscv_attribute_table, riscv_output_return),
3448 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3449 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3450 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3452 (riscv_expand_prologue): Add early return for naked function.
3453 (riscv_expand_epilogue): Likewise.
3454 (riscv_function_ok_for_sibcall): Return false for naked function.
3455 (riscv_set_current_function): New.
3456 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3457 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3458 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3459 * doc/extend.texi (RISC-V Function Attributes): New.
3461 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3463 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3464 check for 128-bit long double before checking TCmode.
3465 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3466 128-bit long doubles before checking TFmode or TCmode.
3467 (FLOAT128_IBM_P): Likewise.
3469 2018-01-10 Martin Sebor <msebor@redhat.com>
3471 PR tree-optimization/83671
3472 * builtins.c (c_strlen): Unconditionally return zero for the empty
3474 Use -Warray-bounds for warnings.
3475 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3476 for non-constant array indices with COMPONENT_REF, arrays of
3477 arrays, and pointers to arrays.
3478 (gimple_fold_builtin_strlen): Determine and set length range for
3479 non-constant character arrays.
3481 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3484 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3487 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3489 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3491 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3494 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3495 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3496 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3497 indexed_or_indirect_operand predicate.
3498 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3499 (*vsx_le_perm_load_v8hi): Likewise.
3500 (*vsx_le_perm_load_v16qi): Likewise.
3501 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3502 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3503 (*vsx_le_perm_store_v8hi): Likewise.
3504 (*vsx_le_perm_store_v16qi): Likewise.
3505 (eight unnamed splitters): Likewise.
3507 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3509 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3510 * config/rs6000/emmintrin.h: Likewise.
3511 * config/rs6000/mmintrin.h: Likewise.
3512 * config/rs6000/xmmintrin.h: Likewise.
3514 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3517 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3519 * tree.c (tree_nop_conversion): Return true for location wrapper
3521 (maybe_wrap_with_location): New function.
3522 (selftest::check_strip_nops): New function.
3523 (selftest::test_location_wrappers): New function.
3524 (selftest::tree_c_tests): Call it.
3525 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3526 (maybe_wrap_with_location): New decl.
3527 (EXPR_LOCATION_WRAPPER_P): New macro.
3528 (location_wrapper_p): New inline function.
3529 (tree_strip_any_location_wrapper): New inline function.
3531 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3534 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3535 stack_realign_offset for the largest alignment of stack slot
3537 (ix86_find_max_used_stack_alignment): New function.
3538 (ix86_finalize_stack_frame_flags): Use it. Set
3539 max_used_stack_alignment if we don't realign stack.
3540 * config/i386/i386.h (machine_function): Add
3541 max_used_stack_alignment.
3543 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3545 * config/arm/arm.opt (-mbranch-cost): New option.
3546 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3549 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3552 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3553 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3555 2018-01-10 Richard Biener <rguenther@suse.de>
3558 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3559 early out so it also covers the case where we have a non-NULL
3562 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3564 PR tree-optimization/83753
3565 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3566 for non-strided grouped accesses if the number of elements is 1.
3568 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3571 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3572 * i386.h (TARGET_USE_GATHER): Define.
3573 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3575 2018-01-10 Martin Liska <mliska@suse.cz>
3578 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3579 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3581 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3582 CLEANUP_NO_PARTITIONING is not set.
3584 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3586 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3587 for vectors, as a partial revert of r254296.
3588 * rtl.h (const_vec_p): Delete.
3589 (const_vec_duplicate_p): Don't test for vector CONSTs.
3590 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3591 * expmed.c (make_tree): Likewise.
3594 * common.md (E, F): Use CONSTANT_P instead of checking for
3596 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3597 checking for CONST_VECTOR.
3599 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3602 * predict.c (force_edge_cold): Handle in more sane way edges
3605 2018-01-09 Carl Love <cel@us.ibm.com>
3607 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3609 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3610 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3611 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3612 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3613 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3614 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3615 * config/rs6000/rs6000-protos.h: Add extern defition for
3616 rs6000_generate_float2_double_code.
3617 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3619 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3620 (float2_v2df): Add define_expand.
3622 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3625 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3626 op_mode in the force_to_mode call.
3628 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3630 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3631 instead of checking each element individually.
3632 (aarch64_evpc_uzp): Likewise.
3633 (aarch64_evpc_zip): Likewise.
3634 (aarch64_evpc_ext): Likewise.
3635 (aarch64_evpc_rev): Likewise.
3636 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3637 instead of checking each element individually. Return true without
3639 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3640 whether all selected elements come from the same input, instead of
3641 checking each element individually. Remove calls to gen_rtx_REG,
3642 start_sequence and end_sequence and instead assert that no rtl is
3645 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3647 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3648 order of HIGH and CONST checks.
3650 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3652 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3653 if the destination isn't an SSA_NAME.
3655 2018-01-09 Richard Biener <rguenther@suse.de>
3657 PR tree-optimization/83668
3658 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3660 (canonicalize_loop_form): ... here, renamed from ...
3661 (canonicalize_loop_closed_ssa_form): ... this and amended to
3662 swap successor edges for loop exit blocks to make us use
3663 the RPO order we need for initial schedule generation.
3665 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3667 PR tree-optimization/64811
3668 * match.pd: When optimizing comparisons with Inf, avoid
3669 introducing or losing exceptions from comparisons with NaN.
3671 2018-01-09 Martin Liska <mliska@suse.cz>
3674 * asan.c (shadow_mem_size): Add gcc_assert.
3676 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3678 Don't save registers in main().
3681 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3682 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3683 * config/avr/avr.c (avr_set_current_function): Don't error if
3684 naked, OS_task or OS_main are specified at the same time.
3685 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3687 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3689 * common/config/avr/avr-common.c (avr_option_optimization_table):
3690 Switch on -mmain-is-OS_task for optimizing compilations.
3692 2018-01-09 Richard Biener <rguenther@suse.de>
3694 PR tree-optimization/83572
3695 * graphite.c: Include cfganal.h.
3696 (graphite_transform_loops): Connect infinite loops to exit
3697 and remove fake edges at the end.
3699 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3701 * ipa-inline.c (edge_badness): Revert accidental checkin.
3703 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3706 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3707 symbols; not inline clones.
3709 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3712 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3713 hard registers. Formatting fixes.
3715 PR preprocessor/83722
3716 * gcc.c (try_generate_repro): Pass
3717 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3718 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3721 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3722 Kito Cheng <kito.cheng@gmail.com>
3724 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3725 (riscv_leaf_function_p): Delete.
3726 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3728 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3730 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3732 (do_ifelse): New function.
3733 (do_isel): New function.
3734 (do_sub3): New function.
3735 (do_add3): New function.
3736 (do_load_mask_compare): New function.
3737 (do_overlap_load_compare): New function.
3738 (expand_compare_loop): New function.
3739 (expand_block_compare): Call expand_compare_loop() when appropriate.
3740 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3742 (-mblock-compare-inline-loop-limit): New option.
3744 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3747 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3748 Reverse order of second and third operands in first alternative.
3749 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3750 of first and second elements in UNSPEC_VPERMR vector.
3751 (altivec_expand_vec_perm_le): Likewise.
3753 2017-01-08 Jeff Law <law@redhat.com>
3755 PR rtl-optimizatin/81308
3756 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3757 (process_switch): If group_case_labels makes a change, then set
3759 (pass_convert_switch::execute): If a switch is converted, then
3760 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3762 PR rtl-optimization/81308
3763 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3766 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3768 PR target/83663 - Revert r255946
3769 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3770 generation for cases where splatting a value is not useful.
3771 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3772 across a vec_duplicate and a paradoxical subreg forming a vector
3773 mode to a vec_concat.
3775 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3777 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3778 -march=armv8.3-a variants.
3779 * config/arm/t-multilib: Likewise.
3780 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3782 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3784 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3786 (cceq_ior_compare_complement): Give it a name so I can use it, and
3787 change boolean_or_operator predicate to boolean_operator so it can
3788 be used to generate a crand.
3789 (eqne): New code iterator.
3790 (bd/bd_neg): New code_attrs.
3791 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3792 a single define_insn.
3793 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3794 decrement (bdnzt/bdnzf/bdzt/bdzf).
