1 2018-01-26 Jim Wilson <jimw@sifive.com>
3 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
5 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
8 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
10 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
11 and CMP + SUB-immediate -> SUBS.
13 2018-01-26 Martin Sebor <msebor@redhat.com>
15 PR tree-optimization/83896
16 * tree-ssa-strlen.c (get_string_len): Rename...
17 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
18 Avoid assuming length is constant.
19 (handle_char_store): Use HOST_WIDE_INT for string length.
21 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
24 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
25 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
27 2018-01-26 Richard Biener <rguenther@suse.de>
29 PR rtl-optimization/84003
30 * dse.c (record_store): Only record redundant stores when
31 the earlier store aliases at least all accesses the later one does.
33 2018-01-26 Jakub Jelinek <jakub@redhat.com>
35 PR rtl-optimization/83985
36 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
37 REG_CFA_RESTORE insns.
38 (delete_unmarked_insns): Don't ignore separate shrink wrapping
39 REG_CFA_RESTORE insns here.
42 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
43 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
45 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
47 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
48 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
50 (arc_override_options): Likewise.
51 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
53 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
55 * config/arc/arc.h (TARGET_DBNZ): Define.
56 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
57 properly set the tune attribute.
58 (dbnz): Use TARGET_DBNZ guard.
59 * config/arc/arc.opt (mtune): Add core3 option.
61 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
63 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
64 recognize new pic like addresses.
65 (arc_delegitimize_address): Clean up.
67 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
69 * config/arc/arc-arches.def: Option mrf16 valid for all
71 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
72 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
73 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
74 * config/arc/arc-tables.opt: Regenerate.
75 * config/arc/arc.c (arc_conditional_register_usage): Handle
76 reduced register file case.
77 (arc_file_start): Set must have build attributes.
78 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
80 * config/arc/arc.opt (mrf16): Add new option.
81 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
82 * config/arc/genmultilib.awk: Handle new mrf16 option.
83 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
84 * config/arc/t-multilib: Regenerate.
85 * doc/invoke.texi (ARC Options): Document mrf16 option.
87 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
89 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
90 * config/arc/arc.c (arc_handle_secure_attribute): New function.
91 (arc_attribute_table): Add 'secure_call' attribute.
92 (arc_print_operand): Print secure call operand.
93 (arc_function_ok_for_sibcall): Don't optimize tail calls when
95 (arc_is_secure_call_p): New function. * config/arc/arc.md
96 (call_i): Add support for sjli instruction.
97 (call_value_i): Likewise.
98 * config/arc/constraints.md (Csc): New constraint.
100 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
101 John Eric Martin <John.Martin@emmicro-us.com>
103 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
104 * config/arc/arc.c (_arc_jli_section): New struct.
105 (arc_jli_section): New type.
106 (rc_jli_sections): New static variable.
107 (arc_handle_jli_attribute): New function.
108 (arc_attribute_table): Add jli_always and jli_fixed attribute.
109 (arc_file_end): New function.
110 (TARGET_ASM_FILE_END): Define.
111 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
112 (arc_add_jli_section): New function.
113 (jli_call_scan): Likewise.
114 (arc_reorg): Call jli_call_scan.
115 (arc_output_addsi): Remove 'S' from printing asm operand.
116 (arc_is_jli_call_p): New function.
117 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
119 (movhi_insn): Likewise.
120 (movsi_insn): Likewise.
121 (movsi_set_cc_insn): Likewise.
122 (loadqi_update): Likewise.
123 (load_zeroextendqisi_update): Likewise.
124 (load_signextendqisi_update): Likewise.
125 (loadhi_update): Likewise.
126 (load_zeroextendhisi_update): Likewise.
127 (load_signextendhisi_update): Likewise.
128 (loadsi_update): Likewise.
129 (loadsf_update): Likewise.
130 (movsicc_insn): Likewise.
131 (bset_insn): Likewise.
132 (bxor_insn): Likewise.
133 (bclr_insn): Likewise.
134 (bmsk_insn): Likewise.
135 (bicsi3_insn): Likewise.
136 (cmpsi_cc_c_insn): Likewise.
137 (movsi_ne): Likewise.
138 (movsi_cond_exec): Likewise.
139 (clrsbsi2): Likewise.
150 (call_i): Remove 'S' asm letter, add jli instruction.
151 (call_value_i): Likewise.
152 * config/arc/arc.op (mjli-always): New option.
153 * config/arc/constraints.md (Cji): New constraint.
154 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
156 (subsf3_fpx): Likewise.
157 (mulsf3_fpx): Likewise.
158 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
160 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
162 * doc/invoke.texi (ARC): Document mjli-always option.
164 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
166 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
167 and use incw and decw where possible
168 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
170 2018-01-26 Richard Biener <rguenther@suse.de>
172 PR tree-optimization/81082
173 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
174 association if it requires casting to unsigned.
175 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
176 from fold_plusminus_mult_expr to catch important cases late when
177 range info is available.
179 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
181 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
182 * configure.ac (hidden_linkonce): New test.
183 * configure: Regenerate.
184 * config.in: Regenerate.
186 2018-01-26 Julia Koval <julia.koval@intel.com>
188 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
189 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
190 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
191 _mm_mask_bitshuffle_epi64_mask): Fix type.
192 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
193 USI_FTYPE_V4DI_V4DI_USI): Remove.
194 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
195 __builtin_ia32_vpshufbitqmb256_mask,
196 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
197 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
198 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
200 2018-01-26 Alan Modra <amodra@gmail.com>
203 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
204 UNSPEC_VBPERMQ. Sort other unspecs.
206 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
208 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
210 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
213 * predict.c (drop_profile): Do not push/pop cfun; update also
215 (handle_missing_profiles): Fix logic looking for zero profiles.
217 2018-01-25 Jakub Jelinek <jakub@redhat.com>
220 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
221 on functions with #pragma omp declare simd or functions with simd
223 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
224 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
225 Remove trailing \n from warning_at calls.
227 2018-01-25 Tom de Vries <tom@codesourcery.com>
230 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
231 for neutered workers.
233 2018-01-24 Joseph Myers <joseph@codesourcery.com>
236 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
237 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
239 2017-01-08 Jeff Law <law@redhat.com>
242 * i386.c (get_probe_interval): Move to earlier point.
243 (ix86_compute_frame_layout): If -fstack-clash-protection and
244 the frame is larger than the probe interval, then use pushes
245 to save registers rather than reg->mem moves.
246 (ix86_expand_prologue): Remove conditional for int_registers_saved
249 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
252 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
253 min/max for never referenced object.
255 2018-01-24 Jakub Jelinek <jakub@redhat.com>
258 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
260 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
261 attributes from DECL_ATTRIBUTES (decl) without affecting
262 DECL_ATTRIBUTES (current_function_decl).
263 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
264 functions with non-NULL DECL_ABSTRACT_ORIGIN.
266 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
268 PR tree-optimization/83979
269 * fold-const.c (fold_comparison): Use constant_boolean_node
270 instead of boolean_{true,false}_node.
272 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
274 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
277 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
279 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
280 Simplify the clause that sets the length attribute.
281 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
282 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
283 clause that sets the length attribute.
284 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
286 2018-01-24 Tom de Vries <tom@codesourcery.com>
289 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
290 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
291 Add strict parameter.
292 (prevent_branch_around_nothing): Insert dummy insn between branch to
293 label and label with no ptx insn inbetween.
294 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
296 2018-01-24 Tom de Vries <tom@codesourcery.com>
299 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
300 for neutered threads in warp.
301 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
303 2018-01-24 Richard Biener <rguenther@suse.de>
305 PR tree-optimization/83176
306 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
309 2018-01-24 Richard Biener <rguenther@suse.de>
311 PR tree-optimization/82819
312 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
313 code generating pluses that are no-ops in the target precision.
315 2018-01-24 Richard Biener <rguenther@suse.de>
318 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
320 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
322 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
323 to merge probabilities.
324 * predict.c (probably_never_executed): Also mark as cold functions
325 with global 0 profile and guessed local profile.
326 * profile-count.c (profile_probability::combine_with_count): New
328 * profile-count.h (profile_probability::operator*,
329 profile_probability::operator*=, profile_probability::operator/,
330 profile_probability::operator/=): Reduce precision to adjusted
331 and set value to guessed on contradictory divisions.
332 (profile_probability::combine_with_freq): Remove.
333 (profile_probability::combine_wiht_count): Declare.
334 (profile_count::force_nonzero):: Set to adjusted.
335 (profile_count::probability_in):: Set quality to adjusted.
336 * tree-ssa-tail-merge.c (replace_block_by): Use
339 2018-01-23 Andrew Waterman <andrew@sifive.com>
340 Jim Wilson <jimw@sifive.com>
342 * config/riscv/riscv.c (riscv_stack_boundary): New.
343 (riscv_option_override): Set riscv_stack_boundary. Handle
344 riscv_preferred_stack_boundary_arg.
345 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
346 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
347 (STACK_BOUNDARY): Set to riscv_stack_boundary.
348 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
349 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
350 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
352 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
355 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
356 of struct ix86_frame.
357 (ix86_expand_epilogue): Likewise. Add a local variable for
358 the reg_save_offset field in struct ix86_frame.
360 2018-01-23 Bin Cheng <bin.cheng@arm.com>
362 PR tree-optimization/82604
363 * tree-loop-distribution.c (enum partition_kind): New enum item
364 PKIND_PARTIAL_MEMSET.
365 (partition_builtin_p): Support above new enum item.
366 (generate_code_for_partition): Ditto.
367 (compute_access_range): Differentiate cases that equality can be
368 proven at all loops, the innermost loops or no loops.
369 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
370 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
371 (finalize_partitions, distribute_loop): Don't fuse partition of
372 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
373 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
376 2018-01-23 Martin Liska <mliska@suse.cz>
378 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
379 order to ignore the predictor.
380 (PRED_POLYMORPHIC_CALL): Likewise.
381 (PRED_RECURSIVE_CALL): Likewise.
383 2018-01-23 Martin Liska <mliska@suse.cz>
385 * tree-profile.c (tree_profiling): Print function header to
386 aware reader which function we are working on.
387 * value-prof.c (gimple_find_values_to_profile): Do not print
388 not interesting value histograms.
390 2018-01-23 Martin Liska <mliska@suse.cz>
392 * profile-count.h (enum profile_quality): Add
393 profile_uninitialized as the first value. Do not number values
394 as they are zero based.
395 (profile_count::verify): Update sanity check.
396 (profile_probability::verify): Likewise.
398 2018-01-23 Nathan Sidwell <nathan@acm.org>
400 * doc/invoke.texi (ffor-scope): Deprecate.
402 2018-01-23 David Malcolm <dmalcolm@redhat.com>
404 PR tree-optimization/83510
405 * domwalk.c (set_all_edges_as_executable): New function.
406 (dom_walker::dom_walker): Convert bool param
407 "skip_unreachable_blocks" to enum reachability. Move setup of
408 edge flags to set_all_edges_as_executable and only do it when
409 reachability is REACHABLE_BLOCKS.
410 * domwalk.h (enum dom_walker::reachability): New enum.
411 (dom_walker::dom_walker): Convert bool param
412 "skip_unreachable_blocks" to enum reachability.
413 (set_all_edges_as_executable): New decl.
414 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
415 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
417 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
418 but converting true to REACHABLE_BLOCKS.
419 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
421 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
422 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
423 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
425 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
426 if check_all_array_refs will be called.
428 2018-01-23 David Malcolm <dmalcolm@redhat.com>
430 * tree.c (selftest::test_location_wrappers): Add more test
433 2018-01-23 David Malcolm <dmalcolm@redhat.com>
435 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
436 (selftest::test_bit_in_range): Likewise.
438 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
441 * doc/sourcebuild.texi (vect_float): Say that the selector
442 only describes the situation when -funsafe-math-optimizations is on.
443 (vect_float_strict): Document.
445 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
447 PR tree-optimization/83965
448 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
449 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
450 instead of checking only for a reduction.
451 (vect_recog_widen_sum_pattern): Likewise.
453 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
455 * predict.c (probably_never_executed): Only use precise profile info.
456 (compute_function_frequency): Skip after inlining hack since we now
457 have quality checking.
459 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
461 * profile-count.h (profile_probability::very_unlikely,
462 profile_probability::unlikely, profile_probability::even): Set
463 precision to guessed.
465 2018-01-23 Richard Biener <rguenther@suse.de>
467 PR tree-optimization/83963
468 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
469 Properly terminate dominator walk when crossing the exit edge not
470 when visiting its source block.
472 2018-01-23 Jakub Jelinek <jakub@redhat.com>
475 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
476 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
478 2018-01-22 Jakub Jelinek <jakub@redhat.com>
480 PR tree-optimization/83957
481 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
482 semicolon after for body surrounded by braces.
484 PR tree-optimization/83081
485 * profile-count.h (profile_probability::split): New method.
486 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
487 Use profile_probability::split.
488 (do_compare_rtx_and_jump): Fix adjustment of probabilities
489 when splitting a single conditional jump into 2.
491 2018-01-22 David Malcolm <dmalcolm@redhat.com>
493 PR tree-optimization/69452
494 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
497 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
499 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
500 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
501 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
503 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
505 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
506 * config/rl78/rl78.md: New define_expand "movdi"
507 * config/rl78/rl78.c: New function definition rl78_split_movdi
509 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
512 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
514 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
515 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
516 128-bit to produce an UNSPEC move to get the double word with the
517 signbit and then a shift directly to do signbit.
518 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
519 implementation with a new version that just does either a direct
520 move or a regular move. Move memory interface to separate insns.
521 Move insns so they are next to the expander.
522 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
523 with signbit move. Split big and little endian case.
524 (signbit<mode>2_dm_mem_le): Likewise.
525 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
526 (signbit<mode>2_dm2): Likewise.
528 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
530 * config/rl78/rl78.md: New define_expand "anddi3".
532 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
534 * config/rl78/rl78.md: New define_expand "umindi3".
536 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
538 * config/rl78/rl78.md: New define_expand "smindi3".
540 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
542 * config/rl78/rl78.md: New define_expand "smaxdi3".
544 2018-01-22 Carl Love <cel@us.ibm.com>
546 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
547 LVX_V1TI): Add macro expansion.
548 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
549 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
550 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
551 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
552 Change check to determine if the instruction is a byte reversing
553 entry. Fix typo in comment.
554 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
555 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
556 Add def_builtin calls for new builtins.
557 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
558 Add define_insn expansion.
560 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
562 * config/rl78/rl78.md: New define_expand "umaxdi3".
564 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
566 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
567 for non-QImode registers
569 2018-01-22 Richard Biener <rguenther@suse.de>
571 PR tree-optimization/83963
572 * graphite-scop-detection.c (scop_detection::get_sese): Delay
573 including the loop exit block.
574 (scop_detection::merge_sese): Likewise.
575 (scop_detection::add_scop): Do it here instead.
577 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
579 * doc/sourcebuild.texi (arm_softfloat): Document.
581 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
584 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
585 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
586 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
588 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
589 David Edelsohn <dje.gcc@gmail.com>
592 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
593 Change "crset eq" to "crset 2".
594 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
595 (*call_indirect_aix<mode>_nospec): Likewise.
596 (*call_value_indirect_aix<mode>_nospec): Likewise.
597 (*call_indirect_elfv2<mode>_nospec): Likewise.
598 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
599 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
600 change assembly output from . to $.
601 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
602 (indirect_jump<mode>_nospec): Change assembly output from . to $.
603 (*tablejump<mode>_internal1_nospec): Likewise.
605 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
608 * config/sh/sh_optimize_sett_clrt.cc:
609 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
611 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
613 PR tree-optimization/83940
614 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
615 offset_dt to vect_constant_def rather than vect_unknown_def_type.
616 (vect_check_load_store_mask): Add a mask_dt_out parameter and
617 use it to pass back the definition type.
618 (vect_check_store_rhs): Likewise rhs_dt_out.
619 (vect_build_gather_load_calls): Add a mask_dt argument and use
620 it instead of a call to vect_is_simple_use.
621 (vectorizable_store): Update calls to vect_check_load_store_mask
622 and vect_check_store_rhs. Use the dt returned by the latter instead
623 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
624 instead of calls to vect_is_simple_use. Pass the scalar rather
625 than the vector operand to vect_is_simple_use when handling
626 second and subsequent copies of an rhs value.
627 (vectorizable_load): Update calls to vect_check_load_store_mask
628 and vect_build_gather_load_calls. Use the cached mask_dt and
629 gs_info.offset_dt instead of calls to vect_is_simple_use.
631 2018-01-20 Jakub Jelinek <jakub@redhat.com>
634 * tree-emutls.c: Include gimplify.h.
635 (lower_emutls_2): New function.
636 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
637 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
638 it before further processing.
641 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
642 UINTVAL (trueop1) instead of INTVAL (op1).
644 2018-01-19 Jakub Jelinek <jakub@redhat.com>
648 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
649 INCOMING_FRAME_SP_OFFSET if not defined.
650 (scan_trace): Add ENTRY argument. If true and
651 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
652 emit a note to adjust the CFA offset.
653 (create_cfi_notes): Adjust scan_trace callers.
654 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
655 INCOMING_FRAME_SP_OFFSET in the CIE.
656 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
657 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
659 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
660 * doc/tm.texi: Regenerated.
662 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
664 PR rtl-optimization/83147
665 * lra-constraints.c (remove_inheritance_pseudos): Use
666 lra_substitute_pseudo_within_insn.
668 2018-01-19 Tom de Vries <tom@codesourcery.com>
669 Cesar Philippidis <cesar@codesourcery.com>
672 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
674 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
677 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
678 spaces for function labels.
680 2018-01-19 Martin Liska <mliska@suse.cz>
682 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
683 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
684 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
685 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
686 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
687 (PRED_CONST_RETURN): Change from 69 to 65.
688 (PRED_NULL_RETURN): Change from 91 to 71.
689 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
690 (PRED_LOOP_GUARD): Change from 66 to 73.
692 2018-01-19 Martin Liska <mliska@suse.cz>
694 * predict.c (predict_insn_def): Add new assert.
695 (struct branch_predictor): Change type to signed integer.
696 (test_prediction_value_range): Amend test to cover
698 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
699 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
700 (PRED_LOOP_ITERATIONS_MAX): Likewise.
701 (PRED_LOOP_IV_COMPARE): Likewise.
702 * predict.h (PROB_UNINITIALIZED): Define new constant.
704 2018-01-19 Martin Liska <mliska@suse.cz>
706 * predict.c (dump_prediction): Add new format for
707 analyze_brprob.py script which is enabled with -details
709 * profile-count.h (precise_p): New function.
711 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
713 PR tree-optimization/83922
714 * tree-vect-loop.c (vect_verify_full_masking): Return false if
715 there are no statements that need masking.
716 (vect_active_double_reduction_p): New function.
717 (vect_analyze_loop_operations): Use it when handling phis that
718 are not in the loop header.
720 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
722 PR tree-optimization/83914
723 * tree-vect-loop.c (vectorizable_induction): Don't convert
724 init_expr or apply the peeling adjustment for inductions
725 that are nested within the vectorized loop.
727 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
729 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
732 2018-01-18 Jakub Jelinek <jakub@redhat.com>
736 * function.h (gimplify_parameters): Add gimple_seq * argument.
737 * function.c: Include gimple.h and options.h.
738 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
739 for the added local temporaries if needed.
740 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
741 if there are any parameter cleanups, wrap whole body into a
742 try/finally with the cleanups.
744 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
747 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
748 Use GET_MODE_CLASS for scalar floating point.
750 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
754 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
755 Fix call of call_cgraph_insertion_hooks.
757 2018-01-18 Martin Sebor <msebor@redhat.com>
759 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
761 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
764 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
767 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
770 * common.opt: (-ffile-prefix-map): New option.
771 * opts.c (common_handle_option): Defer it.
772 * opts-global.c (handle_common_deferred_options): Handle it.
773 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
774 * file-prefix-map.h: New file.
775 (remap_debug_filename, add_debug_prefix_map): ...here.
776 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
777 * final.c (debug_prefix_map, add_debug_prefix_map
778 remap_debug_filename): Move to...
779 * file-prefix-map.c: New file.
780 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
781 generalize, get rid of alloca(), use strrchr() instead of strchr().
782 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
783 Implement in terms of add_prefix_map().
784 (remap_macro_filename, remap_debug_filename): Implement in term of
786 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
787 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
788 * dbxout.c: Include file-prefix-map.h.
789 * varasm.c: Likewise.
790 * vmsdbgout.c: Likewise.
791 * xcoffout.c: Likewise.
792 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
793 * doc/cppopts.texi (-fmacro-prefix-map): Document.
794 * doc/invoke.texi (-ffile-prefix-map): Document.
