| blob | 4aecf47e45a5127326085683a43e3d4792ec55f3 |
1 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
3 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
4 of usage.
6 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
8 PR target/81550
9 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
10 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
11 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
12 flags. This restores the settings used before the 2017-07-24.
13 Turning off pre increment/decrement/modify allows IVOPTS to
14 optimize DF/SF loops where the index is an int.
16 2018-01-29 Richard Biener <rguenther@suse.de>
17 Kelvin Nilsen <kelvin@gcc.gnu.org>
19 PR bootstrap/80867
20 * tree-vect-stmts.c (vectorizable_call): Don't call
21 targetm.vectorize_builtin_md_vectorized_function if callee is
22 NULL.
24 2018-01-22 Carl Love <cel@us.ibm.com>
26 * doc/extend.tex: Fix typo in second arg in
27 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
29 2018-01-29 Richard Biener <rguenther@suse.de>
31 PR tree-optimization/84086
32 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
33 (flush_ssaname_freelist): When SSA names were released reset
34 the SCEV hash table.
36 2018-01-29 Richard Biener <rguenther@suse.de>
38 PR tree-optimization/84057
39 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
40 removed paths when removing edges.
42 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
44 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
45 -mfunction-return=@var{choice}.
47 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
49 PR diagnostic/84034
50 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
51 Handle CR like TAB.
52 (layout::print_source_line): Likewise.
53 (test_get_line_width_without_trailing_whitespace): Add test cases.
55 2018-01-27 Jakub Jelinek <jakub@redhat.com>
57 PR middle-end/84040
58 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
59 debug insns.
61 2018-01-26 Jim Wilson <jimw@sifive.com>
63 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
65 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
66 specified.
68 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
70 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
71 and CMP + SUB-immediate -> SUBS.
73 2018-01-26 Martin Sebor <msebor@redhat.com>
75 PR tree-optimization/83896
76 * tree-ssa-strlen.c (get_string_len): Rename...
77 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
78 Avoid assuming length is constant.
79 (handle_char_store): Use HOST_WIDE_INT for string length.
81 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
83 PR target/81763
84 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
85 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
87 2018-01-26 Richard Biener <rguenther@suse.de>
89 PR rtl-optimization/84003
90 * dse.c (record_store): Only record redundant stores when
91 the earlier store aliases at least all accesses the later one does.
93 2018-01-26 Jakub Jelinek <jakub@redhat.com>
95 PR rtl-optimization/83985
96 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
97 REG_CFA_RESTORE insns.
98 (delete_unmarked_insns): Don't ignore separate shrink wrapping
99 REG_CFA_RESTORE insns here.
101 PR c/83989
102 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
103 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
105 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
107 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
108 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
109 (arc_init): Likewise.
110 (arc_override_options): Likewise.
111 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
112 value.
113 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
114 support.
115 * config/arc/arc.h (TARGET_DBNZ): Define.
116 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
117 properly set the tune attribute.
118 (dbnz): Use TARGET_DBNZ guard.
119 * config/arc/arc.opt (mtune): Add core3 option.
121 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
123 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
124 recognize new pic like addresses.
125 (arc_delegitimize_address): Clean up.
127 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
129 * config/arc/arc-arches.def: Option mrf16 valid for all
130 architectures.
131 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
132 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
133 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
134 * config/arc/arc-tables.opt: Regenerate.
135 * config/arc/arc.c (arc_conditional_register_usage): Handle
136 reduced register file case.
137 (arc_file_start): Set must have build attributes.
138 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
139 mrf16 option value.
140 * config/arc/arc.opt (mrf16): Add new option.
141 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
142 * config/arc/genmultilib.awk: Handle new mrf16 option.
143 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
144 * config/arc/t-multilib: Regenerate.
145 * doc/invoke.texi (ARC Options): Document mrf16 option.
147 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
149 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
150 * config/arc/arc.c (arc_handle_secure_attribute): New function.
151 (arc_attribute_table): Add 'secure_call' attribute.
152 (arc_print_operand): Print secure call operand.
153 (arc_function_ok_for_sibcall): Don't optimize tail calls when
154 secure.
155 (arc_is_secure_call_p): New function. * config/arc/arc.md
156 (call_i): Add support for sjli instruction.
157 (call_value_i): Likewise.
158 * config/arc/constraints.md (Csc): New constraint.
160 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
161 John Eric Martin <John.Martin@emmicro-us.com>
163 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
164 * config/arc/arc.c (_arc_jli_section): New struct.
165 (arc_jli_section): New type.
166 (rc_jli_sections): New static variable.
167 (arc_handle_jli_attribute): New function.
168 (arc_attribute_table): Add jli_always and jli_fixed attribute.
169 (arc_file_end): New function.
170 (TARGET_ASM_FILE_END): Define.
171 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
172 (arc_add_jli_section): New function.
173 (jli_call_scan): Likewise.
174 (arc_reorg): Call jli_call_scan.
175 (arc_output_addsi): Remove 'S' from printing asm operand.
176 (arc_is_jli_call_p): New function.
177 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
178 operand.
179 (movhi_insn): Likewise.
180 (movsi_insn): Likewise.
181 (movsi_set_cc_insn): Likewise.
182 (loadqi_update): Likewise.
183 (load_zeroextendqisi_update): Likewise.
184 (load_signextendqisi_update): Likewise.
185 (loadhi_update): Likewise.
186 (load_zeroextendhisi_update): Likewise.
187 (load_signextendhisi_update): Likewise.
188 (loadsi_update): Likewise.
189 (loadsf_update): Likewise.
190 (movsicc_insn): Likewise.
191 (bset_insn): Likewise.
192 (bxor_insn): Likewise.
193 (bclr_insn): Likewise.
194 (bmsk_insn): Likewise.
195 (bicsi3_insn): Likewise.
196 (cmpsi_cc_c_insn): Likewise.
197 (movsi_ne): Likewise.
198 (movsi_cond_exec): Likewise.
199 (clrsbsi2): Likewise.
200 (norm_f): Likewise.
201 (normw): Likewise.
202 (swap): Likewise.
203 (divaw): Likewise.
204 (flag): Likewise.
205 (sr): Likewise.
206 (kflag): Likewise.
207 (ffs): Likewise.
208 (ffs_f): Likewise.
209 (fls): Likewise.
210 (call_i): Remove 'S' asm letter, add jli instruction.
211 (call_value_i): Likewise.
212 * config/arc/arc.op (mjli-always): New option.
213 * config/arc/constraints.md (Cji): New constraint.
214 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
215 operand.
216 (subsf3_fpx): Likewise.
217 (mulsf3_fpx): Likewise.
218 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
219 asm operand.
220 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
221 function attrbutes.
222 * doc/invoke.texi (ARC): Document mjli-always option.
224 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
226 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
227 and use incw and decw where possible
228 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
230 2018-01-26 Richard Biener <rguenther@suse.de>
232 PR tree-optimization/81082
233 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
234 association if it requires casting to unsigned.
235 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
236 from fold_plusminus_mult_expr to catch important cases late when
237 range info is available.
239 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
241 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
242 * configure.ac (hidden_linkonce): New test.
243 * configure: Regenerate.
244 * config.in: Regenerate.
246 2018-01-26 Julia Koval <julia.koval@intel.com>
248 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
249 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
250 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
251 _mm_mask_bitshuffle_epi64_mask): Fix type.
252 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
253 USI_FTYPE_V4DI_V4DI_USI): Remove.
254 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
255 __builtin_ia32_vpshufbitqmb256_mask,
256 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
257 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
258 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
260 2018-01-26 Alan Modra <amodra@gmail.com>
262 PR target/84033
263 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
264 UNSPEC_VBPERMQ. Sort other unspecs.
266 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
268 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
270 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
272 PR middle-end/83055
273 * predict.c (drop_profile): Do not push/pop cfun; update also
274 node->count.
275 (handle_missing_profiles): Fix logic looking for zero profiles.
277 2018-01-25 Jakub Jelinek <jakub@redhat.com>
279 PR middle-end/83977
280 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
281 on functions with #pragma omp declare simd or functions with simd
282 attribute.
283 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
284 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
285 Remove trailing \n from warning_at calls.
287 2018-01-25 Tom de Vries <tom@codesourcery.com>
289 PR target/84028
290 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
291 for neutered workers.
293 2018-01-24 Joseph Myers <joseph@codesourcery.com>
295 PR target/68467
296 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
297 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
299 2017-01-08 Jeff Law <law@redhat.com>
301 PR target/83994
302 * i386.c (get_probe_interval): Move to earlier point.
303 (ix86_compute_frame_layout): If -fstack-clash-protection and
304 the frame is larger than the probe interval, then use pushes
305 to save registers rather than reg->mem moves.
306 (ix86_expand_prologue): Remove conditional for int_registers_saved
307 assertion.
309 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
311 PR target/84014
312 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
313 min/max for never referenced object.
315 2018-01-24 Jakub Jelinek <jakub@redhat.com>
317 PR middle-end/83977
318 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
319 here.
320 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
321 attributes from DECL_ATTRIBUTES (decl) without affecting
322 DECL_ATTRIBUTES (current_function_decl).
323 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
324 functions with non-NULL DECL_ABSTRACT_ORIGIN.
326 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
328 PR tree-optimization/83979
329 * fold-const.c (fold_comparison): Use constant_boolean_node
330 instead of boolean_{true,false}_node.
332 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
334 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
335 with zero counts.
337 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
339 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
340 Simplify the clause that sets the length attribute.
341 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
342 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
343 clause that sets the length attribute.
344 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
346 2018-01-24 Tom de Vries <tom@codesourcery.com>
348 PR target/83589
349 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
350 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
351 Add strict parameter.
352 (prevent_branch_around_nothing): Insert dummy insn between branch to
353 label and label with no ptx insn inbetween.
354 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
356 2018-01-24 Tom de Vries <tom@codesourcery.com>
358 PR target/81352
359 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
360 for neutered threads in warp.
361 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
363 2018-01-24 Richard Biener <rguenther@suse.de>
365 PR tree-optimization/83176
366 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
367 operands.
369 2018-01-24 Richard Biener <rguenther@suse.de>
371 PR tree-optimization/82819
372 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
373 code generating pluses that are no-ops in the target precision.
375 2018-01-24 Richard Biener <rguenther@suse.de>
377 PR middle-end/84000
378 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
380 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
382 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
383 to merge probabilities.
384 * predict.c (probably_never_executed): Also mark as cold functions
385 with global 0 profile and guessed local profile.
386 * profile-count.c (profile_probability::combine_with_count): New
387 member function.
388 * profile-count.h (profile_probability::operator*,
389 profile_probability::operator*=, profile_probability::operator/,
390 profile_probability::operator/=): Reduce precision to adjusted
391 and set value to guessed on contradictory divisions.
392 (profile_probability::combine_with_freq): Remove.
393 (profile_probability::combine_wiht_count): Declare.
394 (profile_count::force_nonzero):: Set to adjusted.
395 (profile_count::probability_in):: Set quality to adjusted.
396 * tree-ssa-tail-merge.c (replace_block_by): Use
397 combine_with_count.
399 2018-01-23 Andrew Waterman <andrew@sifive.com>
400 Jim Wilson <jimw@sifive.com>
402 * config/riscv/riscv.c (riscv_stack_boundary): New.
403 (riscv_option_override): Set riscv_stack_boundary. Handle
404 riscv_preferred_stack_boundary_arg.
405 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
406 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
407 (STACK_BOUNDARY): Set to riscv_stack_boundary.
408 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
409 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
410 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
412 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
414 PR target/83905
415 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
416 of struct ix86_frame.
417 (ix86_expand_epilogue): Likewise. Add a local variable for
418 the reg_save_offset field in struct ix86_frame.
420 2018-01-23 Bin Cheng <bin.cheng@arm.com>
422 PR tree-optimization/82604
423 * tree-loop-distribution.c (enum partition_kind): New enum item
424 PKIND_PARTIAL_MEMSET.
425 (partition_builtin_p): Support above new enum item.
426 (generate_code_for_partition): Ditto.
427 (compute_access_range): Differentiate cases that equality can be
428 proven at all loops, the innermost loops or no loops.
429 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
430 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
431 (finalize_partitions, distribute_loop): Don't fuse partition of
432 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
433 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
434 parloop is enabled.
436 2018-01-23 Martin Liska <mliska@suse.cz>
438 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
439 order to ignore the predictor.
440 (PRED_POLYMORPHIC_CALL): Likewise.
441 (PRED_RECURSIVE_CALL): Likewise.
443 2018-01-23 Martin Liska <mliska@suse.cz>
445 * tree-profile.c (tree_profiling): Print function header to
446 aware reader which function we are working on.
447 * value-prof.c (gimple_find_values_to_profile): Do not print
448 not interesting value histograms.
450 2018-01-23 Martin Liska <mliska@suse.cz>
452 * profile-count.h (enum profile_quality): Add
453 profile_uninitialized as the first value. Do not number values
454 as they are zero based.
455 (profile_count::verify): Update sanity check.
456 (profile_probability::verify): Likewise.
458 2018-01-23 Nathan Sidwell <nathan@acm.org>
460 * doc/invoke.texi (ffor-scope): Deprecate.
462 2018-01-23 David Malcolm <dmalcolm@redhat.com>
464 PR tree-optimization/83510
465 * domwalk.c (set_all_edges_as_executable): New function.
466 (dom_walker::dom_walker): Convert bool param
467 "skip_unreachable_blocks" to enum reachability. Move setup of
468 edge flags to set_all_edges_as_executable and only do it when
469 reachability is REACHABLE_BLOCKS.
470 * domwalk.h (enum dom_walker::reachability): New enum.
471 (dom_walker::dom_walker): Convert bool param
472 "skip_unreachable_blocks" to enum reachability.
473 (set_all_edges_as_executable): New decl.
474 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
475 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
476 "reachability".
477 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
478 but converting true to REACHABLE_BLOCKS.
479 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
480 * tree-vrp.c
481 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
482 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
483 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
484 REACHABLE_BLOCKS.
485 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
486 if check_all_array_refs will be called.
488 2018-01-23 David Malcolm <dmalcolm@redhat.com>
490 * tree.c (selftest::test_location_wrappers): Add more test
491 coverage.
493 2018-01-23 David Malcolm <dmalcolm@redhat.com>
495 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
496 (selftest::test_bit_in_range): Likewise.
498 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
500 PR testsuite/83888
501 * doc/sourcebuild.texi (vect_float): Say that the selector
502 only describes the situation when -funsafe-math-optimizations is on.
503 (vect_float_strict): Document.
505 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
507 PR tree-optimization/83965
508 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
509 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
510 instead of checking only for a reduction.
511 (vect_recog_widen_sum_pattern): Likewise.
513 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
515 * predict.c (probably_never_executed): Only use precise profile info.
516 (compute_function_frequency): Skip after inlining hack since we now
517 have quality checking.
519 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
521 * profile-count.h (profile_probability::very_unlikely,
522 profile_probability::unlikely, profile_probability::even): Set
523 precision to guessed.
525 2018-01-23 Richard Biener <rguenther@suse.de>
527 PR tree-optimization/83963
528 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
529 Properly terminate dominator walk when crossing the exit edge not
530 when visiting its source block.
532 2018-01-23 Jakub Jelinek <jakub@redhat.com>
534 PR c++/83918
535 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
536 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
538 2018-01-22 Jakub Jelinek <jakub@redhat.com>
540 PR tree-optimization/83957
541 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
542 semicolon after for body surrounded by braces.
544 PR tree-optimization/83081
545 * profile-count.h (profile_probability::split): New method.
546 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
547 Use profile_probability::split.
548 (do_compare_rtx_and_jump): Fix adjustment of probabilities
549 when splitting a single conditional jump into 2.
551 2018-01-22 David Malcolm <dmalcolm@redhat.com>
553 PR tree-optimization/69452
554 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
555 decl.
557 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
559 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
560 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
561 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
563 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
565 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
566 * config/rl78/rl78.md: New define_expand "movdi"
567 * config/rl78/rl78.c: New function definition rl78_split_movdi
569 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
571 PR target/83862
572 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
573 no longer used.
574 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
575 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
576 128-bit to produce an UNSPEC move to get the double word with the
577 signbit and then a shift directly to do signbit.
578 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
579 implementation with a new version that just does either a direct
580 move or a regular move. Move memory interface to separate insns.
581 Move insns so they are next to the expander.
582 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
583 with signbit move. Split big and little endian case.
584 (signbit<mode>2_dm_mem_le): Likewise.
585 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
586 (signbit<mode>2_dm2): Likewise.
588 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
590 * config/rl78/rl78.md: New define_expand "anddi3".
592 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
594 * config/rl78/rl78.md: New define_expand "umindi3".
596 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
598 * config/rl78/rl78.md: New define_expand "smindi3".
600 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
602 * config/rl78/rl78.md: New define_expand "smaxdi3".
604 2018-01-22 Carl Love <cel@us.ibm.com>
606 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
607 LVX_V1TI): Add macro expansion.
608 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
609 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
610 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
611 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
612 Change check to determine if the instruction is a byte reversing
613 entry. Fix typo in comment.
614 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
615 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
616 Add def_builtin calls for new builtins.
617 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
618 Add define_insn expansion.
620 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
622 * config/rl78/rl78.md: New define_expand "umaxdi3".
624 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
626 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
627 for non-QImode registers
629 2018-01-22 Richard Biener <rguenther@suse.de>
631 PR tree-optimization/83963
632 * graphite-scop-detection.c (scop_detection::get_sese): Delay
633 including the loop exit block.
634 (scop_detection::merge_sese): Likewise.
635 (scop_detection::add_scop): Do it here instead.
637 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
639 * doc/sourcebuild.texi (arm_softfloat): Document.
641 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
643 PR gcc/77734
644 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
645 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
646 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
648 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
649 David Edelsohn <dje.gcc@gmail.com>
651 PR target/83946
652 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
653 Change "crset eq" to "crset 2".
654 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
655 (*call_indirect_aix<mode>_nospec): Likewise.
656 (*call_value_indirect_aix<mode>_nospec): Likewise.
657 (*call_indirect_elfv2<mode>_nospec): Likewise.
658 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
659 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
660 change assembly output from . to $.
661 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
662 (indirect_jump<mode>_nospec): Change assembly output from . to $.
663 (*tablejump<mode>_internal1_nospec): Likewise.
665 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
667 PR target/80870
668 * config/sh/sh_optimize_sett_clrt.cc:
669 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
671 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
673 PR tree-optimization/83940
674 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
675 offset_dt to vect_constant_def rather than vect_unknown_def_type.
676 (vect_check_load_store_mask): Add a mask_dt_out parameter and
677 use it to pass back the definition type.
678 (vect_check_store_rhs): Likewise rhs_dt_out.
679 (vect_build_gather_load_calls): Add a mask_dt argument and use
680 it instead of a call to vect_is_simple_use.
681 (vectorizable_store): Update calls to vect_check_load_store_mask
682 and vect_check_store_rhs. Use the dt returned by the latter instead
683 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
684 instead of calls to vect_is_simple_use. Pass the scalar rather
685 than the vector operand to vect_is_simple_use when handling
686 second and subsequent copies of an rhs value.
687 (vectorizable_load): Update calls to vect_check_load_store_mask
688 and vect_build_gather_load_calls. Use the cached mask_dt and
689 gs_info.offset_dt instead of calls to vect_is_simple_use.
691 2018-01-20 Jakub Jelinek <jakub@redhat.com>
693 PR middle-end/83945
694 * tree-emutls.c: Include gimplify.h.
695 (lower_emutls_2): New function.
696 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
697 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
698 it before further processing.
700 PR target/83930
701 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
702 UINTVAL (trueop1) instead of INTVAL (op1).
704 2018-01-19 Jakub Jelinek <jakub@redhat.com>
706 PR debug/81570
707 PR debug/83728
708 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
709 INCOMING_FRAME_SP_OFFSET if not defined.
710 (scan_trace): Add ENTRY argument. If true and
711 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
712 emit a note to adjust the CFA offset.
713 (create_cfi_notes): Adjust scan_trace callers.
714 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
715 INCOMING_FRAME_SP_OFFSET in the CIE.
716 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
717 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
718 Likewise.
719 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
720 * doc/tm.texi: Regenerated.
722 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
724 PR rtl-optimization/83147
725 * lra-constraints.c (remove_inheritance_pseudos): Use
726 lra_substitute_pseudo_within_insn.
728 2018-01-19 Tom de Vries <tom@codesourcery.com>
729 Cesar Philippidis <cesar@codesourcery.com>
731 PR target/83920
732 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
734 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
736 PR target/83790
737 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
738 spaces for function labels.
740 2018-01-19 Martin Liska <mliska@suse.cz>
742 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
743 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
744 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
745 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
746 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
747 (PRED_CONST_RETURN): Change from 69 to 65.
748 (PRED_NULL_RETURN): Change from 91 to 71.
749 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
750 (PRED_LOOP_GUARD): Change from 66 to 73.
752 2018-01-19 Martin Liska <mliska@suse.cz>
754 * predict.c (predict_insn_def): Add new assert.
755 (struct branch_predictor): Change type to signed integer.
756 (test_prediction_value_range): Amend test to cover
757 PROB_UNINITIALIZED.
758 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
759 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
760 (PRED_LOOP_ITERATIONS_MAX): Likewise.
761 (PRED_LOOP_IV_COMPARE): Likewise.
762 * predict.h (PROB_UNINITIALIZED): Define new constant.
764 2018-01-19 Martin Liska <mliska@suse.cz>
766 * predict.c (dump_prediction): Add new format for
767 analyze_brprob.py script which is enabled with -details
768 suboption.