3795 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3796 with the new names of the branch decrement patterns, and added the
3797 names of the branch decrement conditional patterns.
3799 2018-01-08 Richard Biener <rguenther@suse.de>
3801 PR tree-optimization/83563
3802 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3805 2018-01-08 Richard Biener <rguenther@suse.de>
3808 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3810 2018-01-08 Richard Biener <rguenther@suse.de>
3812 PR tree-optimization/83685
3813 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3814 references to abnormals.
3816 2018-01-08 Richard Biener <rguenther@suse.de>
3819 * dwarf2out.c (output_indirect_strings): Handle empty
3820 skeleton_debug_str_hash.
3821 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3823 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3825 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3826 (emit_store_direct): Likewise.
3827 (arc_trampoline_adjust_address): Likewise.
3828 (arc_asm_trampoline_template): New function.
3829 (arc_initialize_trampoline): Use asm_trampoline_template.
3830 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3831 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3832 * config/arc/arc.md (flush_icache): Delete pattern.
3834 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3836 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3837 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3840 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3843 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3844 by not USED_FOR_TARGET.
3845 (make_pass_resolve_sw_modes): Likewise.
3847 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3849 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3852 2018-01-08 Richard Biener <rguenther@suse.de>
3855 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3857 2018-01-08 Richard Biener <rguenther@suse.de>
3860 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3862 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3865 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3866 basic blocks with a small number of successors.
3867 (convert_control_dep_chain_into_preds): Improve handling of
3869 (dump_predicates): Split apart into...
3870 (dump_pred_chain): ...here...
3871 (dump_pred_info): ...and here.
3872 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3873 (can_chain_union_be_invalidated_p): Improve check for invalidation
3875 (uninit_uses_cannot_happen): Avoid unnecessary if
3876 convert_control_dep_chain_into_preds yielded nothing.
3878 2018-01-06 Martin Sebor <msebor@redhat.com>
3880 PR tree-optimization/83640
3881 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3882 subtracting negative offset from size.
3883 (builtin_access::overlap): Adjust offset bounds of the access to fall
3884 within the size of the object if possible.
3886 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3888 PR rtl-optimization/83699
3889 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3890 extract_bit_field_as_subreg to cases in which the extracted
3891 value is also a vector.
3893 * lra-constraints.c (process_alt_operands): Test for the equivalence
3894 substitutions when detecting a possible reload cycle.
3896 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3899 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3900 by default if flag_selective_schedling{,2}. Formatting fixes.
3902 PR rtl-optimization/83682
3903 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
3904 if it has non-VECTOR_MODE element mode.
3905 (vec_duplicate_p): Likewise.
3908 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
3909 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
3911 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3914 * config/i386/i386-builtin.def
3915 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
3916 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
3917 Require also OPTION_MASK_ISA_AVX512F in addition to
3918 OPTION_MASK_ISA_GFNI.
3919 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
3920 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
3921 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
3922 to OPTION_MASK_ISA_GFNI.
3923 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
3924 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
3925 OPTION_MASK_ISA_AVX512BW.
3926 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
3927 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
3928 addition to OPTION_MASK_ISA_GFNI.
3929 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
3930 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
3931 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
3932 to OPTION_MASK_ISA_GFNI.
3933 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
3934 a requirement for all ISAs rather than any of them with a few
3936 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
3938 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
3939 bitmasks to be enabled with 3 exceptions, instead of requiring any
3940 enabled ISA with lots of exceptions.
3941 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
3942 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
3943 Change avx512bw in isa attribute to avx512f.
3944 * config/i386/sgxintrin.h: Add license boilerplate.
3945 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
3946 to __AVX512F__ and __AVX512VL to __AVX512VL__.
3947 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
3948 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
3950 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
3951 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
3952 temporarily sse2 rather than sse if not enabled already.
3955 * config/i386/sse.md (VI248_VLBW): Rename to ...
3956 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
3957 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
3958 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
3959 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
3960 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
3961 mode iterator instead of VI248_VLBW.
3963 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
3965 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
3966 (record_modified): Skip clobbers; add debug output.
3967 (param_change_prob): Use sreal frequencies.
3969 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3971 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
3972 punt for user-aligned variables.
3974 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3976 * tree-chrec.c (chrec_contains_symbols): Return true for
3979 2018-01-05 Sudakshina Das <sudi.das@arm.com>
3982 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
3983 of (x|y) == x for BICS pattern.
3985 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3987 PR tree-optimization/83605
3988 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
3989 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
3992 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
3994 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
3995 * config/epiphany/rtems.h: New file.
3997 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3998 Uros Bizjak <ubizjak@gmail.com>
4001 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4002 QIreg_operand instead of register_operand predicate.
4003 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4004 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4005 comments instead of -fmitigate[-_]rop.
4007 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4010 * cgraphunit.c (symbol_table::compile): Switch to text_section
4011 before calling assembly_start debug hook.
4012 * run-rtl-passes.c (run_rtl_passes): Likewise.
4015 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4017 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4018 range_int_cst_p rather than !symbolic_range_p before calling
4019 extract_range_from_multiplicative_op_1.
4021 2017-01-04 Jeff Law <law@redhat.com>
4023 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4024 redundant test in assertion.
4026 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4028 * doc/rtl.texi: Document machine_mode wrapper classes.
4030 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4032 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4035 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4037 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4038 the VEC_PERM_EXPR fold to fail.
4040 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4043 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4044 to switched_sections.
4046 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4049 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4052 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4055 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4056 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4058 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4061 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4062 is BLKmode and bitpos not zero or mode change is needed.
4064 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4067 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4070 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4073 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4074 instead of MULT rtx. Update all corresponding splitters.
4076 (*ssub<modesuffix>): Ditto.
4078 (*cmp_sadd_di): Update split patterns.
4079 (*cmp_sadd_si): Ditto.
4080 (*cmp_sadd_sidi): Ditto.
4081 (*cmp_ssub_di): Ditto.
4082 (*cmp_ssub_si): Ditto.
4083 (*cmp_ssub_sidi): Ditto.
4084 * config/alpha/predicates.md (const23_operand): New predicate.
4085 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4086 Look for ASHIFT, not MULT inner operand.
4087 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4089 2018-01-04 Martin Liska <mliska@suse.cz>
4091 PR gcov-profile/83669
4092 * gcov.c (output_intermediate_file): Add version to intermediate
4094 * doc/gcov.texi: Document new field 'version' in intermediate
4095 file format. Fix location of '-k' option of gcov command.
4097 2018-01-04 Martin Liska <mliska@suse.cz>
4100 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4102 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4104 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4106 2018-01-03 Martin Sebor <msebor@redhat.com>
4108 PR tree-optimization/83655
4109 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4110 checking calls with invalid arguments.
4112 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4114 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4115 (vectorizable_mask_load_store): Delete.
4116 (vectorizable_call): Return false for masked loads and stores.
4117 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4118 instead of gimple_assign_rhs1.
4119 (vectorizable_load): Handle IFN_MASK_LOAD.
4120 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4122 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4124 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4126 (vectorizable_mask_load_store): ...here.
4127 (vectorizable_load): ...and here.
4129 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4131 * tree-vect-stmts.c (vect_build_all_ones_mask)
4132 (vect_build_zero_merge_argument): New functions, split out from...
4133 (vectorizable_load): ...here.
4135 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4137 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4139 (vectorizable_mask_load_store): ...here.
4140 (vectorizable_store): ...and here.
4142 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4144 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4146 (vectorizable_mask_load_store): ...here.
4148 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4150 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4151 (vect_model_store_cost): Take a vec_load_store_type instead of a
4153 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4154 (vect_model_store_cost): Take a vec_load_store_type instead of a
4156 (vectorizable_mask_load_store): Update accordingly.
4157 (vectorizable_store): Likewise.
4158 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4160 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4162 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4163 IFN_MASK_LOAD calls here rather than...
4164 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4166 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4167 Alan Hayward <alan.hayward@arm.com>
4168 David Sherwood <david.sherwood@arm.com>
4170 * expmed.c (extract_bit_field_1): For vector extracts,
4171 fall back to extract_bit_field_as_subreg if vec_extract
4174 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4175 Alan Hayward <alan.hayward@arm.com>
4176 David Sherwood <david.sherwood@arm.com>
4178 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4179 they are variable or constant sized.