795 (-fdebug-prefix-map): Update description.
797 2018-01-18 Martin Liska <mliska@suse.cz>
799 * config/i386/i386.c (indirect_thunk_name): Document that also
801 (output_indirect_thunk): Document why both instructions
802 (pause and lfence) are generated.
804 2018-01-18 Richard Biener <rguenther@suse.de>
806 PR tree-optimization/83887
807 * graphite-scop-detection.c
808 (scop_detection::get_nearest_dom_with_single_entry): Remove.
809 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
810 (scop_detection::merge_sese): Re-implement with a flood-fill
811 algorithm that properly finds a SESE region if it exists.
813 2018-01-18 Jakub Jelinek <jakub@redhat.com>
816 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
817 pointer_diff optimizations use view_convert instead of convert.
819 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
821 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
822 Generate different code for -mno-speculate-indirect-jumps.
823 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
824 (*call_indirect_aix<mode>): Disable for
825 -mno-speculate-indirect-jumps.
826 (*call_indirect_aix<mode>_nospec): New define_insn.
827 (*call_value_indirect_aix<mode>): Disable for
828 -mno-speculate-indirect-jumps.
829 (*call_value_indirect_aix<mode>_nospec): New define_insn.
830 (*sibcall_nonlocal_sysv<mode>): Generate different code for
831 -mno-speculate-indirect-jumps.
832 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
834 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
836 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
837 long double type, set the flags for noting the default long double
838 type, even if we don't pass or return a long double type.
840 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
843 * ipa-inline.c (flatten_function): Do not overwrite final inlining
846 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
848 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
849 support for merge[hl].
850 (fold_mergehl_helper): New helper function.
851 (tree-vector-builder.h): New #include for tree_vector_builder usage.
852 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
853 (altivec_vmrglw_direct): Add xxmrglw insn.
855 2018-01-17 Andrew Waterman <andrew@sifive.com>
857 * config/riscv/riscv.c (riscv_conditional_register_usage): If
858 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
860 2018-01-17 David Malcolm <dmalcolm@redhat.com>
863 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
864 call the lto_location_cache before reading the
865 DECL_SOURCE_LOCATION of the types.
867 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
868 Richard Sandiford <richard.sandiford@linaro.org>
870 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
871 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
872 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
873 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
874 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
876 * config/aarch64/constraints.md (aarch64_movti_operand):
878 * config/aarch64/predicates.md (Uti): Add new constraint.
880 2018-01-17 Carl Love <cel@us.ibm.com>
881 * config/rs6000/vsx.md (define_expand xl_len_r,
882 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
883 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
885 (define_expand, define_insn): Move the shift left from the
886 define_insn to the define_expand for lxvl and stxvl instructions.
887 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
888 and XL_LEN_R definitions to PURE.
890 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
892 * config/i386/i386.c (indirect_thunk_name): Declare regno
893 as unsigned int. Compare regno with INVALID_REGNUM.
894 (output_indirect_thunk): Ditto.
895 (output_indirect_thunk_function): Ditto.
896 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
897 in the call to output_indirect_thunk_function.
899 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
902 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
903 rather than the size of inner_type to determine the stack slot size
904 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
906 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
909 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
912 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
914 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
915 endian Linux systems to optionally enable multilibs for selecting
916 the long double type if the user configured an explicit type.
917 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
918 have no long double multilibs if not defined.
919 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
920 warn if the user used -mabi={ieee,ibm}longdouble and we built
921 multilibs for long double.
922 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
923 appropriate multilib option.
924 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
926 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
927 for building long double multilibs.
928 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
930 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
932 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
935 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
937 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
940 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
943 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
946 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
948 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
949 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
951 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
953 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
954 different rtl trees depending on TARGET_64BIT.
955 (rs6000_gen_lvx): Likewise.
957 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
959 * config/visium/visium.md (nop): Tweak comment.
960 (hazard_nop): Likewise.
962 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
964 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
965 -mspeculate-indirect-jumps.
966 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
967 for -mno-speculate-indirect-jumps.
968 (*call_indirect_elfv2<mode>_nospec): New define_insn.
969 (*call_value_indirect_elfv2<mode>): Disable for
970 -mno-speculate-indirect-jumps.
971 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
972 (indirect_jump): Emit different RTL for
973 -mno-speculate-indirect-jumps.
974 (*indirect_jump<mode>): Disable for
975 -mno-speculate-indirect-jumps.
976 (*indirect_jump<mode>_nospec): New define_insn.
977 (tablejump): Emit different RTL for
978 -mno-speculate-indirect-jumps.
979 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
980 (tablejumpsi_nospec): New define_expand.
981 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
982 (tablejumpdi_nospec): New define_expand.
983 (*tablejump<mode>_internal1): Disable for
984 -mno-speculate-indirect-jumps.
985 (*tablejump<mode>_internal1_nospec): New define_insn.
986 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
989 2018-01-16 Artyom Skrobov tyomitch@gmail.com
991 * caller-save.c (insert_save): Drop unnecessary parameter. All
994 2018-01-16 Jakub Jelinek <jakub@redhat.com>
995 Richard Biener <rguenth@suse.de>
998 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
999 return early, inline manually is_gimple_sizepos. Make sure if we
1000 call gimplify_expr we don't end up with a gimple constant.
1001 * tree.c (variably_modified_type_p): Don't return true for
1002 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1003 * gimplify.h (is_gimple_sizepos): Remove.
1005 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1007 PR tree-optimization/83857
1008 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1009 vectorizable_live_operation for pure SLP statements.
1010 (vectorizable_live_operation): Handle PHIs.
1012 2018-01-16 Richard Biener <rguenther@suse.de>
1014 PR tree-optimization/83867
1015 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1016 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1018 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1021 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1022 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1023 If off is not INTEGER_CST, issue a may not be aligned warning
1024 rather than isn't aligned. Use isn%'t rather than isn't.
1025 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1027 <case MULT_EXPR>: Improve the case when bottom and one of the
1028 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1029 operand, in that case check if the other operand is multiple of
1030 bottom divided by the INTEGER_CST operand.
1032 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1035 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1036 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1037 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1038 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1039 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1040 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1041 (pa_function_arg_size): New function.
1043 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1045 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1046 in a separate statement.
1048 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1050 PR tree-optimization/83847
1051 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1052 group gathers and scatters.
1054 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1056 PR rtl-optimization/86620
1057 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1059 PR rtl-optimization/83213
1060 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1061 to last if both are JUMP_INSNs.
1063 PR tree-optimization/83843
1064 * gimple-ssa-store-merging.c
1065 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1066 store_immediate_info for bswap/nop orig_stores.
1068 2018-01-15 Andrew Waterman <andrew@sifive.com>
1070 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1072 <UDIV>: Increase cost if !TARGET_DIV.
1074 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1076 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1077 (define_attr "cr_logical_3op"): New.
1078 (cceq_ior_compare): Adjust.
1079 (cceq_ior_compare_complement): Adjust.
1080 (*cceq_rev_compare): Adjust.
1081 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1082 (is_cracked_insn): Adjust.
1083 (insn_must_be_first_in_group): Adjust.
1084 * config/rs6000/40x.md: Adjust.
1085 * config/rs6000/440.md: Adjust.
1086 * config/rs6000/476.md: Adjust.
1087 * config/rs6000/601.md: Adjust.
1088 * config/rs6000/603.md: Adjust.
1089 * config/rs6000/6xx.md: Adjust.
1090 * config/rs6000/7450.md: Adjust.
1091 * config/rs6000/7xx.md: Adjust.
1092 * config/rs6000/8540.md: Adjust.
1093 * config/rs6000/cell.md: Adjust.
1094 * config/rs6000/e300c2c3.md: Adjust.
1095 * config/rs6000/e500mc.md: Adjust.
1096 * config/rs6000/e500mc64.md: Adjust.
1097 * config/rs6000/e5500.md: Adjust.
1098 * config/rs6000/e6500.md: Adjust.
1099 * config/rs6000/mpc.md: Adjust.
1100 * config/rs6000/power4.md: Adjust.
1101 * config/rs6000/power5.md: Adjust.
1102 * config/rs6000/power6.md: Adjust.
1103 * config/rs6000/power7.md: Adjust.
1104 * config/rs6000/power8.md: Adjust.
1105 * config/rs6000/power9.md: Adjust.
1106 * config/rs6000/rs64.md: Adjust.
1107 * config/rs6000/titan.md: Adjust.
1109 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1111 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1112 ix86_indirect_branch_register logic.
1114 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1116 * config/i386/constraints.md (Bs): Update
1117 ix86_indirect_branch_register check. Don't check
1118 ix86_indirect_branch_register with GOT_memory_operand.
1120 * config/i386/predicates.md (GOT_memory_operand): Don't check
1121 ix86_indirect_branch_register here.
1122 (GOT32_symbol_operand): Likewise.
1124 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1126 * config/i386/predicates.md (constant_call_address_operand):
1127 Rewrite ix86_indirect_branch_register logic.
1128 (sibcall_insn_operand): Likewise.
1130 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1132 * config/i386/constraints.md (Bs): Replace
1133 ix86_indirect_branch_thunk_register with
1134 ix86_indirect_branch_register.
1136 * config/i386/i386.md (indirect_jump): Likewise.
1137 (tablejump): Likewise.
1138 (*sibcall_memory): Likewise.
1139 (*sibcall_value_memory): Likewise.
1140 Peepholes of indirect call and jump via memory: Likewise.
1141 * config/i386/i386.opt: Likewise.
1142 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1143 (GOT_memory_operand): Likewise.
1144 (call_insn_operand): Likewise.
1145 (sibcall_insn_operand): Likewise.
1146 (GOT32_symbol_operand): Likewise.
1148 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1151 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1152 type rather than type addr's type points to.
1153 (expand_omp_atomic_mutex): Likewise.
1154 (expand_omp_atomic): Likewise.
1156 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1159 * config/i386/i386.c (output_indirect_thunk_function): Use
1160 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1161 for __x86_return_thunk.
1163 2018-01-15 Richard Biener <rguenther@suse.de>
1166 * expmed.c (extract_bit_field_1): Fix typo.
1168 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1171 * config/arm/iterators.md (VF): New mode iterator.
1172 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1173 Remove integer-related logic from pattern.
1174 (neon_vabd<mode>_3): Likewise.
1176 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1179 * common.opt (fstrict-overflow): No longer an alias.
1180 (fwrapv-pointer): New option.
1181 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1182 also for pointer types based on flag_wrapv_pointer.
1183 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1184 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1185 opts->x_flag_wrapv got set.
1186 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1187 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1188 POINTER_TYPE_OVERFLOW_UNDEFINED.
1189 * match.pd: Likewise in address comparison pattern.
1190 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1192 2018-01-15 Richard Biener <rguenther@suse.de>
1195 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1196 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1197 Reset type names to their identifier if their TYPE_DECL doesn't
1198 have linkage (and thus is used for ODR and devirt).
1199 (save_debug_info_for_decl): Remove.
1200 (save_debug_info_for_type): Likewise.
1201 (add_tree_to_fld_list): Adjust.
1202 * tree-pretty-print.c (dump_generic_node): Make dumping of
1203 type names more robust.
1205 2018-01-15 Richard Biener <rguenther@suse.de>
1207 * BASE-VER: Bump to 8.0.1.
1209 2018-01-14 Martin Sebor <msebor@redhat.com>
1212 * builtins.c (check_access): Avoid warning when the no-warning bit
1215 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1217 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1218 * ira-color (allocno_hard_regs_compare): Likewise.
1220 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1223 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1224 Use .pushsection/.popsection.
1226 2018-01-14 Martin Sebor <msebor@redhat.com>
1229 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1231 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1233 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1234 entry from extra_headers.
1235 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1236 extra_headers, make the list bitwise identical to the i?86-*-* one.
1238 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1240 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1241 -mcmodel=large with -mindirect-branch=thunk,
1242 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1243 -mfunction-return=thunk-extern.
1244 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1245 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1246 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1248 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1250 * config/i386/i386.c (print_reg): Print the name of the full
1251 integer register without '%'.
1252 (ix86_print_operand): Handle 'V'.
1253 * doc/extend.texi: Document 'V' modifier.
1255 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1257 * config/i386/constraints.md (Bs): Disallow memory operand for
1258 -mindirect-branch-register.
1260 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1261 (GOT_memory_operand): Likewise.
1262 (call_insn_operand): Likewise.
1263 (sibcall_insn_operand): Likewise.
1264 (GOT32_symbol_operand): Likewise.
1265 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1266 for -mindirect-branch-register.
1267 (tablejump): Likewise.
1268 (*sibcall_memory): Likewise.
1269 (*sibcall_value_memory): Likewise.
1270 Disallow peepholes of indirect call and jump via memory for
1271 -mindirect-branch-register.
1272 (*call_pop): Replace m with Bw.
1273 (*call_value_pop): Likewise.
1274 (*sibcall_pop_memory): Replace m with Bs.
1275 * config/i386/i386.opt (mindirect-branch-register): New option.
1276 * doc/invoke.texi: Document -mindirect-branch-register option.
1278 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1280 * config/i386/i386-protos.h (ix86_output_function_return): New.
1281 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1282 set function_return_type.
1283 (indirect_thunk_name): Add ret_p to indicate thunk for function
1285 (output_indirect_thunk_function): Pass false to
1286 indirect_thunk_name.
1287 (ix86_output_indirect_branch_via_reg): Likewise.
1288 (ix86_output_indirect_branch_via_push): Likewise.
1289 (output_indirect_thunk_function): Create alias for function
1290 return thunk if regno < 0.
1291 (ix86_output_function_return): New function.
1292 (ix86_handle_fndecl_attribute): Handle function_return.
1293 (ix86_attribute_table): Add function_return.
1294 * config/i386/i386.h (machine_function): Add
1295 function_return_type.
1296 * config/i386/i386.md (simple_return_internal): Use
1297 ix86_output_function_return.
1298 (simple_return_internal_long): Likewise.
1299 * config/i386/i386.opt (mfunction-return=): New option.
1300 (indirect_branch): Mention -mfunction-return=.
1301 * doc/extend.texi: Document function_return function attribute.
1302 * doc/invoke.texi: Document -mfunction-return= option.
1304 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1306 * config/i386/i386-opts.h (indirect_branch): New.
1307 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1308 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1309 with local indirect jump when converting indirect call and jump.
1310 (ix86_set_indirect_branch_type): New.
1311 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1312 (indirectlabelno): New.
1313 (indirect_thunk_needed): Likewise.
1314 (indirect_thunk_bnd_needed): Likewise.
1315 (indirect_thunks_used): Likewise.
1316 (indirect_thunks_bnd_used): Likewise.
1317 (INDIRECT_LABEL): Likewise.
1318 (indirect_thunk_name): Likewise.
1319 (output_indirect_thunk): Likewise.
1320 (output_indirect_thunk_function): Likewise.
1321 (ix86_output_indirect_branch_via_reg): Likewise.
1322 (ix86_output_indirect_branch_via_push): Likewise.
1323 (ix86_output_indirect_branch): Likewise.
1324 (ix86_output_indirect_jmp): Likewise.
1325 (ix86_code_end): Call output_indirect_thunk_function if needed.
1326 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1328 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1329 (ix86_attribute_table): Add indirect_branch.
1330 * config/i386/i386.h (machine_function): Add indirect_branch_type
1331 and has_local_indirect_jump.
1332 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1334 (tablejump): Likewise.
1335 (*indirect_jump): Use ix86_output_indirect_jmp.
1336 (*tablejump_1): Likewise.
1337 (simple_return_indirect_internal): Likewise.
1338 * config/i386/i386.opt (mindirect-branch=): New option.
1339 (indirect_branch): New.
1342 (thunk-inline): Likewise.
1343 (thunk-extern): Likewise.
1344 * doc/extend.texi: Document indirect_branch function attribute.
1345 * doc/invoke.texi: Document -mindirect-branch= option.
1347 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1350 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1352 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1354 * ipa-inline.c (want_inline_small_function_p): Return false if
1355 inlining has already failed with CIF_FINAL_ERROR.
1356 (update_caller_keys): Call want_inline_small_function_p before
1358 (update_callee_keys): Likewise.
1360 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1362 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1364 (rs6000_quadword_masked_address_p): Likewise.
1365 (quad_aligned_load_p): Likewise.
1366 (quad_aligned_store_p): Likewise.
1367 (const_load_sequence_p): Add comment to describe the outer-most loop.
1368 (mimic_memory_attributes_and_flags): New function.
1369 (rs6000_gen_stvx): Likewise.
1370 (replace_swapped_aligned_store): Likewise.
1371 (rs6000_gen_lvx): Likewise.
1372 (replace_swapped_aligned_load): Likewise.
1373 (replace_swapped_load_constant): Capitalize argument name in
1374 comment describing this function.
1375 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1376 and stores that access quad-word aligned addresses and replace
1377 with stvx or lvx instructions when appropriate.
1378 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1379 New function prototype.
1380 (rs6000_quadword_masked_address_p): Likewise.
1381 (rs6000_gen_lvx): Likewise.
1382 (rs6000_gen_stvx): Likewise.
1383 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1384 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1385 when memory address is aligned.
1386 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1387 this split to select lvx instruction when memory address is aligned.
1388 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1389 instruction when memory address is aligned.
1390 (*vsx_le_perm_load_v16qi): Likewise.
1391 (four unnamed splitters): Modify to select the stvx instruction
1392 when memory is aligned.
1394 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1396 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1397 which appears in the queue multiple times.
1399 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1400 Alan Hayward <alan.hayward@arm.com>
1401 David Sherwood <david.sherwood@arm.com>
1403 * tree-vectorizer.h (vec_lower_bound): New structure.
1404 (_loop_vec_info): Add check_nonzero and lower_bounds.
1405 (LOOP_VINFO_CHECK_NONZERO): New macro.
1406 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1407 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1408 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1409 fields. Make seg_len the distance travelled, not including the
1411 (dr_direction_indicator): Declare.
1412 (dr_zero_step_indicator): Likewise.
1413 (dr_known_forward_stride_p): Likewise.
1414 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1416 (runtime_alias_check_p): Allow runtime alias checks with
1418 (operator ==): Compare access_size and align.
1419 (prune_runtime_alias_test_list): Rework for new distinction between
1420 the access_size and seg_len.
1421 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1423 (get_segment_min_max): New function.
1424 (create_intersect_range_checks): Use it.
1425 (dr_step_indicator): New function.
1426 (dr_direction_indicator): Likewise.
1427 (dr_zero_step_indicator): Likewise.
1428 (dr_known_forward_stride_p): Likewise.
1429 * tree-loop-distribution.c (data_ref_segment_size): Return
1430 DR_STEP * (niters - 1).
1431 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1433 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1434 (vect_preserves_scalar_order_p): New function, split out from...
1435 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1436 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1437 (vect_vfa_access_size): New function.
1438 (vect_vfa_align): Likewise.
1439 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1440 (dump_lower_bound): New function.
1441 (vect_check_lower_bound): Likewise.
1442 (vect_small_gap_p): Likewise.
1443 (vectorizable_with_step_bound_p): Likewise.
1444 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1445 depencies if the vectorization factor is 1. Convert the checks
1446 for nonzero steps into checks on the bounds of DR_STEP. Try using
1447 a bunds check for variable steps if the minimum required step is
1448 relatively small. Update calls to the dr_with_seg_len
1449 constructor and to vect_compile_time_alias.
1450 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1452 (vect_loop_versioning): Call it.
1453 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1455 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1457 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1458 Alan Hayward <alan.hayward@arm.com>
1459 David Sherwood <david.sherwood@arm.com>
1461 * doc/sourcebuild.texi (vect_scatter_store): Document.
1462 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1464 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1466 * genopinit.c (main): Add supports_vec_scatter_store and
1467 supports_vec_scatter_store_cached to target_optabs.
1468 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1469 IFN_MASK_SCATTER_STORE.
1470 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1472 * internal-fn.h (internal_store_fn_p): Declare.
1473 (internal_fn_stored_value_index): Likewise.
1474 * internal-fn.c (scatter_store_direct): New macro.
1475 (expand_scatter_store_optab_fn): New function.
1476 (direct_scatter_store_optab_supported_p): New macro.
1477 (internal_store_fn_p): New function.
1478 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1479 IFN_MASK_SCATTER_STORE.
1480 (internal_fn_mask_index): Likewise.
1481 (internal_fn_stored_value_index): New function.
1482 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1484 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1485 * optabs-query.c (supports_vec_scatter_store_p): New function.
1486 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1487 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1488 true for scatter stores.
1489 (vect_gather_scatter_fn_p): Handle scatter stores too.
1490 (vect_check_gather_scatter): Consider using scatter stores if
1491 supports_vec_scatter_store_p.