769 * profile-count.h (precise_p): New function.
771 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
773 PR tree-optimization/83922
774 * tree-vect-loop.c (vect_verify_full_masking): Return false if
775 there are no statements that need masking.
776 (vect_active_double_reduction_p): New function.
777 (vect_analyze_loop_operations): Use it when handling phis that
778 are not in the loop header.
780 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
782 PR tree-optimization/83914
783 * tree-vect-loop.c (vectorizable_induction): Don't convert
784 init_expr or apply the peeling adjustment for inductions
785 that are nested within the vectorized loop.
787 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
789 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
790 instead of NEG.
792 2018-01-18 Jakub Jelinek <jakub@redhat.com>
794 PR sanitizer/81715
795 PR testsuite/83882
796 * function.h (gimplify_parameters): Add gimple_seq * argument.
797 * function.c: Include gimple.h and options.h.
798 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
799 for the added local temporaries if needed.
800 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
801 if there are any parameter cleanups, wrap whole body into a
802 try/finally with the cleanups.
804 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
806 PR target/82964
807 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
808 Use GET_MODE_CLASS for scalar floating point.
810 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
812 PR ipa/82256
813 patch by PaX Team
814 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
815 Fix call of call_cgraph_insertion_hooks.
817 2018-01-18 Martin Sebor <msebor@redhat.com>
819 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
821 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
823 PR ipa/83619
824 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
825 frequencies.
827 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
829 PR other/70268
830 * common.opt: (-ffile-prefix-map): New option.
831 * opts.c (common_handle_option): Defer it.
832 * opts-global.c (handle_common_deferred_options): Handle it.
833 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
834 * file-prefix-map.h: New file.
835 (remap_debug_filename, add_debug_prefix_map): ...here.
836 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
837 * final.c (debug_prefix_map, add_debug_prefix_map
838 remap_debug_filename): Move to...
839 * file-prefix-map.c: New file.
840 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
841 generalize, get rid of alloca(), use strrchr() instead of strchr().
842 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
843 Implement in terms of add_prefix_map().
844 (remap_macro_filename, remap_debug_filename): Implement in term of
845 remap_filename().
846 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
847 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
848 * dbxout.c: Include file-prefix-map.h.
849 * varasm.c: Likewise.
850 * vmsdbgout.c: Likewise.
851 * xcoffout.c: Likewise.
852 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
853 * doc/cppopts.texi (-fmacro-prefix-map): Document.
854 * doc/invoke.texi (-ffile-prefix-map): Document.
855 (-fdebug-prefix-map): Update description.
857 2018-01-18 Martin Liska <mliska@suse.cz>
859 * config/i386/i386.c (indirect_thunk_name): Document that also
860 lfence is emitted.
861 (output_indirect_thunk): Document why both instructions
862 (pause and lfence) are generated.
864 2018-01-18 Richard Biener <rguenther@suse.de>
866 PR tree-optimization/83887
867 * graphite-scop-detection.c
868 (scop_detection::get_nearest_dom_with_single_entry): Remove.
869 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
870 (scop_detection::merge_sese): Re-implement with a flood-fill
871 algorithm that properly finds a SESE region if it exists.
873 2018-01-18 Jakub Jelinek <jakub@redhat.com>
875 PR c/61240
876 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
877 pointer_diff optimizations use view_convert instead of convert.
879 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
881 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
882 Generate different code for -mno-speculate-indirect-jumps.
883 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
884 (*call_indirect_aix<mode>): Disable for
885 -mno-speculate-indirect-jumps.
886 (*call_indirect_aix<mode>_nospec): New define_insn.
887 (*call_value_indirect_aix<mode>): Disable for
888 -mno-speculate-indirect-jumps.
889 (*call_value_indirect_aix<mode>_nospec): New define_insn.
890 (*sibcall_nonlocal_sysv<mode>): Generate different code for
891 -mno-speculate-indirect-jumps.
892 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
894 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
896 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
897 long double type, set the flags for noting the default long double
898 type, even if we don't pass or return a long double type.
900 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
902 PR ipa/83051
903 * ipa-inline.c (flatten_function): Do not overwrite final inlining
904 failure.
906 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
908 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
909 support for merge[hl].
910 (fold_mergehl_helper): New helper function.
911 (tree-vector-builder.h): New #include for tree_vector_builder usage.
912 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
913 (altivec_vmrglw_direct): Add xxmrglw insn.
915 2018-01-17 Andrew Waterman <andrew@sifive.com>
917 * config/riscv/riscv.c (riscv_conditional_register_usage): If
918 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
920 2018-01-17 David Malcolm <dmalcolm@redhat.com>
922 PR lto/83121
923 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
924 call the lto_location_cache before reading the
925 DECL_SOURCE_LOCATION of the types.
927 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
928 Richard Sandiford <richard.sandiford@linaro.org>
930 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
931 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
932 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
933 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
934 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
935 Add declaration.
936 * config/aarch64/constraints.md (aarch64_movti_operand):
937 Limit immediates.
938 * config/aarch64/predicates.md (Uti): Add new constraint.
940 2018-01-17 Carl Love <cel@us.ibm.com>
941 * config/rs6000/vsx.md (define_expand xl_len_r,
942 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
943 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
944 lxvll.
945 (define_expand, define_insn): Move the shift left from the
946 define_insn to the define_expand for lxvl and stxvl instructions.
947 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
948 and XL_LEN_R definitions to PURE.
950 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
952 * config/i386/i386.c (indirect_thunk_name): Declare regno
953 as unsigned int. Compare regno with INVALID_REGNUM.
954 (output_indirect_thunk): Ditto.
955 (output_indirect_thunk_function): Ditto.
956 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
957 in the call to output_indirect_thunk_function.
959 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
961 PR middle-end/83884
962 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
963 rather than the size of inner_type to determine the stack slot size
964 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
966 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
968 PR target/83546
969 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
970 to PTA_SILVERMONT.
972 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
974 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
975 endian Linux systems to optionally enable multilibs for selecting
976 the long double type if the user configured an explicit type.
977 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
978 have no long double multilibs if not defined.
979 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
980 warn if the user used -mabi={ieee,ibm}longdouble and we built
981 multilibs for long double.
982 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
983 appropriate multilib option.
984 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
985 multilib options.
986 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
987 for building long double multilibs.
988 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
990 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
992 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
993 copies.
995 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
996 64 bits.
997 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
998 128 bits.
1000 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1001 variables.
1003 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1004 return value.
1006 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1008 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1009 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1011 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1013 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1014 different rtl trees depending on TARGET_64BIT.
1015 (rs6000_gen_lvx): Likewise.
1017 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1019 * config/visium/visium.md (nop): Tweak comment.
1020 (hazard_nop): Likewise.
1022 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1024 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1025 -mspeculate-indirect-jumps.
1026 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1027 for -mno-speculate-indirect-jumps.
1028 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1029 (*call_value_indirect_elfv2<mode>): Disable for
1030 -mno-speculate-indirect-jumps.
1031 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1032 (indirect_jump): Emit different RTL for
1033 -mno-speculate-indirect-jumps.
1034 (*indirect_jump<mode>): Disable for
1035 -mno-speculate-indirect-jumps.
1036 (*indirect_jump<mode>_nospec): New define_insn.
1037 (tablejump): Emit different RTL for
1038 -mno-speculate-indirect-jumps.
1039 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1040 (tablejumpsi_nospec): New define_expand.
1041 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1042 (tablejumpdi_nospec): New define_expand.
1043 (*tablejump<mode>_internal1): Disable for
1044 -mno-speculate-indirect-jumps.
1045 (*tablejump<mode>_internal1_nospec): New define_insn.
1046 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1047 option.
1049 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1051 * caller-save.c (insert_save): Drop unnecessary parameter. All
1052 callers updated.
1054 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1055 Richard Biener <rguenth@suse.de>
1057 PR libgomp/83590
1058 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1059 return early, inline manually is_gimple_sizepos. Make sure if we
1060 call gimplify_expr we don't end up with a gimple constant.
1061 * tree.c (variably_modified_type_p): Don't return true for
1062 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1063 * gimplify.h (is_gimple_sizepos): Remove.
1065 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1067 PR tree-optimization/83857
1068 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1069 vectorizable_live_operation for pure SLP statements.
1070 (vectorizable_live_operation): Handle PHIs.
1072 2018-01-16 Richard Biener <rguenther@suse.de>
1074 PR tree-optimization/83867
1075 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1076 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1078 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1080 PR c/83844
1081 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1082 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1083 If off is not INTEGER_CST, issue a may not be aligned warning
1084 rather than isn't aligned. Use isn%'t rather than isn't.
1085 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1086 into MULT_EXPR.
1087 <case MULT_EXPR>: Improve the case when bottom and one of the
1088 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1089 operand, in that case check if the other operand is multiple of
1090 bottom divided by the INTEGER_CST operand.
1092 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1094 PR target/83858
1095 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1096 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1097 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1098 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1099 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1100 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1101 (pa_function_arg_size): New function.
1103 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1105 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1106 in a separate statement.
1108 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1110 PR tree-optimization/83847
1111 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1112 group gathers and scatters.
1114 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1116 PR rtl-optimization/86620
1117 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1119 PR rtl-optimization/83213
1120 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1121 to last if both are JUMP_INSNs.
1123 PR tree-optimization/83843
1124 * gimple-ssa-store-merging.c
1125 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1126 store_immediate_info for bswap/nop orig_stores.
1128 2018-01-15 Andrew Waterman <andrew@sifive.com>
1130 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1131 !TARGET_MUL.
1132 <UDIV>: Increase cost if !TARGET_DIV.
1134 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1136 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1137 (define_attr "cr_logical_3op"): New.
1138 (cceq_ior_compare): Adjust.
1139 (cceq_ior_compare_complement): Adjust.
1140 (*cceq_rev_compare): Adjust.
1141 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1142 (is_cracked_insn): Adjust.
1143 (insn_must_be_first_in_group): Adjust.
1144 * config/rs6000/40x.md: Adjust.
1145 * config/rs6000/440.md: Adjust.
1146 * config/rs6000/476.md: Adjust.
1147 * config/rs6000/601.md: Adjust.
1148 * config/rs6000/603.md: Adjust.
1149 * config/rs6000/6xx.md: Adjust.
1150 * config/rs6000/7450.md: Adjust.
1151 * config/rs6000/7xx.md: Adjust.
1152 * config/rs6000/8540.md: Adjust.
1153 * config/rs6000/cell.md: Adjust.
1154 * config/rs6000/e300c2c3.md: Adjust.
1155 * config/rs6000/e500mc.md: Adjust.
1156 * config/rs6000/e500mc64.md: Adjust.
1157 * config/rs6000/e5500.md: Adjust.
1158 * config/rs6000/e6500.md: Adjust.
1159 * config/rs6000/mpc.md: Adjust.
1160 * config/rs6000/power4.md: Adjust.
1161 * config/rs6000/power5.md: Adjust.
1162 * config/rs6000/power6.md: Adjust.
1163 * config/rs6000/power7.md: Adjust.
1164 * config/rs6000/power8.md: Adjust.
1165 * config/rs6000/power9.md: Adjust.
1166 * config/rs6000/rs64.md: Adjust.
1167 * config/rs6000/titan.md: Adjust.
1169 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1171 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1172 ix86_indirect_branch_register logic.
1174 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1176 * config/i386/constraints.md (Bs): Update
1177 ix86_indirect_branch_register check. Don't check
1178 ix86_indirect_branch_register with GOT_memory_operand.
1179 (Bw): Likewise.
1180 * config/i386/predicates.md (GOT_memory_operand): Don't check
1181 ix86_indirect_branch_register here.
1182 (GOT32_symbol_operand): Likewise.
1184 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1186 * config/i386/predicates.md (constant_call_address_operand):
1187 Rewrite ix86_indirect_branch_register logic.
1188 (sibcall_insn_operand): Likewise.
1190 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1192 * config/i386/constraints.md (Bs): Replace
1193 ix86_indirect_branch_thunk_register with
1194 ix86_indirect_branch_register.
1195 (Bw): Likewise.
1196 * config/i386/i386.md (indirect_jump): Likewise.
1197 (tablejump): Likewise.
1198 (*sibcall_memory): Likewise.
1199 (*sibcall_value_memory): Likewise.
1200 Peepholes of indirect call and jump via memory: Likewise.
1201 * config/i386/i386.opt: Likewise.
1202 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1203 (GOT_memory_operand): Likewise.
1204 (call_insn_operand): Likewise.
1205 (sibcall_insn_operand): Likewise.
1206 (GOT32_symbol_operand): Likewise.
1208 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1210 PR middle-end/83837
1211 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1212 type rather than type addr's type points to.
1213 (expand_omp_atomic_mutex): Likewise.
1214 (expand_omp_atomic): Likewise.
1216 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1218 PR target/83839
1219 * config/i386/i386.c (output_indirect_thunk_function): Use
1220 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1221 for __x86_return_thunk.
1223 2018-01-15 Richard Biener <rguenther@suse.de>
1225 PR middle-end/83850
1226 * expmed.c (extract_bit_field_1): Fix typo.
1228 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1230 PR target/83687
1231 * config/arm/iterators.md (VF): New mode iterator.
1232 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1233 Remove integer-related logic from pattern.
1234 (neon_vabd<mode>_3): Likewise.
1236 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1238 PR middle-end/82694
1239 * common.opt (fstrict-overflow): No longer an alias.
1240 (fwrapv-pointer): New option.
1241 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1242 also for pointer types based on flag_wrapv_pointer.
1243 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1244 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1245 opts->x_flag_wrapv got set.
1246 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1247 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1248 POINTER_TYPE_OVERFLOW_UNDEFINED.
1249 * match.pd: Likewise in address comparison pattern.
1250 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1252 2018-01-15 Richard Biener <rguenther@suse.de>
1254 PR lto/83804
1255 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1256 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1257 Reset type names to their identifier if their TYPE_DECL doesn't
1258 have linkage (and thus is used for ODR and devirt).
1259 (save_debug_info_for_decl): Remove.
1260 (save_debug_info_for_type): Likewise.
1261 (add_tree_to_fld_list): Adjust.
1262 * tree-pretty-print.c (dump_generic_node): Make dumping of
1263 type names more robust.
1265 2018-01-15 Richard Biener <rguenther@suse.de>
1267 * BASE-VER: Bump to 8.0.1.
1269 2018-01-14 Martin Sebor <msebor@redhat.com>
1271 PR other/83508
1272 * builtins.c (check_access): Avoid warning when the no-warning bit
1273 is set.
1275 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1277 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1278 * ira-color (allocno_hard_regs_compare): Likewise.
1280 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1282 PR target/83013
1283 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1284 Use .pushsection/.popsection.
1286 2018-01-14 Martin Sebor <msebor@redhat.com>
1288 PR c++/81327
1289 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1291 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1293 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1294 entry from extra_headers.
1295 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1296 extra_headers, make the list bitwise identical to the i?86-*-* one.
1298 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1300 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1301 -mcmodel=large with -mindirect-branch=thunk,
1302 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1303 -mfunction-return=thunk-extern.
1304 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1305 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1306 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1308 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1310 * config/i386/i386.c (print_reg): Print the name of the full
1311 integer register without '%'.
1312 (ix86_print_operand): Handle 'V'.
1313 * doc/extend.texi: Document 'V' modifier.
1315 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1317 * config/i386/constraints.md (Bs): Disallow memory operand for
1318 -mindirect-branch-register.
1319 (Bw): Likewise.
1320 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1321 (GOT_memory_operand): Likewise.
1322 (call_insn_operand): Likewise.
1323 (sibcall_insn_operand): Likewise.
1324 (GOT32_symbol_operand): Likewise.
1325 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1326 for -mindirect-branch-register.
1327 (tablejump): Likewise.
1328 (*sibcall_memory): Likewise.
1329 (*sibcall_value_memory): Likewise.
1330 Disallow peepholes of indirect call and jump via memory for
1331 -mindirect-branch-register.
1332 (*call_pop): Replace m with Bw.
1333 (*call_value_pop): Likewise.
1334 (*sibcall_pop_memory): Replace m with Bs.
1335 * config/i386/i386.opt (mindirect-branch-register): New option.
1336 * doc/invoke.texi: Document -mindirect-branch-register option.
1338 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1340 * config/i386/i386-protos.h (ix86_output_function_return): New.
1341 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1342 set function_return_type.
1343 (indirect_thunk_name): Add ret_p to indicate thunk for function
1344 return.
1345 (output_indirect_thunk_function): Pass false to
1346 indirect_thunk_name.
1347 (ix86_output_indirect_branch_via_reg): Likewise.
1348 (ix86_output_indirect_branch_via_push): Likewise.
1349 (output_indirect_thunk_function): Create alias for function
1350 return thunk if regno < 0.
1351 (ix86_output_function_return): New function.
1352 (ix86_handle_fndecl_attribute): Handle function_return.
1353 (ix86_attribute_table): Add function_return.
1354 * config/i386/i386.h (machine_function): Add
1355 function_return_type.
1356 * config/i386/i386.md (simple_return_internal): Use
1357 ix86_output_function_return.
1358 (simple_return_internal_long): Likewise.
1359 * config/i386/i386.opt (mfunction-return=): New option.
1360 (indirect_branch): Mention -mfunction-return=.
1361 * doc/extend.texi: Document function_return function attribute.
1362 * doc/invoke.texi: Document -mfunction-return= option.
1364 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1366 * config/i386/i386-opts.h (indirect_branch): New.
1367 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1368 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1369 with local indirect jump when converting indirect call and jump.
1370 (ix86_set_indirect_branch_type): New.
1371 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1372 (indirectlabelno): New.
1373 (indirect_thunk_needed): Likewise.
1374 (indirect_thunk_bnd_needed): Likewise.
1375 (indirect_thunks_used): Likewise.
1376 (indirect_thunks_bnd_used): Likewise.
1377 (INDIRECT_LABEL): Likewise.
1378 (indirect_thunk_name): Likewise.
1379 (output_indirect_thunk): Likewise.
1380 (output_indirect_thunk_function): Likewise.
1381 (ix86_output_indirect_branch_via_reg): Likewise.
1382 (ix86_output_indirect_branch_via_push): Likewise.
1383 (ix86_output_indirect_branch): Likewise.
1384 (ix86_output_indirect_jmp): Likewise.
1385 (ix86_code_end): Call output_indirect_thunk_function if needed.
1386 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1387 needed.
1388 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1389 (ix86_attribute_table): Add indirect_branch.
1390 * config/i386/i386.h (machine_function): Add indirect_branch_type
1391 and has_local_indirect_jump.
1392 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1393 to true.
1394 (tablejump): Likewise.
1395 (*indirect_jump): Use ix86_output_indirect_jmp.
1396 (*tablejump_1): Likewise.
1397 (simple_return_indirect_internal): Likewise.
1398 * config/i386/i386.opt (mindirect-branch=): New option.
1399 (indirect_branch): New.
1400 (keep): Likewise.
1401 (thunk): Likewise.
1402 (thunk-inline): Likewise.
1403 (thunk-extern): Likewise.
1404 * doc/extend.texi: Document indirect_branch function attribute.
1405 * doc/invoke.texi: Document -mindirect-branch= option.
1407 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1409 PR ipa/83051
1410 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1412 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1414 * ipa-inline.c (want_inline_small_function_p): Return false if
1415 inlining has already failed with CIF_FINAL_ERROR.
1416 (update_caller_keys): Call want_inline_small_function_p before
1417 can_inline_edge_p.
1418 (update_callee_keys): Likewise.
1420 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1422 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1423 New function.
1424 (rs6000_quadword_masked_address_p): Likewise.
1425 (quad_aligned_load_p): Likewise.
1426 (quad_aligned_store_p): Likewise.
1427 (const_load_sequence_p): Add comment to describe the outer-most loop.
1428 (mimic_memory_attributes_and_flags): New function.
1429 (rs6000_gen_stvx): Likewise.
1430 (replace_swapped_aligned_store): Likewise.
1431 (rs6000_gen_lvx): Likewise.
1432 (replace_swapped_aligned_load): Likewise.
1433 (replace_swapped_load_constant): Capitalize argument name in
1434 comment describing this function.
1435 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1436 and stores that access quad-word aligned addresses and replace
1437 with stvx or lvx instructions when appropriate.
1438 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1439 New function prototype.
1440 (rs6000_quadword_masked_address_p): Likewise.
1441 (rs6000_gen_lvx): Likewise.
1442 (rs6000_gen_stvx): Likewise.
1443 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1444 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1445 when memory address is aligned.
1446 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1447 this split to select lvx instruction when memory address is aligned.
1448 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1449 instruction when memory address is aligned.
1450 (*vsx_le_perm_load_v16qi): Likewise.
1451 (four unnamed splitters): Modify to select the stvx instruction
1452 when memory is aligned.
1454 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1456 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1457 which appears in the queue multiple times.
1459 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1460 Alan Hayward <alan.hayward@arm.com>
1461 David Sherwood <david.sherwood@arm.com>
1463 * tree-vectorizer.h (vec_lower_bound): New structure.
1464 (_loop_vec_info): Add check_nonzero and lower_bounds.
1465 (LOOP_VINFO_CHECK_NONZERO): New macro.
1466 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1467 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1468 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1469 fields. Make seg_len the distance travelled, not including the
1470 access size.
1471 (dr_direction_indicator): Declare.
1472 (dr_zero_step_indicator): Likewise.
1473 (dr_known_forward_stride_p): Likewise.
1474 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1475 tree-ssanames.h.
1476 (runtime_alias_check_p): Allow runtime alias checks with
1477 variable strides.