4180 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4181 slots for constant-sized data.
4183 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4184 Alan Hayward <alan.hayward@arm.com>
4185 David Sherwood <david.sherwood@arm.com>
4187 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4188 handling COND_EXPRs with boolean comparisons, try to find a better
4189 basis for the mask type than the boolean itself.
4191 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4193 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4194 is calculated and how it can be overridden.
4195 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4196 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4198 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4201 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4202 Alan Hayward <alan.hayward@arm.com>
4203 David Sherwood <david.sherwood@arm.com>
4205 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4206 Remove the mode argument.
4207 (aarch64_simd_valid_immediate): Remove the mode and inverse
4209 * config/aarch64/iterators.md (bitsize): New iterator.
4210 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4211 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4212 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4213 aarch64_simd_valid_immediate.
4214 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4215 (aarch64_reg_or_bic_imm): Likewise.
4216 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4217 with an insn_type enum and msl with a modifier_type enum.
4218 Replace element_width with a scalar_mode. Change the shift
4219 to unsigned int. Add constructors for scalar_float_mode and
4220 scalar_int_mode elements.
4221 (aarch64_vect_float_const_representable_p): Delete.
4222 (aarch64_can_const_movi_rtx_p)
4223 (aarch64_simd_scalar_immediate_valid_for_move)
4224 (aarch64_simd_make_constant): Update call to
4225 aarch64_simd_valid_immediate.
4226 (aarch64_advsimd_valid_immediate_hs): New function.
4227 (aarch64_advsimd_valid_immediate): Likewise.
4228 (aarch64_simd_valid_immediate): Remove mode and inverse
4229 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4230 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4231 and aarch64_float_const_representable_p on the result.
4232 (aarch64_output_simd_mov_immediate): Remove mode argument.
4233 Update call to aarch64_simd_valid_immediate and use of
4234 simd_immediate_info.
4235 (aarch64_output_scalar_simd_mov_immediate): Update call
4238 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4239 Alan Hayward <alan.hayward@arm.com>
4240 David Sherwood <david.sherwood@arm.com>
4242 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4243 (mode_nunits): Likewise CONST_MODE_NUNITS.
4244 * machmode.def (ADJUST_NUNITS): Document.
4245 * genmodes.c (mode_data::need_nunits_adj): New field.
4246 (blank_mode): Update accordingly.
4247 (adj_nunits): New variable.
4248 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4250 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4251 listed in adj_nunits.
4252 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4253 listed in adj_nunits. Don't emit case statements for such modes.
4254 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4255 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4256 nothing if adj_nunits is nonnull.
4257 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4258 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4259 (emit_mode_fbit): Update use of print_maybe_const_decl.
4260 (emit_move_size): Likewise. Treat the array as non-const
4262 (emit_mode_adjustments): Handle adj_nunits.
4264 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4266 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4267 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4268 (VECTOR_MODES): Use it.
4269 (make_vector_modes): Take the prefix as an argument.
4271 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4272 Alan Hayward <alan.hayward@arm.com>
4273 David Sherwood <david.sherwood@arm.com>
4275 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4276 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4277 for MODE_VECTOR_BOOL.
4278 * machmode.def (VECTOR_BOOL_MODE): Document.
4279 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4280 (make_vector_bool_mode): New function.
4281 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4283 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4284 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4286 * stor-layout.c (int_mode_for_mode): Likewise.
4287 * tree.c (build_vector_type_for_mode): Likewise.
4288 * varasm.c (output_constant_pool_2): Likewise.
4289 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4290 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4291 for MODE_VECTOR_BOOL.
4292 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4293 of mode class checks.
4294 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4295 instead of a list of mode class checks.
4296 (expand_vector_scalar_condition): Likewise.
4297 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4299 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4300 Alan Hayward <alan.hayward@arm.com>
4301 David Sherwood <david.sherwood@arm.com>
4303 * machmode.h (mode_size): Change from unsigned short to
4305 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4306 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4307 or if measurement_type is not polynomial.
4308 (fixed_size_mode::includes_p): Check for constant-sized modes.
4309 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4310 return a poly_uint16 rather than an unsigned short.
4311 (emit_mode_size): Change the type of mode_size from unsigned short
4312 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4313 (emit_mode_adjustments): Cope with polynomial vector sizes.
4314 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4316 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4318 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4319 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4320 * caller-save.c (setup_save_areas): Likewise.
4321 (replace_reg_with_saved_mem): Likewise.
4322 * calls.c (emit_library_call_value_1): Likewise.
4323 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4324 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4325 (gen_lowpart_for_combine): Likewise.
4326 * convert.c (convert_to_integer_1): Likewise.
4327 * cse.c (equiv_constant, cse_insn): Likewise.
4328 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4329 (cselib_subst_to_values): Likewise.
4330 * dce.c (word_dce_process_block): Likewise.
4331 * df-problems.c (df_word_lr_mark_ref): Likewise.
4332 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4333 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4334 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4335 (rtl_for_decl_location): Likewise.
4336 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4337 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4338 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4339 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4340 (expand_expr_real_1): Likewise.
4341 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4342 (pad_below): Likewise.
4343 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4344 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4345 * ira.c (get_subreg_tracking_sizes): Likewise.
4346 * ira-build.c (ira_create_allocno_objects): Likewise.
4347 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4348 (ira_sort_regnos_for_alter_reg): Likewise.
4349 * ira-costs.c (record_operand_costs): Likewise.
4350 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4351 (resolve_simple_move): Likewise.
4352 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4353 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4354 (lra_constraints): Likewise.
4355 (CONST_POOL_OK_P): Reject variable-sized modes.
4356 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4357 (add_pseudo_to_slot, lra_spill): Likewise.
4358 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4359 * optabs-query.c (get_best_extraction_insn): Likewise.
4360 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4361 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4362 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4363 * recog.c (offsettable_address_addr_space_p): Likewise.
4364 * regcprop.c (maybe_mode_change): Likewise.
4365 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4366 * regrename.c (build_def_use): Likewise.
4367 * regstat.c (dump_reg_info): Likewise.
4368 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4369 (find_reloads, find_reloads_subreg_address): Likewise.
4370 * reload1.c (eliminate_regs_1): Likewise.
4371 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4372 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4373 (simplify_binary_operation_1, simplify_subreg): Likewise.
4374 * targhooks.c (default_function_arg_padding): Likewise.
4375 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4376 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4377 (verify_gimple_assign_ternary): Likewise.
4378 * tree-inline.c (estimate_move_cost): Likewise.
4379 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4380 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4381 (get_address_cost_ainc): Likewise.
4382 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4383 (vect_supportable_dr_alignment): Likewise.
4384 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4385 (vectorizable_reduction): Likewise.
4386 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4387 (vectorizable_operation, vectorizable_load): Likewise.
4388 * tree.c (build_same_sized_truth_vector_type): Likewise.
4389 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4390 * var-tracking.c (emit_note_insn_var_location): Likewise.
4391 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4392 (ADDR_VEC_ALIGN): Likewise.
4394 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4395 Alan Hayward <alan.hayward@arm.com>
4396 David Sherwood <david.sherwood@arm.com>
4398 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4400 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4401 or if measurement_type is polynomial.
4402 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4403 * combine.c (make_extraction): Likewise.
4404 * dse.c (find_shift_sequence): Likewise.
4405 * dwarf2out.c (mem_loc_descriptor): Likewise.
4406 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4407 (extract_bit_field, extract_low_bits): Likewise.
4408 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4409 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4410 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4411 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4412 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4413 * reload.c (find_reloads): Likewise.
4414 * reload1.c (alter_reg): Likewise.
4415 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4416 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4417 * tree-if-conv.c (predicate_mem_writes): Likewise.
4418 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4419 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4420 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4421 * valtrack.c (dead_debug_insert_temp): Likewise.
4422 * varasm.c (mergeable_constant_section): Likewise.
4423 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4425 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4426 Alan Hayward <alan.hayward@arm.com>
4427 David Sherwood <david.sherwood@arm.com>
4429 * expr.c (expand_assignment): Cope with polynomial mode sizes
4430 when assigning to a CONCAT.
4432 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4433 Alan Hayward <alan.hayward@arm.com>
4434 David Sherwood <david.sherwood@arm.com>
4436 * machmode.h (mode_precision): Change from unsigned short to
4438 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4440 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4441 or if measurement_type is not polynomial.