1492 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1494 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1495 internal_fn_stored_value_index.
1496 (check_load_store_masking): Handle scatter stores too.
1497 (vect_get_store_rhs): Make public.
1498 (vectorizable_call): Use internal_store_fn_p.
1499 (vectorizable_store): Handle scatter store internal functions.
1500 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1501 when deciding whether the end of the group has been reached.
1502 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1503 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1504 (mask_scatter_store<mode>): New insns.
1506 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1507 Alan Hayward <alan.hayward@arm.com>
1508 David Sherwood <david.sherwood@arm.com>
1510 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1511 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1512 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1514 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1515 Use vect_truncate_gather_scatter_offset if we can't treat the
1516 operation as a normal gather load or scatter store.
1517 (get_group_load_store_type): Take the gather_scatter_info
1518 as argument. Try using a gather load or scatter store for
1519 single-element groups.
1520 (get_load_store_type): Update calls to get_group_load_store_type
1521 and vect_use_strided_gather_scatters_p.
1523 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1524 Alan Hayward <alan.hayward@arm.com>
1525 David Sherwood <david.sherwood@arm.com>
1527 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1528 optional tree argument.
1529 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1531 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1532 but continue to use the current value as a fallback.
1533 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1534 to compare the updates.
1535 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1536 (get_load_store_type): Use it when handling a strided access.
1537 (vect_get_strided_load_store_ops): New function.
1538 (vect_get_data_ptr_increment): Likewise.
1539 (vectorizable_load): Handle strided gather loads. Always pass
1540 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1542 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1543 Alan Hayward <alan.hayward@arm.com>
1544 David Sherwood <david.sherwood@arm.com>
1546 * doc/md.texi (gather_load@var{m}): Document.
1547 (mask_gather_load@var{m}): Likewise.
1548 * genopinit.c (main): Add supports_vec_gather_load and
1549 supports_vec_gather_load_cached to target_optabs.
1550 * optabs-tree.c (init_tree_optimization_optabs): Use
1551 ggc_cleared_alloc to allocate target_optabs.
1552 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1553 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1555 * internal-fn.h (internal_load_fn_p): Declare.
1556 (internal_gather_scatter_fn_p): Likewise.
1557 (internal_fn_mask_index): Likewise.
1558 (internal_gather_scatter_fn_supported_p): Likewise.
1559 * internal-fn.c (gather_load_direct): New macro.
1560 (expand_gather_load_optab_fn): New function.
1561 (direct_gather_load_optab_supported_p): New macro.
1562 (direct_internal_fn_optab): New function.
1563 (internal_load_fn_p): Likewise.
1564 (internal_gather_scatter_fn_p): Likewise.
1565 (internal_fn_mask_index): Likewise.
1566 (internal_gather_scatter_fn_supported_p): Likewise.
1567 * optabs-query.c (supports_at_least_one_mode_p): New function.
1568 (supports_vec_gather_load_p): Likewise.
1569 * optabs-query.h (supports_vec_gather_load_p): Declare.
1570 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1571 and memory_type field.
1572 (NUM_PATTERNS): Bump to 15.
1573 * tree-vect-data-refs.c: Include internal-fn.h.
1574 (vect_gather_scatter_fn_p): New function.
1575 (vect_describe_gather_scatter_call): Likewise.
1576 (vect_check_gather_scatter): Try using internal functions for
1577 gather loads. Recognize existing calls to a gather load function.
1578 (vect_analyze_data_refs): Consider using gather loads if
1579 supports_vec_gather_load_p.
1580 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1581 (vect_get_gather_scatter_offset_type): Likewise.
1582 (vect_convert_mask_for_vectype): Likewise.
1583 (vect_add_conversion_to_patterm): Likewise.
1584 (vect_try_gather_scatter_pattern): Likewise.
1585 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1586 (vect_vect_recog_func_ptrs): Add it.
1587 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1588 internal_fn_mask_index and internal_gather_scatter_fn_p.
1589 (check_load_store_masking): Take the gather_scatter_info as an
1590 argument and handle gather loads.
1591 (vect_get_gather_scatter_ops): New function.
1592 (vectorizable_call): Check internal_load_fn_p.
1593 (vectorizable_load): Likewise. Handle gather load internal
1595 (vectorizable_store): Update call to check_load_store_masking.
1596 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1597 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1598 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1599 (aarch64_gather_scale_operand_d): New predicates.
1600 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1601 (mask_gather_load<mode>): New insns.
1603 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1604 Alan Hayward <alan.hayward@arm.com>
1605 David Sherwood <david.sherwood@arm.com>
1607 * optabs.def (fold_left_plus_optab): New optab.
1608 * doc/md.texi (fold_left_plus_@var{m}): Document.
1609 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1610 * internal-fn.c (fold_left_direct): Define.
1611 (expand_fold_left_optab_fn): Likewise.
1612 (direct_fold_left_optab_supported_p): Likewise.
1613 * fold-const-call.c (fold_const_fold_left): New function.
1614 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1615 * tree-parloops.c (valid_reduction_p): New function.
1616 (gather_scalar_reductions): Use it.
1617 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1618 (vect_finish_replace_stmt): Declare.
1619 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1620 (needs_fold_left_reduction_p): New function, split out from...
1621 (vect_is_simple_reduction): ...here. Accept reductions that
1622 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1623 (vect_force_simple_reduction): Also store the reduction type in
1624 the assignment's STMT_VINFO_REDUC_TYPE.
1625 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1626 (merge_with_identity): New function.
1627 (vect_expand_fold_left): Likewise.
1628 (vectorize_fold_left_reduction): Likewise.
1629 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1630 scalar phi in place for it. Check for target support and reject
1631 cases that would reassociate the operation. Defer the transform
1632 phase to vectorize_fold_left_reduction.
1633 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1634 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1635 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1637 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1639 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1640 call to ifc_temp_var.
1642 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1643 Alan Hayward <alan.hayward@arm.com>
1644 David Sherwood <david.sherwood@arm.com>
1646 * target.def (legitimize_address_displacement): Take the original
1647 offset as a poly_int.
1648 * targhooks.h (default_legitimize_address_displacement): Update
1650 * targhooks.c (default_legitimize_address_displacement): Likewise.
1651 * doc/tm.texi: Regenerate.
1652 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1653 as an argument, moving assert of ad->disp == ad->disp_term to...
1654 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1655 Try calling targetm.legitimize_address_displacement before expanding
1656 the address rather than afterwards, and adjust for the new interface.
1657 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1658 Match the new hook interface. Handle SVE addresses.
1659 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1662 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1664 * Makefile.in (OBJS): Add early-remat.o.
1665 * target.def (select_early_remat_modes): New hook.
1666 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1667 * doc/tm.texi: Regenerate.
1668 * targhooks.h (default_select_early_remat_modes): Declare.
1669 * targhooks.c (default_select_early_remat_modes): New function.
1670 * timevar.def (TV_EARLY_REMAT): New timevar.
1671 * passes.def (pass_early_remat): New pass.
1672 * tree-pass.h (make_pass_early_remat): Declare.
1673 * early-remat.c: New file.
1674 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1676 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1678 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1679 Alan Hayward <alan.hayward@arm.com>
1680 David Sherwood <david.sherwood@arm.com>
1682 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1683 vfm1 with a bound_epilog parameter.
1684 (vect_do_peeling): Update calls accordingly, and move the prologue
1685 call earlier in the function. Treat the base bound_epilog as 0 for
1686 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1687 this base when peeling for gaps.
1688 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1689 with fully-masked loops.
1690 (vect_estimate_min_profitable_iters): Handle the single peeled
1691 iteration in that case.
1693 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1694 Alan Hayward <alan.hayward@arm.com>
1695 David Sherwood <david.sherwood@arm.com>
1697 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1698 single-element interleaving even if the size is not a power of 2.
1699 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1700 accesses for single-element interleaving if the group size is
1703 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1704 Alan Hayward <alan.hayward@arm.com>
1705 David Sherwood <david.sherwood@arm.com>
1707 * doc/md.texi (fold_extract_last_@var{m}): Document.
1708 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1709 * optabs.def (fold_extract_last_optab): New optab.
1710 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1711 * internal-fn.c (fold_extract_direct): New macro.
1712 (expand_fold_extract_optab_fn): Likewise.
1713 (direct_fold_extract_optab_supported_p): Likewise.
1714 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1715 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1716 EXTRACT_LAST_REDUCTION.
1717 (get_initial_def_for_reduction): Do not create an initial vector
1718 for EXTRACT_LAST_REDUCTION reductions.
1719 (vectorizable_reduction): Leave the scalar phi in place for
1720 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1721 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1722 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1723 transform phase to vectorizable_condition.
1724 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1726 (vect_finish_stmt_generation): ...here.
1727 (vect_finish_replace_stmt): New function.
1728 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1729 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1731 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1733 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1734 Alan Hayward <alan.hayward@arm.com>
1735 David Sherwood <david.sherwood@arm.com>
1737 * doc/md.texi (extract_last_@var{m}): Document.
1738 * optabs.def (extract_last_optab): New optab.
1739 * internal-fn.def (EXTRACT_LAST): New internal function.
1740 * internal-fn.c (cond_unary_direct): New macro.
1741 (expand_cond_unary_optab_fn): Likewise.
1742 (direct_cond_unary_optab_supported_p): Likewise.
1743 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1744 loops using EXTRACT_LAST.
1745 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1746 (extract_last_<mode>): ...this optab.
1747 (vec_extract<mode><Vel>): Update accordingly.
1749 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1750 Alan Hayward <alan.hayward@arm.com>
1751 David Sherwood <david.sherwood@arm.com>
1753 * target.def (empty_mask_is_expensive): New hook.
1754 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
1755 * doc/tm.texi: Regenerate.
1756 * targhooks.h (default_empty_mask_is_expensive): Declare.
1757 * targhooks.c (default_empty_mask_is_expensive): New function.
1758 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
1759 if the target says that empty masks are expensive.
1760 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
1762 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
1764 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1765 Alan Hayward <alan.hayward@arm.com>
1766 David Sherwood <david.sherwood@arm.com>
1768 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
1769 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
1770 (vect_use_loop_mask_for_alignment_p): New function.
1771 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
1772 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
1773 niters_skip argument. Make sure that the first niters_skip elements
1774 of the first iteration are inactive.
1775 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
1776 Update call to vect_set_loop_masks_directly.
1777 (get_misalign_in_elems): New function, split out from...
1778 (vect_gen_prolog_loop_niters): ...here.
1779 (vect_update_init_of_dr): Take a code argument that specifies whether
1780 the adjustment should be added or subtracted.
1781 (vect_update_init_of_drs): Likewise.
1782 (vect_prepare_for_masked_peels): New function.
1783 (vect_do_peeling): Skip prologue peeling if we're using a mask
1784 instead. Update call to vect_update_inits_of_drs.
1785 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1787 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
1788 alignment. Do not include the number of peeled iterations in
1789 the minimum threshold in that case.
1790 (vectorizable_induction): Adjust the start value down by
1791 LOOP_VINFO_MASK_SKIP_NITERS iterations.
1792 (vect_transform_loop): Call vect_prepare_for_masked_peels.
1793 Take the number of skipped iterations into account when calculating
1795 * tree-vect-stmts.c (vect_gen_while_not): New function.
1797 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1798 Alan Hayward <alan.hayward@arm.com>
1799 David Sherwood <david.sherwood@arm.com>
1801 * doc/sourcebuild.texi (vect_fully_masked): Document.
1802 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
1804 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
1806 (vect_analyze_loop_2): ...here. Don't check the vectorization
1807 factor against the number of loop iterations if the loop is
1810 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1811 Alan Hayward <alan.hayward@arm.com>
1812 David Sherwood <david.sherwood@arm.com>
1814 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
1815 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
1816 (dump_groups): Update accordingly.
1817 (iv_use::mem_type): New member variable.
1818 (address_p): New function.
1819 (record_use): Add a mem_type argument and initialize the new
1821 (record_group_use): Add a mem_type argument. Use address_p.
1822 Remove obsolete null checks of base_object. Update call to record_use.
1823 (find_interesting_uses_op): Update call to record_group_use.
1824 (find_interesting_uses_cond): Likewise.
1825 (find_interesting_uses_address): Likewise.
1826 (get_mem_type_for_internal_fn): New function.
1827 (find_address_like_use): Likewise.
1828 (find_interesting_uses_stmt): Try find_address_like_use before
1829 calling find_interesting_uses_op.
1830 (addr_offset_valid_p): Use the iv mem_type field as the type
1831 of the addressed memory.
1832 (add_autoinc_candidates): Likewise.
1833 (get_address_cost): Likewise.
1834 (split_small_address_groups_p): Use address_p.
1835 (split_address_groups): Likewise.
1836 (add_iv_candidate_for_use): Likewise.
1837 (autoinc_possible_for_pair): Likewise.
1838 (rewrite_groups): Likewise.
1839 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
1840 (determine_group_iv_cost): Update after split of USE_ADDRESS.
1841 (get_alias_ptr_type_for_ptr_address): New function.
1842 (rewrite_use_address): Rewrite address uses in calls that were
1843 identified by find_address_like_use.
1845 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1846 Alan Hayward <alan.hayward@arm.com>
1847 David Sherwood <david.sherwood@arm.com>
1849 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
1851 * gimple-expr.h (is_gimple_addressable: Likewise.
1852 * gimple-expr.c (is_gimple_address): Likewise.
1853 * internal-fn.c (expand_call_mem_ref): New function.
1854 (expand_mask_load_optab_fn): Use it.
1855 (expand_mask_store_optab_fn): Likewise.
1857 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1858 Alan Hayward <alan.hayward@arm.com>
1859 David Sherwood <david.sherwood@arm.com>
1861 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
1862 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
1863 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
1864 (cond_umax@var{mode}): Document.
1865 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
1866 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
1867 (cond_umin_optab, cond_umax_optab): New optabs.
1868 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
1869 (COND_IOR, COND_XOR): New internal functions.
1870 * internal-fn.h (get_conditional_internal_fn): Declare.
1871 * internal-fn.c (cond_binary_direct): New macro.
1872 (expand_cond_binary_optab_fn): Likewise.
1873 (direct_cond_binary_optab_supported_p): Likewise.
1874 (get_conditional_internal_fn): New function.
1875 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
1876 Cope with reduction statements that are vectorized as calls rather
1878 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
1879 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
1880 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
1881 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
1882 (UNSPEC_COND_EOR): New unspecs.
1883 (optab): Add mappings for them.
1884 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
1885 (sve_int_op, sve_fp_op): New int attributes.
1887 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1888 Alan Hayward <alan.hayward@arm.com>
1889 David Sherwood <david.sherwood@arm.com>
1891 * optabs.def (while_ult_optab): New optab.
1892 * doc/md.texi (while_ult@var{m}@var{n}): Document.
1893 * internal-fn.def (WHILE_ULT): New internal function.
1894 * internal-fn.h (direct_internal_fn_supported_p): New override
1895 that takes two types as argument.
1896 * internal-fn.c (while_direct): New macro.
1897 (expand_while_optab_fn): New function.
1898 (convert_optab_supported_p): Likewise.
1899 (direct_while_optab_supported_p): New macro.
1900 * wide-int.h (wi::udiv_ceil): New function.
1901 * tree-vectorizer.h (rgroup_masks): New structure.
1902 (vec_loop_masks): New typedef.
1903 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
1905 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
1906 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
1907 (vect_max_vf): New function.
1908 (slpeel_make_loop_iterate_ntimes): Delete.
1909 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
1910 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
1911 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
1912 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
1913 internal-fn.h, stor-layout.h and optabs-query.h.
1914 (vect_set_loop_mask): New function.
1915 (add_preheader_seq): Likewise.
1916 (add_header_seq): Likewise.
1917 (interleave_supported_p): Likewise.
1918 (vect_maybe_permute_loop_masks): Likewise.
1919 (vect_set_loop_masks_directly): Likewise.
1920 (vect_set_loop_condition_masked): Likewise.
1921 (vect_set_loop_condition_unmasked): New function, split out from
1922 slpeel_make_loop_iterate_ntimes.
1923 (slpeel_make_loop_iterate_ntimes): Rename to..
1924 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
1925 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
1926 (vect_do_peeling): Update call accordingly.
1927 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
1929 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1930 mask_compare_type, can_fully_mask_p and fully_masked_p.
1931 (release_vec_loop_masks): New function.
1932 (_loop_vec_info): Use it to free the loop masks.
1933 (can_produce_all_loop_masks_p): New function.
1934 (vect_get_max_nscalars_per_iter): Likewise.
1935 (vect_verify_full_masking): Likewise.
1936 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
1937 retries, and free the mask rgroups before retrying. Check loop-wide
1938 reasons for disallowing fully-masked loops. Make the final decision
1939 about whether use a fully-masked loop or not.
1940 (vect_estimate_min_profitable_iters): Do not assume that peeling
1941 for the number of iterations will be needed for fully-masked loops.
1942 (vectorizable_reduction): Disable fully-masked loops.
1943 (vectorizable_live_operation): Likewise.
1944 (vect_halve_mask_nunits): New function.
1945 (vect_double_mask_nunits): Likewise.
1946 (vect_record_loop_mask): Likewise.
1947 (vect_get_loop_mask): Likewise.
1948 (vect_transform_loop): Handle the case in which the final loop
1949 iteration might handle a partial vector. Call vect_set_loop_condition
1950 instead of slpeel_make_loop_iterate_ntimes.
1951 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
1952 (check_load_store_masking): New function.
1953 (prepare_load_store_mask): Likewise.
1954 (vectorizable_store): Handle fully-masked loops.
1955 (vectorizable_load): Likewise.
1956 (supportable_widening_operation): Use vect_halve_mask_nunits for
1958 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
1959 (vect_gen_while): New function.
1960 * config/aarch64/aarch64.md (umax<mode>3): New expander.
1961 (aarch64_uqdec<mode>): New insn.
1963 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1964 Alan Hayward <alan.hayward@arm.com>
1965 David Sherwood <david.sherwood@arm.com>
1967 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
1968 (reduc_xor_scal_optab): New optabs.
1969 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
1970 (reduc_xor_scal_@var{m}): Document.
1971 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
1972 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
1974 * fold-const-call.c (fold_const_call): Handle them.
1975 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
1976 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
1977 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
1978 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
1979 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
1980 (UNSPEC_XORV): New unspecs.
1981 (optab): Add entries for them.
1982 (BITWISEV): New int iterator.
1983 (bit_reduc_op): New int attributes.
1985 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1986 Alan Hayward <alan.hayward@arm.com>
1987 David Sherwood <david.sherwood@arm.com>
1989 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
1990 * internal-fn.def (VEC_SHL_INSERT): New internal function.
1991 * optabs.def (vec_shl_insert_optab): New optab.
1992 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
1993 (duplicate_and_interleave): Likewise.
1994 * tree-vect-loop.c: Include internal-fn.h.
1995 (neutral_op_for_slp_reduction): New function, split out from
1996 get_initial_defs_for_reduction.
1997 (get_initial_def_for_reduction): Handle option 2 for variable-length
1998 vectors by loading the neutral value into a vector and then shifting
1999 the initial value into element 0.
2000 (get_initial_defs_for_reduction): Replace the code argument with
2001 the neutral value calculated by neutral_op_for_slp_reduction.
2002 Use gimple_build_vector for constant-length vectors.
2003 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2004 but the first group_size elements have a neutral value.
2005 Use duplicate_and_interleave otherwise.
2006 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2007 Update call to get_initial_defs_for_reduction. Handle SLP
2008 reductions for variable-length vectors by creating one vector
2009 result for each scalar result, with the elements associated
2010 with other scalar results stubbed out with the neutral value.
2011 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2012 Require IFN_VEC_SHL_INSERT for double reductions on
2013 variable-length vectors, or SLP reductions that have
2014 a neutral value. Require can_duplicate_and_interleave_p
2015 support for variable-length unchained SLP reductions if there
2016 is no neutral value, such as for MIN/MAX reductions. Also require
2017 the number of vector elements to be a multiple of the number of
2018 SLP statements when doing variable-length unchained SLP reductions.
2019 Update call to vect_create_epilog_for_reduction.
2020 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2021 and remove initial values.
2022 (duplicate_and_interleave): Make public.
2023 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2024 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2026 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2027 Alan Hayward <alan.hayward@arm.com>
2028 David Sherwood <david.sherwood@arm.com>
2030 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2031 (can_duplicate_and_interleave_p): New function.
2032 (vect_get_and_check_slp_defs): Take the vector of statements
2033 rather than just the current one. Remove excess parentheses.
2034 Restriction rejectinon of vect_constant_def and vect_external_def
2035 for variable-length vectors to boolean types, or types for which
2036 can_duplicate_and_interleave_p is false.