1478 (operator ==): Compare access_size and align.
1479 (prune_runtime_alias_test_list): Rework for new distinction between
1480 the access_size and seg_len.
1481 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1482 segment lengths.
1483 (get_segment_min_max): New function.
1484 (create_intersect_range_checks): Use it.
1485 (dr_step_indicator): New function.
1486 (dr_direction_indicator): Likewise.
1487 (dr_zero_step_indicator): Likewise.
1488 (dr_known_forward_stride_p): Likewise.
1489 * tree-loop-distribution.c (data_ref_segment_size): Return
1490 DR_STEP * (niters - 1).
1491 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1492 constructor.
1493 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1494 (vect_preserves_scalar_order_p): New function, split out from...
1495 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1496 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1497 (vect_vfa_access_size): New function.
1498 (vect_vfa_align): Likewise.
1499 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1500 (dump_lower_bound): New function.
1501 (vect_check_lower_bound): Likewise.
1502 (vect_small_gap_p): Likewise.
1503 (vectorizable_with_step_bound_p): Likewise.
1504 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1505 depencies if the vectorization factor is 1. Convert the checks
1506 for nonzero steps into checks on the bounds of DR_STEP. Try using
1507 a bunds check for variable steps if the minimum required step is
1508 relatively small. Update calls to the dr_with_seg_len
1509 constructor and to vect_compile_time_alias.
1510 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1511 function.
1512 (vect_loop_versioning): Call it.
1513 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1514 when retrying.
1515 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1517 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1518 Alan Hayward <alan.hayward@arm.com>
1519 David Sherwood <david.sherwood@arm.com>
1521 * doc/sourcebuild.texi (vect_scatter_store): Document.
1522 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1523 optabs.
1524 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1525 Document.
1526 * genopinit.c (main): Add supports_vec_scatter_store and
1527 supports_vec_scatter_store_cached to target_optabs.
1528 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1529 IFN_MASK_SCATTER_STORE.
1530 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1531 functions.
1532 * internal-fn.h (internal_store_fn_p): Declare.
1533 (internal_fn_stored_value_index): Likewise.
1534 * internal-fn.c (scatter_store_direct): New macro.
1535 (expand_scatter_store_optab_fn): New function.
1536 (direct_scatter_store_optab_supported_p): New macro.
1537 (internal_store_fn_p): New function.
1538 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1539 IFN_MASK_SCATTER_STORE.
1540 (internal_fn_mask_index): Likewise.
1541 (internal_fn_stored_value_index): New function.
1542 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1543 for scatter stores.
1544 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1545 * optabs-query.c (supports_vec_scatter_store_p): New function.
1546 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1547 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1548 true for scatter stores.
1549 (vect_gather_scatter_fn_p): Handle scatter stores too.
1550 (vect_check_gather_scatter): Consider using scatter stores if
1551 supports_vec_scatter_store_p.
1552 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1553 scatter stores too.
1554 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1555 internal_fn_stored_value_index.
1556 (check_load_store_masking): Handle scatter stores too.
1557 (vect_get_store_rhs): Make public.
1558 (vectorizable_call): Use internal_store_fn_p.
1559 (vectorizable_store): Handle scatter store internal functions.
1560 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1561 when deciding whether the end of the group has been reached.
1562 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1563 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1564 (mask_scatter_store<mode>): New insns.
1566 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1567 Alan Hayward <alan.hayward@arm.com>
1568 David Sherwood <david.sherwood@arm.com>
1570 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1571 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1572 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1573 function.
1574 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1575 Use vect_truncate_gather_scatter_offset if we can't treat the
1576 operation as a normal gather load or scatter store.
1577 (get_group_load_store_type): Take the gather_scatter_info
1578 as argument. Try using a gather load or scatter store for
1579 single-element groups.
1580 (get_load_store_type): Update calls to get_group_load_store_type
1581 and vect_use_strided_gather_scatters_p.
1583 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1584 Alan Hayward <alan.hayward@arm.com>
1585 David Sherwood <david.sherwood@arm.com>
1587 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1588 optional tree argument.
1589 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1590 null target hooks.
1591 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1592 but continue to use the current value as a fallback.
1593 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1594 to compare the updates.
1595 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1596 (get_load_store_type): Use it when handling a strided access.
1597 (vect_get_strided_load_store_ops): New function.
1598 (vect_get_data_ptr_increment): Likewise.
1599 (vectorizable_load): Handle strided gather loads. Always pass
1600 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1602 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1603 Alan Hayward <alan.hayward@arm.com>
1604 David Sherwood <david.sherwood@arm.com>
1606 * doc/md.texi (gather_load@var{m}): Document.
1607 (mask_gather_load@var{m}): Likewise.
1608 * genopinit.c (main): Add supports_vec_gather_load and
1609 supports_vec_gather_load_cached to target_optabs.
1610 * optabs-tree.c (init_tree_optimization_optabs): Use
1611 ggc_cleared_alloc to allocate target_optabs.
1612 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1613 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1614 functions.
1615 * internal-fn.h (internal_load_fn_p): Declare.
1616 (internal_gather_scatter_fn_p): Likewise.
1617 (internal_fn_mask_index): Likewise.
1618 (internal_gather_scatter_fn_supported_p): Likewise.
1619 * internal-fn.c (gather_load_direct): New macro.
1620 (expand_gather_load_optab_fn): New function.
1621 (direct_gather_load_optab_supported_p): New macro.
1622 (direct_internal_fn_optab): New function.
1623 (internal_load_fn_p): Likewise.
1624 (internal_gather_scatter_fn_p): Likewise.
1625 (internal_fn_mask_index): Likewise.
1626 (internal_gather_scatter_fn_supported_p): Likewise.
1627 * optabs-query.c (supports_at_least_one_mode_p): New function.
1628 (supports_vec_gather_load_p): Likewise.
1629 * optabs-query.h (supports_vec_gather_load_p): Declare.
1630 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1631 and memory_type field.
1632 (NUM_PATTERNS): Bump to 15.
1633 * tree-vect-data-refs.c: Include internal-fn.h.
1634 (vect_gather_scatter_fn_p): New function.
1635 (vect_describe_gather_scatter_call): Likewise.
1636 (vect_check_gather_scatter): Try using internal functions for
1637 gather loads. Recognize existing calls to a gather load function.
1638 (vect_analyze_data_refs): Consider using gather loads if
1639 supports_vec_gather_load_p.
1640 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1641 (vect_get_gather_scatter_offset_type): Likewise.
1642 (vect_convert_mask_for_vectype): Likewise.
1643 (vect_add_conversion_to_patterm): Likewise.
1644 (vect_try_gather_scatter_pattern): Likewise.
1645 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1646 (vect_vect_recog_func_ptrs): Add it.
1647 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1648 internal_fn_mask_index and internal_gather_scatter_fn_p.
1649 (check_load_store_masking): Take the gather_scatter_info as an
1650 argument and handle gather loads.
1651 (vect_get_gather_scatter_ops): New function.
1652 (vectorizable_call): Check internal_load_fn_p.
1653 (vectorizable_load): Likewise. Handle gather load internal
1654 functions.
1655 (vectorizable_store): Update call to check_load_store_masking.
1656 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1657 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1658 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1659 (aarch64_gather_scale_operand_d): New predicates.
1660 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1661 (mask_gather_load<mode>): New insns.
1663 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1664 Alan Hayward <alan.hayward@arm.com>
1665 David Sherwood <david.sherwood@arm.com>
1667 * optabs.def (fold_left_plus_optab): New optab.
1668 * doc/md.texi (fold_left_plus_@var{m}): Document.
1669 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1670 * internal-fn.c (fold_left_direct): Define.
1671 (expand_fold_left_optab_fn): Likewise.
1672 (direct_fold_left_optab_supported_p): Likewise.
1673 * fold-const-call.c (fold_const_fold_left): New function.
1674 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1675 * tree-parloops.c (valid_reduction_p): New function.
1676 (gather_scalar_reductions): Use it.
1677 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1678 (vect_finish_replace_stmt): Declare.
1679 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1680 (needs_fold_left_reduction_p): New function, split out from...
1681 (vect_is_simple_reduction): ...here. Accept reductions that
1682 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1683 (vect_force_simple_reduction): Also store the reduction type in
1684 the assignment's STMT_VINFO_REDUC_TYPE.
1685 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1686 (merge_with_identity): New function.
1687 (vect_expand_fold_left): Likewise.
1688 (vectorize_fold_left_reduction): Likewise.
1689 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1690 scalar phi in place for it. Check for target support and reject
1691 cases that would reassociate the operation. Defer the transform
1692 phase to vectorize_fold_left_reduction.
1693 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1694 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1695 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1697 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1699 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1700 call to ifc_temp_var.
1702 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1703 Alan Hayward <alan.hayward@arm.com>
1704 David Sherwood <david.sherwood@arm.com>
1706 * target.def (legitimize_address_displacement): Take the original
1707 offset as a poly_int.
1708 * targhooks.h (default_legitimize_address_displacement): Update
1709 accordingly.
1710 * targhooks.c (default_legitimize_address_displacement): Likewise.
1711 * doc/tm.texi: Regenerate.
1712 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1713 as an argument, moving assert of ad->disp == ad->disp_term to...
1714 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1715 Try calling targetm.legitimize_address_displacement before expanding
1716 the address rather than afterwards, and adjust for the new interface.
1717 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1718 Match the new hook interface. Handle SVE addresses.
1719 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1720 new hook interface.
1722 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1724 * Makefile.in (OBJS): Add early-remat.o.
1725 * target.def (select_early_remat_modes): New hook.
1726 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1727 * doc/tm.texi: Regenerate.
1728 * targhooks.h (default_select_early_remat_modes): Declare.
1729 * targhooks.c (default_select_early_remat_modes): New function.
1730 * timevar.def (TV_EARLY_REMAT): New timevar.
1731 * passes.def (pass_early_remat): New pass.
1732 * tree-pass.h (make_pass_early_remat): Declare.
1733 * early-remat.c: New file.
1734 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1735 function.
1736 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1738 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1739 Alan Hayward <alan.hayward@arm.com>
1740 David Sherwood <david.sherwood@arm.com>
1742 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1743 vfm1 with a bound_epilog parameter.
1744 (vect_do_peeling): Update calls accordingly, and move the prologue
1745 call earlier in the function. Treat the base bound_epilog as 0 for
1746 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1747 this base when peeling for gaps.
1748 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1749 with fully-masked loops.
1750 (vect_estimate_min_profitable_iters): Handle the single peeled
1751 iteration in that case.
1753 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1754 Alan Hayward <alan.hayward@arm.com>
1755 David Sherwood <david.sherwood@arm.com>
1757 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1758 single-element interleaving even if the size is not a power of 2.
1759 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1760 accesses for single-element interleaving if the group size is
1761 not a power of 2.
1763 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1764 Alan Hayward <alan.hayward@arm.com>
1765 David Sherwood <david.sherwood@arm.com>
1767 * doc/md.texi (fold_extract_last_@var{m}): Document.
1768 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1769 * optabs.def (fold_extract_last_optab): New optab.
1770 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1771 * internal-fn.c (fold_extract_direct): New macro.
1772 (expand_fold_extract_optab_fn): Likewise.
1773 (direct_fold_extract_optab_supported_p): Likewise.
1774 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1775 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1776 EXTRACT_LAST_REDUCTION.
1777 (get_initial_def_for_reduction): Do not create an initial vector
1778 for EXTRACT_LAST_REDUCTION reductions.
1779 (vectorizable_reduction): Leave the scalar phi in place for
1780 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1781 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1782 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1783 transform phase to vectorizable_condition.
1784 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1785 split out from...
1786 (vect_finish_stmt_generation): ...here.
1787 (vect_finish_replace_stmt): New function.
1788 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1789 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1790 pattern.
1791 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1793 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1794 Alan Hayward <alan.hayward@arm.com>
1795 David Sherwood <david.sherwood@arm.com>
1797 * doc/md.texi (extract_last_@var{m}): Document.
1798 * optabs.def (extract_last_optab): New optab.
1799 * internal-fn.def (EXTRACT_LAST): New internal function.
1800 * internal-fn.c (cond_unary_direct): New macro.
1801 (expand_cond_unary_optab_fn): Likewise.
1802 (direct_cond_unary_optab_supported_p): Likewise.
1803 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1804 loops using EXTRACT_LAST.
1805 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1806 (extract_last_<mode>): ...this optab.
1807 (vec_extract<mode><Vel>): Update accordingly.
1809 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1810 Alan Hayward <alan.hayward@arm.com>
1811 David Sherwood <david.sherwood@arm.com>
1813 * target.def (empty_mask_is_expensive): New hook.
1814 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
1815 * doc/tm.texi: Regenerate.
1816 * targhooks.h (default_empty_mask_is_expensive): Declare.
1817 * targhooks.c (default_empty_mask_is_expensive): New function.
1818 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
1819 if the target says that empty masks are expensive.
1820 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
1821 New function.
1822 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
1824 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1825 Alan Hayward <alan.hayward@arm.com>
1826 David Sherwood <david.sherwood@arm.com>
1828 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
1829 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
1830 (vect_use_loop_mask_for_alignment_p): New function.
1831 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
1832 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
1833 niters_skip argument. Make sure that the first niters_skip elements
1834 of the first iteration are inactive.
1835 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
1836 Update call to vect_set_loop_masks_directly.
1837 (get_misalign_in_elems): New function, split out from...
1838 (vect_gen_prolog_loop_niters): ...here.
1839 (vect_update_init_of_dr): Take a code argument that specifies whether
1840 the adjustment should be added or subtracted.
1841 (vect_update_init_of_drs): Likewise.
1842 (vect_prepare_for_masked_peels): New function.
1843 (vect_do_peeling): Skip prologue peeling if we're using a mask
1844 instead. Update call to vect_update_inits_of_drs.
1845 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1846 mask_skip_niters.
1847 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
1848 alignment. Do not include the number of peeled iterations in
1849 the minimum threshold in that case.
1850 (vectorizable_induction): Adjust the start value down by
1851 LOOP_VINFO_MASK_SKIP_NITERS iterations.
1852 (vect_transform_loop): Call vect_prepare_for_masked_peels.
1853 Take the number of skipped iterations into account when calculating
1854 the loop bounds.
1855 * tree-vect-stmts.c (vect_gen_while_not): New function.
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/sourcebuild.texi (vect_fully_masked): Document.
1862 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
1863 default value to 0.
1864 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
1865 split out from...
1866 (vect_analyze_loop_2): ...here. Don't check the vectorization
1867 factor against the number of loop iterations if the loop is
1868 fully-masked.
1870 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1871 Alan Hayward <alan.hayward@arm.com>
1872 David Sherwood <david.sherwood@arm.com>
1874 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
1875 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
1876 (dump_groups): Update accordingly.
1877 (iv_use::mem_type): New member variable.
1878 (address_p): New function.
1879 (record_use): Add a mem_type argument and initialize the new
1880 mem_type field.
1881 (record_group_use): Add a mem_type argument. Use address_p.
1882 Remove obsolete null checks of base_object. Update call to record_use.
1883 (find_interesting_uses_op): Update call to record_group_use.
1884 (find_interesting_uses_cond): Likewise.
1885 (find_interesting_uses_address): Likewise.
1886 (get_mem_type_for_internal_fn): New function.
1887 (find_address_like_use): Likewise.
1888 (find_interesting_uses_stmt): Try find_address_like_use before
1889 calling find_interesting_uses_op.
1890 (addr_offset_valid_p): Use the iv mem_type field as the type
1891 of the addressed memory.
1892 (add_autoinc_candidates): Likewise.
1893 (get_address_cost): Likewise.
1894 (split_small_address_groups_p): Use address_p.
1895 (split_address_groups): Likewise.
1896 (add_iv_candidate_for_use): Likewise.
1897 (autoinc_possible_for_pair): Likewise.
1898 (rewrite_groups): Likewise.
1899 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
1900 (determine_group_iv_cost): Update after split of USE_ADDRESS.
1901 (get_alias_ptr_type_for_ptr_address): New function.
1902 (rewrite_use_address): Rewrite address uses in calls that were
1903 identified by find_address_like_use.
1905 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1906 Alan Hayward <alan.hayward@arm.com>
1907 David Sherwood <david.sherwood@arm.com>
1909 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
1910 TARGET_MEM_REFs.
1911 * gimple-expr.h (is_gimple_addressable: Likewise.
1912 * gimple-expr.c (is_gimple_address): Likewise.
1913 * internal-fn.c (expand_call_mem_ref): New function.
1914 (expand_mask_load_optab_fn): Use it.
1915 (expand_mask_store_optab_fn): Likewise.
1917 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1918 Alan Hayward <alan.hayward@arm.com>
1919 David Sherwood <david.sherwood@arm.com>
1921 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
1922 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
1923 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
1924 (cond_umax@var{mode}): Document.
1925 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
1926 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
1927 (cond_umin_optab, cond_umax_optab): New optabs.
1928 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
1929 (COND_IOR, COND_XOR): New internal functions.
1930 * internal-fn.h (get_conditional_internal_fn): Declare.
1931 * internal-fn.c (cond_binary_direct): New macro.
1932 (expand_cond_binary_optab_fn): Likewise.
1933 (direct_cond_binary_optab_supported_p): Likewise.
1934 (get_conditional_internal_fn): New function.
1935 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
1936 Cope with reduction statements that are vectorized as calls rather
1937 than assignments.
1938 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
1939 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
1940 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
1941 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
1942 (UNSPEC_COND_EOR): New unspecs.
1943 (optab): Add mappings for them.
1944 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
1945 (sve_int_op, sve_fp_op): New int attributes.
1947 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1948 Alan Hayward <alan.hayward@arm.com>
1949 David Sherwood <david.sherwood@arm.com>
1951 * optabs.def (while_ult_optab): New optab.
1952 * doc/md.texi (while_ult@var{m}@var{n}): Document.
1953 * internal-fn.def (WHILE_ULT): New internal function.
1954 * internal-fn.h (direct_internal_fn_supported_p): New override
1955 that takes two types as argument.
1956 * internal-fn.c (while_direct): New macro.
1957 (expand_while_optab_fn): New function.
1958 (convert_optab_supported_p): Likewise.
1959 (direct_while_optab_supported_p): New macro.
1960 * wide-int.h (wi::udiv_ceil): New function.
1961 * tree-vectorizer.h (rgroup_masks): New structure.
1962 (vec_loop_masks): New typedef.
1963 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
1964 and fully_masked_p.
1965 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
1966 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
1967 (vect_max_vf): New function.
1968 (slpeel_make_loop_iterate_ntimes): Delete.
1969 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
1970 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
1971 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
1972 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
1973 internal-fn.h, stor-layout.h and optabs-query.h.
1974 (vect_set_loop_mask): New function.
1975 (add_preheader_seq): Likewise.
1976 (add_header_seq): Likewise.
1977 (interleave_supported_p): Likewise.
1978 (vect_maybe_permute_loop_masks): Likewise.
1979 (vect_set_loop_masks_directly): Likewise.
1980 (vect_set_loop_condition_masked): Likewise.
1981 (vect_set_loop_condition_unmasked): New function, split out from
1982 slpeel_make_loop_iterate_ntimes.
1983 (slpeel_make_loop_iterate_ntimes): Rename to..
1984 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
1985 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
1986 (vect_do_peeling): Update call accordingly.
1987 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
1988 loops.
1989 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1990 mask_compare_type, can_fully_mask_p and fully_masked_p.
1991 (release_vec_loop_masks): New function.
1992 (_loop_vec_info): Use it to free the loop masks.
1993 (can_produce_all_loop_masks_p): New function.
1994 (vect_get_max_nscalars_per_iter): Likewise.
1995 (vect_verify_full_masking): Likewise.
1996 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
1997 retries, and free the mask rgroups before retrying. Check loop-wide
1998 reasons for disallowing fully-masked loops. Make the final decision
1999 about whether use a fully-masked loop or not.
2000 (vect_estimate_min_profitable_iters): Do not assume that peeling
2001 for the number of iterations will be needed for fully-masked loops.
2002 (vectorizable_reduction): Disable fully-masked loops.
2003 (vectorizable_live_operation): Likewise.
2004 (vect_halve_mask_nunits): New function.
2005 (vect_double_mask_nunits): Likewise.
2006 (vect_record_loop_mask): Likewise.
2007 (vect_get_loop_mask): Likewise.
2008 (vect_transform_loop): Handle the case in which the final loop
2009 iteration might handle a partial vector. Call vect_set_loop_condition
2010 instead of slpeel_make_loop_iterate_ntimes.
2011 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2012 (check_load_store_masking): New function.
2013 (prepare_load_store_mask): Likewise.
2014 (vectorizable_store): Handle fully-masked loops.
2015 (vectorizable_load): Likewise.
2016 (supportable_widening_operation): Use vect_halve_mask_nunits for
2017 booleans.
2018 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2019 (vect_gen_while): New function.
2020 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2021 (aarch64_uqdec<mode>): New insn.
2023 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2024 Alan Hayward <alan.hayward@arm.com>
2025 David Sherwood <david.sherwood@arm.com>
2027 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2028 (reduc_xor_scal_optab): New optabs.
2029 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2030 (reduc_xor_scal_@var{m}): Document.
2031 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2032 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2033 internal functions.
2034 * fold-const-call.c (fold_const_call): Handle them.
2035 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2036 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2037 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2038 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2039 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2040 (UNSPEC_XORV): New unspecs.
2041 (optab): Add entries for them.
2042 (BITWISEV): New int iterator.
2043 (bit_reduc_op): New int attributes.