4442 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4443 in which the mode is already known to be a scalar_int_mode.
4444 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4445 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4447 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4448 for GET_MODE_PRECISION.
4449 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4450 for GET_MODE_PRECISION.
4451 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4453 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4454 (expand_field_assignment, make_extraction): Likewise.
4455 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4456 (get_last_value): Likewise.
4457 * convert.c (convert_to_integer_1): Likewise.
4458 * cse.c (cse_insn): Likewise.
4459 * expr.c (expand_expr_real_1): Likewise.
4460 * lra-constraints.c (simplify_operand_subreg): Likewise.
4461 * optabs-query.c (can_atomic_load_p): Likewise.
4462 * optabs.c (expand_atomic_load): Likewise.
4463 (expand_atomic_store): Likewise.
4464 * ree.c (combine_reaching_defs): Likewise.
4465 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4466 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4467 * tree.h (type_has_mode_precision_p): Likewise.
4468 * ubsan.c (instrument_si_overflow): Likewise.
4470 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4471 Alan Hayward <alan.hayward@arm.com>
4472 David Sherwood <david.sherwood@arm.com>
4474 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4475 polynomial numbers of units.
4476 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4477 (valid_vector_subparts_p): New function.
4478 (build_vector_type): Remove temporary shim and take the number
4479 of units as a poly_uint64 rather than an int.
4480 (build_opaque_vector_type): Take the number of units as a
4481 poly_uint64 rather than an int.
4482 * tree.c (build_vector_from_ctor): Handle polynomial
4483 TYPE_VECTOR_SUBPARTS.
4484 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4485 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4486 (build_vector_from_val): If the number of units is variable,
4487 use build_vec_duplicate_cst for constant operands and
4488 VEC_DUPLICATE_EXPR otherwise.
4489 (make_vector_type): Remove temporary is_constant ().
4490 (build_vector_type, build_opaque_vector_type): Take the number of
4491 units as a poly_uint64 rather than an int.
4492 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4494 * cfgexpand.c (expand_debug_expr): Likewise.
4495 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4496 (store_constructor, expand_expr_real_1): Likewise.
4497 (const_scalar_mask_from_tree): Likewise.
4498 * fold-const-call.c (fold_const_reduction): Likewise.
4499 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4500 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4501 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4502 (fold_relational_const): Likewise.
4503 (native_interpret_vector): Likewise. Change the size from an
4504 int to an unsigned int.
4505 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4506 TYPE_VECTOR_SUBPARTS.
4507 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4508 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4509 duplicating a non-constant operand into a variable-length vector.
4510 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4511 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4512 * ipa-icf.c (sem_variable::equals): Likewise.
4513 * match.pd: Likewise.
4514 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4515 * print-tree.c (print_node): Likewise.
4516 * stor-layout.c (layout_type): Likewise.
4517 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4518 * tree-cfg.c (verify_gimple_comparison): Likewise.
4519 (verify_gimple_assign_binary): Likewise.
4520 (verify_gimple_assign_ternary): Likewise.
4521 (verify_gimple_assign_single): Likewise.
4522 * tree-pretty-print.c (dump_generic_node): Likewise.
4523 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4524 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4525 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4526 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4527 (vect_shift_permute_load_chain): Likewise.
4528 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4529 (expand_vector_condition, optimize_vector_constructor): Likewise.
4530 (lower_vec_perm, get_compute_type): Likewise.
4531 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4532 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4533 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4534 (vect_recog_mask_conversion_pattern): Likewise.
4535 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4536 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4537 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4538 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4539 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4540 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4541 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4542 (supportable_widening_operation): Likewise.
4543 (supportable_narrowing_operation): Likewise.
4544 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4546 * varasm.c (output_constant): Likewise.
4548 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4549 Alan Hayward <alan.hayward@arm.com>
4550 David Sherwood <david.sherwood@arm.com>
4552 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4553 so that both the length == 3 and length != 3 cases set up their
4554 own permute vectors. Add comments explaining why we know the
4555 number of elements is constant.
4556 (vect_permute_load_chain): Likewise.
4558 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4559 Alan Hayward <alan.hayward@arm.com>
4560 David Sherwood <david.sherwood@arm.com>
4562 * machmode.h (mode_nunits): Change from unsigned char to
4564 (ONLY_FIXED_SIZE_MODES): New macro.
4565 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4566 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4567 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4569 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4570 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4571 or if measurement_type is not polynomial.
4572 * genmodes.c (ZERO_COEFFS): New macro.
4573 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4575 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4576 Use ZERO_COEFFS when emitting initializers.
4577 * data-streamer.h (bp_pack_poly_value): New function.
4578 (bp_unpack_poly_value): Likewise.
4579 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4580 for GET_MODE_NUNITS.
4581 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4582 for GET_MODE_NUNITS.
4583 * tree.c (make_vector_type): Remove temporary shim and make
4584 the real function take the number of units as a poly_uint64
4586 (build_vector_type_for_mode): Handle polynomial nunits.
4587 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4588 * emit-rtl.c (const_vec_series_p_1): Likewise.
4589 (gen_rtx_CONST_VECTOR): Likewise.
4590 * fold-const.c (test_vec_duplicate_folding): Likewise.
4591 * genrecog.c (validate_pattern): Likewise.
4592 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4593 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4594 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4595 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4596 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4597 * rtlanal.c (subreg_get_info): Likewise.
4598 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4599 (vect_grouped_load_supported): Likewise.
4600 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4601 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4602 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4603 (simplify_const_unary_operation, simplify_binary_operation_1)
4604 (simplify_const_binary_operation, simplify_ternary_operation)
4605 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4606 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4607 instead of CONST_VECTOR_NUNITS.
4608 * varasm.c (output_constant_pool_2): Likewise.
4609 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4610 explicit-encoded elements in the XVEC for variable-length vectors.
4612 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4614 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4616 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4617 Alan Hayward <alan.hayward@arm.com>
4618 David Sherwood <david.sherwood@arm.com>
4620 * coretypes.h (fixed_size_mode): Declare.
4621 (fixed_size_mode_pod): New typedef.
4622 * builtins.h (target_builtins::x_apply_args_mode)
4623 (target_builtins::x_apply_result_mode): Change type to
4624 fixed_size_mode_pod.
4625 * builtins.c (apply_args_size, apply_result_size, result_vector)
4626 (expand_builtin_apply_args_1, expand_builtin_apply)
4627 (expand_builtin_return): Update accordingly.
4629 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4631 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4632 * cselib.c (cselib_hash_rtx): Likewise.
4633 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4634 CONST_VECTOR encoding.
4636 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4637 Jeff Law <law@redhat.com>
4640 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4641 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4642 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4643 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4646 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4647 explicitly probe *sp in a noreturn function if there were any callee
4648 register saves or frame pointer is needed.
4650 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4653 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4654 BLKmode for ternary, binary or unary expressions.
4657 * var-tracking.c (delete_vta_debug_insn): New inline function.
4658 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4659 insns from get_insns () to NULL instead of each bb separately.
4660 Use delete_vta_debug_insn. No longer static.
4661 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4662 delete_vta_debug_insns callers.
4663 * rtl.h (delete_vta_debug_insns): Declare.
4664 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4665 instead of variable_tracking_main.
4667 2018-01-03 Martin Sebor <msebor@redhat.com>
4669 PR tree-optimization/83603
4670 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4671 arguments past the endof the argument list in functions declared
4672 without a prototype.
4673 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4674 Avoid checking when arguments are null.
4676 2018-01-03 Martin Sebor <msebor@redhat.com>
4679 * doc/extend.texi (attribute const): Fix a typo.
4680 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4681 issuing -Wsuggest-attribute for void functions.
4683 2018-01-03 Martin Sebor <msebor@redhat.com>
4685 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4686 offset_int::from instead of wide_int::to_shwi.
4687 (maybe_diag_overlap): Remove assertion.
4688 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4689 * gimple-ssa-sprintf.c (format_directive): Same.
4690 (parse_directive): Same.
4691 (sprintf_dom_walker::compute_format_length): Same.
4692 (try_substitute_return_value): Same.