2037 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2038 (duplicate_and_interleave): New function.
2039 (vect_get_constant_vectors): Use gimple_build_vector for
2040 constant-length vectors and suitable variable-length constant
2041 vectors. Use duplicate_and_interleave for other variable-length
2042 vectors. Don't defer the update when inserting new statements.
2044 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2045 Alan Hayward <alan.hayward@arm.com>
2046 David Sherwood <david.sherwood@arm.com>
2048 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2049 min_profitable_iters doesn't go negative.
2051 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2052 Alan Hayward <alan.hayward@arm.com>
2053 David Sherwood <david.sherwood@arm.com>
2055 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2056 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2057 * optabs.def (vec_mask_load_lanes_optab): New optab.
2058 (vec_mask_store_lanes_optab): Likewise.
2059 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2060 (MASK_STORE_LANES): Likewise.
2061 * internal-fn.c (mask_load_lanes_direct): New macro.
2062 (mask_store_lanes_direct): Likewise.
2063 (expand_mask_load_optab_fn): Handle masked operations.
2064 (expand_mask_load_lanes_optab_fn): New macro.
2065 (expand_mask_store_optab_fn): Handle masked operations.
2066 (expand_mask_store_lanes_optab_fn): New macro.
2067 (direct_mask_load_lanes_optab_supported_p): Likewise.
2068 (direct_mask_store_lanes_optab_supported_p): Likewise.
2069 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2071 (vect_load_lanes_supported): Likewise.
2072 * tree-vect-data-refs.c (strip_conversion): New function.
2073 (can_group_stmts_p): Likewise.
2074 (vect_analyze_data_ref_accesses): Use it instead of checking
2075 for a pair of assignments.
2076 (vect_store_lanes_supported): Take a masked_p parameter.
2077 (vect_load_lanes_supported): Likewise.
2078 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2079 vect_store_lanes_supported and vect_load_lanes_supported.
2080 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2081 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2082 parameter. Don't allow gaps for masked accesses.
2083 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2084 and vect_load_lanes_supported.
2085 (get_load_store_type): Take a masked_p parameter and update
2086 call to get_group_load_store_type.
2087 (vectorizable_store): Update call to get_load_store_type.
2088 Handle IFN_MASK_STORE_LANES.
2089 (vectorizable_load): Update call to get_load_store_type.
2090 Handle IFN_MASK_LOAD_LANES.
2092 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2093 Alan Hayward <alan.hayward@arm.com>
2094 David Sherwood <david.sherwood@arm.com>
2096 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2098 * config/aarch64/aarch64-protos.h
2099 (aarch64_sve_struct_memory_operand_p): Declare.
2100 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2101 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2102 (VPRED, vpred): Handle SVE structure modes.
2103 * config/aarch64/constraints.md (Utx): New constraint.
2104 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2105 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2106 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2107 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2108 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2109 structure modes. Split into pieces after RA.
2110 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2111 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2113 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2114 SVE structure modes.
2115 (aarch64_classify_address): Likewise.
2116 (sizetochar): Move earlier in file.
2117 (aarch64_print_operand): Handle SVE register lists.
2118 (aarch64_array_mode): New function.
2119 (aarch64_sve_struct_memory_operand_p): Likewise.
2120 (TARGET_ARRAY_MODE): Redefine.
2122 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2123 Alan Hayward <alan.hayward@arm.com>
2124 David Sherwood <david.sherwood@arm.com>
2126 * target.def (array_mode): New target hook.
2127 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2128 * doc/tm.texi: Regenerate.
2129 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2130 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2131 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2133 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2136 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2137 Alan Hayward <alan.hayward@arm.com>
2138 David Sherwood <david.sherwood@arm.com>
2140 * fold-const.c (fold_binary_loc): Check the argument types
2141 rather than the result type when testing for a vector operation.
2143 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2145 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2146 * doc/tm.texi: Regenerate.
2148 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2149 Alan Hayward <alan.hayward@arm.com>
2150 David Sherwood <david.sherwood@arm.com>
2152 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2153 (sve): Document new AArch64 extension.
2154 * doc/md.texi (w): Extend the description of the AArch64
2155 constraint to include SVE vectors.
2156 (Upl, Upa): Document new AArch64 predicate constraints.
2157 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2159 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2160 (msve-vector-bits=): New option.
2161 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2162 SVE when these are disabled.
2163 (sve): New extension.
2164 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2165 modes. Adjust their number of units based on aarch64_sve_vg.
2166 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2167 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2168 aarch64_addr_query_type.
2169 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2170 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2171 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2172 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2173 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2174 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2175 (aarch64_simd_imm_zero_p): Delete.
2176 (aarch64_check_zero_based_sve_index_immediate): Declare.
2177 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2178 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2179 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2180 (aarch64_sve_float_mul_immediate_p): Likewise.
2181 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2183 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2184 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2185 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2186 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2187 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2188 (aarch64_regmode_natural_size): Likewise.
2189 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2190 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2192 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2193 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2194 for VG and the SVE predicate registers.
2195 (V_ALIASES): Add a "z"-prefixed alias.
2196 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2197 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2198 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2199 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2200 (REG_CLASS_NAMES): Add entries for them.
2201 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2202 and the predicate registers.
2203 (aarch64_sve_vg): Declare.
2204 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2205 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2206 (REGMODE_NATURAL_SIZE): Define.
2207 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2209 * config/aarch64/aarch64.c: Include cfgrtl.h.
2210 (simd_immediate_info): Add a constructor for series vectors,
2211 and an associated step field.
2212 (aarch64_sve_vg): New variable.
2213 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2214 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2215 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2216 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2217 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2218 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2219 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2220 (aarch64_get_mask_mode): New functions.
2221 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2222 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2223 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2224 predicate modes and predicate registers. Explicitly restrict
2225 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2226 to store a vector mode if it is recognized by
2227 aarch64_classify_vector_mode.
2228 (aarch64_regmode_natural_size): New function.
2229 (aarch64_hard_regno_caller_save_mode): Return the original mode
2231 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2232 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2233 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2234 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2236 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2237 does not overlap dest if the function is frame-related. Handle
2239 (aarch64_split_add_offset): New function.
2240 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2241 them aarch64_add_offset.
2242 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2243 and update call to aarch64_sub_sp.
2244 (aarch64_add_cfa_expression): New function.
2245 (aarch64_expand_prologue): Pass extra temporary registers to the
2246 functions above. Handle the case in which we need to emit new
2247 DW_CFA_expressions for registers that were originally saved
2248 relative to the stack pointer, but now have to be expressed
2249 relative to the frame pointer.
2250 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2252 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2253 IP0 and IP1 values for SVE frames.
2254 (aarch64_expand_vec_series): New function.
2255 (aarch64_expand_sve_widened_duplicate): Likewise.
2256 (aarch64_expand_sve_const_vector): Likewise.
2257 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2258 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2259 into the register, rather than emitting a SET directly.
2260 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2261 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2262 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2263 (offset_9bit_signed_scaled_p): New functions.
2264 (aarch64_replicate_bitmask_imm): New function.
2265 (aarch64_bitmask_imm): Use it.
2266 (aarch64_cannot_force_const_mem): Reject expressions involving
2267 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2268 (aarch64_classify_index): Handle SVE indices, by requiring
2269 a plain register index with a scale that matches the element size.
2270 (aarch64_classify_address): Handle SVE addresses. Assert that
2271 the mode of the address is VOIDmode or an integer mode.
2272 Update call to aarch64_classify_symbol.
2273 (aarch64_classify_symbolic_expression): Update call to
2274 aarch64_classify_symbol.
2275 (aarch64_const_vec_all_in_range_p): New function.
2276 (aarch64_print_vector_float_operand): Likewise.
2277 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2278 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2279 and the FP immediates 1.0 and 0.5.
2280 (aarch64_print_address_internal): Handle SVE addresses.
2281 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2282 (aarch64_regno_regclass): Handle predicate registers.
2283 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2285 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2286 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2287 (aarch64_convert_sve_vector_bits): New function.
2288 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2289 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2291 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2292 Handle SVE vector and predicate modes. Accept VL-based constants
2293 that need only one temporary register, and VL offsets that require
2294 no temporary registers.
2295 (aarch64_conditional_register_usage): Mark the predicate registers
2296 as fixed if SVE isn't available.
2297 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2298 Return true for SVE vector and predicate modes.
2299 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2300 rather than an unsigned int. Handle SVE modes.
2301 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2303 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2305 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2306 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2307 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2308 (aarch64_sve_float_mul_immediate_p): New functions.
2309 (aarch64_sve_valid_immediate): New function.
2310 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2311 Explicitly reject structure modes. Check for INDEX constants.
2312 Handle PTRUE and PFALSE constants.
2313 (aarch64_check_zero_based_sve_index_immediate): New function.
2314 (aarch64_simd_imm_zero_p): Delete.
2315 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2316 vector modes. Accept constants in the range of CNT[BHWD].
2317 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2318 ask for an Advanced SIMD mode.
2319 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2320 (aarch64_simd_vector_alignment): Handle SVE predicates.
2321 (aarch64_vectorize_preferred_vector_alignment): New function.
2322 (aarch64_simd_vector_alignment_reachable): Use it instead of
2324 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2325 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2327 (MAX_VECT_LEN): Delete.
2328 (expand_vec_perm_d): Add a vec_flags field.
2329 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2330 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2331 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2333 (aarch64_evpc_rev): Rename to...
2334 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2335 (aarch64_evpc_rev_global): New function.
2336 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2337 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2339 (aarch64_evpc_sve_tbl): New function.
2340 (aarch64_expand_vec_perm_const_1): Update after rename of
2341 aarch64_evpc_rev. Handle SVE permutes too, trying
2342 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2343 than aarch64_evpc_tbl.
2344 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2345 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2346 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2347 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2348 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2349 (aarch64_expand_sve_vcond): New functions.
2350 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2351 of aarch64_vector_mode_p.
2352 (aarch64_dwarf_poly_indeterminate_value): New function.
2353 (aarch64_compute_pressure_classes): Likewise.
2354 (aarch64_can_change_mode_class): Likewise.
2355 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2356 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2357 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2358 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2359 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2360 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2361 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2362 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2364 (Dn, Dl, Dr): Accept const as well as const_vector.
2365 (Dz): Likewise. Compare against CONST0_RTX.
2366 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2367 of "vector" where appropriate.
2368 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2369 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2370 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2371 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2372 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2373 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2374 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2375 (v_int_equiv): Extend to SVE modes.
2376 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2378 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2379 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2380 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2381 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2382 (SVE_COND_FP_CMP): New int iterators.
2383 (perm_hilo): Handle the new unpack unspecs.
2384 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2386 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2387 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2388 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2389 (aarch64_equality_operator, aarch64_constant_vector_operand)
2390 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2391 (aarch64_sve_nonimmediate_operand): Likewise.
2392 (aarch64_sve_general_operand): Likewise.
2393 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2394 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2395 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2396 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2397 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2398 (aarch64_sve_float_arith_immediate): Likewise.
2399 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2400 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2401 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2402 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2403 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2404 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2405 (aarch64_sve_float_arith_operand): Likewise.
2406 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2407 (aarch64_sve_float_mul_operand): Likewise.
2408 (aarch64_sve_vec_perm_operand): Likewise.
2409 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2410 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2411 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2412 as well as const_vector.
2413 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2414 in file. Use CONST0_RTX and CONSTM1_RTX.
2415 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2416 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2417 Use aarch64_simd_imm_zero.
2418 * config/aarch64/aarch64-sve.md: New file.
2419 * config/aarch64/aarch64.md: Include it.
2420 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2421 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2422 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2423 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2424 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2425 (sve): New attribute.
2426 (enabled): Disable instructions with the sve attribute unless
2428 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2429 aarch64_expand_mov_immediate.
2430 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2431 CNT[BHSD] immediates.
2432 (movti): Split CONST_POLY_INT moves into two halves.
2433 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2434 Split additions that need a temporary here if the destination
2435 is the stack pointer.
2436 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2437 (*add<mode>3_poly_1): New instruction.
2438 (set_clobber_cc): New expander.
2440 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2442 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2443 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2444 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2445 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2446 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2447 Change innermode from fixed_mode_size to machine_mode.
2448 (simplify_subreg): Update call accordingly. Handle a constant-sized
2449 subreg of a variable-length CONST_VECTOR.
2451 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2452 Alan Hayward <alan.hayward@arm.com>
2453 David Sherwood <david.sherwood@arm.com>
2455 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2456 (add_offset_to_base): New function, split out from...
2457 (create_mem_ref): ...here. When handling a scale other than 1,
2458 check first whether the address is valid without the offset.
2459 Add it into the base if so, leaving the index and scale as-is.
2461 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2464 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2465 fold_for_warn before checking if arg2 is INTEGER_CST.
2467 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2469 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2470 (store_multiple_operation): Delete.
2471 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2472 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2473 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2474 guarded by TARGET_STRING.
2475 (rs6000_output_load_multiple): Delete.
2476 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2477 OPTION_MASK_STRING / TARGET_STRING handling.
2478 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2479 (const rs6000_opt_masks) <"string">: Change mask to 0.
2480 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2481 (MASK_STRING): Delete.
2482 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2484 (load_multiple): Delete.
2491 (store_multiple): Delete.
2498 (movmemsi_8reg): Delete.
2499 (corresponding unnamed define_insn): Delete.
2500 (movmemsi_6reg): Delete.
2501 (corresponding unnamed define_insn): Delete.
2502 (movmemsi_4reg): Delete.
2503 (corresponding unnamed define_insn): Delete.
2504 (movmemsi_2reg): Delete.
2505 (corresponding unnamed define_insn): Delete.
2506 (movmemsi_1reg): Delete.
2507 (corresponding unnamed define_insn): Delete.
2508 * config/rs6000/rs6000.opt (mno-string): New.
2509 (mstring): Replace by deprecation warning stub.
2510 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2512 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2514 * regrename.c (regrename_do_replace): If replacing the same
2515 reg multiple times, try to reuse last created gen_raw_REG.
2518 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2519 main to workaround a bug in GDB.
2521 2018-01-12 Tom de Vries <tom@codesourcery.com>
2524 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2526 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2528 PR rtl-optimization/80481
2529 * ira-color.c (get_cap_member): New function.
2530 (allocnos_conflict_by_live_ranges_p): Use it.
2531 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2532 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2534 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2537 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2538 (*saddl_se_1): Ditto.
2540 (*ssubl_se_1): Ditto.
2542 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2544 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2545 rather than wi::to_widest for DR_INITs.
2546 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2547 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2548 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2550 (vect_analyze_group_access_1): Note that here.
2552 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2554 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2555 polynomial type sizes.
2557 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2559 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2560 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2561 (gimple_add_tmp_var): Likewise.
2563 2018-01-12 Martin Liska <mliska@suse.cz>
2565 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2566 (gimple_alloc_sizes): Likewise.
2567 (dump_gimple_statistics): Use PRIu64 in printf format.
2568 * gimple.h: Change uint64_t to int.
2570 2018-01-12 Martin Liska <mliska@suse.cz>
2572 * tree-core.h: Use uint64_t instead of int.
2573 * tree.c (tree_node_counts): Likewise.
2574 (tree_node_sizes): Likewise.
2575 (dump_tree_statistics): Use PRIu64 in printf format.
2577 2018-01-12 Martin Liska <mliska@suse.cz>
2579 * Makefile.in: As qsort_chk is implemented in vec.c, add
2580 vec.o to linkage of gencfn-macros.
2581 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2582 passing the info to record_node_allocation_statistics.
2583 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2585 * ggc-common.c (struct ggc_usage): Add operator== and use
2586 it in operator< and compare function.
2587 * mem-stats.h (struct mem_usage): Likewise.
2588 * vec.c (struct vec_usage): Remove operator< and compare
2589 function. Can be simply inherited.
2591 2018-01-12 Martin Jambor <mjambor@suse.cz>
2594 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2595 * tree-ssa-math-opts.c: Include domwalk.h.
2596 (convert_mult_to_fma_1): New function.
2597 (fma_transformation_info): New type.
2598 (fma_deferring_state): Likewise.
2599 (cancel_fma_deferring): New function.
2600 (result_of_phi): Likewise.
2601 (last_fma_candidate_feeds_initial_phi): Likewise.
2602 (convert_mult_to_fma): Added deferring logic, split actual
2603 transformation to convert_mult_to_fma_1.
2604 (math_opts_dom_walker): New type.
2605 (math_opts_dom_walker::after_dom_children): New method, body moved
2606 here from pass_optimize_widening_mul::execute, added deferring logic
2608 (pass_optimize_widening_mul::execute): Moved most of code to
2609 math_opts_dom_walker::after_dom_children.
2610 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2611 * config/i386/i386.c (ix86_option_override_internal): Added
2612 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2614 2018-01-12 Richard Biener <rguenther@suse.de>
2617 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2618 inline instance vars.
2620 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2623 * config/rx/rx.c (rx_is_restricted_memory_address):
2626 2018-01-12 Richard Biener <rguenther@suse.de>
2628 PR tree-optimization/80846
2629 * target.def (split_reduction): New target hook.
2630 * targhooks.c (default_split_reduction): New function.
2631 * targhooks.h (default_split_reduction): Declare.
2632 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2633 target requests first reduce vectors by combining low and high
2635 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2636 (get_vectype_for_scalar_type_and_size): Export.
2637 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2638 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2639 * doc/tm.texi: Regenerate.
2640 * config/i386/i386.c (ix86_split_reduction): Implement
2641 TARGET_VECTORIZE_SPLIT_REDUCTION.
2643 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2646 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2647 in PIC mode except for TARGET_VXWORKS_RTP.
2648 * config/sparc/sparc.c: Include cfgrtl.h.
2649 (TARGET_INIT_PIC_REG): Define.
2650 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2651 (sparc_pic_register_p): New predicate.
2652 (sparc_legitimate_address_p): Use it.
2653 (sparc_legitimize_pic_address): Likewise.
2654 (sparc_delegitimize_address): Likewise.
2655 (sparc_mode_dependent_address_p): Likewise.
2656 (gen_load_pcrel_sym): Remove 4th parameter.
2657 (load_got_register): Adjust call to above. Remove obsolete stuff.
2658 (sparc_expand_prologue): Do not call load_got_register here.
2659 (sparc_flat_expand_prologue): Likewise.
2660 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2661 (sparc_use_pseudo_pic_reg): New function.
2662 (sparc_init_pic_reg): Likewise.
2663 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2664 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2666 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2668 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2669 Add item for branch_cost.
2671 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2673 PR rtl-optimization/83565
2674 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2675 not extend the result to a larger mode for rotate operations.
2676 (num_sign_bit_copies1): Likewise.
2678 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2681 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2683 Use values-Xc.o for -pedantic.
2684 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2686 2018-01-12 Martin Liska <mliska@suse.cz>
2689 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2691 (possible_polymorphic_call_targets): Use it.
2692 (ipa_devirt): Likewise.
2694 2018-01-12 Martin Liska <mliska@suse.cz>
2696 * profile-count.h (enum profile_quality): Use 0 as invalid
2697 enum value of profile_quality.
2699 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2701 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2702 -mext-string options.
2704 2018-01-12 Richard Biener <rguenther@suse.de>
2706 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2707 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2708 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2710 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2712 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2714 * configure.ac (--with-long-double-format): Add support for the
2715 configuration option to change the default long double format on
2717 * config.gcc (powerpc*-linux*-*): Likewise.
2718 * configure: Regenerate.
2719 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2720 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2721 used without modification.
2723 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2725 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2726 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2727 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2728 MISC_BUILTIN_SPEC_BARRIER.
2729 (rs6000_init_builtins): Likewise.
2730 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2732 (speculation_barrier): New define_insn.
2733 * doc/extend.texi: Document __builtin_speculation_barrier.
2735 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2738 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2739 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2740 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2742 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2743 integral modes instead of "ss" and "sd".
2744 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2745 vectors with 32-bit and 64-bit elements.
2746 (vecdupssescalarmodesuffix): New mode attribute.
2747 (vec_dup<mode>): Use it.
2749 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2752 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2753 frame if argument is passed on stack.
2755 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2758 * ree.c (combine_reaching_defs): Optimize also
2759 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
2760 reg2=any_extend(exp); reg1=reg2;, formatting fix.
2762 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2765 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
2767 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2770 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
2771 after they are computed.
2773 2018-01-11 Bin Cheng <bin.cheng@arm.com>
2775 PR tree-optimization/83695
2776 * gimple-loop-linterchange.cc
2777 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
2778 reset cached scev information after interchange.
2779 (pass_linterchange::execute): Remove call to scev_reset_htab.