2045 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2046 Alan Hayward <alan.hayward@arm.com>
2047 David Sherwood <david.sherwood@arm.com>
2049 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2050 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2051 * optabs.def (vec_shl_insert_optab): New optab.
2052 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2053 (duplicate_and_interleave): Likewise.
2054 * tree-vect-loop.c: Include internal-fn.h.
2055 (neutral_op_for_slp_reduction): New function, split out from
2056 get_initial_defs_for_reduction.
2057 (get_initial_def_for_reduction): Handle option 2 for variable-length
2058 vectors by loading the neutral value into a vector and then shifting
2059 the initial value into element 0.
2060 (get_initial_defs_for_reduction): Replace the code argument with
2061 the neutral value calculated by neutral_op_for_slp_reduction.
2062 Use gimple_build_vector for constant-length vectors.
2063 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2064 but the first group_size elements have a neutral value.
2065 Use duplicate_and_interleave otherwise.
2066 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2067 Update call to get_initial_defs_for_reduction. Handle SLP
2068 reductions for variable-length vectors by creating one vector
2069 result for each scalar result, with the elements associated
2070 with other scalar results stubbed out with the neutral value.
2071 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2072 Require IFN_VEC_SHL_INSERT for double reductions on
2073 variable-length vectors, or SLP reductions that have
2074 a neutral value. Require can_duplicate_and_interleave_p
2075 support for variable-length unchained SLP reductions if there
2076 is no neutral value, such as for MIN/MAX reductions. Also require
2077 the number of vector elements to be a multiple of the number of
2078 SLP statements when doing variable-length unchained SLP reductions.
2079 Update call to vect_create_epilog_for_reduction.
2080 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2081 and remove initial values.
2082 (duplicate_and_interleave): Make public.
2083 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2084 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2086 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2087 Alan Hayward <alan.hayward@arm.com>
2088 David Sherwood <david.sherwood@arm.com>
2090 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2091 (can_duplicate_and_interleave_p): New function.
2092 (vect_get_and_check_slp_defs): Take the vector of statements
2093 rather than just the current one. Remove excess parentheses.
2094 Restriction rejectinon of vect_constant_def and vect_external_def
2095 for variable-length vectors to boolean types, or types for which
2096 can_duplicate_and_interleave_p is false.
2097 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2098 (duplicate_and_interleave): New function.
2099 (vect_get_constant_vectors): Use gimple_build_vector for
2100 constant-length vectors and suitable variable-length constant
2101 vectors. Use duplicate_and_interleave for other variable-length
2102 vectors. Don't defer the update when inserting new statements.
2104 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2105 Alan Hayward <alan.hayward@arm.com>
2106 David Sherwood <david.sherwood@arm.com>
2108 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2109 min_profitable_iters doesn't go negative.
2111 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2112 Alan Hayward <alan.hayward@arm.com>
2113 David Sherwood <david.sherwood@arm.com>
2115 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2116 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2117 * optabs.def (vec_mask_load_lanes_optab): New optab.
2118 (vec_mask_store_lanes_optab): Likewise.
2119 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2120 (MASK_STORE_LANES): Likewise.
2121 * internal-fn.c (mask_load_lanes_direct): New macro.
2122 (mask_store_lanes_direct): Likewise.
2123 (expand_mask_load_optab_fn): Handle masked operations.
2124 (expand_mask_load_lanes_optab_fn): New macro.
2125 (expand_mask_store_optab_fn): Handle masked operations.
2126 (expand_mask_store_lanes_optab_fn): New macro.
2127 (direct_mask_load_lanes_optab_supported_p): Likewise.
2128 (direct_mask_store_lanes_optab_supported_p): Likewise.
2129 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2130 parameter.
2131 (vect_load_lanes_supported): Likewise.
2132 * tree-vect-data-refs.c (strip_conversion): New function.
2133 (can_group_stmts_p): Likewise.
2134 (vect_analyze_data_ref_accesses): Use it instead of checking
2135 for a pair of assignments.
2136 (vect_store_lanes_supported): Take a masked_p parameter.
2137 (vect_load_lanes_supported): Likewise.
2138 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2139 vect_store_lanes_supported and vect_load_lanes_supported.
2140 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2141 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2142 parameter. Don't allow gaps for masked accesses.
2143 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2144 and vect_load_lanes_supported.
2145 (get_load_store_type): Take a masked_p parameter and update
2146 call to get_group_load_store_type.
2147 (vectorizable_store): Update call to get_load_store_type.
2148 Handle IFN_MASK_STORE_LANES.
2149 (vectorizable_load): Update call to get_load_store_type.
2150 Handle IFN_MASK_LOAD_LANES.
2152 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2153 Alan Hayward <alan.hayward@arm.com>
2154 David Sherwood <david.sherwood@arm.com>
2156 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2157 modes for SVE.
2158 * config/aarch64/aarch64-protos.h
2159 (aarch64_sve_struct_memory_operand_p): Declare.
2160 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2161 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2162 (VPRED, vpred): Handle SVE structure modes.
2163 * config/aarch64/constraints.md (Utx): New constraint.
2164 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2165 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2166 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2167 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2168 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2169 structure modes. Split into pieces after RA.
2170 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2171 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2172 New patterns.
2173 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2174 SVE structure modes.
2175 (aarch64_classify_address): Likewise.
2176 (sizetochar): Move earlier in file.
2177 (aarch64_print_operand): Handle SVE register lists.
2178 (aarch64_array_mode): New function.
2179 (aarch64_sve_struct_memory_operand_p): Likewise.
2180 (TARGET_ARRAY_MODE): Redefine.
2182 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2183 Alan Hayward <alan.hayward@arm.com>
2184 David Sherwood <david.sherwood@arm.com>
2186 * target.def (array_mode): New target hook.
2187 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2188 * doc/tm.texi: Regenerate.
2189 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2190 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2191 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2192 targetm.array_mode.
2193 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2194 type sizes.
2196 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2197 Alan Hayward <alan.hayward@arm.com>
2198 David Sherwood <david.sherwood@arm.com>
2200 * fold-const.c (fold_binary_loc): Check the argument types
2201 rather than the result type when testing for a vector operation.
2203 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2205 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2206 * doc/tm.texi: Regenerate.
2208 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2209 Alan Hayward <alan.hayward@arm.com>
2210 David Sherwood <david.sherwood@arm.com>
2212 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2213 (sve): Document new AArch64 extension.
2214 * doc/md.texi (w): Extend the description of the AArch64
2215 constraint to include SVE vectors.
2216 (Upl, Upa): Document new AArch64 predicate constraints.
2217 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2218 enum.
2219 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2220 (msve-vector-bits=): New option.
2221 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2222 SVE when these are disabled.
2223 (sve): New extension.
2224 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2225 modes. Adjust their number of units based on aarch64_sve_vg.
2226 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2227 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2228 aarch64_addr_query_type.
2229 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2230 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2231 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2232 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2233 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2234 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2235 (aarch64_simd_imm_zero_p): Delete.
2236 (aarch64_check_zero_based_sve_index_immediate): Declare.
2237 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2238 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2239 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2240 (aarch64_sve_float_mul_immediate_p): Likewise.
2241 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2242 rather than an rtx.
2243 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2244 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2245 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2246 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2247 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2248 (aarch64_regmode_natural_size): Likewise.
2249 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2250 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2251 left one place.
2252 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2253 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2254 for VG and the SVE predicate registers.
2255 (V_ALIASES): Add a "z"-prefixed alias.
2256 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2257 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2258 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2259 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2260 (REG_CLASS_NAMES): Add entries for them.
2261 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2262 and the predicate registers.
2263 (aarch64_sve_vg): Declare.
2264 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2265 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2266 (REGMODE_NATURAL_SIZE): Define.
2267 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2268 SVE macros.
2269 * config/aarch64/aarch64.c: Include cfgrtl.h.
2270 (simd_immediate_info): Add a constructor for series vectors,
2271 and an associated step field.
2272 (aarch64_sve_vg): New variable.
2273 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2274 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2275 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2276 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2277 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2278 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2279 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2280 (aarch64_get_mask_mode): New functions.
2281 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2282 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2283 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2284 predicate modes and predicate registers. Explicitly restrict
2285 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2286 to store a vector mode if it is recognized by
2287 aarch64_classify_vector_mode.
2288 (aarch64_regmode_natural_size): New function.
2289 (aarch64_hard_regno_caller_save_mode): Return the original mode
2290 for predicates.
2291 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2292 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2293 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2294 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2295 functions.
2296 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2297 does not overlap dest if the function is frame-related. Handle
2298 SVE constants.
2299 (aarch64_split_add_offset): New function.
2300 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2301 them aarch64_add_offset.
2302 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2303 and update call to aarch64_sub_sp.
2304 (aarch64_add_cfa_expression): New function.
2305 (aarch64_expand_prologue): Pass extra temporary registers to the
2306 functions above. Handle the case in which we need to emit new
2307 DW_CFA_expressions for registers that were originally saved
2308 relative to the stack pointer, but now have to be expressed
2309 relative to the frame pointer.
2310 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2311 functions above.
2312 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2313 IP0 and IP1 values for SVE frames.
2314 (aarch64_expand_vec_series): New function.
2315 (aarch64_expand_sve_widened_duplicate): Likewise.
2316 (aarch64_expand_sve_const_vector): Likewise.
2317 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2318 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2319 into the register, rather than emitting a SET directly.
2320 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2321 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2322 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2323 (offset_9bit_signed_scaled_p): New functions.
2324 (aarch64_replicate_bitmask_imm): New function.
2325 (aarch64_bitmask_imm): Use it.
2326 (aarch64_cannot_force_const_mem): Reject expressions involving
2327 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2328 (aarch64_classify_index): Handle SVE indices, by requiring
2329 a plain register index with a scale that matches the element size.
2330 (aarch64_classify_address): Handle SVE addresses. Assert that
2331 the mode of the address is VOIDmode or an integer mode.
2332 Update call to aarch64_classify_symbol.
2333 (aarch64_classify_symbolic_expression): Update call to
2334 aarch64_classify_symbol.
2335 (aarch64_const_vec_all_in_range_p): New function.
2336 (aarch64_print_vector_float_operand): Likewise.
2337 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2338 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2339 and the FP immediates 1.0 and 0.5.
2340 (aarch64_print_address_internal): Handle SVE addresses.
2341 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2342 (aarch64_regno_regclass): Handle predicate registers.
2343 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2344 data modes.
2345 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2346 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2347 (aarch64_convert_sve_vector_bits): New function.
2348 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2349 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2350 rather than an rtx.
2351 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2352 Handle SVE vector and predicate modes. Accept VL-based constants
2353 that need only one temporary register, and VL offsets that require
2354 no temporary registers.
2355 (aarch64_conditional_register_usage): Mark the predicate registers
2356 as fixed if SVE isn't available.
2357 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2358 Return true for SVE vector and predicate modes.
2359 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2360 rather than an unsigned int. Handle SVE modes.
2361 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2362 SVE modes.
2363 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2364 if SVE is enabled.
2365 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2366 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2367 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2368 (aarch64_sve_float_mul_immediate_p): New functions.
2369 (aarch64_sve_valid_immediate): New function.
2370 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2371 Explicitly reject structure modes. Check for INDEX constants.
2372 Handle PTRUE and PFALSE constants.
2373 (aarch64_check_zero_based_sve_index_immediate): New function.
2374 (aarch64_simd_imm_zero_p): Delete.
2375 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2376 vector modes. Accept constants in the range of CNT[BHWD].
2377 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2378 ask for an Advanced SIMD mode.
2379 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2380 (aarch64_simd_vector_alignment): Handle SVE predicates.
2381 (aarch64_vectorize_preferred_vector_alignment): New function.
2382 (aarch64_simd_vector_alignment_reachable): Use it instead of
2383 the vector size.
2384 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2385 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2386 functions.
2387 (MAX_VECT_LEN): Delete.
2388 (expand_vec_perm_d): Add a vec_flags field.
2389 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2390 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2391 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2392 for SVE modes.
2393 (aarch64_evpc_rev): Rename to...
2394 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2395 (aarch64_evpc_rev_global): New function.
2396 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2397 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2398 MAX_VECT_LEN.
2399 (aarch64_evpc_sve_tbl): New function.
2400 (aarch64_expand_vec_perm_const_1): Update after rename of
2401 aarch64_evpc_rev. Handle SVE permutes too, trying
2402 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2403 than aarch64_evpc_tbl.
2404 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2405 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2406 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2407 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2408 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2409 (aarch64_expand_sve_vcond): New functions.
2410 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2411 of aarch64_vector_mode_p.
2412 (aarch64_dwarf_poly_indeterminate_value): New function.
2413 (aarch64_compute_pressure_classes): Likewise.
2414 (aarch64_can_change_mode_class): Likewise.
2415 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2416 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2417 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2418 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2419 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2420 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2421 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2422 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2423 constraints.
2424 (Dn, Dl, Dr): Accept const as well as const_vector.
2425 (Dz): Likewise. Compare against CONST0_RTX.
2426 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2427 of "vector" where appropriate.
2428 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2429 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2430 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2431 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2432 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2433 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2434 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2435 (v_int_equiv): Extend to SVE modes.
2436 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2437 mode attributes.
2438 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2439 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2440 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2441 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2442 (SVE_COND_FP_CMP): New int iterators.
2443 (perm_hilo): Handle the new unpack unspecs.
2444 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2445 attributes.
2446 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2447 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2448 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2449 (aarch64_equality_operator, aarch64_constant_vector_operand)
2450 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2451 (aarch64_sve_nonimmediate_operand): Likewise.
2452 (aarch64_sve_general_operand): Likewise.
2453 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2454 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2455 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2456 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2457 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2458 (aarch64_sve_float_arith_immediate): Likewise.
2459 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2460 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2461 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2462 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2463 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2464 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2465 (aarch64_sve_float_arith_operand): Likewise.
2466 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2467 (aarch64_sve_float_mul_operand): Likewise.
2468 (aarch64_sve_vec_perm_operand): Likewise.
2469 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2470 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2471 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2472 as well as const_vector.
2473 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2474 in file. Use CONST0_RTX and CONSTM1_RTX.
2475 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2476 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2477 Use aarch64_simd_imm_zero.
2478 * config/aarch64/aarch64-sve.md: New file.
2479 * config/aarch64/aarch64.md: Include it.
2480 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2481 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2482 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2483 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2484 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2485 (sve): New attribute.
2486 (enabled): Disable instructions with the sve attribute unless
2487 TARGET_SVE.
2488 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2489 aarch64_expand_mov_immediate.
2490 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2491 CNT[BHSD] immediates.
2492 (movti): Split CONST_POLY_INT moves into two halves.
2493 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2494 Split additions that need a temporary here if the destination
2495 is the stack pointer.
2496 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2497 (*add<mode>3_poly_1): New instruction.
2498 (set_clobber_cc): New expander.
2500 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2502 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2503 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2504 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2505 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2506 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2507 Change innermode from fixed_mode_size to machine_mode.
2508 (simplify_subreg): Update call accordingly. Handle a constant-sized
2509 subreg of a variable-length CONST_VECTOR.
2511 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2512 Alan Hayward <alan.hayward@arm.com>
2513 David Sherwood <david.sherwood@arm.com>
2515 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2516 (add_offset_to_base): New function, split out from...
2517 (create_mem_ref): ...here. When handling a scale other than 1,
2518 check first whether the address is valid without the offset.
2519 Add it into the base if so, leaving the index and scale as-is.
2521 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2523 PR c++/83778
2524 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2525 fold_for_warn before checking if arg2 is INTEGER_CST.
2527 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2529 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2530 (store_multiple_operation): Delete.
2531 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2532 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2533 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2534 guarded by TARGET_STRING.
2535 (rs6000_output_load_multiple): Delete.
2536 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2537 OPTION_MASK_STRING / TARGET_STRING handling.
2538 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2539 (const rs6000_opt_masks) <"string">: Change mask to 0.
2540 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2541 (MASK_STRING): Delete.
2542 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2543 parts. Simplify.
2544 (load_multiple): Delete.
2545 (*ldmsi8): Delete.
2546 (*ldmsi7): Delete.
2547 (*ldmsi6): Delete.
2548 (*ldmsi5): Delete.
2549 (*ldmsi4): Delete.
2550 (*ldmsi3): Delete.
2551 (store_multiple): Delete.
2552 (*stmsi8): Delete.
2553 (*stmsi7): Delete.
2554 (*stmsi6): Delete.
2555 (*stmsi5): Delete.
2556 (*stmsi4): Delete.
2557 (*stmsi3): Delete.
2558 (movmemsi_8reg): Delete.
2559 (corresponding unnamed define_insn): Delete.
2560 (movmemsi_6reg): Delete.
2561 (corresponding unnamed define_insn): Delete.
2562 (movmemsi_4reg): Delete.
2563 (corresponding unnamed define_insn): Delete.
2564 (movmemsi_2reg): Delete.
2565 (corresponding unnamed define_insn): Delete.
2566 (movmemsi_1reg): Delete.
2567 (corresponding unnamed define_insn): Delete.
2568 * config/rs6000/rs6000.opt (mno-string): New.
2569 (mstring): Replace by deprecation warning stub.
2570 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2572 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2574 * regrename.c (regrename_do_replace): If replacing the same
2575 reg multiple times, try to reuse last created gen_raw_REG.
2577 PR debug/81155
2578 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2579 main to workaround a bug in GDB.
2581 2018-01-12 Tom de Vries <tom@codesourcery.com>
2583 PR target/83737
2584 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2586 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2588 PR rtl-optimization/80481
2589 * ira-color.c (get_cap_member): New function.
2590 (allocnos_conflict_by_live_ranges_p): Use it.
2591 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2592 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2594 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2596 PR target/83628
2597 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2598 (*saddl_se_1): Ditto.
2599 (*ssubsi_1): Ditto.
2600 (*ssubl_se_1): Ditto.
2602 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2604 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2605 rather than wi::to_widest for DR_INITs.
2606 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2607 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2608 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2609 INTEGER_CSTs.
2610 (vect_analyze_group_access_1): Note that here.
2612 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2614 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2615 polynomial type sizes.
2617 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2619 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2620 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2621 (gimple_add_tmp_var): Likewise.
2623 2018-01-12 Martin Liska <mliska@suse.cz>
2625 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2626 (gimple_alloc_sizes): Likewise.
2627 (dump_gimple_statistics): Use PRIu64 in printf format.
2628 * gimple.h: Change uint64_t to int.
2630 2018-01-12 Martin Liska <mliska@suse.cz>
2632 * tree-core.h: Use uint64_t instead of int.
2633 * tree.c (tree_node_counts): Likewise.
2634 (tree_node_sizes): Likewise.
2635 (dump_tree_statistics): Use PRIu64 in printf format.
2637 2018-01-12 Martin Liska <mliska@suse.cz>
2639 * Makefile.in: As qsort_chk is implemented in vec.c, add
2640 vec.o to linkage of gencfn-macros.
2641 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2642 passing the info to record_node_allocation_statistics.
2643 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2644 and pass the info.
2645 * ggc-common.c (struct ggc_usage): Add operator== and use
2646 it in operator< and compare function.
2647 * mem-stats.h (struct mem_usage): Likewise.
2648 * vec.c (struct vec_usage): Remove operator< and compare
2649 function. Can be simply inherited.
2651 2018-01-12 Martin Jambor <mjambor@suse.cz>
2653 PR target/81616
2654 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2655 * tree-ssa-math-opts.c: Include domwalk.h.
2656 (convert_mult_to_fma_1): New function.
2657 (fma_transformation_info): New type.
2658 (fma_deferring_state): Likewise.
2659 (cancel_fma_deferring): New function.
2660 (result_of_phi): Likewise.
2661 (last_fma_candidate_feeds_initial_phi): Likewise.
2662 (convert_mult_to_fma): Added deferring logic, split actual
2663 transformation to convert_mult_to_fma_1.
2664 (math_opts_dom_walker): New type.
2665 (math_opts_dom_walker::after_dom_children): New method, body moved
2666 here from pass_optimize_widening_mul::execute, added deferring logic
2667 bits.
2668 (pass_optimize_widening_mul::execute): Moved most of code to
2669 math_opts_dom_walker::after_dom_children.
2670 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2671 * config/i386/i386.c (ix86_option_override_internal): Added
2672 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2674 2018-01-12 Richard Biener <rguenther@suse.de>
2676 PR debug/83157
2677 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2678 inline instance vars.
2680 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2682 PR target/81819
2683 * config/rx/rx.c (rx_is_restricted_memory_address):
2684 Handle SUBREG case.
2686 2018-01-12 Richard Biener <rguenther@suse.de>
2688 PR tree-optimization/80846
2689 * target.def (split_reduction): New target hook.
2690 * targhooks.c (default_split_reduction): New function.
2691 * targhooks.h (default_split_reduction): Declare.
2692 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2693 target requests first reduce vectors by combining low and high
2694 parts.
2695 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2696 (get_vectype_for_scalar_type_and_size): Export.
2697 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2698 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2699 * doc/tm.texi: Regenerate.
2700 * config/i386/i386.c (ix86_split_reduction): Implement
2701 TARGET_VECTORIZE_SPLIT_REDUCTION.
2703 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2705 PR target/83368
2706 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2707 in PIC mode except for TARGET_VXWORKS_RTP.
2708 * config/sparc/sparc.c: Include cfgrtl.h.
2709 (TARGET_INIT_PIC_REG): Define.
2710 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2711 (sparc_pic_register_p): New predicate.
2712 (sparc_legitimate_address_p): Use it.
2713 (sparc_legitimize_pic_address): Likewise.
2714 (sparc_delegitimize_address): Likewise.