4694 2017-01-03 Jeff Law <law@redhat.com>
4697 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4698 non-constant residual for zero at runtime and avoid probing in
4699 that case. Reorganize code for trailing problem to mirror handling
4702 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4704 PR tree-optimization/83501
4705 * tree-ssa-strlen.c (get_string_cst): New.
4706 (handle_char_store): Call get_string_cst.
4708 2018-01-03 Martin Liska <mliska@suse.cz>
4710 PR tree-optimization/83593
4711 * tree-ssa-strlen.c: Include tree-cfg.h.
4712 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4713 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4714 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4716 (strlen_dom_walker::before_dom_children): Call
4717 gimple_purge_dead_eh_edges. Dump tranformation with details
4719 (strlen_dom_walker::before_dom_children): Update call by adding
4720 new argument cleanup_eh.
4721 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4723 2018-01-03 Martin Liska <mliska@suse.cz>
4726 * cif-code.def (VARIADIC_THUNK): New enum value.
4727 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4730 2018-01-03 Jan Beulich <jbeulich@suse.com>
4732 * sse.md (mov<mode>_internal): Tighten condition for when to use
4733 vmovdqu<ssescalarsize> for TI and OI modes.
4735 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4737 Update copyright years.
4739 2018-01-03 Martin Liska <mliska@suse.cz>
4742 * ipa-visibility.c (function_and_variable_visibility): Skip
4743 functions with noipa attribure.
4745 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4747 * gcc.c (process_command): Update copyright notice dates.
4748 * gcov-dump.c (print_version): Ditto.
4749 * gcov.c (print_version): Ditto.
4750 * gcov-tool.c (print_version): Ditto.
4751 * gengtype.c (create_file): Ditto.
4752 * doc/cpp.texi: Bump @copying's copyright year.
4753 * doc/cppinternals.texi: Ditto.
4754 * doc/gcc.texi: Ditto.
4755 * doc/gccint.texi: Ditto.
4756 * doc/gcov.texi: Ditto.
4757 * doc/install.texi: Ditto.
4758 * doc/invoke.texi: Ditto.
4760 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4762 * vector-builder.h (vector_builder::m_full_nelts): Change from
4763 unsigned int to poly_uint64.
4764 (vector_builder::full_nelts): Update prototype accordingly.
4765 (vector_builder::new_vector): Likewise.
4766 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4767 (vector_builder::operator ==): Likewise.
4768 (vector_builder::finalize): Likewise.
4769 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4770 Take the number of elements as a poly_uint64 rather than an
4772 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4773 from unsigned int to poly_uint64.
4774 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4775 (vec_perm_indices::new_vector): Likewise.
4776 (vec_perm_indices::length): Likewise.
4777 (vec_perm_indices::nelts_per_input): Likewise.
4778 (vec_perm_indices::input_nelts): Likewise.
4779 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4780 number of elements per input as a poly_uint64 rather than an
4781 unsigned int. Use the original encoding for variable-length
4782 vectors, rather than clamping each individual element.
4783 For the second and subsequent elements in each pattern,
4784 clamp the step and base before clamping their sum.
4785 (vec_perm_indices::series_p): Handle polynomial element counts.
4786 (vec_perm_indices::all_in_range_p): Likewise.
4787 (vec_perm_indices_to_tree): Likewise.
4788 (vec_perm_indices_to_rtx): Likewise.
4789 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4790 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4791 (tree_vector_builder::new_binary_operation): Handle polynomial
4792 element counts. Return false if we need to know the number
4793 of elements at compile time.
4794 * fold-const.c (fold_vec_perm): Punt if the number of elements
4795 isn't known at compile time.
4797 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4799 * vec-perm-indices.h (vec_perm_builder): Change element type
4800 from HOST_WIDE_INT to poly_int64.
4801 (vec_perm_indices::element_type): Update accordingly.
4802 (vec_perm_indices::clamp): Handle polynomial element_types.
4803 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4804 (vec_perm_indices::all_in_range_p): Likewise.
4805 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4807 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4808 polynomial vec_perm_indices element types.
4809 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4810 * fold-const.c (fold_vec_perm): Likewise.
4811 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4812 * tree-vect-generic.c (lower_vec_perm): Likewise.
4813 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4814 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4815 element type to HOST_WIDE_INT.
4817 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4818 Alan Hayward <alan.hayward@arm.com>
4819 David Sherwood <david.sherwood@arm.com>
4821 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4822 rather than an int. Use plus_constant.
4823 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4824 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4826 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4827 Alan Hayward <alan.hayward@arm.com>
4828 David Sherwood <david.sherwood@arm.com>
4830 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4831 a HOST_WIDE_INT to a poly_int64.
4833 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4834 Alan Hayward <alan.hayward@arm.com>
4835 David Sherwood <david.sherwood@arm.com>
4837 * calls.c (load_register_parameters): Cope with polynomial
4838 mode sizes. Require a constant size for BLKmode parameters
4839 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4840 forces a parameter to be padded at the lsb end in order to
4841 fill a complete number of words, require the parameter size
4842 to be ordered wrt UNITS_PER_WORD.
4844 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4845 Alan Hayward <alan.hayward@arm.com>
4846 David Sherwood <david.sherwood@arm.com>
4848 * reload1.c (spill_stack_slot_width): Change element type
4849 from unsigned int to poly_uint64_pod.
4850 (alter_reg): Treat mode sizes as polynomial.
4852 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4853 Alan Hayward <alan.hayward@arm.com>
4854 David Sherwood <david.sherwood@arm.com>
4856 * reload.c (complex_word_subreg_p): New function.
4857 (reload_inner_reg_of_subreg, push_reload): Use it.
4859 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4860 Alan Hayward <alan.hayward@arm.com>
4861 David Sherwood <david.sherwood@arm.com>
4863 * lra-constraints.c (process_alt_operands): Reject matched
4864 operands whose sizes aren't ordered.
4865 (match_reload): Refer to this check here.
4867 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4868 Alan Hayward <alan.hayward@arm.com>
4869 David Sherwood <david.sherwood@arm.com>
4871 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4872 that the mode size is in the set {1, 2, 4, 8, 16}.
4874 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4875 Alan Hayward <alan.hayward@arm.com>
4876 David Sherwood <david.sherwood@arm.com>
4878 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4879 Use plus_constant instead of gen_rtx_PLUS.
4881 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4882 Alan Hayward <alan.hayward@arm.com>
4883 David Sherwood <david.sherwood@arm.com>
4885 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4886 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4887 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4888 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4889 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4890 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4891 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4892 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4893 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4894 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4896 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4897 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4898 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4899 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4900 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4901 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4902 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4903 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
4904 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
4905 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
4907 * expr.c (emit_move_resolve_push): Treat the input and result
4908 of PUSH_ROUNDING as a poly_int64.
4909 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
4910 (emit_push_insn): Likewise.
4911 * lra-eliminations.c (mark_not_eliminable): Likewise.
4912 * recog.c (push_operand): Likewise.
4913 * reload1.c (elimination_effects): Likewise.
4914 * rtlanal.c (nonzero_bits1): Likewise.
4915 * calls.c (store_one_arg): Likewise. Require the padding to be
4916 known at compile time.
4918 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4919 Alan Hayward <alan.hayward@arm.com>
4920 David Sherwood <david.sherwood@arm.com>
4922 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
4923 Use plus_constant instead of gen_rtx_PLUS.
4925 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4926 Alan Hayward <alan.hayward@arm.com>
4927 David Sherwood <david.sherwood@arm.com>
4929 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
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 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
4937 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
4938 via stack temporaries. Treat the mode size as polynomial too.
4940 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4941 Alan Hayward <alan.hayward@arm.com>
4942 David Sherwood <david.sherwood@arm.com>
4944 * expr.c (expand_expr_real_2): When handling conversions involving
4945 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
4946 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
4947 as a poly_uint64 too.
4949 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4950 Alan Hayward <alan.hayward@arm.com>
4951 David Sherwood <david.sherwood@arm.com>
4953 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
4955 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4956 Alan Hayward <alan.hayward@arm.com>
4957 David Sherwood <david.sherwood@arm.com>
4959 * combine.c (can_change_dest_mode): Handle polynomial
4960 REGMODE_NATURAL_SIZE.
4961 * expmed.c (store_bit_field_1): Likewise.