2781 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2783 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
2784 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
2785 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
2786 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
2787 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
2788 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
2789 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
2790 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
2791 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
2792 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
2793 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
2794 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
2795 (V_lane_reg): Likewise.
2796 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
2798 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
2799 (vfmal_lane_low<mode>_intrinsic,
2800 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
2801 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
2802 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
2803 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
2804 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
2805 vfmsl_lane_high<mode>_intrinsic): New define_insns.
2807 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2809 * config/arm/arm-cpus.in (fp16fml): New feature.
2810 (ALL_SIMD): Add fp16fml.
2811 (armv8.2-a): Add fp16fml as an option.
2812 (armv8.3-a): Likewise.
2813 (armv8.4-a): Add fp16fml as part of fp16.
2814 * config/arm/arm.h (TARGET_FP16FML): Define.
2815 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
2817 * config/arm/arm-modes.def (V2HF): Define.
2818 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
2819 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
2820 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
2821 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
2822 vfmsl_low, vfmsl_high): New set of builtins.
2823 * config/arm/iterators.md (PLUSMINUS): New code iterator.
2824 (vfml_op): New code attribute.
2825 (VFMLHALVES): New int iterator.
2826 (VFML, VFMLSEL): New mode attributes.
2827 (V_reg): Define mapping for V2HF.
2828 (V_hi, V_lo): New mode attributes.
2829 (VF_constraint): Likewise.
2830 (vfml_half, vfml_half_selector): New int attributes.
2831 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
2833 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
2834 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
2836 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
2837 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
2838 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
2839 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
2841 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
2842 Document new effective target and option set.
2844 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2846 * config/arm/arm-cpus.in (armv8_4): New feature.
2847 (ARMv8_4a): New fgroup.
2848 (armv8.4-a): New arch.
2849 * config/arm/arm-tables.opt: Regenerate.
2850 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
2851 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
2852 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
2853 Add matching rules for -march=armv8.4-a and extensions.
2854 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
2856 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
2859 * config/rx/rx.md (BW): New mode attribute.
2860 (sync_lock_test_and_setsi): Add mode suffix to insn output.
2862 2018-01-11 Richard Biener <rguenther@suse.de>
2864 PR tree-optimization/83435
2865 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
2866 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
2867 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
2869 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2870 Alan Hayward <alan.hayward@arm.com>
2871 David Sherwood <david.sherwood@arm.com>
2873 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
2875 (aarch64_classify_address): Initialize it. Track polynomial offsets.
2876 (aarch64_print_address_internal): Use it to check for a zero offset.
2878 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2879 Alan Hayward <alan.hayward@arm.com>
2880 David Sherwood <david.sherwood@arm.com>
2882 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
2883 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
2884 Return a poly_int64 rather than a HOST_WIDE_INT.
2885 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
2886 rather than a HOST_WIDE_INT.
2887 * config/aarch64/aarch64.h (aarch64_frame): Protect with
2888 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
2889 hard_fp_offset, frame_size, initial_adjust, callee_offset and
2890 final_offset from HOST_WIDE_INT to poly_int64.
2891 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
2892 to_constant when getting the number of units in an Advanced SIMD
2894 (aarch64_builtin_vectorized_function): Check for a constant number
2896 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
2898 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
2899 attribute instead of GET_MODE_NUNITS.
2900 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
2901 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
2902 GET_MODE_SIZE for fixed-size registers.
2903 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
2904 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
2905 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
2906 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
2907 (aarch64_print_operand, aarch64_print_address_internal)
2908 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
2909 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
2910 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
2911 Handle polynomial GET_MODE_SIZE.
2912 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
2913 wider than SImode without modification.
2914 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
2915 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
2916 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
2917 passing and returning SVE modes.
2918 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
2919 rather than GEN_INT.
2920 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
2921 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
2922 (aarch64_allocate_and_probe_stack_space): Likewise.
2923 (aarch64_layout_frame): Cope with polynomial offsets.
2924 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
2925 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
2927 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
2928 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
2929 poly_int64 rather than a HOST_WIDE_INT.
2930 (aarch64_get_separate_components, aarch64_process_components)
2931 (aarch64_expand_prologue, aarch64_expand_epilogue)
2932 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
2933 (aarch64_anchor_offset): New function, split out from...
2934 (aarch64_legitimize_address): ...here.
2935 (aarch64_builtin_vectorization_cost): Handle polynomial
2936 TYPE_VECTOR_SUBPARTS.
2937 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
2939 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
2940 number of elements from the PARALLEL rather than the mode.
2941 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
2942 rather than GET_MODE_BITSIZE.
2943 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
2944 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
2945 (aarch64_expand_vec_perm_const_1): Handle polynomial
2946 d->perm.length () and d->perm elements.
2947 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
2948 Apply to_constant to d->perm elements.
2949 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
2950 polynomial CONST_VECTOR_NUNITS.
2951 (aarch64_move_pointer): Take amount as a poly_int64 rather
2953 (aarch64_progress_pointer): Avoid temporary variable.
2954 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
2955 the mode attribute instead of GET_MODE.
2957 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2958 Alan Hayward <alan.hayward@arm.com>
2959 David Sherwood <david.sherwood@arm.com>
2961 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
2962 x exists before using it.
2963 (aarch64_add_constant_internal): Rename to...
2964 (aarch64_add_offset_1): ...this. Replace regnum with separate
2965 src and dest rtxes. Handle the case in which they're different,
2966 including when the offset is zero. Replace scratchreg with an rtx.
2967 Use 2 additions if there is no spare register into which we can
2968 move a 16-bit constant.
2969 (aarch64_add_constant): Delete.
2970 (aarch64_add_offset): Replace reg with separate src and dest
2971 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
2972 Use aarch64_add_offset_1.
2973 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
2974 an rtx rather than an int. Take the delta as a poly_int64
2975 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
2976 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
2977 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
2978 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
2979 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
2981 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
2982 aarch64_add_constant.
2984 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2986 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
2987 Use scalar_float_mode.
2989 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2991 * config/aarch64/aarch64-simd.md
2992 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
2993 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
2994 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
2995 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
2996 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
2997 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
2998 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
2999 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3000 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3001 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3003 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3006 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3007 targ_options->x_arm_arch_string is non NULL.
3009 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3011 * config/aarch64/aarch64.h
3012 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3014 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3017 * expmed.c (emit_store_flag_force): Swap if const op0
3018 and change VOIDmode to mode of op0.
3020 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3022 PR rtl-optimization/83761
3023 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3024 than bytes to mode_for_size.
3026 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3029 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3030 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3033 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3036 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3037 when in layout mode.
3038 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3039 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3042 2018-01-10 Michael Collison <michael.collison@arm.com>
3044 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3045 * config/aarch64/aarch64-option-extension.def: Add
3046 AARCH64_OPT_EXTENSION of 'fp16fml'.
3047 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3048 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3049 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3050 * config/aarch64/constraints.md (Ui7): New constraint.
3051 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3052 (VFMLA_SEL_W): Ditto.
3055 (VFMLA16_LOW): New int iterator.
3056 (VFMLA16_HIGH): Ditto.
3057 (UNSPEC_FMLAL): New unspec.
3058 (UNSPEC_FMLSL): Ditto.
3059 (UNSPEC_FMLAL2): Ditto.
3060 (UNSPEC_FMLSL2): Ditto.
3061 (f16mac): New code attribute.
3062 * config/aarch64/aarch64-simd-builtins.def
3063 (aarch64_fmlal_lowv2sf): Ditto.
3064 (aarch64_fmlsl_lowv2sf): Ditto.
3065 (aarch64_fmlalq_lowv4sf): Ditto.
3066 (aarch64_fmlslq_lowv4sf): Ditto.
3067 (aarch64_fmlal_highv2sf): Ditto.
3068 (aarch64_fmlsl_highv2sf): Ditto.
3069 (aarch64_fmlalq_highv4sf): Ditto.
3070 (aarch64_fmlslq_highv4sf): Ditto.
3071 (aarch64_fmlal_lane_lowv2sf): Ditto.
3072 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3073 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3074 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3075 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3076 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3077 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3078 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3079 (aarch64_fmlal_lane_highv2sf): Ditto.
3080 (aarch64_fmlsl_lane_highv2sf): Ditto.
3081 (aarch64_fmlal_laneq_highv2sf): Ditto.
3082 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3083 (aarch64_fmlalq_lane_highv4sf): Ditto.
3084 (aarch64_fmlsl_lane_highv4sf): Ditto.
3085 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3086 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3087 * config/aarch64/aarch64-simd.md:
3088 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3089 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3090 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3091 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3092 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3093 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3094 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3095 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3096 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3097 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3098 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3099 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3100 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3101 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3102 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3103 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3104 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3105 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3106 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3107 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3108 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3109 (vfmlsl_low_u32): Ditto.
3110 (vfmlalq_low_u32): Ditto.
3111 (vfmlslq_low_u32): Ditto.
3112 (vfmlal_high_u32): Ditto.
3113 (vfmlsl_high_u32): Ditto.
3114 (vfmlalq_high_u32): Ditto.
3115 (vfmlslq_high_u32): Ditto.
3116 (vfmlal_lane_low_u32): Ditto.
3117 (vfmlsl_lane_low_u32): Ditto.
3118 (vfmlal_laneq_low_u32): Ditto.
3119 (vfmlsl_laneq_low_u32): Ditto.
3120 (vfmlalq_lane_low_u32): Ditto.
3121 (vfmlslq_lane_low_u32): Ditto.
3122 (vfmlalq_laneq_low_u32): Ditto.
3123 (vfmlslq_laneq_low_u32): Ditto.
3124 (vfmlal_lane_high_u32): Ditto.
3125 (vfmlsl_lane_high_u32): Ditto.
3126 (vfmlal_laneq_high_u32): Ditto.
3127 (vfmlsl_laneq_high_u32): Ditto.
3128 (vfmlalq_lane_high_u32): Ditto.
3129 (vfmlslq_lane_high_u32): Ditto.
3130 (vfmlalq_laneq_high_u32): Ditto.
3131 (vfmlslq_laneq_high_u32): Ditto.
3132 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3133 (AARCH64_FL_FOR_ARCH8_4): New.
3134 (AARCH64_ISA_F16FML): New ISA flag.
3135 (TARGET_F16FML): New feature flag for fp16fml.
3136 (doc/invoke.texi): Document new fp16fml option.
3138 2018-01-10 Michael Collison <michael.collison@arm.com>
3140 * config/aarch64/aarch64-builtins.c:
3141 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3142 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3143 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3144 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3145 (AARCH64_ISA_SHA3): New ISA flag.
3146 (TARGET_SHA3): New feature flag for sha3.
3147 * config/aarch64/iterators.md (sha512_op): New int attribute.
3148 (CRYPTO_SHA512): New int iterator.
3149 (UNSPEC_SHA512H): New unspec.
3150 (UNSPEC_SHA512H2): Ditto.
3151 (UNSPEC_SHA512SU0): Ditto.
3152 (UNSPEC_SHA512SU1): Ditto.
3153 * config/aarch64/aarch64-simd-builtins.def
3154 (aarch64_crypto_sha512hqv2di): New builtin.
3155 (aarch64_crypto_sha512h2qv2di): Ditto.
3156 (aarch64_crypto_sha512su0qv2di): Ditto.
3157 (aarch64_crypto_sha512su1qv2di): Ditto.
3158 (aarch64_eor3qv8hi): Ditto.
3159 (aarch64_rax1qv2di): Ditto.
3160 (aarch64_xarqv2di): Ditto.
3161 (aarch64_bcaxqv8hi): Ditto.
3162 * config/aarch64/aarch64-simd.md:
3163 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3164 (aarch64_crypto_sha512su0qv2di): Ditto.
3165 (aarch64_crypto_sha512su1qv2di): Ditto.
3166 (aarch64_eor3qv8hi): Ditto.
3167 (aarch64_rax1qv2di): Ditto.
3168 (aarch64_xarqv2di): Ditto.
3169 (aarch64_bcaxqv8hi): Ditto.
3170 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3171 (vsha512h2q_u64): Ditto.
3172 (vsha512su0q_u64): Ditto.
3173 (vsha512su1q_u64): Ditto.
3174 (veor3q_u16): Ditto.
3175 (vrax1q_u64): Ditto.
3177 (vbcaxq_u16): Ditto.
3178 * config/arm/types.md (crypto_sha512): New type attribute.
3179 (crypto_sha3): Ditto.
3180 (doc/invoke.texi): Document new sha3 option.
3182 2018-01-10 Michael Collison <michael.collison@arm.com>
3184 * config/aarch64/aarch64-builtins.c:
3185 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3186 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3187 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3188 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3189 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3190 (AARCH64_ISA_SM4): New ISA flag.
3191 (TARGET_SM4): New feature flag for sm4.
3192 * config/aarch64/aarch64-simd-builtins.def
3193 (aarch64_sm3ss1qv4si): Ditto.
3194 (aarch64_sm3tt1aq4si): Ditto.
3195 (aarch64_sm3tt1bq4si): Ditto.
3196 (aarch64_sm3tt2aq4si): Ditto.
3197 (aarch64_sm3tt2bq4si): Ditto.
3198 (aarch64_sm3partw1qv4si): Ditto.
3199 (aarch64_sm3partw2qv4si): Ditto.
3200 (aarch64_sm4eqv4si): Ditto.
3201 (aarch64_sm4ekeyqv4si): Ditto.
3202 * config/aarch64/aarch64-simd.md:
3203 (aarch64_sm3ss1qv4si): Ditto.
3204 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3205 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3206 (aarch64_sm4eqv4si): Ditto.
3207 (aarch64_sm4ekeyqv4si): Ditto.
3208 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3209 (sm3part_op): Ditto.
3210 (CRYPTO_SM3TT): Ditto.
3211 (CRYPTO_SM3PART): Ditto.
3212 (UNSPEC_SM3SS1): New unspec.
3213 (UNSPEC_SM3TT1A): Ditto.
3214 (UNSPEC_SM3TT1B): Ditto.
3215 (UNSPEC_SM3TT2A): Ditto.
3216 (UNSPEC_SM3TT2B): Ditto.
3217 (UNSPEC_SM3PARTW1): Ditto.
3218 (UNSPEC_SM3PARTW2): Ditto.
3219 (UNSPEC_SM4E): Ditto.
3220 (UNSPEC_SM4EKEY): Ditto.
3221 * config/aarch64/constraints.md (Ui2): New constraint.
3222 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3223 * config/arm/types.md (crypto_sm3): New type attribute.
3224 (crypto_sm4): Ditto.
3225 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3226 (vsm3tt1aq_u32): Ditto.
3227 (vsm3tt1bq_u32): Ditto.
3228 (vsm3tt2aq_u32): Ditto.
3229 (vsm3tt2bq_u32): Ditto.
3230 (vsm3partw1q_u32): Ditto.
3231 (vsm3partw2q_u32): Ditto.
3232 (vsm4eq_u32): Ditto.
3233 (vsm4ekeyq_u32): Ditto.
3234 (doc/invoke.texi): Document new sm4 option.
3236 2018-01-10 Michael Collison <michael.collison@arm.com>
3238 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3239 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3240 (AARCH64_FL_FOR_ARCH8_4): New.
3241 (AARCH64_FL_V8_4): New flag.
3242 (doc/invoke.texi): Document new armv8.4-a option.
3244 2018-01-10 Michael Collison <michael.collison@arm.com>
3246 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3247 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3248 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3249 * config/aarch64/aarch64-option-extension.def: Add
3250 AARCH64_OPT_EXTENSION of 'sha2'.
3251 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3252 (crypto): Disable sha2 and aes if crypto disabled.
3253 (crypto): Enable aes and sha2 if enabled.
3254 (simd): Disable sha2 and aes if simd disabled.
3255 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3257 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3258 (TARGET_SHA2): New feature flag for sha2.
3259 (TARGET_AES): New feature flag for aes.
3260 * config/aarch64/aarch64-simd.md:
3261 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3262 conditional on TARGET_AES.
3263 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3264 (aarch64_crypto_sha1hsi): Make pattern conditional
3266 (aarch64_crypto_sha1hv4si): Ditto.
3267 (aarch64_be_crypto_sha1hv4si): Ditto.
3268 (aarch64_crypto_sha1su1v4si): Ditto.
3269 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3270 (aarch64_crypto_sha1su0v4si): Ditto.
3271 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3272 (aarch64_crypto_sha256su0v4si): Ditto.
3273 (aarch64_crypto_sha256su1v4si): Ditto.
3274 (doc/invoke.texi): Document new aes and sha2 options.
3276 2018-01-10 Martin Sebor <msebor@redhat.com>
3278 PR tree-optimization/83781
3279 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3282 2018-01-11 Martin Sebor <msebor@gmail.com>
3283 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3285 PR tree-optimization/83501
3286 PR tree-optimization/81703
3288 * tree-ssa-strlen.c (get_string_cst): Rename...
3289 (get_string_len): ...to this. Handle global constants.
3290 (handle_char_store): Adjust.
3292 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3293 Jim Wilson <jimw@sifive.com>
3295 * config/riscv/riscv-protos.h (riscv_output_return): New.
3296 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3297 (riscv_attribute_table, riscv_output_return),
3298 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3299 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3300 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3302 (riscv_expand_prologue): Add early return for naked function.
3303 (riscv_expand_epilogue): Likewise.
3304 (riscv_function_ok_for_sibcall): Return false for naked function.
3305 (riscv_set_current_function): New.
3306 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3307 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3308 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3309 * doc/extend.texi (RISC-V Function Attributes): New.
3311 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3313 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3314 check for 128-bit long double before checking TCmode.
3315 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3316 128-bit long doubles before checking TFmode or TCmode.
3317 (FLOAT128_IBM_P): Likewise.
3319 2018-01-10 Martin Sebor <msebor@redhat.com>
3321 PR tree-optimization/83671
3322 * builtins.c (c_strlen): Unconditionally return zero for the empty
3324 Use -Warray-bounds for warnings.
3325 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3326 for non-constant array indices with COMPONENT_REF, arrays of
3327 arrays, and pointers to arrays.
3328 (gimple_fold_builtin_strlen): Determine and set length range for
3329 non-constant character arrays.
3331 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3334 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3337 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3339 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3341 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3344 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3345 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3346 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3347 indexed_or_indirect_operand predicate.
3348 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3349 (*vsx_le_perm_load_v8hi): Likewise.
3350 (*vsx_le_perm_load_v16qi): Likewise.
3351 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3352 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3353 (*vsx_le_perm_store_v8hi): Likewise.
3354 (*vsx_le_perm_store_v16qi): Likewise.
3355 (eight unnamed splitters): Likewise.
3357 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3359 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3360 * config/rs6000/emmintrin.h: Likewise.
3361 * config/rs6000/mmintrin.h: Likewise.
3362 * config/rs6000/xmmintrin.h: Likewise.
3364 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3367 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3369 * tree.c (tree_nop_conversion): Return true for location wrapper
3371 (maybe_wrap_with_location): New function.
3372 (selftest::check_strip_nops): New function.
3373 (selftest::test_location_wrappers): New function.
3374 (selftest::tree_c_tests): Call it.
3375 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3376 (maybe_wrap_with_location): New decl.
3377 (EXPR_LOCATION_WRAPPER_P): New macro.
3378 (location_wrapper_p): New inline function.
3379 (tree_strip_any_location_wrapper): New inline function.
3381 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3384 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3385 stack_realign_offset for the largest alignment of stack slot
3387 (ix86_find_max_used_stack_alignment): New function.
3388 (ix86_finalize_stack_frame_flags): Use it. Set
3389 max_used_stack_alignment if we don't realign stack.
3390 * config/i386/i386.h (machine_function): Add
3391 max_used_stack_alignment.
3393 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3395 * config/arm/arm.opt (-mbranch-cost): New option.
3396 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3399 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3402 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3403 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3405 2018-01-10 Richard Biener <rguenther@suse.de>
3408 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3409 early out so it also covers the case where we have a non-NULL
3412 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3414 PR tree-optimization/83753
3415 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3416 for non-strided grouped accesses if the number of elements is 1.
3418 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3421 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3422 * i386.h (TARGET_USE_GATHER): Define.
3423 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3425 2018-01-10 Martin Liska <mliska@suse.cz>
3428 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3429 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3431 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3432 CLEANUP_NO_PARTITIONING is not set.
3434 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3436 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3437 for vectors, as a partial revert of r254296.
3438 * rtl.h (const_vec_p): Delete.
3439 (const_vec_duplicate_p): Don't test for vector CONSTs.
3440 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3441 * expmed.c (make_tree): Likewise.
3444 * common.md (E, F): Use CONSTANT_P instead of checking for
3446 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3447 checking for CONST_VECTOR.