2715 (sparc_mode_dependent_address_p): Likewise.
2716 (gen_load_pcrel_sym): Remove 4th parameter.
2717 (load_got_register): Adjust call to above. Remove obsolete stuff.
2718 (sparc_expand_prologue): Do not call load_got_register here.
2719 (sparc_flat_expand_prologue): Likewise.
2720 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2721 (sparc_use_pseudo_pic_reg): New function.
2722 (sparc_init_pic_reg): Likewise.
2723 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2724 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2726 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2728 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2729 Add item for branch_cost.
2731 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2733 PR rtl-optimization/83565
2734 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2735 not extend the result to a larger mode for rotate operations.
2736 (num_sign_bit_copies1): Likewise.
2738 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2740 PR target/40411
2741 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2742 -symbolic.
2743 Use values-Xc.o for -pedantic.
2744 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2746 2018-01-12 Martin Liska <mliska@suse.cz>
2748 PR ipa/83054
2749 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2750 New function.
2751 (possible_polymorphic_call_targets): Use it.
2752 (ipa_devirt): Likewise.
2754 2018-01-12 Martin Liska <mliska@suse.cz>
2756 * profile-count.h (enum profile_quality): Use 0 as invalid
2757 enum value of profile_quality.
2759 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2761 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2762 -mext-string options.
2764 2018-01-12 Richard Biener <rguenther@suse.de>
2766 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2767 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2768 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2769 Likewise.
2770 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2772 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2774 * configure.ac (--with-long-double-format): Add support for the
2775 configuration option to change the default long double format on
2776 PowerPC systems.
2777 * config.gcc (powerpc*-linux*-*): Likewise.
2778 * configure: Regenerate.
2779 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2780 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2781 used without modification.
2783 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2785 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2786 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2787 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2788 MISC_BUILTIN_SPEC_BARRIER.
2789 (rs6000_init_builtins): Likewise.
2790 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2791 enum value.
2792 (speculation_barrier): New define_insn.
2793 * doc/extend.texi: Document __builtin_speculation_barrier.
2795 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2797 PR target/83203
2798 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2799 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2800 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2801 iterators.
2802 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2803 integral modes instead of "ss" and "sd".
2804 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2805 vectors with 32-bit and 64-bit elements.
2806 (vecdupssescalarmodesuffix): New mode attribute.
2807 (vec_dup<mode>): Use it.
2809 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2811 PR target/83330
2812 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2813 frame if argument is passed on stack.
2815 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2817 PR target/82682
2818 * ree.c (combine_reaching_defs): Optimize also
2819 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
2820 reg2=any_extend(exp); reg1=reg2;, formatting fix.
2822 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2824 PR middle-end/83189
2825 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
2827 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2829 PR middle-end/83718
2830 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
2831 after they are computed.
2833 2018-01-11 Bin Cheng <bin.cheng@arm.com>
2835 PR tree-optimization/83695
2836 * gimple-loop-linterchange.cc
2837 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
2838 reset cached scev information after interchange.
2839 (pass_linterchange::execute): Remove call to scev_reset_htab.
2841 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2843 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
2844 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
2845 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
2846 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
2847 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
2848 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
2849 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
2850 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
2851 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
2852 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
2853 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
2854 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
2855 (V_lane_reg): Likewise.
2856 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
2857 New define_expand.
2858 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
2859 (vfmal_lane_low<mode>_intrinsic,
2860 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
2861 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
2862 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
2863 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
2864 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
2865 vfmsl_lane_high<mode>_intrinsic): New define_insns.
2867 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2869 * config/arm/arm-cpus.in (fp16fml): New feature.
2870 (ALL_SIMD): Add fp16fml.
2871 (armv8.2-a): Add fp16fml as an option.
2872 (armv8.3-a): Likewise.
2873 (armv8.4-a): Add fp16fml as part of fp16.
2874 * config/arm/arm.h (TARGET_FP16FML): Define.
2875 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
2876 when appropriate.
2877 * config/arm/arm-modes.def (V2HF): Define.
2878 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
2879 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
2880 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
2881 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
2882 vfmsl_low, vfmsl_high): New set of builtins.
2883 * config/arm/iterators.md (PLUSMINUS): New code iterator.
2884 (vfml_op): New code attribute.
2885 (VFMLHALVES): New int iterator.
2886 (VFML, VFMLSEL): New mode attributes.
2887 (V_reg): Define mapping for V2HF.
2888 (V_hi, V_lo): New mode attributes.
2889 (VF_constraint): Likewise.
2890 (vfml_half, vfml_half_selector): New int attributes.
2891 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
2892 define_expand.
2893 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
2894 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
2895 New define_insn.
2896 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
2897 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
2898 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
2899 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
2900 documentation.
2901 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
2902 Document new effective target and option set.
2904 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2906 * config/arm/arm-cpus.in (armv8_4): New feature.
2907 (ARMv8_4a): New fgroup.
2908 (armv8.4-a): New arch.
2909 * config/arm/arm-tables.opt: Regenerate.
2910 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
2911 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
2912 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
2913 Add matching rules for -march=armv8.4-a and extensions.
2914 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
2916 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
2918 PR target/81821
2919 * config/rx/rx.md (BW): New mode attribute.
2920 (sync_lock_test_and_setsi): Add mode suffix to insn output.
2922 2018-01-11 Richard Biener <rguenther@suse.de>
2924 PR tree-optimization/83435
2925 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
2926 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
2927 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
2929 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2930 Alan Hayward <alan.hayward@arm.com>
2931 David Sherwood <david.sherwood@arm.com>
2933 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
2934 field.
2935 (aarch64_classify_address): Initialize it. Track polynomial offsets.
2936 (aarch64_print_address_internal): Use it to check for a zero offset.
2938 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2939 Alan Hayward <alan.hayward@arm.com>
2940 David Sherwood <david.sherwood@arm.com>
2942 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
2943 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
2944 Return a poly_int64 rather than a HOST_WIDE_INT.
2945 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
2946 rather than a HOST_WIDE_INT.
2947 * config/aarch64/aarch64.h (aarch64_frame): Protect with
2948 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
2949 hard_fp_offset, frame_size, initial_adjust, callee_offset and
2950 final_offset from HOST_WIDE_INT to poly_int64.
2951 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
2952 to_constant when getting the number of units in an Advanced SIMD
2953 mode.
2954 (aarch64_builtin_vectorized_function): Check for a constant number
2955 of units.
2956 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
2957 GET_MODE_SIZE.
2958 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
2959 attribute instead of GET_MODE_NUNITS.
2960 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
2961 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
2962 GET_MODE_SIZE for fixed-size registers.
2963 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
2964 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
2965 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
2966 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
2967 (aarch64_print_operand, aarch64_print_address_internal)
2968 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
2969 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
2970 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
2971 Handle polynomial GET_MODE_SIZE.
2972 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
2973 wider than SImode without modification.
2974 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
2975 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
2976 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
2977 passing and returning SVE modes.
2978 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
2979 rather than GEN_INT.
2980 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
2981 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
2982 (aarch64_allocate_and_probe_stack_space): Likewise.
2983 (aarch64_layout_frame): Cope with polynomial offsets.
2984 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
2985 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
2986 polynomial offsets.
2987 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
2988 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
2989 poly_int64 rather than a HOST_WIDE_INT.
2990 (aarch64_get_separate_components, aarch64_process_components)
2991 (aarch64_expand_prologue, aarch64_expand_epilogue)
2992 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
2993 (aarch64_anchor_offset): New function, split out from...
2994 (aarch64_legitimize_address): ...here.
2995 (aarch64_builtin_vectorization_cost): Handle polynomial
2996 TYPE_VECTOR_SUBPARTS.
2997 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
2998 GET_MODE_NUNITS.
2999 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3000 number of elements from the PARALLEL rather than the mode.
3001 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3002 rather than GET_MODE_BITSIZE.
3003 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3004 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3005 (aarch64_expand_vec_perm_const_1): Handle polynomial
3006 d->perm.length () and d->perm elements.
3007 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3008 Apply to_constant to d->perm elements.
3009 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3010 polynomial CONST_VECTOR_NUNITS.
3011 (aarch64_move_pointer): Take amount as a poly_int64 rather
3012 than an int.
3013 (aarch64_progress_pointer): Avoid temporary variable.
3014 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3015 the mode attribute instead of GET_MODE.
3017 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3018 Alan Hayward <alan.hayward@arm.com>
3019 David Sherwood <david.sherwood@arm.com>
3021 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3022 x exists before using it.
3023 (aarch64_add_constant_internal): Rename to...
3024 (aarch64_add_offset_1): ...this. Replace regnum with separate
3025 src and dest rtxes. Handle the case in which they're different,
3026 including when the offset is zero. Replace scratchreg with an rtx.
3027 Use 2 additions if there is no spare register into which we can
3028 move a 16-bit constant.
3029 (aarch64_add_constant): Delete.
3030 (aarch64_add_offset): Replace reg with separate src and dest
3031 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3032 Use aarch64_add_offset_1.
3033 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3034 an rtx rather than an int. Take the delta as a poly_int64
3035 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3036 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3037 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3038 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3039 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3040 and aarch64_add_sp.
3041 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3042 aarch64_add_constant.
3044 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3046 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3047 Use scalar_float_mode.
3049 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3051 * config/aarch64/aarch64-simd.md
3052 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3053 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3054 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3055 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3056 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3057 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3058 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3059 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3060 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3061 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3063 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3065 PR target/83514
3066 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3067 targ_options->x_arm_arch_string is non NULL.
3069 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3071 * config/aarch64/aarch64.h
3072 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3074 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3076 PR target/82096
3077 * expmed.c (emit_store_flag_force): Swap if const op0
3078 and change VOIDmode to mode of op0.
3080 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3082 PR rtl-optimization/83761
3083 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3084 than bytes to mode_for_size.
3086 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3088 PR middle-end/83189
3089 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3090 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3091 profile.
3093 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3095 PR middle-end/83575
3096 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3097 when in layout mode.
3098 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3099 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3100 partition fixup.
3102 2018-01-10 Michael Collison <michael.collison@arm.com>
3104 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3105 * config/aarch64/aarch64-option-extension.def: Add
3106 AARCH64_OPT_EXTENSION of 'fp16fml'.
3107 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3108 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3109 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3110 * config/aarch64/constraints.md (Ui7): New constraint.
3111 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3112 (VFMLA_SEL_W): Ditto.
3113 (f16quad): Ditto.
3114 (f16mac1): Ditto.
3115 (VFMLA16_LOW): New int iterator.
3116 (VFMLA16_HIGH): Ditto.
3117 (UNSPEC_FMLAL): New unspec.
3118 (UNSPEC_FMLSL): Ditto.
3119 (UNSPEC_FMLAL2): Ditto.
3120 (UNSPEC_FMLSL2): Ditto.
3121 (f16mac): New code attribute.
3122 * config/aarch64/aarch64-simd-builtins.def
3123 (aarch64_fmlal_lowv2sf): Ditto.
3124 (aarch64_fmlsl_lowv2sf): Ditto.
3125 (aarch64_fmlalq_lowv4sf): Ditto.
3126 (aarch64_fmlslq_lowv4sf): Ditto.
3127 (aarch64_fmlal_highv2sf): Ditto.
3128 (aarch64_fmlsl_highv2sf): Ditto.
3129 (aarch64_fmlalq_highv4sf): Ditto.
3130 (aarch64_fmlslq_highv4sf): Ditto.
3131 (aarch64_fmlal_lane_lowv2sf): Ditto.
3132 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3133 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3134 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3135 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3136 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3137 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3138 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3139 (aarch64_fmlal_lane_highv2sf): Ditto.
3140 (aarch64_fmlsl_lane_highv2sf): Ditto.
3141 (aarch64_fmlal_laneq_highv2sf): Ditto.
3142 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3143 (aarch64_fmlalq_lane_highv4sf): Ditto.
3144 (aarch64_fmlsl_lane_highv4sf): Ditto.
3145 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3146 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3147 * config/aarch64/aarch64-simd.md:
3148 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3149 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3150 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3151 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3152 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3153 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3154 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3155 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3156 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3157 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3158 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3159 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3160 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3161 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3162 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3163 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3164 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3165 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3166 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3167 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3168 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3169 (vfmlsl_low_u32): Ditto.
3170 (vfmlalq_low_u32): Ditto.
3171 (vfmlslq_low_u32): Ditto.
3172 (vfmlal_high_u32): Ditto.
3173 (vfmlsl_high_u32): Ditto.
3174 (vfmlalq_high_u32): Ditto.
3175 (vfmlslq_high_u32): Ditto.
3176 (vfmlal_lane_low_u32): Ditto.
3177 (vfmlsl_lane_low_u32): Ditto.
3178 (vfmlal_laneq_low_u32): Ditto.
3179 (vfmlsl_laneq_low_u32): Ditto.
3180 (vfmlalq_lane_low_u32): Ditto.
3181 (vfmlslq_lane_low_u32): Ditto.
3182 (vfmlalq_laneq_low_u32): Ditto.
3183 (vfmlslq_laneq_low_u32): Ditto.
3184 (vfmlal_lane_high_u32): Ditto.
3185 (vfmlsl_lane_high_u32): Ditto.
3186 (vfmlal_laneq_high_u32): Ditto.
3187 (vfmlsl_laneq_high_u32): Ditto.
3188 (vfmlalq_lane_high_u32): Ditto.
3189 (vfmlslq_lane_high_u32): Ditto.
3190 (vfmlalq_laneq_high_u32): Ditto.
3191 (vfmlslq_laneq_high_u32): Ditto.
3192 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3193 (AARCH64_FL_FOR_ARCH8_4): New.
3194 (AARCH64_ISA_F16FML): New ISA flag.
3195 (TARGET_F16FML): New feature flag for fp16fml.
3196 (doc/invoke.texi): Document new fp16fml option.
3198 2018-01-10 Michael Collison <michael.collison@arm.com>
3200 * config/aarch64/aarch64-builtins.c:
3201 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3202 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3203 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3204 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3205 (AARCH64_ISA_SHA3): New ISA flag.
3206 (TARGET_SHA3): New feature flag for sha3.
3207 * config/aarch64/iterators.md (sha512_op): New int attribute.
3208 (CRYPTO_SHA512): New int iterator.
3209 (UNSPEC_SHA512H): New unspec.
3210 (UNSPEC_SHA512H2): Ditto.
3211 (UNSPEC_SHA512SU0): Ditto.
3212 (UNSPEC_SHA512SU1): Ditto.
3213 * config/aarch64/aarch64-simd-builtins.def
3214 (aarch64_crypto_sha512hqv2di): New builtin.
3215 (aarch64_crypto_sha512h2qv2di): Ditto.
3216 (aarch64_crypto_sha512su0qv2di): Ditto.
3217 (aarch64_crypto_sha512su1qv2di): Ditto.
3218 (aarch64_eor3qv8hi): Ditto.
3219 (aarch64_rax1qv2di): Ditto.
3220 (aarch64_xarqv2di): Ditto.
3221 (aarch64_bcaxqv8hi): Ditto.
3222 * config/aarch64/aarch64-simd.md:
3223 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3224 (aarch64_crypto_sha512su0qv2di): Ditto.
3225 (aarch64_crypto_sha512su1qv2di): Ditto.
3226 (aarch64_eor3qv8hi): Ditto.
3227 (aarch64_rax1qv2di): Ditto.
3228 (aarch64_xarqv2di): Ditto.
3229 (aarch64_bcaxqv8hi): Ditto.
3230 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3231 (vsha512h2q_u64): Ditto.
3232 (vsha512su0q_u64): Ditto.
3233 (vsha512su1q_u64): Ditto.
3234 (veor3q_u16): Ditto.
3235 (vrax1q_u64): Ditto.
3236 (vxarq_u64): Ditto.
3237 (vbcaxq_u16): Ditto.
3238 * config/arm/types.md (crypto_sha512): New type attribute.
3239 (crypto_sha3): Ditto.
3240 (doc/invoke.texi): Document new sha3 option.
3242 2018-01-10 Michael Collison <michael.collison@arm.com>
3244 * config/aarch64/aarch64-builtins.c:
3245 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3246 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3247 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3248 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3249 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3250 (AARCH64_ISA_SM4): New ISA flag.
3251 (TARGET_SM4): New feature flag for sm4.
3252 * config/aarch64/aarch64-simd-builtins.def
3253 (aarch64_sm3ss1qv4si): Ditto.
3254 (aarch64_sm3tt1aq4si): Ditto.
3255 (aarch64_sm3tt1bq4si): Ditto.
3256 (aarch64_sm3tt2aq4si): Ditto.
3257 (aarch64_sm3tt2bq4si): Ditto.
3258 (aarch64_sm3partw1qv4si): Ditto.
3259 (aarch64_sm3partw2qv4si): Ditto.
3260 (aarch64_sm4eqv4si): Ditto.
3261 (aarch64_sm4ekeyqv4si): Ditto.
3262 * config/aarch64/aarch64-simd.md:
3263 (aarch64_sm3ss1qv4si): Ditto.
3264 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3265 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3266 (aarch64_sm4eqv4si): Ditto.
3267 (aarch64_sm4ekeyqv4si): Ditto.
3268 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3269 (sm3part_op): Ditto.
3270 (CRYPTO_SM3TT): Ditto.
3271 (CRYPTO_SM3PART): Ditto.
3272 (UNSPEC_SM3SS1): New unspec.
3273 (UNSPEC_SM3TT1A): Ditto.
3274 (UNSPEC_SM3TT1B): Ditto.
3275 (UNSPEC_SM3TT2A): Ditto.
3276 (UNSPEC_SM3TT2B): Ditto.
3277 (UNSPEC_SM3PARTW1): Ditto.
3278 (UNSPEC_SM3PARTW2): Ditto.
3279 (UNSPEC_SM4E): Ditto.
3280 (UNSPEC_SM4EKEY): Ditto.
3281 * config/aarch64/constraints.md (Ui2): New constraint.
3282 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3283 * config/arm/types.md (crypto_sm3): New type attribute.
3284 (crypto_sm4): Ditto.
3285 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3286 (vsm3tt1aq_u32): Ditto.
3287 (vsm3tt1bq_u32): Ditto.
3288 (vsm3tt2aq_u32): Ditto.
3289 (vsm3tt2bq_u32): Ditto.
3290 (vsm3partw1q_u32): Ditto.
3291 (vsm3partw2q_u32): Ditto.
3292 (vsm4eq_u32): Ditto.
3293 (vsm4ekeyq_u32): Ditto.
3294 (doc/invoke.texi): Document new sm4 option.
3296 2018-01-10 Michael Collison <michael.collison@arm.com>
3298 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3299 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3300 (AARCH64_FL_FOR_ARCH8_4): New.
3301 (AARCH64_FL_V8_4): New flag.
3302 (doc/invoke.texi): Document new armv8.4-a option.
3304 2018-01-10 Michael Collison <michael.collison@arm.com>
3306 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3307 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3308 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3309 * config/aarch64/aarch64-option-extension.def: Add
3310 AARCH64_OPT_EXTENSION of 'sha2'.
3311 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3312 (crypto): Disable sha2 and aes if crypto disabled.
3313 (crypto): Enable aes and sha2 if enabled.
3314 (simd): Disable sha2 and aes if simd disabled.
3315 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3316 New flags.
3317 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3318 (TARGET_SHA2): New feature flag for sha2.
3319 (TARGET_AES): New feature flag for aes.
3320 * config/aarch64/aarch64-simd.md:
3321 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3322 conditional on TARGET_AES.
3323 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3324 (aarch64_crypto_sha1hsi): Make pattern conditional
3325 on TARGET_SHA2.
3326 (aarch64_crypto_sha1hv4si): Ditto.
3327 (aarch64_be_crypto_sha1hv4si): Ditto.
3328 (aarch64_crypto_sha1su1v4si): Ditto.
3329 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3330 (aarch64_crypto_sha1su0v4si): Ditto.
3331 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3332 (aarch64_crypto_sha256su0v4si): Ditto.
3333 (aarch64_crypto_sha256su1v4si): Ditto.
3334 (doc/invoke.texi): Document new aes and sha2 options.
3336 2018-01-10 Martin Sebor <msebor@redhat.com>
3338 PR tree-optimization/83781
3339 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3340 as string arrays.
3342 2018-01-11 Martin Sebor <msebor@gmail.com>
3343 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3345 PR tree-optimization/83501
3346 PR tree-optimization/81703
3348 * tree-ssa-strlen.c (get_string_cst): Rename...
3349 (get_string_len): ...to this. Handle global constants.
3350 (handle_char_store): Adjust.
3352 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3353 Jim Wilson <jimw@sifive.com>
3355 * config/riscv/riscv-protos.h (riscv_output_return): New.
3356 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3357 (riscv_attribute_table, riscv_output_return),
3358 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3359 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3360 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3361 function.
3362 (riscv_expand_prologue): Add early return for naked function.
3363 (riscv_expand_epilogue): Likewise.
3364 (riscv_function_ok_for_sibcall): Return false for naked function.
3365 (riscv_set_current_function): New.
3366 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3367 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3368 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3369 * doc/extend.texi (RISC-V Function Attributes): New.
3371 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3373 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3374 check for 128-bit long double before checking TCmode.
3375 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3376 128-bit long doubles before checking TFmode or TCmode.
3377 (FLOAT128_IBM_P): Likewise.
3379 2018-01-10 Martin Sebor <msebor@redhat.com>
3381 PR tree-optimization/83671
3382 * builtins.c (c_strlen): Unconditionally return zero for the empty
3383 string.
3384 Use -Warray-bounds for warnings.
3385 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3386 for non-constant array indices with COMPONENT_REF, arrays of
3387 arrays, and pointers to arrays.