4962 * expr.c (store_constructor): Likewise.
4963 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
4964 and polynomial REGMODE_NATURAL_SIZE.
4965 (gen_lowpart_common): Likewise.
4966 * reginfo.c (record_subregs_of_mode): Likewise.
4967 * rtlanal.c (read_modify_subreg_p): Likewise.
4969 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4970 Alan Hayward <alan.hayward@arm.com>
4971 David Sherwood <david.sherwood@arm.com>
4973 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
4974 numbers of elements.
4976 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4977 Alan Hayward <alan.hayward@arm.com>
4978 David Sherwood <david.sherwood@arm.com>
4980 * match.pd: Cope with polynomial numbers of vector elements.
4982 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4983 Alan Hayward <alan.hayward@arm.com>
4984 David Sherwood <david.sherwood@arm.com>
4986 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
4987 in a POINTER_PLUS_EXPR.
4989 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4990 Alan Hayward <alan.hayward@arm.com>
4991 David Sherwood <david.sherwood@arm.com>
4993 * omp-simd-clone.c (simd_clone_subparts): New function.
4994 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
4995 (ipa_simd_modify_function_body): Likewise.
4997 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4998 Alan Hayward <alan.hayward@arm.com>
4999 David Sherwood <david.sherwood@arm.com>
5001 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5002 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5003 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5004 (expand_vector_condition, vector_element): Likewise.
5005 (subparts_gt): New function.
5006 (get_compute_type): Use subparts_gt.
5007 (count_type_subparts): Delete.
5008 (expand_vector_operations_1): Use subparts_gt instead of
5009 count_type_subparts.
5011 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5012 Alan Hayward <alan.hayward@arm.com>
5013 David Sherwood <david.sherwood@arm.com>
5015 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5016 (vect_compile_time_alias): ...this new function. Do the calculation
5017 on poly_ints rather than trees.
5018 (vect_prune_runtime_alias_test_list): Update call accordingly.
5020 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5021 Alan Hayward <alan.hayward@arm.com>
5022 David Sherwood <david.sherwood@arm.com>
5024 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5026 (vect_schedule_slp_instance): Likewise.
5028 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5029 Alan Hayward <alan.hayward@arm.com>
5030 David Sherwood <david.sherwood@arm.com>
5032 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5033 constant and extern definitions for variable-length vectors.
5034 (vect_get_constant_vectors): Note that the number of units
5035 is known to be constant.
5037 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5038 Alan Hayward <alan.hayward@arm.com>
5039 David Sherwood <david.sherwood@arm.com>
5041 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5042 of units as polynomial. Choose between WIDE and NARROW based
5045 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5046 Alan Hayward <alan.hayward@arm.com>
5047 David Sherwood <david.sherwood@arm.com>
5049 * tree-vect-stmts.c (simd_clone_subparts): New function.
5050 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5052 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5053 Alan Hayward <alan.hayward@arm.com>
5054 David Sherwood <david.sherwood@arm.com>
5056 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5057 vectors as polynomial. Use build_index_vector for
5060 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5061 Alan Hayward <alan.hayward@arm.com>
5062 David Sherwood <david.sherwood@arm.com>
5064 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5065 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5066 for variable-length vectors.
5067 (vectorizable_mask_load_store): Treat the number of units as
5068 polynomial, asserting that it is constant if the condition has
5069 already been enforced.
5070 (vectorizable_store, vectorizable_load): Likewise.
5072 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5073 Alan Hayward <alan.hayward@arm.com>
5074 David Sherwood <david.sherwood@arm.com>
5076 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5077 of units as polynomial. Punt if we can't tell at compile time
5078 which vector contains the final result.
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-vect-loop.c (vectorizable_induction): Treat the number
5085 of units as polynomial. Punt on SLP inductions. Use an integer
5086 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5087 cast of such a series for variable-length floating-point
5090 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5091 Alan Hayward <alan.hayward@arm.com>
5092 David Sherwood <david.sherwood@arm.com>
5094 * tree.h (build_index_vector): Declare.
5095 * tree.c (build_index_vector): New function.
5096 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5097 of units as polynomial, forcibly converting it to a constant if
5098 vectorizable_reduction has already enforced the condition.
5099 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5100 to create a {1,2,3,...} vector.
5101 (vectorizable_reduction): Treat the number of units as polynomial.
5102 Choose vectype_in based on the largest scalar element size rather
5103 than the smallest number of units. Enforce the restrictions
5106 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5107 Alan Hayward <alan.hayward@arm.com>
5108 David Sherwood <david.sherwood@arm.com>
5110 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5111 number of units as polynomial.
5113 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5114 Alan Hayward <alan.hayward@arm.com>
5115 David Sherwood <david.sherwood@arm.com>
5117 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5118 * target.def (autovectorize_vector_sizes): Return the vector sizes
5119 by pointer, using vector_sizes rather than a bitmask.
5120 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5121 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5122 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5124 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5125 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5126 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5127 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5128 * omp-general.c (omp_max_vf): Likewise.
5129 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5130 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5131 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5132 * tree-vect-slp.c (vect_slp_bb): Likewise.
5133 * doc/tm.texi: Regenerate.
5134 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5136 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5137 the vector size as a poly_uint64 rather than an unsigned int.
5138 (current_vector_size): Change from an unsigned int to a poly_uint64.
5139 (get_vectype_for_scalar_type): Update accordingly.
5140 * tree.h (build_truth_vector_type): Take the size and number of
5141 units as a poly_uint64 rather than an unsigned int.
5142 (build_vector_type): Add a temporary overload that takes
5143 the number of units as a poly_uint64 rather than an unsigned int.
5144 * tree.c (make_vector_type): Likewise.
5145 (build_truth_vector_type): Take the number of units as a poly_uint64
5146 rather than an unsigned int.
5148 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5149 Alan Hayward <alan.hayward@arm.com>
5150 David Sherwood <david.sherwood@arm.com>
5152 * target.def (get_mask_mode): Take the number of units and length
5153 as poly_uint64s rather than unsigned ints.
5154 * targhooks.h (default_get_mask_mode): Update accordingly.
5155 * targhooks.c (default_get_mask_mode): Likewise.
5156 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5157 * doc/tm.texi: Regenerate.
5159 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5160 Alan Hayward <alan.hayward@arm.com>
5161 David Sherwood <david.sherwood@arm.com>
5163 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5164 * omp-general.c (omp_max_vf): Likewise.
5165 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5166 (expand_omp_simd): Handle polynomial safelen.
5167 * omp-low.c (omplow_simd_context): Add a default constructor.
5168 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5169 (lower_rec_simd_input_clauses): Update accordingly.
5170 (lower_rec_input_clauses): Likewise.
5172 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5173 Alan Hayward <alan.hayward@arm.com>
5174 David Sherwood <david.sherwood@arm.com>
5176 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5177 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5178 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5179 (vect_analyze_slp_cost): Likewise.
5180 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5181 (vect_model_load_cost): Likewise.
5183 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5184 Alan Hayward <alan.hayward@arm.com>
5185 David Sherwood <david.sherwood@arm.com>
5187 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5188 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5189 from an unsigned int * to a poly_uint64_pod *.
5190 (calculate_unrolling_factor): New function.
5191 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5193 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5194 Alan Hayward <alan.hayward@arm.com>
5195 David Sherwood <david.sherwood@arm.com>
5197 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5198 from an unsigned int to a poly_uint64.
5199 (_loop_vec_info::slp_unrolling_factor): Likewise.
5200 (_loop_vec_info::vectorization_factor): Change from an int
5202 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5203 (vect_get_num_vectors): New function.
5204 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5205 (vect_get_num_copies): Use vect_get_num_vectors.
5206 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5207 to an unsigned int *.
5208 (vect_analyze_data_refs): Change min_vf from an int * to a
5210 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5211 than an unsigned HOST_WIDE_INT.
5212 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5213 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5214 to an unsigned int *.
5215 (vect_analyze_data_ref_dependences): Likewise.
5216 (vect_compute_data_ref_alignment): Handle polynomial vf.
5217 (vect_enhance_data_refs_alignment): Likewise.
5218 (vect_prune_runtime_alias_test_list): Likewise.
5219 (vect_shift_permute_load_chain): Likewise.