3449 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3452 * predict.c (force_edge_cold): Handle in more sane way edges
3455 2018-01-09 Carl Love <cel@us.ibm.com>
3457 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3459 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3460 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3461 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3462 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3463 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3464 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3465 * config/rs6000/rs6000-protos.h: Add extern defition for
3466 rs6000_generate_float2_double_code.
3467 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3469 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3470 (float2_v2df): Add define_expand.
3472 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3475 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3476 op_mode in the force_to_mode call.
3478 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3480 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3481 instead of checking each element individually.
3482 (aarch64_evpc_uzp): Likewise.
3483 (aarch64_evpc_zip): Likewise.
3484 (aarch64_evpc_ext): Likewise.
3485 (aarch64_evpc_rev): Likewise.
3486 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3487 instead of checking each element individually. Return true without
3489 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3490 whether all selected elements come from the same input, instead of
3491 checking each element individually. Remove calls to gen_rtx_REG,
3492 start_sequence and end_sequence and instead assert that no rtl is
3495 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3497 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3498 order of HIGH and CONST checks.
3500 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3502 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3503 if the destination isn't an SSA_NAME.
3505 2018-01-09 Richard Biener <rguenther@suse.de>
3507 PR tree-optimization/83668
3508 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3510 (canonicalize_loop_form): ... here, renamed from ...
3511 (canonicalize_loop_closed_ssa_form): ... this and amended to
3512 swap successor edges for loop exit blocks to make us use
3513 the RPO order we need for initial schedule generation.
3515 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3517 PR tree-optimization/64811
3518 * match.pd: When optimizing comparisons with Inf, avoid
3519 introducing or losing exceptions from comparisons with NaN.
3521 2018-01-09 Martin Liska <mliska@suse.cz>
3524 * asan.c (shadow_mem_size): Add gcc_assert.
3526 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3528 Don't save registers in main().
3531 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3532 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3533 * config/avr/avr.c (avr_set_current_function): Don't error if
3534 naked, OS_task or OS_main are specified at the same time.
3535 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3537 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3539 * common/config/avr/avr-common.c (avr_option_optimization_table):
3540 Switch on -mmain-is-OS_task for optimizing compilations.
3542 2018-01-09 Richard Biener <rguenther@suse.de>
3544 PR tree-optimization/83572
3545 * graphite.c: Include cfganal.h.
3546 (graphite_transform_loops): Connect infinite loops to exit
3547 and remove fake edges at the end.
3549 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3551 * ipa-inline.c (edge_badness): Revert accidental checkin.
3553 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3556 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3557 symbols; not inline clones.
3559 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3562 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3563 hard registers. Formatting fixes.
3565 PR preprocessor/83722
3566 * gcc.c (try_generate_repro): Pass
3567 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3568 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3571 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3572 Kito Cheng <kito.cheng@gmail.com>
3574 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3575 (riscv_leaf_function_p): Delete.
3576 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3578 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3580 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3582 (do_ifelse): New function.
3583 (do_isel): New function.
3584 (do_sub3): New function.
3585 (do_add3): New function.
3586 (do_load_mask_compare): New function.
3587 (do_overlap_load_compare): New function.
3588 (expand_compare_loop): New function.
3589 (expand_block_compare): Call expand_compare_loop() when appropriate.
3590 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3592 (-mblock-compare-inline-loop-limit): New option.
3594 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3597 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3598 Reverse order of second and third operands in first alternative.
3599 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3600 of first and second elements in UNSPEC_VPERMR vector.
3601 (altivec_expand_vec_perm_le): Likewise.
3603 2017-01-08 Jeff Law <law@redhat.com>
3605 PR rtl-optimizatin/81308
3606 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3607 (process_switch): If group_case_labels makes a change, then set
3609 (pass_convert_switch::execute): If a switch is converted, then
3610 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3612 PR rtl-optimization/81308
3613 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3616 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3618 PR target/83663 - Revert r255946
3619 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3620 generation for cases where splatting a value is not useful.
3621 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3622 across a vec_duplicate and a paradoxical subreg forming a vector
3623 mode to a vec_concat.
3625 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3627 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3628 -march=armv8.3-a variants.
3629 * config/arm/t-multilib: Likewise.
3630 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3632 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3634 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3636 (cceq_ior_compare_complement): Give it a name so I can use it, and
3637 change boolean_or_operator predicate to boolean_operator so it can
3638 be used to generate a crand.
3639 (eqne): New code iterator.
3640 (bd/bd_neg): New code_attrs.
3641 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3642 a single define_insn.
3643 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3644 decrement (bdnzt/bdnzf/bdzt/bdzf).
3645 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3646 with the new names of the branch decrement patterns, and added the
3647 names of the branch decrement conditional patterns.
3649 2018-01-08 Richard Biener <rguenther@suse.de>
3651 PR tree-optimization/83563
3652 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3655 2018-01-08 Richard Biener <rguenther@suse.de>
3658 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3660 2018-01-08 Richard Biener <rguenther@suse.de>
3662 PR tree-optimization/83685
3663 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3664 references to abnormals.
3666 2018-01-08 Richard Biener <rguenther@suse.de>
3669 * dwarf2out.c (output_indirect_strings): Handle empty
3670 skeleton_debug_str_hash.
3671 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3673 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3675 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3676 (emit_store_direct): Likewise.
3677 (arc_trampoline_adjust_address): Likewise.
3678 (arc_asm_trampoline_template): New function.
3679 (arc_initialize_trampoline): Use asm_trampoline_template.
3680 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3681 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3682 * config/arc/arc.md (flush_icache): Delete pattern.
3684 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3686 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3687 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3690 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3693 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3694 by not USED_FOR_TARGET.
3695 (make_pass_resolve_sw_modes): Likewise.
3697 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3699 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3702 2018-01-08 Richard Biener <rguenther@suse.de>
3705 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3707 2018-01-08 Richard Biener <rguenther@suse.de>
3710 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3712 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3715 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3716 basic blocks with a small number of successors.
3717 (convert_control_dep_chain_into_preds): Improve handling of
3719 (dump_predicates): Split apart into...
3720 (dump_pred_chain): ...here...
3721 (dump_pred_info): ...and here.
3722 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3723 (can_chain_union_be_invalidated_p): Improve check for invalidation
3725 (uninit_uses_cannot_happen): Avoid unnecessary if
3726 convert_control_dep_chain_into_preds yielded nothing.
3728 2018-01-06 Martin Sebor <msebor@redhat.com>
3730 PR tree-optimization/83640
3731 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3732 subtracting negative offset from size.
3733 (builtin_access::overlap): Adjust offset bounds of the access to fall
3734 within the size of the object if possible.
3736 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3738 PR rtl-optimization/83699
3739 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3740 extract_bit_field_as_subreg to cases in which the extracted
3741 value is also a vector.
3743 * lra-constraints.c (process_alt_operands): Test for the equivalence
3744 substitutions when detecting a possible reload cycle.
3746 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3749 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3750 by default if flag_selective_schedling{,2}. Formatting fixes.
3752 PR rtl-optimization/83682
3753 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
3754 if it has non-VECTOR_MODE element mode.
3755 (vec_duplicate_p): Likewise.
3758 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
3759 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
3761 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3764 * config/i386/i386-builtin.def
3765 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
3766 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
3767 Require also OPTION_MASK_ISA_AVX512F in addition to
3768 OPTION_MASK_ISA_GFNI.
3769 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
3770 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
3771 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
3772 to OPTION_MASK_ISA_GFNI.
3773 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
3774 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
3775 OPTION_MASK_ISA_AVX512BW.
3776 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
3777 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
3778 addition to OPTION_MASK_ISA_GFNI.
3779 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
3780 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
3781 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
3782 to OPTION_MASK_ISA_GFNI.
3783 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
3784 a requirement for all ISAs rather than any of them with a few
3786 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
3788 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
3789 bitmasks to be enabled with 3 exceptions, instead of requiring any
3790 enabled ISA with lots of exceptions.
3791 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
3792 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
3793 Change avx512bw in isa attribute to avx512f.
3794 * config/i386/sgxintrin.h: Add license boilerplate.
3795 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
3796 to __AVX512F__ and __AVX512VL to __AVX512VL__.
3797 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
3798 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
3800 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
3801 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
3802 temporarily sse2 rather than sse if not enabled already.
3805 * config/i386/sse.md (VI248_VLBW): Rename to ...
3806 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
3807 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
3808 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
3809 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
3810 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
3811 mode iterator instead of VI248_VLBW.
3813 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
3815 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
3816 (record_modified): Skip clobbers; add debug output.
3817 (param_change_prob): Use sreal frequencies.
3819 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3821 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
3822 punt for user-aligned variables.
3824 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3826 * tree-chrec.c (chrec_contains_symbols): Return true for
3829 2018-01-05 Sudakshina Das <sudi.das@arm.com>
3832 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
3833 of (x|y) == x for BICS pattern.
3835 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3837 PR tree-optimization/83605
3838 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
3839 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
3842 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
3844 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
3845 * config/epiphany/rtems.h: New file.
3847 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3848 Uros Bizjak <ubizjak@gmail.com>
3851 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
3852 QIreg_operand instead of register_operand predicate.
3853 * config/i386/i386.c (ix86_rop_should_change_byte_p,
3854 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
3855 comments instead of -fmitigate[-_]rop.
3857 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3860 * cgraphunit.c (symbol_table::compile): Switch to text_section
3861 before calling assembly_start debug hook.
3862 * run-rtl-passes.c (run_rtl_passes): Likewise.
3865 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3867 * tree-vrp.c (extract_range_from_binary_expr_1): Check
3868 range_int_cst_p rather than !symbolic_range_p before calling
3869 extract_range_from_multiplicative_op_1.
3871 2017-01-04 Jeff Law <law@redhat.com>
3873 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
3874 redundant test in assertion.
3876 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3878 * doc/rtl.texi: Document machine_mode wrapper classes.
3880 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3882 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
3885 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3887 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
3888 the VEC_PERM_EXPR fold to fail.
3890 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3893 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
3894 to switched_sections.
3896 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3899 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
3902 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
3905 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
3906 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
3908 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3911 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
3912 is BLKmode and bitpos not zero or mode change is needed.
3914 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3917 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
3920 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
3923 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
3924 instead of MULT rtx. Update all corresponding splitters.
3926 (*ssub<modesuffix>): Ditto.
3928 (*cmp_sadd_di): Update split patterns.
3929 (*cmp_sadd_si): Ditto.
3930 (*cmp_sadd_sidi): Ditto.
3931 (*cmp_ssub_di): Ditto.
3932 (*cmp_ssub_si): Ditto.
3933 (*cmp_ssub_sidi): Ditto.
3934 * config/alpha/predicates.md (const23_operand): New predicate.
3935 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
3936 Look for ASHIFT, not MULT inner operand.
3937 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
3939 2018-01-04 Martin Liska <mliska@suse.cz>
3941 PR gcov-profile/83669
3942 * gcov.c (output_intermediate_file): Add version to intermediate
3944 * doc/gcov.texi: Document new field 'version' in intermediate
3945 file format. Fix location of '-k' option of gcov command.
3947 2018-01-04 Martin Liska <mliska@suse.cz>
3950 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
3952 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3954 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
3956 2018-01-03 Martin Sebor <msebor@redhat.com>
3958 PR tree-optimization/83655
3959 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
3960 checking calls with invalid arguments.
3962 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3964 * tree-vect-stmts.c (vect_get_store_rhs): New function.
3965 (vectorizable_mask_load_store): Delete.
3966 (vectorizable_call): Return false for masked loads and stores.
3967 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
3968 instead of gimple_assign_rhs1.
3969 (vectorizable_load): Handle IFN_MASK_LOAD.
3970 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
3972 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3974 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
3976 (vectorizable_mask_load_store): ...here.
3977 (vectorizable_load): ...and here.
3979 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3981 * tree-vect-stmts.c (vect_build_all_ones_mask)
3982 (vect_build_zero_merge_argument): New functions, split out from...
3983 (vectorizable_load): ...here.
3985 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3987 * tree-vect-stmts.c (vect_check_store_rhs): New function,
3989 (vectorizable_mask_load_store): ...here.
3990 (vectorizable_store): ...and here.
3992 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3994 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
3996 (vectorizable_mask_load_store): ...here.
3998 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4000 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4001 (vect_model_store_cost): Take a vec_load_store_type instead of a
4003 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4004 (vect_model_store_cost): Take a vec_load_store_type instead of a
4006 (vectorizable_mask_load_store): Update accordingly.
4007 (vectorizable_store): Likewise.
4008 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4010 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4012 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4013 IFN_MASK_LOAD calls here rather than...
4014 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4016 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4017 Alan Hayward <alan.hayward@arm.com>
4018 David Sherwood <david.sherwood@arm.com>
4020 * expmed.c (extract_bit_field_1): For vector extracts,
4021 fall back to extract_bit_field_as_subreg if vec_extract
4024 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4025 Alan Hayward <alan.hayward@arm.com>
4026 David Sherwood <david.sherwood@arm.com>
4028 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4029 they are variable or constant sized.
4030 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4031 slots for constant-sized data.
4033 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4034 Alan Hayward <alan.hayward@arm.com>
4035 David Sherwood <david.sherwood@arm.com>
4037 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4038 handling COND_EXPRs with boolean comparisons, try to find a better
4039 basis for the mask type than the boolean itself.
4041 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4043 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4044 is calculated and how it can be overridden.
4045 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4046 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4048 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4051 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4052 Alan Hayward <alan.hayward@arm.com>
4053 David Sherwood <david.sherwood@arm.com>
4055 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4056 Remove the mode argument.
4057 (aarch64_simd_valid_immediate): Remove the mode and inverse
4059 * config/aarch64/iterators.md (bitsize): New iterator.
4060 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4061 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4062 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4063 aarch64_simd_valid_immediate.
4064 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4065 (aarch64_reg_or_bic_imm): Likewise.
4066 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4067 with an insn_type enum and msl with a modifier_type enum.
4068 Replace element_width with a scalar_mode. Change the shift
4069 to unsigned int. Add constructors for scalar_float_mode and
4070 scalar_int_mode elements.
4071 (aarch64_vect_float_const_representable_p): Delete.
4072 (aarch64_can_const_movi_rtx_p)
4073 (aarch64_simd_scalar_immediate_valid_for_move)
4074 (aarch64_simd_make_constant): Update call to
4075 aarch64_simd_valid_immediate.
4076 (aarch64_advsimd_valid_immediate_hs): New function.
4077 (aarch64_advsimd_valid_immediate): Likewise.
4078 (aarch64_simd_valid_immediate): Remove mode and inverse
4079 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4080 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4081 and aarch64_float_const_representable_p on the result.
4082 (aarch64_output_simd_mov_immediate): Remove mode argument.
4083 Update call to aarch64_simd_valid_immediate and use of
4084 simd_immediate_info.
4085 (aarch64_output_scalar_simd_mov_immediate): Update call
4088 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4089 Alan Hayward <alan.hayward@arm.com>
4090 David Sherwood <david.sherwood@arm.com>
4092 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4093 (mode_nunits): Likewise CONST_MODE_NUNITS.
4094 * machmode.def (ADJUST_NUNITS): Document.
4095 * genmodes.c (mode_data::need_nunits_adj): New field.
4096 (blank_mode): Update accordingly.
4097 (adj_nunits): New variable.
4098 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4100 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4101 listed in adj_nunits.
4102 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4103 listed in adj_nunits. Don't emit case statements for such modes.
4104 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4105 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4106 nothing if adj_nunits is nonnull.
4107 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4108 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4109 (emit_mode_fbit): Update use of print_maybe_const_decl.
4110 (emit_move_size): Likewise. Treat the array as non-const
4112 (emit_mode_adjustments): Handle adj_nunits.
4114 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4116 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4117 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4118 (VECTOR_MODES): Use it.
4119 (make_vector_modes): Take the prefix as an argument.
4121 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4122 Alan Hayward <alan.hayward@arm.com>
4123 David Sherwood <david.sherwood@arm.com>
4125 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4126 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4127 for MODE_VECTOR_BOOL.
4128 * machmode.def (VECTOR_BOOL_MODE): Document.
4129 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4130 (make_vector_bool_mode): New function.
4131 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4133 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4134 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4136 * stor-layout.c (int_mode_for_mode): Likewise.
4137 * tree.c (build_vector_type_for_mode): Likewise.
4138 * varasm.c (output_constant_pool_2): Likewise.
4139 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4140 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4141 for MODE_VECTOR_BOOL.
4142 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4143 of mode class checks.
4144 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4145 instead of a list of mode class checks.
4146 (expand_vector_scalar_condition): Likewise.
4147 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4149 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4150 Alan Hayward <alan.hayward@arm.com>
4151 David Sherwood <david.sherwood@arm.com>
4153 * machmode.h (mode_size): Change from unsigned short to
4155 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4156 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4157 or if measurement_type is not polynomial.
4158 (fixed_size_mode::includes_p): Check for constant-sized modes.
4159 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4160 return a poly_uint16 rather than an unsigned short.
4161 (emit_mode_size): Change the type of mode_size from unsigned short
4162 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4163 (emit_mode_adjustments): Cope with polynomial vector sizes.
4164 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4166 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4168 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4169 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4170 * caller-save.c (setup_save_areas): Likewise.
4171 (replace_reg_with_saved_mem): Likewise.
4172 * calls.c (emit_library_call_value_1): Likewise.
4173 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4174 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4175 (gen_lowpart_for_combine): Likewise.
4176 * convert.c (convert_to_integer_1): Likewise.
4177 * cse.c (equiv_constant, cse_insn): Likewise.
4178 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4179 (cselib_subst_to_values): Likewise.
4180 * dce.c (word_dce_process_block): Likewise.
4181 * df-problems.c (df_word_lr_mark_ref): Likewise.
4182 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4183 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4184 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4185 (rtl_for_decl_location): Likewise.
4186 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4187 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4188 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4189 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4190 (expand_expr_real_1): Likewise.
4191 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4192 (pad_below): Likewise.
4193 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4194 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4195 * ira.c (get_subreg_tracking_sizes): Likewise.
4196 * ira-build.c (ira_create_allocno_objects): Likewise.
4197 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4198 (ira_sort_regnos_for_alter_reg): Likewise.
4199 * ira-costs.c (record_operand_costs): Likewise.
4200 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4201 (resolve_simple_move): Likewise.
4202 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4203 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4204 (lra_constraints): Likewise.
4205 (CONST_POOL_OK_P): Reject variable-sized modes.
4206 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4207 (add_pseudo_to_slot, lra_spill): Likewise.
4208 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4209 * optabs-query.c (get_best_extraction_insn): Likewise.
4210 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4211 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4212 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4213 * recog.c (offsettable_address_addr_space_p): Likewise.
4214 * regcprop.c (maybe_mode_change): Likewise.
4215 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4216 * regrename.c (build_def_use): Likewise.
4217 * regstat.c (dump_reg_info): Likewise.
4218 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4219 (find_reloads, find_reloads_subreg_address): Likewise.
4220 * reload1.c (eliminate_regs_1): Likewise.
4221 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4222 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4223 (simplify_binary_operation_1, simplify_subreg): Likewise.
4224 * targhooks.c (default_function_arg_padding): Likewise.
4225 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4226 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4227 (verify_gimple_assign_ternary): Likewise.
4228 * tree-inline.c (estimate_move_cost): Likewise.
4229 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4230 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4231 (get_address_cost_ainc): Likewise.
4232 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4233 (vect_supportable_dr_alignment): Likewise.
4234 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4235 (vectorizable_reduction): Likewise.
4236 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4237 (vectorizable_operation, vectorizable_load): Likewise.
4238 * tree.c (build_same_sized_truth_vector_type): Likewise.
4239 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4240 * var-tracking.c (emit_note_insn_var_location): Likewise.
4241 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4242 (ADDR_VEC_ALIGN): Likewise.
4244 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4245 Alan Hayward <alan.hayward@arm.com>
4246 David Sherwood <david.sherwood@arm.com>
4248 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4250 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4251 or if measurement_type is polynomial.
4252 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4253 * combine.c (make_extraction): Likewise.
4254 * dse.c (find_shift_sequence): Likewise.
4255 * dwarf2out.c (mem_loc_descriptor): Likewise.
4256 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4257 (extract_bit_field, extract_low_bits): Likewise.
4258 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4259 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4260 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4261 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4262 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4263 * reload.c (find_reloads): Likewise.
4264 * reload1.c (alter_reg): Likewise.
4265 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4266 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4267 * tree-if-conv.c (predicate_mem_writes): Likewise.
4268 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4269 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4270 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4271 * valtrack.c (dead_debug_insert_temp): Likewise.
4272 * varasm.c (mergeable_constant_section): Likewise.