3388 (gimple_fold_builtin_strlen): Determine and set length range for
3389 non-constant character arrays.
3391 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3393 PR middle-end/81897
3394 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3395 empty blocks.
3397 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3399 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3401 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3403 PR target/83399
3404 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3405 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3406 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3407 indexed_or_indirect_operand predicate.
3408 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3409 (*vsx_le_perm_load_v8hi): Likewise.
3410 (*vsx_le_perm_load_v16qi): Likewise.
3411 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3412 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3413 (*vsx_le_perm_store_v8hi): Likewise.
3414 (*vsx_le_perm_store_v16qi): Likewise.
3415 (eight unnamed splitters): Likewise.
3417 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3419 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3420 * config/rs6000/emmintrin.h: Likewise.
3421 * config/rs6000/mmintrin.h: Likewise.
3422 * config/rs6000/xmmintrin.h: Likewise.
3424 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3426 PR c++/43486
3427 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3428 "public_flag".
3429 * tree.c (tree_nop_conversion): Return true for location wrapper
3430 nodes.
3431 (maybe_wrap_with_location): New function.
3432 (selftest::check_strip_nops): New function.
3433 (selftest::test_location_wrappers): New function.
3434 (selftest::tree_c_tests): Call it.
3435 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3436 (maybe_wrap_with_location): New decl.
3437 (EXPR_LOCATION_WRAPPER_P): New macro.
3438 (location_wrapper_p): New inline function.
3439 (tree_strip_any_location_wrapper): New inline function.
3441 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3443 PR target/83735
3444 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3445 stack_realign_offset for the largest alignment of stack slot
3446 actually used.
3447 (ix86_find_max_used_stack_alignment): New function.
3448 (ix86_finalize_stack_frame_flags): Use it. Set
3449 max_used_stack_alignment if we don't realign stack.
3450 * config/i386/i386.h (machine_function): Add
3451 max_used_stack_alignment.
3453 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3455 * config/arm/arm.opt (-mbranch-cost): New option.
3456 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3457 account.
3459 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3461 PR target/83629
3462 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3463 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3465 2018-01-10 Richard Biener <rguenther@suse.de>
3467 PR debug/83765
3468 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3469 early out so it also covers the case where we have a non-NULL
3470 origin.
3472 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3474 PR tree-optimization/83753
3475 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3476 for non-strided grouped accesses if the number of elements is 1.
3478 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3480 PR target/81616
3481 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3482 * i386.h (TARGET_USE_GATHER): Define.
3483 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3485 2018-01-10 Martin Liska <mliska@suse.cz>
3487 PR bootstrap/82831
3488 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3489 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3490 partitioning.
3491 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3492 CLEANUP_NO_PARTITIONING is not set.
3494 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3496 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3497 for vectors, as a partial revert of r254296.
3498 * rtl.h (const_vec_p): Delete.
3499 (const_vec_duplicate_p): Don't test for vector CONSTs.
3500 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3501 * expmed.c (make_tree): Likewise.
3503 Revert:
3504 * common.md (E, F): Use CONSTANT_P instead of checking for
3505 CONST_VECTOR.
3506 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3507 checking for CONST_VECTOR.
3509 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3511 PR middle-end/83575
3512 * predict.c (force_edge_cold): Handle in more sane way edges
3513 with no prediction.
3515 2018-01-09 Carl Love <cel@us.ibm.com>
3517 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3518 V4SI, V4SF types.
3519 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3520 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3521 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3522 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3523 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3524 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3525 * config/rs6000/rs6000-protos.h: Add extern defition for
3526 rs6000_generate_float2_double_code.
3527 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3528 function.
3529 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3530 (float2_v2df): Add define_expand.
3532 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3534 PR target/83628
3535 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3536 op_mode in the force_to_mode call.
3538 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3540 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3541 instead of checking each element individually.
3542 (aarch64_evpc_uzp): Likewise.
3543 (aarch64_evpc_zip): Likewise.
3544 (aarch64_evpc_ext): Likewise.
3545 (aarch64_evpc_rev): Likewise.
3546 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3547 instead of checking each element individually. Return true without
3548 generating rtl if
3549 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3550 whether all selected elements come from the same input, instead of
3551 checking each element individually. Remove calls to gen_rtx_REG,
3552 start_sequence and end_sequence and instead assert that no rtl is
3553 generated.
3555 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3557 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3558 order of HIGH and CONST checks.
3560 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3562 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3563 if the destination isn't an SSA_NAME.
3565 2018-01-09 Richard Biener <rguenther@suse.de>
3567 PR tree-optimization/83668
3568 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3569 move prologue...
3570 (canonicalize_loop_form): ... here, renamed from ...
3571 (canonicalize_loop_closed_ssa_form): ... this and amended to
3572 swap successor edges for loop exit blocks to make us use
3573 the RPO order we need for initial schedule generation.
3575 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3577 PR tree-optimization/64811
3578 * match.pd: When optimizing comparisons with Inf, avoid
3579 introducing or losing exceptions from comparisons with NaN.
3581 2018-01-09 Martin Liska <mliska@suse.cz>
3583 PR sanitizer/82517
3584 * asan.c (shadow_mem_size): Add gcc_assert.
3586 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3588 Don't save registers in main().
3590 PR target/83738
3591 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3592 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3593 * config/avr/avr.c (avr_set_current_function): Don't error if
3594 naked, OS_task or OS_main are specified at the same time.
3595 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3596 OS_main.
3597 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3598 attribute.
3599 * common/config/avr/avr-common.c (avr_option_optimization_table):
3600 Switch on -mmain-is-OS_task for optimizing compilations.
3602 2018-01-09 Richard Biener <rguenther@suse.de>
3604 PR tree-optimization/83572
3605 * graphite.c: Include cfganal.h.
3606 (graphite_transform_loops): Connect infinite loops to exit
3607 and remove fake edges at the end.
3609 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3611 * ipa-inline.c (edge_badness): Revert accidental checkin.
3613 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3615 PR ipa/80763
3616 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3617 symbols; not inline clones.
3619 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3621 PR target/83507
3622 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3623 hard registers. Formatting fixes.
3625 PR preprocessor/83722
3626 * gcc.c (try_generate_repro): Pass
3627 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3628 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3629 do_report_bug.
3631 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3632 Kito Cheng <kito.cheng@gmail.com>
3634 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3635 (riscv_leaf_function_p): Delete.
3636 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3638 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3640 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3641 function.
3642 (do_ifelse): New function.
3643 (do_isel): New function.
3644 (do_sub3): New function.
3645 (do_add3): New function.
3646 (do_load_mask_compare): New function.
3647 (do_overlap_load_compare): New function.
3648 (expand_compare_loop): New function.
3649 (expand_block_compare): Call expand_compare_loop() when appropriate.
3650 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3651 option description.
3652 (-mblock-compare-inline-loop-limit): New option.
3654 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3656 PR target/83677
3657 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3658 Reverse order of second and third operands in first alternative.
3659 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3660 of first and second elements in UNSPEC_VPERMR vector.
3661 (altivec_expand_vec_perm_le): Likewise.
3663 2017-01-08 Jeff Law <law@redhat.com>
3665 PR rtl-optimizatin/81308
3666 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3667 (process_switch): If group_case_labels makes a change, then set
3668 cfg_altered.
3669 (pass_convert_switch::execute): If a switch is converted, then
3670 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3672 PR rtl-optimization/81308
3673 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3674 splitting insns.
3676 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3678 PR target/83663 - Revert r255946
3679 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3680 generation for cases where splatting a value is not useful.
3681 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3682 across a vec_duplicate and a paradoxical subreg forming a vector
3683 mode to a vec_concat.
3685 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3687 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3688 -march=armv8.3-a variants.
3689 * config/arm/t-multilib: Likewise.
3690 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3692 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3694 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3695 to generate rtl.
3696 (cceq_ior_compare_complement): Give it a name so I can use it, and
3697 change boolean_or_operator predicate to boolean_operator so it can
3698 be used to generate a crand.
3699 (eqne): New code iterator.
3700 (bd/bd_neg): New code_attrs.
3701 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3702 a single define_insn.
3703 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3704 decrement (bdnzt/bdnzf/bdzt/bdzf).
3705 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3706 with the new names of the branch decrement patterns, and added the
3707 names of the branch decrement conditional patterns.
3709 2018-01-08 Richard Biener <rguenther@suse.de>
3711 PR tree-optimization/83563
3712 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3713 cache.
3715 2018-01-08 Richard Biener <rguenther@suse.de>
3717 PR middle-end/83713
3718 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3720 2018-01-08 Richard Biener <rguenther@suse.de>
3722 PR tree-optimization/83685
3723 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3724 references to abnormals.
3726 2018-01-08 Richard Biener <rguenther@suse.de>
3728 PR lto/83719
3729 * dwarf2out.c (output_indirect_strings): Handle empty
3730 skeleton_debug_str_hash.
3731 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3733 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3735 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3736 (emit_store_direct): Likewise.
3737 (arc_trampoline_adjust_address): Likewise.
3738 (arc_asm_trampoline_template): New function.
3739 (arc_initialize_trampoline): Use asm_trampoline_template.
3740 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3741 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3742 * config/arc/arc.md (flush_icache): Delete pattern.
3744 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3746 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3747 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3748 munaligned-access.
3750 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3752 PR target/83681
3753 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3754 by not USED_FOR_TARGET.
3755 (make_pass_resolve_sw_modes): Likewise.
3757 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3759 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3760 USED_FOR_TARGET.
3762 2018-01-08 Richard Biener <rguenther@suse.de>
3764 PR middle-end/83580
3765 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3767 2018-01-08 Richard Biener <rguenther@suse.de>
3769 PR middle-end/83517
3770 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3772 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3774 PR middle-end/81897
3775 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3776 basic blocks with a small number of successors.
3777 (convert_control_dep_chain_into_preds): Improve handling of
3778 forwarder blocks.
3779 (dump_predicates): Split apart into...
3780 (dump_pred_chain): ...here...
3781 (dump_pred_info): ...and here.
3782 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3783 (can_chain_union_be_invalidated_p): Improve check for invalidation
3784 of paths.
3785 (uninit_uses_cannot_happen): Avoid unnecessary if
3786 convert_control_dep_chain_into_preds yielded nothing.
3788 2018-01-06 Martin Sebor <msebor@redhat.com>
3790 PR tree-optimization/83640
3791 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3792 subtracting negative offset from size.
3793 (builtin_access::overlap): Adjust offset bounds of the access to fall
3794 within the size of the object if possible.
3796 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3798 PR rtl-optimization/83699
3799 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3800 extract_bit_field_as_subreg to cases in which the extracted
3801 value is also a vector.
3803 * lra-constraints.c (process_alt_operands): Test for the equivalence
3804 substitutions when detecting a possible reload cycle.
3806 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3808 PR debug/83480
3809 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3810 by default if flag_selective_schedling{,2}. Formatting fixes.
3812 PR rtl-optimization/83682
3813 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
3814 if it has non-VECTOR_MODE element mode.
3815 (vec_duplicate_p): Likewise.
3817 PR middle-end/83694
3818 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
3819 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
3821 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3823 PR target/83604
3824 * config/i386/i386-builtin.def
3825 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
3826 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
3827 Require also OPTION_MASK_ISA_AVX512F in addition to
3828 OPTION_MASK_ISA_GFNI.
3829 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
3830 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
3831 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
3832 to OPTION_MASK_ISA_GFNI.
3833 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
3834 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
3835 OPTION_MASK_ISA_AVX512BW.
3836 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
3837 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
3838 addition to OPTION_MASK_ISA_GFNI.
3839 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
3840 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
3841 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
3842 to OPTION_MASK_ISA_GFNI.
3843 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
3844 a requirement for all ISAs rather than any of them with a few
3845 exceptions.
3846 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
3847 processing.
3848 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
3849 bitmasks to be enabled with 3 exceptions, instead of requiring any
3850 enabled ISA with lots of exceptions.
3851 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
3852 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
3853 Change avx512bw in isa attribute to avx512f.
3854 * config/i386/sgxintrin.h: Add license boilerplate.
3855 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
3856 to __AVX512F__ and __AVX512VL to __AVX512VL__.
3857 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
3858 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
3859 defined.
3860 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
3861 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
3862 temporarily sse2 rather than sse if not enabled already.
3864 PR target/83604
3865 * config/i386/sse.md (VI248_VLBW): Rename to ...
3866 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
3867 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
3868 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
3869 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
3870 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
3871 mode iterator instead of VI248_VLBW.
3873 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
3875 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
3876 (record_modified): Skip clobbers; add debug output.
3877 (param_change_prob): Use sreal frequencies.
3879 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3881 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
3882 punt for user-aligned variables.
3884 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3886 * tree-chrec.c (chrec_contains_symbols): Return true for
3887 POLY_INT_CST.
3889 2018-01-05 Sudakshina Das <sudi.das@arm.com>
3891 PR target/82439
3892 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
3893 of (x|y) == x for BICS pattern.
3895 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3897 PR tree-optimization/83605
3898 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
3899 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
3900 can throw.
3902 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
3904 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
3905 * config/epiphany/rtems.h: New file.
3907 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3908 Uros Bizjak <ubizjak@gmail.com>
3910 PR target/83554
3911 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
3912 QIreg_operand instead of register_operand predicate.
3913 * config/i386/i386.c (ix86_rop_should_change_byte_p,
3914 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
3915 comments instead of -fmitigate[-_]rop.
3917 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3919 PR bootstrap/81926
3920 * cgraphunit.c (symbol_table::compile): Switch to text_section
3921 before calling assembly_start debug hook.
3922 * run-rtl-passes.c (run_rtl_passes): Likewise.
3923 Include output.h.
3925 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3927 * tree-vrp.c (extract_range_from_binary_expr_1): Check
3928 range_int_cst_p rather than !symbolic_range_p before calling
3929 extract_range_from_multiplicative_op_1.
3931 2017-01-04 Jeff Law <law@redhat.com>
3933 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
3934 redundant test in assertion.
3936 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3938 * doc/rtl.texi: Document machine_mode wrapper classes.
3940 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3942 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
3943 using tree_to_uhwi.
3945 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3947 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
3948 the VEC_PERM_EXPR fold to fail.
3950 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3952 PR debug/83585
3953 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
3954 to switched_sections.
3956 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3958 PR target/83680
3959 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
3960 test for d.testing.
3962 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
3964 PR target/83387
3965 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
3966 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
3968 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3970 PR debug/83666
3971 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
3972 is BLKmode and bitpos not zero or mode change is needed.
3974 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3976 PR target/83675
3977 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
3978 TARGET_VIS2.
3980 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
3982 PR target/83628
3983 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
3984 instead of MULT rtx. Update all corresponding splitters.
3985 (*saddl_se): Ditto.
3986 (*ssub<modesuffix>): Ditto.
3987 (*ssubl_se): Ditto.
3988 (*cmp_sadd_di): Update split patterns.
3989 (*cmp_sadd_si): Ditto.
3990 (*cmp_sadd_sidi): Ditto.
3991 (*cmp_ssub_di): Ditto.
3992 (*cmp_ssub_si): Ditto.
3993 (*cmp_ssub_sidi): Ditto.
3994 * config/alpha/predicates.md (const23_operand): New predicate.
3995 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
3996 Look for ASHIFT, not MULT inner operand.
3997 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
3999 2018-01-04 Martin Liska <mliska@suse.cz>
4001 PR gcov-profile/83669
4002 * gcov.c (output_intermediate_file): Add version to intermediate
4003 gcov file.
4004 * doc/gcov.texi: Document new field 'version' in intermediate
4005 file format. Fix location of '-k' option of gcov command.
4007 2018-01-04 Martin Liska <mliska@suse.cz>
4009 PR ipa/82352
4010 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4012 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4014 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4016 2018-01-03 Martin Sebor <msebor@redhat.com>
4018 PR tree-optimization/83655
4019 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4020 checking calls with invalid arguments.
4022 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4024 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4025 (vectorizable_mask_load_store): Delete.
4026 (vectorizable_call): Return false for masked loads and stores.
4027 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4028 instead of gimple_assign_rhs1.
4029 (vectorizable_load): Handle IFN_MASK_LOAD.
4030 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4032 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4034 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4035 split out from..,
4036 (vectorizable_mask_load_store): ...here.
4037 (vectorizable_load): ...and here.
4039 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4041 * tree-vect-stmts.c (vect_build_all_ones_mask)
4042 (vect_build_zero_merge_argument): New functions, split out from...
4043 (vectorizable_load): ...here.
4045 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4047 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4048 split out from...
4049 (vectorizable_mask_load_store): ...here.
4050 (vectorizable_store): ...and here.
4052 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4054 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4055 split out from...
4056 (vectorizable_mask_load_store): ...here.
4058 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4060 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4061 (vect_model_store_cost): Take a vec_load_store_type instead of a
4062 vect_def_type.
4063 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4064 (vect_model_store_cost): Take a vec_load_store_type instead of a
4065 vect_def_type.
4066 (vectorizable_mask_load_store): Update accordingly.
4067 (vectorizable_store): Likewise.
4068 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4070 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4072 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4073 IFN_MASK_LOAD calls here rather than...
4074 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4076 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4077 Alan Hayward <alan.hayward@arm.com>
4078 David Sherwood <david.sherwood@arm.com>
4080 * expmed.c (extract_bit_field_1): For vector extracts,
4081 fall back to extract_bit_field_as_subreg if vec_extract
4082 isn't available.
4084 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4085 Alan Hayward <alan.hayward@arm.com>
4086 David Sherwood <david.sherwood@arm.com>
4088 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4089 they are variable or constant sized.
4090 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4091 slots for constant-sized data.
4093 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4094 Alan Hayward <alan.hayward@arm.com>
4095 David Sherwood <david.sherwood@arm.com>
4097 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4098 handling COND_EXPRs with boolean comparisons, try to find a better
4099 basis for the mask type than the boolean itself.
4101 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4103 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4104 is calculated and how it can be overridden.
4105 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4106 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4107 if defined.
4108 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4109 if nonzero.
4111 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4112 Alan Hayward <alan.hayward@arm.com>
4113 David Sherwood <david.sherwood@arm.com>
4115 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4116 Remove the mode argument.
4117 (aarch64_simd_valid_immediate): Remove the mode and inverse
4118 arguments.
4119 * config/aarch64/iterators.md (bitsize): New iterator.
4120 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4121 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4122 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4123 aarch64_simd_valid_immediate.
4124 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4125 (aarch64_reg_or_bic_imm): Likewise.
4126 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4127 with an insn_type enum and msl with a modifier_type enum.
4128 Replace element_width with a scalar_mode. Change the shift
4129 to unsigned int. Add constructors for scalar_float_mode and
4130 scalar_int_mode elements.
4131 (aarch64_vect_float_const_representable_p): Delete.
4132 (aarch64_can_const_movi_rtx_p)
4133 (aarch64_simd_scalar_immediate_valid_for_move)
4134 (aarch64_simd_make_constant): Update call to
4135 aarch64_simd_valid_immediate.
4136 (aarch64_advsimd_valid_immediate_hs): New function.
4137 (aarch64_advsimd_valid_immediate): Likewise.
4138 (aarch64_simd_valid_immediate): Remove mode and inverse
4139 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4140 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4141 and aarch64_float_const_representable_p on the result.
4142 (aarch64_output_simd_mov_immediate): Remove mode argument.
4143 Update call to aarch64_simd_valid_immediate and use of
4144 simd_immediate_info.
4145 (aarch64_output_scalar_simd_mov_immediate): Update call
4146 accordingly.
4148 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4149 Alan Hayward <alan.hayward@arm.com>
4150 David Sherwood <david.sherwood@arm.com>
4152 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4153 (mode_nunits): Likewise CONST_MODE_NUNITS.
4154 * machmode.def (ADJUST_NUNITS): Document.
4155 * genmodes.c (mode_data::need_nunits_adj): New field.
4156 (blank_mode): Update accordingly.
4157 (adj_nunits): New variable.
4158 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4159 parameter.
4160 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4161 listed in adj_nunits.
4162 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4163 listed in adj_nunits. Don't emit case statements for such modes.
4164 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4165 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4166 nothing if adj_nunits is nonnull.
4167 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4168 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4169 (emit_mode_fbit): Update use of print_maybe_const_decl.
4170 (emit_move_size): Likewise. Treat the array as non-const
4171 if adj_nunits.
4172 (emit_mode_adjustments): Handle adj_nunits.
4174 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4176 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4177 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4178 (VECTOR_MODES): Use it.
4179 (make_vector_modes): Take the prefix as an argument.
4181 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4182 Alan Hayward <alan.hayward@arm.com>
4183 David Sherwood <david.sherwood@arm.com>
4185 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4186 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4187 for MODE_VECTOR_BOOL.
4188 * machmode.def (VECTOR_BOOL_MODE): Document.
4189 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4190 (make_vector_bool_mode): New function.
4191 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4192 MODE_VECTOR_BOOL.
4193 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4194 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4195 Likewise.
4196 * stor-layout.c (int_mode_for_mode): Likewise.
4197 * tree.c (build_vector_type_for_mode): Likewise.
4198 * varasm.c (output_constant_pool_2): Likewise.
4199 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4200 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4201 for MODE_VECTOR_BOOL.
4202 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4203 of mode class checks.
4204 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4205 instead of a list of mode class checks.
4206 (expand_vector_scalar_condition): Likewise.
4207 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4209 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4210 Alan Hayward <alan.hayward@arm.com>
4211 David Sherwood <david.sherwood@arm.com>
4213 * machmode.h (mode_size): Change from unsigned short to
4214 poly_uint16_pod.