5220 (vect_supportable_dr_alignment): Likewise.
5221 (dependence_distance_ge_vf): Take the vectorization factor as a
5222 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5223 (vect_analyze_data_refs): Change min_vf from an int * to a
5225 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5226 vfm1 as a poly_uint64 rather than an int. Make the same change
5227 for the returned bound_scalar.
5228 (vect_gen_vector_loop_niters): Handle polynomial vf.
5229 (vect_do_peeling): Likewise. Update call to
5230 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5231 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5233 * tree-vect-loop.c (vect_determine_vectorization_factor)
5234 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5235 (vect_get_known_peeling_cost): Likewise.
5236 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5237 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5238 (vect_transform_loop): Likewise. Use the lowest possible VF when
5239 updating the upper bounds of the loop.
5240 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5242 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5243 polynomial unroll factors.
5244 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5245 (vect_make_slp_decision): Likewise.
5246 (vect_supported_load_permutation_p): Likewise, and polynomial
5248 (vect_analyze_slp_cost): Handle polynomial vf.
5249 (vect_slp_analyze_node_operations): Likewise.
5250 (vect_slp_analyze_bb_1): Likewise.
5251 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5252 than an unsigned HOST_WIDE_INT.
5253 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5254 (vectorizable_load): Handle polynomial vf.
5255 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5257 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5259 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5260 Alan Hayward <alan.hayward@arm.com>
5261 David Sherwood <david.sherwood@arm.com>
5263 * match.pd: Handle bit operations involving three constants
5264 and try to fold one pair.
5266 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5268 * tree-vect-loop-manip.c: Include gimple-fold.h.
5269 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5270 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5271 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5272 Add a path that uses a step of VF instead of 1, but disable it
5274 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5275 and niters_no_overflow parameters. Update calls to
5276 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5277 Create a new SSA name if the latter choses to use a ste other
5278 than zero, and return it via niters_vector_mult_vf_var.
5279 * tree-vect-loop.c (vect_transform_loop): Update calls to
5280 vect_do_peeling, vect_gen_vector_loop_niters and
5281 slpeel_make_loop_iterate_ntimes.
5282 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5283 (vect_gen_vector_loop_niters): Update declarations after above changes.
5285 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5287 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5288 128-bit round to integer instructions.
5289 (ceil<mode>2): Likewise.
5290 (btrunc<mode>2): Likewise.
5291 (round<mode>2): Likewise.
5293 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5295 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5296 unaligned VSX load/store on P8/P9.
5297 (expand_block_clear): Allow the use of unaligned VSX
5298 load/store on P8/P9.
5300 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5302 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5304 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5305 swap associated with both a load and a store.
5307 2018-01-02 Andrew Waterman <andrew@sifive.com>
5309 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5310 * config/riscv/riscv.md (clear_cache): Use it.
5312 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5314 * web.c: Remove out-of-date comment.
5316 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5318 * expr.c (fixup_args_size_notes): Check that any existing
5319 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5320 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5321 (emit_single_push_insn): ...here.
5323 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5325 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5326 (const_vector_encoded_nelts): New function.
5327 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5328 (const_vector_int_elt, const_vector_elt): Declare.
5329 * emit-rtl.c (const_vector_int_elt_1): New function.
5330 (const_vector_elt): Likewise.
5331 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5332 of CONST_VECTOR_ELT.
5334 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5336 * expr.c: Include rtx-vector-builder.h.
5337 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5338 directly on the tree encoding.
5339 (const_vector_from_tree): Likewise.
5340 * optabs.c: Include rtx-vector-builder.h.
5341 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5342 sequence of "u" values.
5343 * vec-perm-indices.c: Include rtx-vector-builder.h.
5344 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5345 directly on the vec_perm_indices encoding.
5347 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5349 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5350 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5351 * rtx-vector-builder.h: New file.
5352 * rtx-vector-builder.c: Likewise.
5353 * rtl.h (rtx_def::u2): Add a const_vector field.
5354 (CONST_VECTOR_NPATTERNS): New macro.
5355 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5356 (CONST_VECTOR_DUPLICATE_P): Likewise.
5357 (CONST_VECTOR_STEPPED_P): Likewise.
5358 (CONST_VECTOR_ENCODED_ELT): Likewise.
5359 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5360 (unwrap_const_vec_duplicate): Likewise.
5361 (const_vec_series_p): Check for a non-duplicated vector encoding.
5362 Say that the function only returns true for integer vectors.
5363 * emit-rtl.c: Include rtx-vector-builder.h.
5364 (gen_const_vec_duplicate_1): Delete.
5365 (gen_const_vector): Call gen_const_vec_duplicate instead of
5366 gen_const_vec_duplicate_1.
5367 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5368 (gen_const_vec_duplicate): Use rtx_vector_builder.
5369 (gen_const_vec_series): Likewise.
5370 (gen_rtx_CONST_VECTOR): Likewise.
5371 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5372 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5373 Build a new vector rather than modifying a CONST_VECTOR in-place.
5374 (handle_special_swappables): Update call accordingly.
5375 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5376 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5377 Build a new vector rather than modifying a CONST_VECTOR in-place.
5378 (handle_special_swappables): Update call accordingly.
5380 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5382 * simplify-rtx.c (simplify_const_binary_operation): Use
5383 CONST_VECTOR_ELT instead of XVECEXP.
5385 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5387 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5388 the selector elements to be different from the data elements
5389 if the selector is a VECTOR_CST.
5390 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5391 ssizetype for the selector.
5393 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5395 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5396 before testing each element individually.
5397 * tree-vect-generic.c (lower_vec_perm): Likewise.
5399 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5401 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5402 * selftest-run-tests.c (selftest::run_tests): Call it.
5403 * vector-builder.h (vector_builder::operator ==): New function.
5404 (vector_builder::operator !=): Likewise.
5405 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5406 (vec_perm_indices::all_from_input_p): New function.
5407 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5408 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5409 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5410 instead of reading the VECTOR_CST directly. Detect whether both
5411 vector inputs are the same before constructing the vec_perm_indices,
5412 and update the number of inputs argument accordingly. Use the
5413 utility functions added above. Only construct sel2 if we need to.
5415 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5417 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5418 the broadcast of the low byte.
5419 (expand_mult_highpart): Use an explicit encoding for the permutes.
5420 * optabs-query.c (can_mult_highpart_p): Likewise.
5421 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5422 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5423 (vectorizable_bswap): Likewise.
5424 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5425 explicit encoding for the power-of-2 permutes.
5426 (vect_permute_store_chain): Likewise.
5427 (vect_grouped_load_supported): Likewise.
5428 (vect_permute_load_chain): Likewise.
5430 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5432 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5433 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5434 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5435 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5436 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5437 (vect_gen_perm_mask_any): Likewise.
5439 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5441 * int-vector-builder.h: New file.
5442 * vec-perm-indices.h: Include int-vector-builder.h.
5443 (vec_perm_indices): Redefine as an int_vector_builder.
5444 (auto_vec_perm_indices): Delete.
5445 (vec_perm_builder): Redefine as a stand-alone class.
5446 (vec_perm_indices::vec_perm_indices): New function.
5447 (vec_perm_indices::clamp): Likewise.
5448 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5449 (vec_perm_indices::new_vector): New function.
5450 (vec_perm_indices::new_expanded_vector): Update for new
5451 vec_perm_indices class.
5452 (vec_perm_indices::rotate_inputs): New function.
5453 (vec_perm_indices::all_in_range_p): Operate directly on the
5454 encoded form, without computing elided elements.
5455 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5456 encoding. Update for new vec_perm_indices class.
5457 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5458 the given vec_perm_builder.
5459 (expand_vec_perm_var): Update vec_perm_builder constructor.
5460 (expand_mult_highpart): Use vec_perm_builder instead of
5461 auto_vec_perm_indices.
5462 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5463 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5464 or double series encoding as appropriate.
5465 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5466 vec_perm_indices instead of auto_vec_perm_indices.
5467 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5468 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5469 (vect_permute_store_chain): Likewise.
5470 (vect_grouped_load_supported): Likewise.
5471 (vect_permute_load_chain): Likewise.
5472 (vect_shift_permute_load_chain): Likewise.
5473 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5474 (vect_transform_slp_perm_load): Likewise.