4273 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4275 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4276 Alan Hayward <alan.hayward@arm.com>
4277 David Sherwood <david.sherwood@arm.com>
4279 * expr.c (expand_assignment): Cope with polynomial mode sizes
4280 when assigning to a CONCAT.
4282 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4283 Alan Hayward <alan.hayward@arm.com>
4284 David Sherwood <david.sherwood@arm.com>
4286 * machmode.h (mode_precision): Change from unsigned short to
4288 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4290 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4291 or if measurement_type is not polynomial.
4292 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4293 in which the mode is already known to be a scalar_int_mode.
4294 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4295 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4297 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4298 for GET_MODE_PRECISION.
4299 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4300 for GET_MODE_PRECISION.
4301 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4303 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4304 (expand_field_assignment, make_extraction): Likewise.
4305 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4306 (get_last_value): Likewise.
4307 * convert.c (convert_to_integer_1): Likewise.
4308 * cse.c (cse_insn): Likewise.
4309 * expr.c (expand_expr_real_1): Likewise.
4310 * lra-constraints.c (simplify_operand_subreg): Likewise.
4311 * optabs-query.c (can_atomic_load_p): Likewise.
4312 * optabs.c (expand_atomic_load): Likewise.
4313 (expand_atomic_store): Likewise.
4314 * ree.c (combine_reaching_defs): Likewise.
4315 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4316 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4317 * tree.h (type_has_mode_precision_p): Likewise.
4318 * ubsan.c (instrument_si_overflow): Likewise.
4320 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4321 Alan Hayward <alan.hayward@arm.com>
4322 David Sherwood <david.sherwood@arm.com>
4324 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4325 polynomial numbers of units.
4326 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4327 (valid_vector_subparts_p): New function.
4328 (build_vector_type): Remove temporary shim and take the number
4329 of units as a poly_uint64 rather than an int.
4330 (build_opaque_vector_type): Take the number of units as a
4331 poly_uint64 rather than an int.
4332 * tree.c (build_vector_from_ctor): Handle polynomial
4333 TYPE_VECTOR_SUBPARTS.
4334 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4335 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4336 (build_vector_from_val): If the number of units is variable,
4337 use build_vec_duplicate_cst for constant operands and
4338 VEC_DUPLICATE_EXPR otherwise.
4339 (make_vector_type): Remove temporary is_constant ().
4340 (build_vector_type, build_opaque_vector_type): Take the number of
4341 units as a poly_uint64 rather than an int.
4342 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4344 * cfgexpand.c (expand_debug_expr): Likewise.
4345 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4346 (store_constructor, expand_expr_real_1): Likewise.
4347 (const_scalar_mask_from_tree): Likewise.
4348 * fold-const-call.c (fold_const_reduction): Likewise.
4349 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4350 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4351 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4352 (fold_relational_const): Likewise.
4353 (native_interpret_vector): Likewise. Change the size from an
4354 int to an unsigned int.
4355 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4356 TYPE_VECTOR_SUBPARTS.
4357 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4358 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4359 duplicating a non-constant operand into a variable-length vector.
4360 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4361 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4362 * ipa-icf.c (sem_variable::equals): Likewise.
4363 * match.pd: Likewise.
4364 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4365 * print-tree.c (print_node): Likewise.
4366 * stor-layout.c (layout_type): Likewise.
4367 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4368 * tree-cfg.c (verify_gimple_comparison): Likewise.
4369 (verify_gimple_assign_binary): Likewise.
4370 (verify_gimple_assign_ternary): Likewise.
4371 (verify_gimple_assign_single): Likewise.
4372 * tree-pretty-print.c (dump_generic_node): Likewise.
4373 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4374 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4375 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4376 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4377 (vect_shift_permute_load_chain): Likewise.
4378 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4379 (expand_vector_condition, optimize_vector_constructor): Likewise.
4380 (lower_vec_perm, get_compute_type): Likewise.
4381 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4382 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4383 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4384 (vect_recog_mask_conversion_pattern): Likewise.
4385 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4386 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4387 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4388 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4389 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4390 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4391 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4392 (supportable_widening_operation): Likewise.
4393 (supportable_narrowing_operation): Likewise.
4394 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4396 * varasm.c (output_constant): Likewise.
4398 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4399 Alan Hayward <alan.hayward@arm.com>
4400 David Sherwood <david.sherwood@arm.com>
4402 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4403 so that both the length == 3 and length != 3 cases set up their
4404 own permute vectors. Add comments explaining why we know the
4405 number of elements is constant.
4406 (vect_permute_load_chain): Likewise.
4408 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4409 Alan Hayward <alan.hayward@arm.com>
4410 David Sherwood <david.sherwood@arm.com>
4412 * machmode.h (mode_nunits): Change from unsigned char to
4414 (ONLY_FIXED_SIZE_MODES): New macro.
4415 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4416 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4417 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4419 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4420 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4421 or if measurement_type is not polynomial.
4422 * genmodes.c (ZERO_COEFFS): New macro.
4423 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4425 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4426 Use ZERO_COEFFS when emitting initializers.
4427 * data-streamer.h (bp_pack_poly_value): New function.
4428 (bp_unpack_poly_value): Likewise.
4429 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4430 for GET_MODE_NUNITS.
4431 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4432 for GET_MODE_NUNITS.
4433 * tree.c (make_vector_type): Remove temporary shim and make
4434 the real function take the number of units as a poly_uint64
4436 (build_vector_type_for_mode): Handle polynomial nunits.
4437 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4438 * emit-rtl.c (const_vec_series_p_1): Likewise.
4439 (gen_rtx_CONST_VECTOR): Likewise.
4440 * fold-const.c (test_vec_duplicate_folding): Likewise.
4441 * genrecog.c (validate_pattern): Likewise.
4442 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4443 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4444 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4445 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4446 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4447 * rtlanal.c (subreg_get_info): Likewise.
4448 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4449 (vect_grouped_load_supported): Likewise.
4450 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4451 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4452 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4453 (simplify_const_unary_operation, simplify_binary_operation_1)
4454 (simplify_const_binary_operation, simplify_ternary_operation)
4455 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4456 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4457 instead of CONST_VECTOR_NUNITS.
4458 * varasm.c (output_constant_pool_2): Likewise.
4459 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4460 explicit-encoded elements in the XVEC for variable-length vectors.
4462 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4464 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4466 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4467 Alan Hayward <alan.hayward@arm.com>
4468 David Sherwood <david.sherwood@arm.com>
4470 * coretypes.h (fixed_size_mode): Declare.
4471 (fixed_size_mode_pod): New typedef.
4472 * builtins.h (target_builtins::x_apply_args_mode)
4473 (target_builtins::x_apply_result_mode): Change type to
4474 fixed_size_mode_pod.
4475 * builtins.c (apply_args_size, apply_result_size, result_vector)
4476 (expand_builtin_apply_args_1, expand_builtin_apply)
4477 (expand_builtin_return): Update accordingly.
4479 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4481 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4482 * cselib.c (cselib_hash_rtx): Likewise.
4483 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4484 CONST_VECTOR encoding.
4486 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4487 Jeff Law <law@redhat.com>
4490 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4491 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4492 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4493 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4496 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4497 explicitly probe *sp in a noreturn function if there were any callee
4498 register saves or frame pointer is needed.
4500 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4503 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4504 BLKmode for ternary, binary or unary expressions.
4507 * var-tracking.c (delete_vta_debug_insn): New inline function.
4508 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4509 insns from get_insns () to NULL instead of each bb separately.
4510 Use delete_vta_debug_insn. No longer static.
4511 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4512 delete_vta_debug_insns callers.
4513 * rtl.h (delete_vta_debug_insns): Declare.
4514 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4515 instead of variable_tracking_main.
4517 2018-01-03 Martin Sebor <msebor@redhat.com>
4519 PR tree-optimization/83603
4520 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4521 arguments past the endof the argument list in functions declared
4522 without a prototype.
4523 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4524 Avoid checking when arguments are null.
4526 2018-01-03 Martin Sebor <msebor@redhat.com>
4529 * doc/extend.texi (attribute const): Fix a typo.
4530 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4531 issuing -Wsuggest-attribute for void functions.
4533 2018-01-03 Martin Sebor <msebor@redhat.com>
4535 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4536 offset_int::from instead of wide_int::to_shwi.
4537 (maybe_diag_overlap): Remove assertion.
4538 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4539 * gimple-ssa-sprintf.c (format_directive): Same.
4540 (parse_directive): Same.
4541 (sprintf_dom_walker::compute_format_length): Same.
4542 (try_substitute_return_value): Same.
4544 2017-01-03 Jeff Law <law@redhat.com>
4547 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4548 non-constant residual for zero at runtime and avoid probing in
4549 that case. Reorganize code for trailing problem to mirror handling
4552 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4554 PR tree-optimization/83501
4555 * tree-ssa-strlen.c (get_string_cst): New.
4556 (handle_char_store): Call get_string_cst.
4558 2018-01-03 Martin Liska <mliska@suse.cz>
4560 PR tree-optimization/83593
4561 * tree-ssa-strlen.c: Include tree-cfg.h.
4562 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4563 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4564 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4566 (strlen_dom_walker::before_dom_children): Call
4567 gimple_purge_dead_eh_edges. Dump tranformation with details
4569 (strlen_dom_walker::before_dom_children): Update call by adding
4570 new argument cleanup_eh.
4571 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4573 2018-01-03 Martin Liska <mliska@suse.cz>
4576 * cif-code.def (VARIADIC_THUNK): New enum value.
4577 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4580 2018-01-03 Jan Beulich <jbeulich@suse.com>
4582 * sse.md (mov<mode>_internal): Tighten condition for when to use
4583 vmovdqu<ssescalarsize> for TI and OI modes.
4585 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4587 Update copyright years.
4589 2018-01-03 Martin Liska <mliska@suse.cz>
4592 * ipa-visibility.c (function_and_variable_visibility): Skip
4593 functions with noipa attribure.
4595 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4597 * gcc.c (process_command): Update copyright notice dates.
4598 * gcov-dump.c (print_version): Ditto.
4599 * gcov.c (print_version): Ditto.
4600 * gcov-tool.c (print_version): Ditto.
4601 * gengtype.c (create_file): Ditto.
4602 * doc/cpp.texi: Bump @copying's copyright year.
4603 * doc/cppinternals.texi: Ditto.
4604 * doc/gcc.texi: Ditto.
4605 * doc/gccint.texi: Ditto.
4606 * doc/gcov.texi: Ditto.
4607 * doc/install.texi: Ditto.
4608 * doc/invoke.texi: Ditto.
4610 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4612 * vector-builder.h (vector_builder::m_full_nelts): Change from
4613 unsigned int to poly_uint64.
4614 (vector_builder::full_nelts): Update prototype accordingly.
4615 (vector_builder::new_vector): Likewise.
4616 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4617 (vector_builder::operator ==): Likewise.
4618 (vector_builder::finalize): Likewise.
4619 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4620 Take the number of elements as a poly_uint64 rather than an
4622 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4623 from unsigned int to poly_uint64.
4624 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4625 (vec_perm_indices::new_vector): Likewise.
4626 (vec_perm_indices::length): Likewise.
4627 (vec_perm_indices::nelts_per_input): Likewise.
4628 (vec_perm_indices::input_nelts): Likewise.
4629 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4630 number of elements per input as a poly_uint64 rather than an
4631 unsigned int. Use the original encoding for variable-length
4632 vectors, rather than clamping each individual element.
4633 For the second and subsequent elements in each pattern,
4634 clamp the step and base before clamping their sum.
4635 (vec_perm_indices::series_p): Handle polynomial element counts.
4636 (vec_perm_indices::all_in_range_p): Likewise.
4637 (vec_perm_indices_to_tree): Likewise.
4638 (vec_perm_indices_to_rtx): Likewise.
4639 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4640 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4641 (tree_vector_builder::new_binary_operation): Handle polynomial
4642 element counts. Return false if we need to know the number
4643 of elements at compile time.
4644 * fold-const.c (fold_vec_perm): Punt if the number of elements
4645 isn't known at compile time.
4647 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4649 * vec-perm-indices.h (vec_perm_builder): Change element type
4650 from HOST_WIDE_INT to poly_int64.
4651 (vec_perm_indices::element_type): Update accordingly.
4652 (vec_perm_indices::clamp): Handle polynomial element_types.
4653 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4654 (vec_perm_indices::all_in_range_p): Likewise.
4655 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4657 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4658 polynomial vec_perm_indices element types.
4659 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4660 * fold-const.c (fold_vec_perm): Likewise.
4661 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4662 * tree-vect-generic.c (lower_vec_perm): Likewise.
4663 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4664 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4665 element type to HOST_WIDE_INT.
4667 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4668 Alan Hayward <alan.hayward@arm.com>
4669 David Sherwood <david.sherwood@arm.com>
4671 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4672 rather than an int. Use plus_constant.
4673 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4674 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4676 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4677 Alan Hayward <alan.hayward@arm.com>
4678 David Sherwood <david.sherwood@arm.com>
4680 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4681 a HOST_WIDE_INT to a poly_int64.
4683 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4684 Alan Hayward <alan.hayward@arm.com>
4685 David Sherwood <david.sherwood@arm.com>
4687 * calls.c (load_register_parameters): Cope with polynomial
4688 mode sizes. Require a constant size for BLKmode parameters
4689 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4690 forces a parameter to be padded at the lsb end in order to
4691 fill a complete number of words, require the parameter size
4692 to be ordered wrt UNITS_PER_WORD.
4694 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4695 Alan Hayward <alan.hayward@arm.com>
4696 David Sherwood <david.sherwood@arm.com>
4698 * reload1.c (spill_stack_slot_width): Change element type
4699 from unsigned int to poly_uint64_pod.
4700 (alter_reg): Treat mode sizes as polynomial.
4702 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4703 Alan Hayward <alan.hayward@arm.com>
4704 David Sherwood <david.sherwood@arm.com>
4706 * reload.c (complex_word_subreg_p): New function.
4707 (reload_inner_reg_of_subreg, push_reload): Use it.
4709 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4710 Alan Hayward <alan.hayward@arm.com>
4711 David Sherwood <david.sherwood@arm.com>
4713 * lra-constraints.c (process_alt_operands): Reject matched
4714 operands whose sizes aren't ordered.
4715 (match_reload): Refer to this check here.
4717 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4718 Alan Hayward <alan.hayward@arm.com>
4719 David Sherwood <david.sherwood@arm.com>
4721 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4722 that the mode size is in the set {1, 2, 4, 8, 16}.
4724 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4725 Alan Hayward <alan.hayward@arm.com>
4726 David Sherwood <david.sherwood@arm.com>
4728 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4729 Use plus_constant instead of gen_rtx_PLUS.
4731 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4732 Alan Hayward <alan.hayward@arm.com>
4733 David Sherwood <david.sherwood@arm.com>
4735 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4736 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4737 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4738 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4739 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4740 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4741 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4742 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4743 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4744 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4746 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4747 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4748 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4749 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4750 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4751 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4752 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4753 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
4754 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
4755 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
4757 * expr.c (emit_move_resolve_push): Treat the input and result
4758 of PUSH_ROUNDING as a poly_int64.
4759 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
4760 (emit_push_insn): Likewise.
4761 * lra-eliminations.c (mark_not_eliminable): Likewise.
4762 * recog.c (push_operand): Likewise.
4763 * reload1.c (elimination_effects): Likewise.
4764 * rtlanal.c (nonzero_bits1): Likewise.
4765 * calls.c (store_one_arg): Likewise. Require the padding to be
4766 known at compile time.
4768 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4769 Alan Hayward <alan.hayward@arm.com>
4770 David Sherwood <david.sherwood@arm.com>
4772 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
4773 Use plus_constant instead of gen_rtx_PLUS.
4775 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4776 Alan Hayward <alan.hayward@arm.com>
4777 David Sherwood <david.sherwood@arm.com>
4779 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
4782 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4783 Alan Hayward <alan.hayward@arm.com>
4784 David Sherwood <david.sherwood@arm.com>
4786 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
4787 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
4788 via stack temporaries. Treat the mode size as polynomial too.
4790 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4791 Alan Hayward <alan.hayward@arm.com>
4792 David Sherwood <david.sherwood@arm.com>
4794 * expr.c (expand_expr_real_2): When handling conversions involving
4795 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
4796 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
4797 as a poly_uint64 too.
4799 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4800 Alan Hayward <alan.hayward@arm.com>
4801 David Sherwood <david.sherwood@arm.com>
4803 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
4805 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4806 Alan Hayward <alan.hayward@arm.com>
4807 David Sherwood <david.sherwood@arm.com>
4809 * combine.c (can_change_dest_mode): Handle polynomial
4810 REGMODE_NATURAL_SIZE.
4811 * expmed.c (store_bit_field_1): Likewise.
4812 * expr.c (store_constructor): Likewise.
4813 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
4814 and polynomial REGMODE_NATURAL_SIZE.
4815 (gen_lowpart_common): Likewise.
4816 * reginfo.c (record_subregs_of_mode): Likewise.
4817 * rtlanal.c (read_modify_subreg_p): Likewise.
4819 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4820 Alan Hayward <alan.hayward@arm.com>
4821 David Sherwood <david.sherwood@arm.com>
4823 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
4824 numbers of elements.
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 * match.pd: Cope with polynomial numbers of vector elements.
4832 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4833 Alan Hayward <alan.hayward@arm.com>
4834 David Sherwood <david.sherwood@arm.com>
4836 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
4837 in a POINTER_PLUS_EXPR.
4839 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4840 Alan Hayward <alan.hayward@arm.com>
4841 David Sherwood <david.sherwood@arm.com>
4843 * omp-simd-clone.c (simd_clone_subparts): New function.
4844 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
4845 (ipa_simd_modify_function_body): Likewise.
4847 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4848 Alan Hayward <alan.hayward@arm.com>
4849 David Sherwood <david.sherwood@arm.com>
4851 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
4852 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
4853 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
4854 (expand_vector_condition, vector_element): Likewise.
4855 (subparts_gt): New function.
4856 (get_compute_type): Use subparts_gt.
4857 (count_type_subparts): Delete.
4858 (expand_vector_operations_1): Use subparts_gt instead of
4859 count_type_subparts.
4861 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4862 Alan Hayward <alan.hayward@arm.com>
4863 David Sherwood <david.sherwood@arm.com>
4865 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
4866 (vect_compile_time_alias): ...this new function. Do the calculation
4867 on poly_ints rather than trees.
4868 (vect_prune_runtime_alias_test_list): Update call accordingly.
4870 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4871 Alan Hayward <alan.hayward@arm.com>
4872 David Sherwood <david.sherwood@arm.com>
4874 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
4876 (vect_schedule_slp_instance): Likewise.
4878 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4879 Alan Hayward <alan.hayward@arm.com>
4880 David Sherwood <david.sherwood@arm.com>
4882 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
4883 constant and extern definitions for variable-length vectors.
4884 (vect_get_constant_vectors): Note that the number of units
4885 is known to be constant.
4887 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4888 Alan Hayward <alan.hayward@arm.com>
4889 David Sherwood <david.sherwood@arm.com>
4891 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
4892 of units as polynomial. Choose between WIDE and NARROW based
4895 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4896 Alan Hayward <alan.hayward@arm.com>
4897 David Sherwood <david.sherwood@arm.com>
4899 * tree-vect-stmts.c (simd_clone_subparts): New function.
4900 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
4902 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4903 Alan Hayward <alan.hayward@arm.com>
4904 David Sherwood <david.sherwood@arm.com>
4906 * tree-vect-stmts.c (vectorizable_call): Treat the number of
4907 vectors as polynomial. Use build_index_vector for
4910 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4911 Alan Hayward <alan.hayward@arm.com>
4912 David Sherwood <david.sherwood@arm.com>
4914 * tree-vect-stmts.c (get_load_store_type): Treat the number of
4915 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
4916 for variable-length vectors.
4917 (vectorizable_mask_load_store): Treat the number of units as
4918 polynomial, asserting that it is constant if the condition has
4919 already been enforced.
4920 (vectorizable_store, vectorizable_load): Likewise.
4922 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4923 Alan Hayward <alan.hayward@arm.com>
4924 David Sherwood <david.sherwood@arm.com>
4926 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
4927 of units as polynomial. Punt if we can't tell at compile time
4928 which vector contains the final result.
4930 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4931 Alan Hayward <alan.hayward@arm.com>
4932 David Sherwood <david.sherwood@arm.com>
4934 * tree-vect-loop.c (vectorizable_induction): Treat the number
4935 of units as polynomial. Punt on SLP inductions. Use an integer
4936 VEC_SERIES_EXPR for variable-length integer reductions. Use a
4937 cast of such a series for variable-length floating-point
4940 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4941 Alan Hayward <alan.hayward@arm.com>
4942 David Sherwood <david.sherwood@arm.com>
4944 * tree.h (build_index_vector): Declare.