4215 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4216 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4217 or if measurement_type is not polynomial.
4218 (fixed_size_mode::includes_p): Check for constant-sized modes.
4219 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4220 return a poly_uint16 rather than an unsigned short.
4221 (emit_mode_size): Change the type of mode_size from unsigned short
4222 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4223 (emit_mode_adjustments): Cope with polynomial vector sizes.
4224 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4225 for GET_MODE_SIZE.
4226 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4227 for GET_MODE_SIZE.
4228 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4229 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4230 * caller-save.c (setup_save_areas): Likewise.
4231 (replace_reg_with_saved_mem): Likewise.
4232 * calls.c (emit_library_call_value_1): Likewise.
4233 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4234 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4235 (gen_lowpart_for_combine): Likewise.
4236 * convert.c (convert_to_integer_1): Likewise.
4237 * cse.c (equiv_constant, cse_insn): Likewise.
4238 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4239 (cselib_subst_to_values): Likewise.
4240 * dce.c (word_dce_process_block): Likewise.
4241 * df-problems.c (df_word_lr_mark_ref): Likewise.
4242 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4243 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4244 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4245 (rtl_for_decl_location): Likewise.
4246 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4247 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4248 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4249 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4250 (expand_expr_real_1): Likewise.
4251 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4252 (pad_below): Likewise.
4253 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4254 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4255 * ira.c (get_subreg_tracking_sizes): Likewise.
4256 * ira-build.c (ira_create_allocno_objects): Likewise.
4257 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4258 (ira_sort_regnos_for_alter_reg): Likewise.
4259 * ira-costs.c (record_operand_costs): Likewise.
4260 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4261 (resolve_simple_move): Likewise.
4262 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4263 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4264 (lra_constraints): Likewise.
4265 (CONST_POOL_OK_P): Reject variable-sized modes.
4266 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4267 (add_pseudo_to_slot, lra_spill): Likewise.
4268 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4269 * optabs-query.c (get_best_extraction_insn): Likewise.
4270 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4271 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4272 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4273 * recog.c (offsettable_address_addr_space_p): Likewise.
4274 * regcprop.c (maybe_mode_change): Likewise.
4275 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4276 * regrename.c (build_def_use): Likewise.
4277 * regstat.c (dump_reg_info): Likewise.
4278 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4279 (find_reloads, find_reloads_subreg_address): Likewise.
4280 * reload1.c (eliminate_regs_1): Likewise.
4281 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4282 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4283 (simplify_binary_operation_1, simplify_subreg): Likewise.
4284 * targhooks.c (default_function_arg_padding): Likewise.
4285 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4286 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4287 (verify_gimple_assign_ternary): Likewise.
4288 * tree-inline.c (estimate_move_cost): Likewise.
4289 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4290 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4291 (get_address_cost_ainc): Likewise.
4292 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4293 (vect_supportable_dr_alignment): Likewise.
4294 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4295 (vectorizable_reduction): Likewise.
4296 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4297 (vectorizable_operation, vectorizable_load): Likewise.
4298 * tree.c (build_same_sized_truth_vector_type): Likewise.
4299 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4300 * var-tracking.c (emit_note_insn_var_location): Likewise.
4301 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4302 (ADDR_VEC_ALIGN): Likewise.
4304 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4305 Alan Hayward <alan.hayward@arm.com>
4306 David Sherwood <david.sherwood@arm.com>
4308 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4309 unsigned short.
4310 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4311 or if measurement_type is polynomial.
4312 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4313 * combine.c (make_extraction): Likewise.
4314 * dse.c (find_shift_sequence): Likewise.
4315 * dwarf2out.c (mem_loc_descriptor): Likewise.
4316 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4317 (extract_bit_field, extract_low_bits): Likewise.
4318 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4319 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4320 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4321 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4322 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4323 * reload.c (find_reloads): Likewise.
4324 * reload1.c (alter_reg): Likewise.
4325 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4326 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4327 * tree-if-conv.c (predicate_mem_writes): Likewise.
4328 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4329 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4330 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4331 * valtrack.c (dead_debug_insert_temp): Likewise.
4332 * varasm.c (mergeable_constant_section): Likewise.
4333 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4335 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4336 Alan Hayward <alan.hayward@arm.com>
4337 David Sherwood <david.sherwood@arm.com>
4339 * expr.c (expand_assignment): Cope with polynomial mode sizes
4340 when assigning to a CONCAT.
4342 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4343 Alan Hayward <alan.hayward@arm.com>
4344 David Sherwood <david.sherwood@arm.com>
4346 * machmode.h (mode_precision): Change from unsigned short to
4347 poly_uint16_pod.
4348 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4349 short.
4350 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4351 or if measurement_type is not polynomial.
4352 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4353 in which the mode is already known to be a scalar_int_mode.
4354 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4355 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4356 initializer.
4357 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4358 for GET_MODE_PRECISION.
4359 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4360 for GET_MODE_PRECISION.
4361 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4362 as polynomial.
4363 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4364 (expand_field_assignment, make_extraction): Likewise.
4365 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4366 (get_last_value): Likewise.
4367 * convert.c (convert_to_integer_1): Likewise.
4368 * cse.c (cse_insn): Likewise.
4369 * expr.c (expand_expr_real_1): Likewise.
4370 * lra-constraints.c (simplify_operand_subreg): Likewise.
4371 * optabs-query.c (can_atomic_load_p): Likewise.
4372 * optabs.c (expand_atomic_load): Likewise.
4373 (expand_atomic_store): Likewise.
4374 * ree.c (combine_reaching_defs): Likewise.
4375 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4376 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4377 * tree.h (type_has_mode_precision_p): Likewise.
4378 * ubsan.c (instrument_si_overflow): Likewise.
4380 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4381 Alan Hayward <alan.hayward@arm.com>
4382 David Sherwood <david.sherwood@arm.com>
4384 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4385 polynomial numbers of units.
4386 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4387 (valid_vector_subparts_p): New function.
4388 (build_vector_type): Remove temporary shim and take the number
4389 of units as a poly_uint64 rather than an int.
4390 (build_opaque_vector_type): Take the number of units as a
4391 poly_uint64 rather than an int.
4392 * tree.c (build_vector_from_ctor): Handle polynomial
4393 TYPE_VECTOR_SUBPARTS.
4394 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4395 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4396 (build_vector_from_val): If the number of units is variable,
4397 use build_vec_duplicate_cst for constant operands and
4398 VEC_DUPLICATE_EXPR otherwise.
4399 (make_vector_type): Remove temporary is_constant ().
4400 (build_vector_type, build_opaque_vector_type): Take the number of
4401 units as a poly_uint64 rather than an int.
4402 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4403 VECTOR_CST_NELTS.
4404 * cfgexpand.c (expand_debug_expr): Likewise.
4405 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4406 (store_constructor, expand_expr_real_1): Likewise.
4407 (const_scalar_mask_from_tree): Likewise.
4408 * fold-const-call.c (fold_const_reduction): Likewise.
4409 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4410 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4411 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4412 (fold_relational_const): Likewise.
4413 (native_interpret_vector): Likewise. Change the size from an
4414 int to an unsigned int.
4415 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4416 TYPE_VECTOR_SUBPARTS.
4417 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4418 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4419 duplicating a non-constant operand into a variable-length vector.
4420 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4421 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4422 * ipa-icf.c (sem_variable::equals): Likewise.
4423 * match.pd: Likewise.
4424 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4425 * print-tree.c (print_node): Likewise.
4426 * stor-layout.c (layout_type): Likewise.
4427 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4428 * tree-cfg.c (verify_gimple_comparison): Likewise.
4429 (verify_gimple_assign_binary): Likewise.
4430 (verify_gimple_assign_ternary): Likewise.
4431 (verify_gimple_assign_single): Likewise.
4432 * tree-pretty-print.c (dump_generic_node): Likewise.
4433 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4434 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4435 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4436 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4437 (vect_shift_permute_load_chain): Likewise.
4438 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4439 (expand_vector_condition, optimize_vector_constructor): Likewise.
4440 (lower_vec_perm, get_compute_type): Likewise.
4441 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4442 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4443 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4444 (vect_recog_mask_conversion_pattern): Likewise.
4445 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4446 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4447 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4448 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4449 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4450 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4451 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4452 (supportable_widening_operation): Likewise.
4453 (supportable_narrowing_operation): Likewise.
4454 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4455 Likewise.
4456 * varasm.c (output_constant): Likewise.
4458 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4459 Alan Hayward <alan.hayward@arm.com>
4460 David Sherwood <david.sherwood@arm.com>
4462 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4463 so that both the length == 3 and length != 3 cases set up their
4464 own permute vectors. Add comments explaining why we know the
4465 number of elements is constant.
4466 (vect_permute_load_chain): Likewise.
4468 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4469 Alan Hayward <alan.hayward@arm.com>
4470 David Sherwood <david.sherwood@arm.com>
4472 * machmode.h (mode_nunits): Change from unsigned char to
4473 poly_uint16_pod.
4474 (ONLY_FIXED_SIZE_MODES): New macro.
4475 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4476 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4477 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4478 New typedefs.
4479 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4480 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4481 or if measurement_type is not polynomial.
4482 * genmodes.c (ZERO_COEFFS): New macro.
4483 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4484 poly_uint16.
4485 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4486 Use ZERO_COEFFS when emitting initializers.
4487 * data-streamer.h (bp_pack_poly_value): New function.
4488 (bp_unpack_poly_value): Likewise.
4489 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4490 for GET_MODE_NUNITS.
4491 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4492 for GET_MODE_NUNITS.
4493 * tree.c (make_vector_type): Remove temporary shim and make
4494 the real function take the number of units as a poly_uint64
4495 rather than an int.
4496 (build_vector_type_for_mode): Handle polynomial nunits.
4497 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4498 * emit-rtl.c (const_vec_series_p_1): Likewise.
4499 (gen_rtx_CONST_VECTOR): Likewise.
4500 * fold-const.c (test_vec_duplicate_folding): Likewise.
4501 * genrecog.c (validate_pattern): Likewise.
4502 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4503 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4504 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4505 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4506 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4507 * rtlanal.c (subreg_get_info): Likewise.
4508 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4509 (vect_grouped_load_supported): Likewise.
4510 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4511 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4512 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4513 (simplify_const_unary_operation, simplify_binary_operation_1)
4514 (simplify_const_binary_operation, simplify_ternary_operation)
4515 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4516 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4517 instead of CONST_VECTOR_NUNITS.
4518 * varasm.c (output_constant_pool_2): Likewise.
4519 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4520 explicit-encoded elements in the XVEC for variable-length vectors.
4522 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4524 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4526 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4527 Alan Hayward <alan.hayward@arm.com>
4528 David Sherwood <david.sherwood@arm.com>
4530 * coretypes.h (fixed_size_mode): Declare.
4531 (fixed_size_mode_pod): New typedef.
4532 * builtins.h (target_builtins::x_apply_args_mode)
4533 (target_builtins::x_apply_result_mode): Change type to
4534 fixed_size_mode_pod.
4535 * builtins.c (apply_args_size, apply_result_size, result_vector)
4536 (expand_builtin_apply_args_1, expand_builtin_apply)
4537 (expand_builtin_return): Update accordingly.
4539 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4541 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4542 * cselib.c (cselib_hash_rtx): Likewise.
4543 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4544 CONST_VECTOR encoding.
4546 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4547 Jeff Law <law@redhat.com>
4549 PR target/83641
4550 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4551 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4552 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4553 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4555 PR target/83641
4556 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4557 explicitly probe *sp in a noreturn function if there were any callee
4558 register saves or frame pointer is needed.
4560 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4562 PR debug/83621
4563 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4564 BLKmode for ternary, binary or unary expressions.
4566 PR debug/83645
4567 * var-tracking.c (delete_vta_debug_insn): New inline function.
4568 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4569 insns from get_insns () to NULL instead of each bb separately.
4570 Use delete_vta_debug_insn. No longer static.
4571 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4572 delete_vta_debug_insns callers.
4573 * rtl.h (delete_vta_debug_insns): Declare.
4574 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4575 instead of variable_tracking_main.
4577 2018-01-03 Martin Sebor <msebor@redhat.com>
4579 PR tree-optimization/83603
4580 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4581 arguments past the endof the argument list in functions declared
4582 without a prototype.
4583 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4584 Avoid checking when arguments are null.
4586 2018-01-03 Martin Sebor <msebor@redhat.com>
4588 PR c/83559
4589 * doc/extend.texi (attribute const): Fix a typo.
4590 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4591 issuing -Wsuggest-attribute for void functions.
4593 2018-01-03 Martin Sebor <msebor@redhat.com>
4595 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4596 offset_int::from instead of wide_int::to_shwi.
4597 (maybe_diag_overlap): Remove assertion.
4598 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4599 * gimple-ssa-sprintf.c (format_directive): Same.
4600 (parse_directive): Same.
4601 (sprintf_dom_walker::compute_format_length): Same.
4602 (try_substitute_return_value): Same.
4604 2017-01-03 Jeff Law <law@redhat.com>
4606 PR middle-end/83654
4607 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4608 non-constant residual for zero at runtime and avoid probing in
4609 that case. Reorganize code for trailing problem to mirror handling
4610 of the residual.
4612 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4614 PR tree-optimization/83501
4615 * tree-ssa-strlen.c (get_string_cst): New.
4616 (handle_char_store): Call get_string_cst.
4618 2018-01-03 Martin Liska <mliska@suse.cz>
4620 PR tree-optimization/83593
4621 * tree-ssa-strlen.c: Include tree-cfg.h.
4622 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4623 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4624 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4625 to false.
4626 (strlen_dom_walker::before_dom_children): Call
4627 gimple_purge_dead_eh_edges. Dump tranformation with details
4628 dump flags.
4629 (strlen_dom_walker::before_dom_children): Update call by adding
4630 new argument cleanup_eh.
4631 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4633 2018-01-03 Martin Liska <mliska@suse.cz>
4635 PR ipa/83549
4636 * cif-code.def (VARIADIC_THUNK): New enum value.
4637 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4638 thunks.
4640 2018-01-03 Jan Beulich <jbeulich@suse.com>
4642 * sse.md (mov<mode>_internal): Tighten condition for when to use
4643 vmovdqu<ssescalarsize> for TI and OI modes.
4645 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4647 Update copyright years.
4649 2018-01-03 Martin Liska <mliska@suse.cz>
4651 PR ipa/83594
4652 * ipa-visibility.c (function_and_variable_visibility): Skip
4653 functions with noipa attribure.
4655 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4657 * gcc.c (process_command): Update copyright notice dates.
4658 * gcov-dump.c (print_version): Ditto.
4659 * gcov.c (print_version): Ditto.
4660 * gcov-tool.c (print_version): Ditto.
4661 * gengtype.c (create_file): Ditto.
4662 * doc/cpp.texi: Bump @copying's copyright year.
4663 * doc/cppinternals.texi: Ditto.
4664 * doc/gcc.texi: Ditto.
4665 * doc/gccint.texi: Ditto.
4666 * doc/gcov.texi: Ditto.
4667 * doc/install.texi: Ditto.
4668 * doc/invoke.texi: Ditto.
4670 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4672 * vector-builder.h (vector_builder::m_full_nelts): Change from
4673 unsigned int to poly_uint64.
4674 (vector_builder::full_nelts): Update prototype accordingly.
4675 (vector_builder::new_vector): Likewise.
4676 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4677 (vector_builder::operator ==): Likewise.
4678 (vector_builder::finalize): Likewise.
4679 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4680 Take the number of elements as a poly_uint64 rather than an
4681 unsigned int.
4682 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4683 from unsigned int to poly_uint64.
4684 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4685 (vec_perm_indices::new_vector): Likewise.
4686 (vec_perm_indices::length): Likewise.
4687 (vec_perm_indices::nelts_per_input): Likewise.
4688 (vec_perm_indices::input_nelts): Likewise.
4689 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4690 number of elements per input as a poly_uint64 rather than an
4691 unsigned int. Use the original encoding for variable-length
4692 vectors, rather than clamping each individual element.
4693 For the second and subsequent elements in each pattern,
4694 clamp the step and base before clamping their sum.
4695 (vec_perm_indices::series_p): Handle polynomial element counts.
4696 (vec_perm_indices::all_in_range_p): Likewise.
4697 (vec_perm_indices_to_tree): Likewise.
4698 (vec_perm_indices_to_rtx): Likewise.
4699 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4700 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4701 (tree_vector_builder::new_binary_operation): Handle polynomial
4702 element counts. Return false if we need to know the number
4703 of elements at compile time.
4704 * fold-const.c (fold_vec_perm): Punt if the number of elements
4705 isn't known at compile time.
4707 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4709 * vec-perm-indices.h (vec_perm_builder): Change element type
4710 from HOST_WIDE_INT to poly_int64.
4711 (vec_perm_indices::element_type): Update accordingly.
4712 (vec_perm_indices::clamp): Handle polynomial element_types.
4713 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4714 (vec_perm_indices::all_in_range_p): Likewise.
4715 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4716 than shwi trees.
4717 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4718 polynomial vec_perm_indices element types.
4719 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4720 * fold-const.c (fold_vec_perm): Likewise.
4721 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4722 * tree-vect-generic.c (lower_vec_perm): Likewise.
4723 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4724 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4725 element type to HOST_WIDE_INT.
4727 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4728 Alan Hayward <alan.hayward@arm.com>
4729 David Sherwood <david.sherwood@arm.com>
4731 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4732 rather than an int. Use plus_constant.
4733 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4734 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4736 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4737 Alan Hayward <alan.hayward@arm.com>
4738 David Sherwood <david.sherwood@arm.com>
4740 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4741 a HOST_WIDE_INT to a poly_int64.
4743 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4744 Alan Hayward <alan.hayward@arm.com>
4745 David Sherwood <david.sherwood@arm.com>
4747 * calls.c (load_register_parameters): Cope with polynomial
4748 mode sizes. Require a constant size for BLKmode parameters
4749 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4750 forces a parameter to be padded at the lsb end in order to
4751 fill a complete number of words, require the parameter size
4752 to be ordered wrt UNITS_PER_WORD.
4754 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4755 Alan Hayward <alan.hayward@arm.com>
4756 David Sherwood <david.sherwood@arm.com>
4758 * reload1.c (spill_stack_slot_width): Change element type
4759 from unsigned int to poly_uint64_pod.
4760 (alter_reg): Treat mode sizes as polynomial.
4762 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4763 Alan Hayward <alan.hayward@arm.com>
4764 David Sherwood <david.sherwood@arm.com>
4766 * reload.c (complex_word_subreg_p): New function.
4767 (reload_inner_reg_of_subreg, push_reload): Use it.
4769 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4770 Alan Hayward <alan.hayward@arm.com>
4771 David Sherwood <david.sherwood@arm.com>
4773 * lra-constraints.c (process_alt_operands): Reject matched
4774 operands whose sizes aren't ordered.
4775 (match_reload): Refer to this check here.
4777 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4778 Alan Hayward <alan.hayward@arm.com>
4779 David Sherwood <david.sherwood@arm.com>
4781 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4782 that the mode size is in the set {1, 2, 4, 8, 16}.
4784 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4785 Alan Hayward <alan.hayward@arm.com>
4786 David Sherwood <david.sherwood@arm.com>
4788 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4789 Use plus_constant instead of gen_rtx_PLUS.
4791 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4792 Alan Hayward <alan.hayward@arm.com>
4793 David Sherwood <david.sherwood@arm.com>
4795 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4796 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4797 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4798 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4799 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4800 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4801 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4802 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4803 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4804 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4805 a poly_int64.
4806 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4807 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4808 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4809 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4810 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4811 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4812 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4813 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
4814 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
4815 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
4816 function.
4817 * expr.c (emit_move_resolve_push): Treat the input and result
4818 of PUSH_ROUNDING as a poly_int64.
4819 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
4820 (emit_push_insn): Likewise.
4821 * lra-eliminations.c (mark_not_eliminable): Likewise.
4822 * recog.c (push_operand): Likewise.
4823 * reload1.c (elimination_effects): Likewise.
4824 * rtlanal.c (nonzero_bits1): Likewise.
4825 * calls.c (store_one_arg): Likewise. Require the padding to be
4826 known at compile time.
4828 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4829 Alan Hayward <alan.hayward@arm.com>
4830 David Sherwood <david.sherwood@arm.com>
4832 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
4833 Use plus_constant instead of gen_rtx_PLUS.
4835 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4836 Alan Hayward <alan.hayward@arm.com>
4837 David Sherwood <david.sherwood@arm.com>
4839 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
4840 rather than an int.
4842 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4843 Alan Hayward <alan.hayward@arm.com>
4844 David Sherwood <david.sherwood@arm.com>
4846 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
4847 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
4848 via stack temporaries. Treat the mode size as polynomial too.
4850 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4851 Alan Hayward <alan.hayward@arm.com>
4852 David Sherwood <david.sherwood@arm.com>
4854 * expr.c (expand_expr_real_2): When handling conversions involving
4855 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
4856 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
4857 as a poly_uint64 too.
4859 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4860 Alan Hayward <alan.hayward@arm.com>
4861 David Sherwood <david.sherwood@arm.com>
4863 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
4865 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4866 Alan Hayward <alan.hayward@arm.com>
4867 David Sherwood <david.sherwood@arm.com>
4869 * combine.c (can_change_dest_mode): Handle polynomial
4870 REGMODE_NATURAL_SIZE.
4871 * expmed.c (store_bit_field_1): Likewise.
4872 * expr.c (store_constructor): Likewise.