5475 (vect_schedule_slp_instance): Likewise.
5476 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5477 (vectorizable_mask_load_store): Likewise.
5478 (vectorizable_bswap): Likewise.
5479 (vectorizable_store): Likewise.
5480 (vectorizable_load): Likewise.
5481 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5482 vec_perm_indices instead of auto_vec_perm_indices. Use
5483 tree_to_vec_perm_builder to read the vector from a tree.
5484 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5485 vec_perm_builder instead of a vec_perm_indices.
5486 (have_whole_vector_shift): Use vec_perm_builder and
5487 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5488 truncation to calc_vec_perm_mask_for_shift.
5489 (vect_create_epilog_for_reduction): Likewise.
5490 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5491 from auto_vec_perm_indices to vec_perm_indices.
5492 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5493 instead of changing individual elements.
5494 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5495 the vector in d.perm.
5496 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5497 from auto_vec_perm_indices to vec_perm_indices.
5498 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5499 instead of changing individual elements.
5500 (arm_vectorize_vec_perm_const): Use new_vector to install
5501 the vector in d.perm.
5502 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5503 Update vec_perm_builder constructor.
5504 (rs6000_expand_interleave): Likewise.
5505 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5506 (rs6000_expand_interleave): Likewise.
5508 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5510 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5511 to qimode could truncate the indices.
5512 * optabs.c (expand_vec_perm_var): Likewise.
5514 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5516 * Makefile.in (OBJS): Add vec-perm-indices.o.
5517 * vec-perm-indices.h: New file.
5518 * vec-perm-indices.c: Likewise.
5519 * target.h (vec_perm_indices): Replace with a forward class
5521 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5522 * optabs.h: Include vec-perm-indices.h.
5523 (expand_vec_perm): Delete.
5524 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5525 (expand_vec_perm_const): Declare.
5526 * target.def (vec_perm_const_ok): Replace with...
5527 (vec_perm_const): ...this new hook.
5528 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5529 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5530 * doc/tm.texi: Regenerate.
5531 * optabs.def (vec_perm_const): Delete.
5532 * doc/md.texi (vec_perm_const): Likewise.
5533 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5534 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5535 expand_vec_perm for constant permutation vectors. Assert that
5536 the mode of variable permutation vectors is the integer equivalent
5537 of the mode that is being permuted.
5538 * optabs-query.h (selector_fits_mode_p): Declare.
5539 * optabs-query.c: Include vec-perm-indices.h.
5540 (selector_fits_mode_p): New function.
5541 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5542 is defined, instead of checking whether the vec_perm_const_optab
5543 exists. Use targetm.vectorize.vec_perm_const instead of
5544 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5545 fit in the vector mode before using a variable permute.
5546 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5547 vec_perm_indices instead of an rtx.
5548 (expand_vec_perm): Replace with...
5549 (expand_vec_perm_const): ...this new function. Take the selector
5550 as a vec_perm_indices rather than an rtx. Also take the mode of
5551 the selector. Update call to shift_amt_for_vec_perm_mask.
5552 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5553 Use vec_perm_indices::new_expanded_vector to expand the original
5554 selector into bytes. Check whether the indices fit in the vector
5555 mode before using a variable permute.
5556 (expand_vec_perm_var): Make global.
5557 (expand_mult_highpart): Use expand_vec_perm_const.
5558 * fold-const.c: Includes vec-perm-indices.h.
5559 * tree-ssa-forwprop.c: Likewise.
5560 * tree-vect-data-refs.c: Likewise.
5561 * tree-vect-generic.c: Likewise.
5562 * tree-vect-loop.c: Likewise.
5563 * tree-vect-slp.c: Likewise.
5564 * tree-vect-stmts.c: Likewise.
5565 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5567 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5568 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5569 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5570 (aarch64_vectorize_vec_perm_const): ...this new function.
5571 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5572 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5573 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5574 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5575 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5576 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5577 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5579 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5580 check for NEON modes.
5581 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5582 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5583 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5584 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5586 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5587 the old VEC_PERM_CONST conditions.
5588 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5589 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5590 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5591 (ia64_vectorize_vec_perm_const_ok): Merge into...
5592 (ia64_vectorize_vec_perm_const): ...this new function.
5593 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5594 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5595 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5596 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5597 * config/mips/mips.c (mips_expand_vec_perm_const)
5598 (mips_vectorize_vec_perm_const_ok): Merge into...
5599 (mips_vectorize_vec_perm_const): ...this new function.
5600 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5601 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5602 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5603 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5604 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5605 (rs6000_expand_vec_perm_const): Delete.
5606 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5608 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5609 (altivec_expand_vec_perm_const_le): Take each operand individually.
5610 Operate on constant selectors rather than rtxes.
5611 (altivec_expand_vec_perm_const): Likewise. Update call to
5612 altivec_expand_vec_perm_const_le.
5613 (rs6000_expand_vec_perm_const): Delete.
5614 (rs6000_vectorize_vec_perm_const_ok): Delete.
5615 (rs6000_vectorize_vec_perm_const): New function.
5616 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5617 an element count and rtx array.
5618 (rs6000_expand_extract_even): Update call accordingly.
5619 (rs6000_expand_interleave): Likewise.
5620 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5621 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5622 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5623 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5624 (rs6000_expand_vec_perm_const): Delete.
5625 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5626 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5627 (altivec_expand_vec_perm_const_le): Take each operand individually.
5628 Operate on constant selectors rather than rtxes.
5629 (altivec_expand_vec_perm_const): Likewise. Update call to
5630 altivec_expand_vec_perm_const_le.
5631 (rs6000_expand_vec_perm_const): Delete.
5632 (rs6000_vectorize_vec_perm_const_ok): Delete.
5633 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5634 reference to the SPE evmerge intructions.
5635 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5636 an element count and rtx array.
5637 (rs6000_expand_extract_even): Update call accordingly.
5638 (rs6000_expand_interleave): Likewise.
5639 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5640 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5642 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5644 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5646 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5647 vector mode and that that mode matches the mode of the data
5649 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5650 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5651 directly using expand_vec_perm_1 when forcing selectors into
5653 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5655 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5657 * optabs-query.h (can_vec_perm_p): Delete.
5658 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5659 * optabs-query.c (can_vec_perm_p): Split into...
5660 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5661 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5662 particular selector is valid.
5663 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5664 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5665 (vect_grouped_load_supported): Likewise.
5666 (vect_shift_permute_load_chain): Likewise.
5667 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5668 (vect_transform_slp_perm_load): Likewise.
5669 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5670 (vectorizable_bswap): Likewise.
5671 (vect_gen_perm_mask_checked): Likewise.
5672 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5673 implementations of variable permutation vectors into account
5674 when deciding which selector to use.
5675 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5676 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5677 with a false third argument.
5678 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5679 to test whether the constant selector is valid and can_vec_perm_var_p
5680 to test whether a variable selector is valid.
5682 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5684 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5685 * optabs-query.c (can_vec_perm_p): Likewise.
5686 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5687 instead of vec_perm_indices.
5688 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5689 (vect_gen_perm_mask_checked): Likewise,
5690 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5691 (vect_gen_perm_mask_checked): Likewise,
5693 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5695 * optabs-query.h (qimode_for_vec_perm): Declare.
5696 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5697 (qimode_for_vec_perm): ...this new function.
5698 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5700 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5702 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5703 does not have a conditional at the top.
5705 2018-01-02 Richard Biener <rguenther@suse.de>
5707 * ipa-inline.c (big_speedup_p): Fix expression.
5709 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5712 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5715 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5719 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5720 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5721 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5722 cond_taken_branch_cost 3->4.
5724 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5726 PR tree-optimization/83581
5727 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5728 TODO_cleanup_cfg if any changes have been made.
5731 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5732 convert_modes if target mode has the right side, but different mode
5736 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5737 last argument when extracting from CONCAT. If either from_real or
5738 from_imag is NULL, use expansion through memory. If result is not
5739 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5740 the parts directly to inner mode, if even that fails, use expansion
5744 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5745 check for bswap in mode rather than HImode and use that in expand_unop
5748 Copyright (C) 2018 Free Software Foundation, Inc.
5750 Copying and distribution of this file, with or without modification,
5751 are permitted in any medium without royalty provided the copyright
5752 notice and this notice are preserved.