4945 * tree.c (build_index_vector): New function.
4946 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
4947 of units as polynomial, forcibly converting it to a constant if
4948 vectorizable_reduction has already enforced the condition.
4949 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
4950 to create a {1,2,3,...} vector.
4951 (vectorizable_reduction): Treat the number of units as polynomial.
4952 Choose vectype_in based on the largest scalar element size rather
4953 than the smallest number of units. Enforce the restrictions
4956 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4957 Alan Hayward <alan.hayward@arm.com>
4958 David Sherwood <david.sherwood@arm.com>
4960 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
4961 number of units as polynomial.
4963 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4964 Alan Hayward <alan.hayward@arm.com>
4965 David Sherwood <david.sherwood@arm.com>
4967 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
4968 * target.def (autovectorize_vector_sizes): Return the vector sizes
4969 by pointer, using vector_sizes rather than a bitmask.
4970 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
4971 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
4972 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
4974 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
4975 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
4976 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
4977 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
4978 * omp-general.c (omp_max_vf): Likewise.
4979 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4980 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
4981 * tree-vect-loop.c (vect_analyze_loop): Likewise.
4982 * tree-vect-slp.c (vect_slp_bb): Likewise.
4983 * doc/tm.texi: Regenerate.
4984 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
4986 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
4987 the vector size as a poly_uint64 rather than an unsigned int.
4988 (current_vector_size): Change from an unsigned int to a poly_uint64.
4989 (get_vectype_for_scalar_type): Update accordingly.
4990 * tree.h (build_truth_vector_type): Take the size and number of
4991 units as a poly_uint64 rather than an unsigned int.
4992 (build_vector_type): Add a temporary overload that takes
4993 the number of units as a poly_uint64 rather than an unsigned int.
4994 * tree.c (make_vector_type): Likewise.
4995 (build_truth_vector_type): Take the number of units as a poly_uint64
4996 rather than an unsigned int.
4998 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4999 Alan Hayward <alan.hayward@arm.com>
5000 David Sherwood <david.sherwood@arm.com>
5002 * target.def (get_mask_mode): Take the number of units and length
5003 as poly_uint64s rather than unsigned ints.
5004 * targhooks.h (default_get_mask_mode): Update accordingly.
5005 * targhooks.c (default_get_mask_mode): Likewise.
5006 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5007 * doc/tm.texi: Regenerate.
5009 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5010 Alan Hayward <alan.hayward@arm.com>
5011 David Sherwood <david.sherwood@arm.com>
5013 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5014 * omp-general.c (omp_max_vf): Likewise.
5015 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5016 (expand_omp_simd): Handle polynomial safelen.
5017 * omp-low.c (omplow_simd_context): Add a default constructor.
5018 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5019 (lower_rec_simd_input_clauses): Update accordingly.
5020 (lower_rec_input_clauses): Likewise.
5022 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5023 Alan Hayward <alan.hayward@arm.com>
5024 David Sherwood <david.sherwood@arm.com>
5026 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5027 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5028 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5029 (vect_analyze_slp_cost): Likewise.
5030 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5031 (vect_model_load_cost): Likewise.
5033 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5034 Alan Hayward <alan.hayward@arm.com>
5035 David Sherwood <david.sherwood@arm.com>
5037 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5038 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5039 from an unsigned int * to a poly_uint64_pod *.
5040 (calculate_unrolling_factor): New function.
5041 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5043 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5044 Alan Hayward <alan.hayward@arm.com>
5045 David Sherwood <david.sherwood@arm.com>
5047 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5048 from an unsigned int to a poly_uint64.
5049 (_loop_vec_info::slp_unrolling_factor): Likewise.
5050 (_loop_vec_info::vectorization_factor): Change from an int
5052 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5053 (vect_get_num_vectors): New function.
5054 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5055 (vect_get_num_copies): Use vect_get_num_vectors.
5056 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5057 to an unsigned int *.
5058 (vect_analyze_data_refs): Change min_vf from an int * to a
5060 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5061 than an unsigned HOST_WIDE_INT.
5062 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5063 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5064 to an unsigned int *.
5065 (vect_analyze_data_ref_dependences): Likewise.
5066 (vect_compute_data_ref_alignment): Handle polynomial vf.
5067 (vect_enhance_data_refs_alignment): Likewise.
5068 (vect_prune_runtime_alias_test_list): Likewise.
5069 (vect_shift_permute_load_chain): Likewise.
5070 (vect_supportable_dr_alignment): Likewise.
5071 (dependence_distance_ge_vf): Take the vectorization factor as a
5072 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5073 (vect_analyze_data_refs): Change min_vf from an int * to a
5075 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5076 vfm1 as a poly_uint64 rather than an int. Make the same change
5077 for the returned bound_scalar.
5078 (vect_gen_vector_loop_niters): Handle polynomial vf.
5079 (vect_do_peeling): Likewise. Update call to
5080 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5081 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5083 * tree-vect-loop.c (vect_determine_vectorization_factor)
5084 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5085 (vect_get_known_peeling_cost): Likewise.
5086 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5087 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5088 (vect_transform_loop): Likewise. Use the lowest possible VF when
5089 updating the upper bounds of the loop.
5090 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5092 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5093 polynomial unroll factors.
5094 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5095 (vect_make_slp_decision): Likewise.
5096 (vect_supported_load_permutation_p): Likewise, and polynomial
5098 (vect_analyze_slp_cost): Handle polynomial vf.
5099 (vect_slp_analyze_node_operations): Likewise.
5100 (vect_slp_analyze_bb_1): Likewise.
5101 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5102 than an unsigned HOST_WIDE_INT.
5103 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5104 (vectorizable_load): Handle polynomial vf.
5105 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5107 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5109 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5110 Alan Hayward <alan.hayward@arm.com>
5111 David Sherwood <david.sherwood@arm.com>
5113 * match.pd: Handle bit operations involving three constants
5114 and try to fold one pair.
5116 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5118 * tree-vect-loop-manip.c: Include gimple-fold.h.
5119 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5120 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5121 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5122 Add a path that uses a step of VF instead of 1, but disable it
5124 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5125 and niters_no_overflow parameters. Update calls to
5126 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5127 Create a new SSA name if the latter choses to use a ste other
5128 than zero, and return it via niters_vector_mult_vf_var.
5129 * tree-vect-loop.c (vect_transform_loop): Update calls to
5130 vect_do_peeling, vect_gen_vector_loop_niters and
5131 slpeel_make_loop_iterate_ntimes.
5132 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5133 (vect_gen_vector_loop_niters): Update declarations after above changes.
5135 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5137 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5138 128-bit round to integer instructions.
5139 (ceil<mode>2): Likewise.
5140 (btrunc<mode>2): Likewise.
5141 (round<mode>2): Likewise.
5143 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5145 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5146 unaligned VSX load/store on P8/P9.
5147 (expand_block_clear): Allow the use of unaligned VSX
5148 load/store on P8/P9.
5150 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5152 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5154 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5155 swap associated with both a load and a store.
5157 2018-01-02 Andrew Waterman <andrew@sifive.com>
5159 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5160 * config/riscv/riscv.md (clear_cache): Use it.
5162 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5164 * web.c: Remove out-of-date comment.
5166 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5168 * expr.c (fixup_args_size_notes): Check that any existing
5169 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5170 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5171 (emit_single_push_insn): ...here.
5173 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5175 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5176 (const_vector_encoded_nelts): New function.
5177 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5178 (const_vector_int_elt, const_vector_elt): Declare.
5179 * emit-rtl.c (const_vector_int_elt_1): New function.
5180 (const_vector_elt): Likewise.
5181 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5182 of CONST_VECTOR_ELT.
5184 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5186 * expr.c: Include rtx-vector-builder.h.
5187 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5188 directly on the tree encoding.
5189 (const_vector_from_tree): Likewise.
5190 * optabs.c: Include rtx-vector-builder.h.
5191 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5192 sequence of "u" values.
5193 * vec-perm-indices.c: Include rtx-vector-builder.h.
5194 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5195 directly on the vec_perm_indices encoding.
5197 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5199 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5200 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5201 * rtx-vector-builder.h: New file.
5202 * rtx-vector-builder.c: Likewise.
5203 * rtl.h (rtx_def::u2): Add a const_vector field.
5204 (CONST_VECTOR_NPATTERNS): New macro.
5205 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5206 (CONST_VECTOR_DUPLICATE_P): Likewise.
5207 (CONST_VECTOR_STEPPED_P): Likewise.
5208 (CONST_VECTOR_ENCODED_ELT): Likewise.
5209 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5210 (unwrap_const_vec_duplicate): Likewise.
5211 (const_vec_series_p): Check for a non-duplicated vector encoding.
5212 Say that the function only returns true for integer vectors.
5213 * emit-rtl.c: Include rtx-vector-builder.h.
5214 (gen_const_vec_duplicate_1): Delete.
5215 (gen_const_vector): Call gen_const_vec_duplicate instead of
5216 gen_const_vec_duplicate_1.
5217 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5218 (gen_const_vec_duplicate): Use rtx_vector_builder.
5219 (gen_const_vec_series): Likewise.
5220 (gen_rtx_CONST_VECTOR): Likewise.
5221 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5222 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5223 Build a new vector rather than modifying a CONST_VECTOR in-place.
5224 (handle_special_swappables): Update call accordingly.
5225 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5226 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5227 Build a new vector rather than modifying a CONST_VECTOR in-place.
5228 (handle_special_swappables): Update call accordingly.
5230 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5232 * simplify-rtx.c (simplify_const_binary_operation): Use
5233 CONST_VECTOR_ELT instead of XVECEXP.
5235 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5237 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5238 the selector elements to be different from the data elements
5239 if the selector is a VECTOR_CST.
5240 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5241 ssizetype for the selector.
5243 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5245 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5246 before testing each element individually.
5247 * tree-vect-generic.c (lower_vec_perm): Likewise.
5249 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5251 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5252 * selftest-run-tests.c (selftest::run_tests): Call it.
5253 * vector-builder.h (vector_builder::operator ==): New function.
5254 (vector_builder::operator !=): Likewise.
5255 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5256 (vec_perm_indices::all_from_input_p): New function.
5257 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5258 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5259 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5260 instead of reading the VECTOR_CST directly. Detect whether both
5261 vector inputs are the same before constructing the vec_perm_indices,
5262 and update the number of inputs argument accordingly. Use the
5263 utility functions added above. Only construct sel2 if we need to.
5265 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5267 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5268 the broadcast of the low byte.
5269 (expand_mult_highpart): Use an explicit encoding for the permutes.
5270 * optabs-query.c (can_mult_highpart_p): Likewise.
5271 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5272 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5273 (vectorizable_bswap): Likewise.
5274 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5275 explicit encoding for the power-of-2 permutes.
5276 (vect_permute_store_chain): Likewise.
5277 (vect_grouped_load_supported): Likewise.
5278 (vect_permute_load_chain): Likewise.
5280 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5282 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5283 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5284 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5285 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5286 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5287 (vect_gen_perm_mask_any): Likewise.
5289 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5291 * int-vector-builder.h: New file.
5292 * vec-perm-indices.h: Include int-vector-builder.h.
5293 (vec_perm_indices): Redefine as an int_vector_builder.
5294 (auto_vec_perm_indices): Delete.
5295 (vec_perm_builder): Redefine as a stand-alone class.
5296 (vec_perm_indices::vec_perm_indices): New function.
5297 (vec_perm_indices::clamp): Likewise.
5298 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5299 (vec_perm_indices::new_vector): New function.
5300 (vec_perm_indices::new_expanded_vector): Update for new
5301 vec_perm_indices class.
5302 (vec_perm_indices::rotate_inputs): New function.
5303 (vec_perm_indices::all_in_range_p): Operate directly on the
5304 encoded form, without computing elided elements.
5305 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5306 encoding. Update for new vec_perm_indices class.
5307 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5308 the given vec_perm_builder.
5309 (expand_vec_perm_var): Update vec_perm_builder constructor.
5310 (expand_mult_highpart): Use vec_perm_builder instead of
5311 auto_vec_perm_indices.
5312 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5313 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5314 or double series encoding as appropriate.
5315 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5316 vec_perm_indices instead of auto_vec_perm_indices.
5317 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5318 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5319 (vect_permute_store_chain): Likewise.
5320 (vect_grouped_load_supported): Likewise.
5321 (vect_permute_load_chain): Likewise.
5322 (vect_shift_permute_load_chain): Likewise.
5323 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5324 (vect_transform_slp_perm_load): Likewise.
5325 (vect_schedule_slp_instance): Likewise.
5326 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5327 (vectorizable_mask_load_store): Likewise.
5328 (vectorizable_bswap): Likewise.
5329 (vectorizable_store): Likewise.
5330 (vectorizable_load): Likewise.
5331 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5332 vec_perm_indices instead of auto_vec_perm_indices. Use
5333 tree_to_vec_perm_builder to read the vector from a tree.
5334 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5335 vec_perm_builder instead of a vec_perm_indices.
5336 (have_whole_vector_shift): Use vec_perm_builder and
5337 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5338 truncation to calc_vec_perm_mask_for_shift.
5339 (vect_create_epilog_for_reduction): Likewise.
5340 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5341 from auto_vec_perm_indices to vec_perm_indices.
5342 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5343 instead of changing individual elements.
5344 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5345 the vector in d.perm.
5346 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5347 from auto_vec_perm_indices to vec_perm_indices.
5348 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5349 instead of changing individual elements.
5350 (arm_vectorize_vec_perm_const): Use new_vector to install
5351 the vector in d.perm.
5352 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5353 Update vec_perm_builder constructor.
5354 (rs6000_expand_interleave): Likewise.
5355 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5356 (rs6000_expand_interleave): Likewise.
5358 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5360 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5361 to qimode could truncate the indices.
5362 * optabs.c (expand_vec_perm_var): Likewise.
5364 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5366 * Makefile.in (OBJS): Add vec-perm-indices.o.
5367 * vec-perm-indices.h: New file.
5368 * vec-perm-indices.c: Likewise.
5369 * target.h (vec_perm_indices): Replace with a forward class
5371 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5372 * optabs.h: Include vec-perm-indices.h.
5373 (expand_vec_perm): Delete.
5374 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5375 (expand_vec_perm_const): Declare.
5376 * target.def (vec_perm_const_ok): Replace with...
5377 (vec_perm_const): ...this new hook.
5378 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5379 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5380 * doc/tm.texi: Regenerate.
5381 * optabs.def (vec_perm_const): Delete.
5382 * doc/md.texi (vec_perm_const): Likewise.
5383 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5384 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5385 expand_vec_perm for constant permutation vectors. Assert that
5386 the mode of variable permutation vectors is the integer equivalent
5387 of the mode that is being permuted.
5388 * optabs-query.h (selector_fits_mode_p): Declare.
5389 * optabs-query.c: Include vec-perm-indices.h.
5390 (selector_fits_mode_p): New function.
5391 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5392 is defined, instead of checking whether the vec_perm_const_optab
5393 exists. Use targetm.vectorize.vec_perm_const instead of
5394 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5395 fit in the vector mode before using a variable permute.
5396 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5397 vec_perm_indices instead of an rtx.
5398 (expand_vec_perm): Replace with...
5399 (expand_vec_perm_const): ...this new function. Take the selector
5400 as a vec_perm_indices rather than an rtx. Also take the mode of
5401 the selector. Update call to shift_amt_for_vec_perm_mask.
5402 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5403 Use vec_perm_indices::new_expanded_vector to expand the original
5404 selector into bytes. Check whether the indices fit in the vector
5405 mode before using a variable permute.
5406 (expand_vec_perm_var): Make global.
5407 (expand_mult_highpart): Use expand_vec_perm_const.
5408 * fold-const.c: Includes vec-perm-indices.h.
5409 * tree-ssa-forwprop.c: Likewise.
5410 * tree-vect-data-refs.c: Likewise.
5411 * tree-vect-generic.c: Likewise.
5412 * tree-vect-loop.c: Likewise.
5413 * tree-vect-slp.c: Likewise.
5414 * tree-vect-stmts.c: Likewise.
5415 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5417 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5418 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5419 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5420 (aarch64_vectorize_vec_perm_const): ...this new function.
5421 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5422 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5423 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5424 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5425 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5426 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5427 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5429 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5430 check for NEON modes.
5431 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5432 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5433 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5434 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5436 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5437 the old VEC_PERM_CONST conditions.
5438 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5439 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5440 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5441 (ia64_vectorize_vec_perm_const_ok): Merge into...
5442 (ia64_vectorize_vec_perm_const): ...this new function.
5443 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5444 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5445 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5446 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5447 * config/mips/mips.c (mips_expand_vec_perm_const)
5448 (mips_vectorize_vec_perm_const_ok): Merge into...
5449 (mips_vectorize_vec_perm_const): ...this new function.
5450 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5451 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5452 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5453 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5454 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5455 (rs6000_expand_vec_perm_const): Delete.
5456 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5458 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5459 (altivec_expand_vec_perm_const_le): Take each operand individually.
5460 Operate on constant selectors rather than rtxes.
5461 (altivec_expand_vec_perm_const): Likewise. Update call to
5462 altivec_expand_vec_perm_const_le.
5463 (rs6000_expand_vec_perm_const): Delete.
5464 (rs6000_vectorize_vec_perm_const_ok): Delete.
5465 (rs6000_vectorize_vec_perm_const): New function.
5466 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5467 an element count and rtx array.
5468 (rs6000_expand_extract_even): Update call accordingly.
5469 (rs6000_expand_interleave): Likewise.
5470 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5471 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5472 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5473 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5474 (rs6000_expand_vec_perm_const): Delete.
5475 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5476 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5477 (altivec_expand_vec_perm_const_le): Take each operand individually.
5478 Operate on constant selectors rather than rtxes.
5479 (altivec_expand_vec_perm_const): Likewise. Update call to
5480 altivec_expand_vec_perm_const_le.
5481 (rs6000_expand_vec_perm_const): Delete.
5482 (rs6000_vectorize_vec_perm_const_ok): Delete.
5483 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5484 reference to the SPE evmerge intructions.
5485 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5486 an element count and rtx array.
5487 (rs6000_expand_extract_even): Update call accordingly.
5488 (rs6000_expand_interleave): Likewise.
5489 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5490 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5492 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5494 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5496 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5497 vector mode and that that mode matches the mode of the data
5499 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5500 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5501 directly using expand_vec_perm_1 when forcing selectors into
5503 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5505 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5507 * optabs-query.h (can_vec_perm_p): Delete.
5508 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5509 * optabs-query.c (can_vec_perm_p): Split into...
5510 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5511 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5512 particular selector is valid.
5513 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5514 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5515 (vect_grouped_load_supported): Likewise.
5516 (vect_shift_permute_load_chain): Likewise.
5517 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5518 (vect_transform_slp_perm_load): Likewise.
5519 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5520 (vectorizable_bswap): Likewise.
5521 (vect_gen_perm_mask_checked): Likewise.
5522 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5523 implementations of variable permutation vectors into account
5524 when deciding which selector to use.
5525 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5526 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5527 with a false third argument.
5528 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5529 to test whether the constant selector is valid and can_vec_perm_var_p
5530 to test whether a variable selector is valid.
5532 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5534 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5535 * optabs-query.c (can_vec_perm_p): Likewise.
5536 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5537 instead of vec_perm_indices.
5538 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5539 (vect_gen_perm_mask_checked): Likewise,
5540 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5541 (vect_gen_perm_mask_checked): Likewise,
5543 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5545 * optabs-query.h (qimode_for_vec_perm): Declare.
5546 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5547 (qimode_for_vec_perm): ...this new function.
5548 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5550 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5552 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5553 does not have a conditional at the top.
5555 2018-01-02 Richard Biener <rguenther@suse.de>
5557 * ipa-inline.c (big_speedup_p): Fix expression.
5559 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5562 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5565 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5569 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5570 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5571 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5572 cond_taken_branch_cost 3->4.
5574 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5576 PR tree-optimization/83581
5577 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5578 TODO_cleanup_cfg if any changes have been made.
5581 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5582 convert_modes if target mode has the right side, but different mode
5586 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5587 last argument when extracting from CONCAT. If either from_real or
5588 from_imag is NULL, use expansion through memory. If result is not
5589 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5590 the parts directly to inner mode, if even that fails, use expansion
5594 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5595 check for bswap in mode rather than HImode and use that in expand_unop
5598 Copyright (C) 2018 Free Software Foundation, Inc.
5600 Copying and distribution of this file, with or without modification,
5601 are permitted in any medium without royalty provided the copyright
5602 notice and this notice are preserved.