4873 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
4874 and polynomial REGMODE_NATURAL_SIZE.
4875 (gen_lowpart_common): Likewise.
4876 * reginfo.c (record_subregs_of_mode): Likewise.
4877 * rtlanal.c (read_modify_subreg_p): Likewise.
4879 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4880 Alan Hayward <alan.hayward@arm.com>
4881 David Sherwood <david.sherwood@arm.com>
4883 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
4884 numbers of elements.
4886 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4887 Alan Hayward <alan.hayward@arm.com>
4888 David Sherwood <david.sherwood@arm.com>
4890 * match.pd: Cope with polynomial numbers of vector elements.
4892 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4893 Alan Hayward <alan.hayward@arm.com>
4894 David Sherwood <david.sherwood@arm.com>
4896 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
4897 in a POINTER_PLUS_EXPR.
4899 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4900 Alan Hayward <alan.hayward@arm.com>
4901 David Sherwood <david.sherwood@arm.com>
4903 * omp-simd-clone.c (simd_clone_subparts): New function.
4904 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
4905 (ipa_simd_modify_function_body): Likewise.
4907 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4908 Alan Hayward <alan.hayward@arm.com>
4909 David Sherwood <david.sherwood@arm.com>
4911 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
4912 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
4913 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
4914 (expand_vector_condition, vector_element): Likewise.
4915 (subparts_gt): New function.
4916 (get_compute_type): Use subparts_gt.
4917 (count_type_subparts): Delete.
4918 (expand_vector_operations_1): Use subparts_gt instead of
4919 count_type_subparts.
4921 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4922 Alan Hayward <alan.hayward@arm.com>
4923 David Sherwood <david.sherwood@arm.com>
4925 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
4926 (vect_compile_time_alias): ...this new function. Do the calculation
4927 on poly_ints rather than trees.
4928 (vect_prune_runtime_alias_test_list): Update call accordingly.
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-slp.c (vect_build_slp_tree_1): Handle polynomial
4935 numbers of units.
4936 (vect_schedule_slp_instance): Likewise.
4938 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4939 Alan Hayward <alan.hayward@arm.com>
4940 David Sherwood <david.sherwood@arm.com>
4942 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
4943 constant and extern definitions for variable-length vectors.
4944 (vect_get_constant_vectors): Note that the number of units
4945 is known to be constant.
4947 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4948 Alan Hayward <alan.hayward@arm.com>
4949 David Sherwood <david.sherwood@arm.com>
4951 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
4952 of units as polynomial. Choose between WIDE and NARROW based
4953 on multiple_p.
4955 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4956 Alan Hayward <alan.hayward@arm.com>
4957 David Sherwood <david.sherwood@arm.com>
4959 * tree-vect-stmts.c (simd_clone_subparts): New function.
4960 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
4962 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4963 Alan Hayward <alan.hayward@arm.com>
4964 David Sherwood <david.sherwood@arm.com>
4966 * tree-vect-stmts.c (vectorizable_call): Treat the number of
4967 vectors as polynomial. Use build_index_vector for
4968 IFN_GOMP_SIMD_LANE.
4970 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4971 Alan Hayward <alan.hayward@arm.com>
4972 David Sherwood <david.sherwood@arm.com>
4974 * tree-vect-stmts.c (get_load_store_type): Treat the number of
4975 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
4976 for variable-length vectors.
4977 (vectorizable_mask_load_store): Treat the number of units as
4978 polynomial, asserting that it is constant if the condition has
4979 already been enforced.
4980 (vectorizable_store, vectorizable_load): Likewise.
4982 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4983 Alan Hayward <alan.hayward@arm.com>
4984 David Sherwood <david.sherwood@arm.com>
4986 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
4987 of units as polynomial. Punt if we can't tell at compile time
4988 which vector contains the final result.
4990 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4991 Alan Hayward <alan.hayward@arm.com>
4992 David Sherwood <david.sherwood@arm.com>
4994 * tree-vect-loop.c (vectorizable_induction): Treat the number
4995 of units as polynomial. Punt on SLP inductions. Use an integer
4996 VEC_SERIES_EXPR for variable-length integer reductions. Use a
4997 cast of such a series for variable-length floating-point
4998 reductions.
5000 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5001 Alan Hayward <alan.hayward@arm.com>
5002 David Sherwood <david.sherwood@arm.com>
5004 * tree.h (build_index_vector): Declare.
5005 * tree.c (build_index_vector): New function.
5006 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5007 of units as polynomial, forcibly converting it to a constant if
5008 vectorizable_reduction has already enforced the condition.
5009 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5010 to create a {1,2,3,...} vector.
5011 (vectorizable_reduction): Treat the number of units as polynomial.
5012 Choose vectype_in based on the largest scalar element size rather
5013 than the smallest number of units. Enforce the restrictions
5014 relied on above.
5016 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5017 Alan Hayward <alan.hayward@arm.com>
5018 David Sherwood <david.sherwood@arm.com>
5020 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5021 number of units as polynomial.
5023 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5024 Alan Hayward <alan.hayward@arm.com>
5025 David Sherwood <david.sherwood@arm.com>
5027 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5028 * target.def (autovectorize_vector_sizes): Return the vector sizes
5029 by pointer, using vector_sizes rather than a bitmask.
5030 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5031 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5032 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5033 Likewise.
5034 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5035 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5036 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5037 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5038 * omp-general.c (omp_max_vf): Likewise.
5039 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5040 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5041 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5042 * tree-vect-slp.c (vect_slp_bb): Likewise.
5043 * doc/tm.texi: Regenerate.
5044 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5045 to a poly_uint64.
5046 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5047 the vector size as a poly_uint64 rather than an unsigned int.
5048 (current_vector_size): Change from an unsigned int to a poly_uint64.
5049 (get_vectype_for_scalar_type): Update accordingly.
5050 * tree.h (build_truth_vector_type): Take the size and number of
5051 units as a poly_uint64 rather than an unsigned int.
5052 (build_vector_type): Add a temporary overload that takes
5053 the number of units as a poly_uint64 rather than an unsigned int.
5054 * tree.c (make_vector_type): Likewise.
5055 (build_truth_vector_type): Take the number of units as a poly_uint64
5056 rather than an unsigned int.
5058 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5059 Alan Hayward <alan.hayward@arm.com>
5060 David Sherwood <david.sherwood@arm.com>
5062 * target.def (get_mask_mode): Take the number of units and length
5063 as poly_uint64s rather than unsigned ints.
5064 * targhooks.h (default_get_mask_mode): Update accordingly.
5065 * targhooks.c (default_get_mask_mode): Likewise.
5066 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5067 * doc/tm.texi: Regenerate.
5069 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5070 Alan Hayward <alan.hayward@arm.com>
5071 David Sherwood <david.sherwood@arm.com>
5073 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5074 * omp-general.c (omp_max_vf): Likewise.
5075 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5076 (expand_omp_simd): Handle polynomial safelen.
5077 * omp-low.c (omplow_simd_context): Add a default constructor.
5078 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5079 (lower_rec_simd_input_clauses): Update accordingly.
5080 (lower_rec_input_clauses): Likewise.
5082 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5083 Alan Hayward <alan.hayward@arm.com>
5084 David Sherwood <david.sherwood@arm.com>
5086 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5087 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5088 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5089 (vect_analyze_slp_cost): Likewise.
5090 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5091 (vect_model_load_cost): Likewise.
5093 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5094 Alan Hayward <alan.hayward@arm.com>
5095 David Sherwood <david.sherwood@arm.com>
5097 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5098 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5099 from an unsigned int * to a poly_uint64_pod *.
5100 (calculate_unrolling_factor): New function.
5101 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5103 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5104 Alan Hayward <alan.hayward@arm.com>
5105 David Sherwood <david.sherwood@arm.com>
5107 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5108 from an unsigned int to a poly_uint64.
5109 (_loop_vec_info::slp_unrolling_factor): Likewise.
5110 (_loop_vec_info::vectorization_factor): Change from an int
5111 to a poly_uint64.
5112 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5113 (vect_get_num_vectors): New function.
5114 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5115 (vect_get_num_copies): Use vect_get_num_vectors.
5116 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5117 to an unsigned int *.
5118 (vect_analyze_data_refs): Change min_vf from an int * to a
5119 poly_uint64 *.
5120 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5121 than an unsigned HOST_WIDE_INT.
5122 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5123 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5124 to an unsigned int *.
5125 (vect_analyze_data_ref_dependences): Likewise.
5126 (vect_compute_data_ref_alignment): Handle polynomial vf.
5127 (vect_enhance_data_refs_alignment): Likewise.
5128 (vect_prune_runtime_alias_test_list): Likewise.
5129 (vect_shift_permute_load_chain): Likewise.
5130 (vect_supportable_dr_alignment): Likewise.
5131 (dependence_distance_ge_vf): Take the vectorization factor as a
5132 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5133 (vect_analyze_data_refs): Change min_vf from an int * to a
5134 poly_uint64 *.
5135 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5136 vfm1 as a poly_uint64 rather than an int. Make the same change
5137 for the returned bound_scalar.
5138 (vect_gen_vector_loop_niters): Handle polynomial vf.
5139 (vect_do_peeling): Likewise. Update call to
5140 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5141 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5142 be constant.
5143 * tree-vect-loop.c (vect_determine_vectorization_factor)
5144 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5145 (vect_get_known_peeling_cost): Likewise.
5146 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5147 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5148 (vect_transform_loop): Likewise. Use the lowest possible VF when
5149 updating the upper bounds of the loop.
5150 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5151 rather than an int.
5152 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5153 polynomial unroll factors.
5154 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5155 (vect_make_slp_decision): Likewise.
5156 (vect_supported_load_permutation_p): Likewise, and polynomial
5157 vf too.
5158 (vect_analyze_slp_cost): Handle polynomial vf.
5159 (vect_slp_analyze_node_operations): Likewise.
5160 (vect_slp_analyze_bb_1): Likewise.
5161 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5162 than an unsigned HOST_WIDE_INT.
5163 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5164 (vectorizable_load): Handle polynomial vf.
5165 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5166 a poly_uint64.
5167 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5169 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5170 Alan Hayward <alan.hayward@arm.com>
5171 David Sherwood <david.sherwood@arm.com>
5173 * match.pd: Handle bit operations involving three constants
5174 and try to fold one pair.
5176 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5178 * tree-vect-loop-manip.c: Include gimple-fold.h.
5179 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5180 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5181 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5182 Add a path that uses a step of VF instead of 1, but disable it
5183 for now.
5184 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5185 and niters_no_overflow parameters. Update calls to
5186 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5187 Create a new SSA name if the latter choses to use a ste other
5188 than zero, and return it via niters_vector_mult_vf_var.
5189 * tree-vect-loop.c (vect_transform_loop): Update calls to
5190 vect_do_peeling, vect_gen_vector_loop_niters and
5191 slpeel_make_loop_iterate_ntimes.
5192 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5193 (vect_gen_vector_loop_niters): Update declarations after above changes.
5195 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5197 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5198 128-bit round to integer instructions.
5199 (ceil<mode>2): Likewise.
5200 (btrunc<mode>2): Likewise.
5201 (round<mode>2): Likewise.
5203 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5205 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5206 unaligned VSX load/store on P8/P9.
5207 (expand_block_clear): Allow the use of unaligned VSX
5208 load/store on P8/P9.
5210 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5212 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5213 New function.
5214 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5215 swap associated with both a load and a store.
5217 2018-01-02 Andrew Waterman <andrew@sifive.com>
5219 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5220 * config/riscv/riscv.md (clear_cache): Use it.
5222 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5224 * web.c: Remove out-of-date comment.
5226 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5228 * expr.c (fixup_args_size_notes): Check that any existing
5229 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5230 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5231 (emit_single_push_insn): ...here.
5233 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5235 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5236 (const_vector_encoded_nelts): New function.
5237 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5238 (const_vector_int_elt, const_vector_elt): Declare.
5239 * emit-rtl.c (const_vector_int_elt_1): New function.
5240 (const_vector_elt): Likewise.
5241 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5242 of CONST_VECTOR_ELT.
5244 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5246 * expr.c: Include rtx-vector-builder.h.
5247 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5248 directly on the tree encoding.
5249 (const_vector_from_tree): Likewise.
5250 * optabs.c: Include rtx-vector-builder.h.
5251 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5252 sequence of "u" values.
5253 * vec-perm-indices.c: Include rtx-vector-builder.h.
5254 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5255 directly on the vec_perm_indices encoding.
5257 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5259 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5260 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5261 * rtx-vector-builder.h: New file.
5262 * rtx-vector-builder.c: Likewise.
5263 * rtl.h (rtx_def::u2): Add a const_vector field.
5264 (CONST_VECTOR_NPATTERNS): New macro.
5265 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5266 (CONST_VECTOR_DUPLICATE_P): Likewise.
5267 (CONST_VECTOR_STEPPED_P): Likewise.
5268 (CONST_VECTOR_ENCODED_ELT): Likewise.
5269 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5270 (unwrap_const_vec_duplicate): Likewise.
5271 (const_vec_series_p): Check for a non-duplicated vector encoding.
5272 Say that the function only returns true for integer vectors.
5273 * emit-rtl.c: Include rtx-vector-builder.h.
5274 (gen_const_vec_duplicate_1): Delete.
5275 (gen_const_vector): Call gen_const_vec_duplicate instead of
5276 gen_const_vec_duplicate_1.
5277 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5278 (gen_const_vec_duplicate): Use rtx_vector_builder.
5279 (gen_const_vec_series): Likewise.
5280 (gen_rtx_CONST_VECTOR): Likewise.
5281 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5282 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5283 Build a new vector rather than modifying a CONST_VECTOR in-place.
5284 (handle_special_swappables): Update call accordingly.
5285 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5286 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5287 Build a new vector rather than modifying a CONST_VECTOR in-place.
5288 (handle_special_swappables): Update call accordingly.
5290 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5292 * simplify-rtx.c (simplify_const_binary_operation): Use
5293 CONST_VECTOR_ELT instead of XVECEXP.
5295 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5297 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5298 the selector elements to be different from the data elements
5299 if the selector is a VECTOR_CST.
5300 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5301 ssizetype for the selector.
5303 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5305 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5306 before testing each element individually.
5307 * tree-vect-generic.c (lower_vec_perm): Likewise.
5309 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5311 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5312 * selftest-run-tests.c (selftest::run_tests): Call it.
5313 * vector-builder.h (vector_builder::operator ==): New function.
5314 (vector_builder::operator !=): Likewise.
5315 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5316 (vec_perm_indices::all_from_input_p): New function.
5317 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5318 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5319 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5320 instead of reading the VECTOR_CST directly. Detect whether both
5321 vector inputs are the same before constructing the vec_perm_indices,
5322 and update the number of inputs argument accordingly. Use the
5323 utility functions added above. Only construct sel2 if we need to.
5325 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5327 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5328 the broadcast of the low byte.
5329 (expand_mult_highpart): Use an explicit encoding for the permutes.
5330 * optabs-query.c (can_mult_highpart_p): Likewise.
5331 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5332 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5333 (vectorizable_bswap): Likewise.
5334 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5335 explicit encoding for the power-of-2 permutes.
5336 (vect_permute_store_chain): Likewise.
5337 (vect_grouped_load_supported): Likewise.
5338 (vect_permute_load_chain): Likewise.
5340 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5342 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5343 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5344 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5345 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5346 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5347 (vect_gen_perm_mask_any): Likewise.
5349 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5351 * int-vector-builder.h: New file.
5352 * vec-perm-indices.h: Include int-vector-builder.h.
5353 (vec_perm_indices): Redefine as an int_vector_builder.
5354 (auto_vec_perm_indices): Delete.
5355 (vec_perm_builder): Redefine as a stand-alone class.
5356 (vec_perm_indices::vec_perm_indices): New function.
5357 (vec_perm_indices::clamp): Likewise.
5358 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5359 (vec_perm_indices::new_vector): New function.
5360 (vec_perm_indices::new_expanded_vector): Update for new
5361 vec_perm_indices class.
5362 (vec_perm_indices::rotate_inputs): New function.
5363 (vec_perm_indices::all_in_range_p): Operate directly on the
5364 encoded form, without computing elided elements.
5365 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5366 encoding. Update for new vec_perm_indices class.
5367 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5368 the given vec_perm_builder.
5369 (expand_vec_perm_var): Update vec_perm_builder constructor.
5370 (expand_mult_highpart): Use vec_perm_builder instead of
5371 auto_vec_perm_indices.
5372 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5373 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5374 or double series encoding as appropriate.
5375 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5376 vec_perm_indices instead of auto_vec_perm_indices.
5377 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5378 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5379 (vect_permute_store_chain): Likewise.
5380 (vect_grouped_load_supported): Likewise.
5381 (vect_permute_load_chain): Likewise.
5382 (vect_shift_permute_load_chain): Likewise.
5383 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5384 (vect_transform_slp_perm_load): Likewise.
5385 (vect_schedule_slp_instance): Likewise.
5386 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5387 (vectorizable_mask_load_store): Likewise.
5388 (vectorizable_bswap): Likewise.
5389 (vectorizable_store): Likewise.
5390 (vectorizable_load): Likewise.
5391 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5392 vec_perm_indices instead of auto_vec_perm_indices. Use
5393 tree_to_vec_perm_builder to read the vector from a tree.
5394 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5395 vec_perm_builder instead of a vec_perm_indices.
5396 (have_whole_vector_shift): Use vec_perm_builder and
5397 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5398 truncation to calc_vec_perm_mask_for_shift.
5399 (vect_create_epilog_for_reduction): Likewise.
5400 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5401 from auto_vec_perm_indices to vec_perm_indices.
5402 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5403 instead of changing individual elements.
5404 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5405 the vector in d.perm.
5406 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5407 from auto_vec_perm_indices to vec_perm_indices.
5408 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5409 instead of changing individual elements.
5410 (arm_vectorize_vec_perm_const): Use new_vector to install
5411 the vector in d.perm.
5412 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5413 Update vec_perm_builder constructor.
5414 (rs6000_expand_interleave): Likewise.
5415 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5416 (rs6000_expand_interleave): Likewise.
5418 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5420 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5421 to qimode could truncate the indices.
5422 * optabs.c (expand_vec_perm_var): Likewise.
5424 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5426 * Makefile.in (OBJS): Add vec-perm-indices.o.
5427 * vec-perm-indices.h: New file.
5428 * vec-perm-indices.c: Likewise.
5429 * target.h (vec_perm_indices): Replace with a forward class
5430 declaration.
5431 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5432 * optabs.h: Include vec-perm-indices.h.
5433 (expand_vec_perm): Delete.
5434 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5435 (expand_vec_perm_const): Declare.
5436 * target.def (vec_perm_const_ok): Replace with...
5437 (vec_perm_const): ...this new hook.
5438 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5439 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5440 * doc/tm.texi: Regenerate.
5441 * optabs.def (vec_perm_const): Delete.
5442 * doc/md.texi (vec_perm_const): Likewise.
5443 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5444 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5445 expand_vec_perm for constant permutation vectors. Assert that
5446 the mode of variable permutation vectors is the integer equivalent
5447 of the mode that is being permuted.
5448 * optabs-query.h (selector_fits_mode_p): Declare.
5449 * optabs-query.c: Include vec-perm-indices.h.
5450 (selector_fits_mode_p): New function.
5451 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5452 is defined, instead of checking whether the vec_perm_const_optab
5453 exists. Use targetm.vectorize.vec_perm_const instead of
5454 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5455 fit in the vector mode before using a variable permute.
5456 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5457 vec_perm_indices instead of an rtx.
5458 (expand_vec_perm): Replace with...
5459 (expand_vec_perm_const): ...this new function. Take the selector
5460 as a vec_perm_indices rather than an rtx. Also take the mode of
5461 the selector. Update call to shift_amt_for_vec_perm_mask.
5462 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5463 Use vec_perm_indices::new_expanded_vector to expand the original
5464 selector into bytes. Check whether the indices fit in the vector
5465 mode before using a variable permute.
5466 (expand_vec_perm_var): Make global.
5467 (expand_mult_highpart): Use expand_vec_perm_const.
5468 * fold-const.c: Includes vec-perm-indices.h.
5469 * tree-ssa-forwprop.c: Likewise.
5470 * tree-vect-data-refs.c: Likewise.
5471 * tree-vect-generic.c: Likewise.
5472 * tree-vect-loop.c: Likewise.
5473 * tree-vect-slp.c: Likewise.
5474 * tree-vect-stmts.c: Likewise.
5475 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5476 Delete.
5477 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5478 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5479 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5480 (aarch64_vectorize_vec_perm_const): ...this new function.
5481 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5482 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5483 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5484 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5485 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5486 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5487 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5488 into...
5489 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5490 check for NEON modes.
5491 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5492 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5493 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5494 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5495 into...
5496 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5497 the old VEC_PERM_CONST conditions.
5498 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5499 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5500 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5501 (ia64_vectorize_vec_perm_const_ok): Merge into...
5502 (ia64_vectorize_vec_perm_const): ...this new function.
5503 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5504 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5505 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5506 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5507 * config/mips/mips.c (mips_expand_vec_perm_const)
5508 (mips_vectorize_vec_perm_const_ok): Merge into...
5509 (mips_vectorize_vec_perm_const): ...this new function.
5510 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5511 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5512 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5513 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5514 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5515 (rs6000_expand_vec_perm_const): Delete.
5516 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5517 Delete.
5518 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5519 (altivec_expand_vec_perm_const_le): Take each operand individually.
5520 Operate on constant selectors rather than rtxes.
5521 (altivec_expand_vec_perm_const): Likewise. Update call to
5522 altivec_expand_vec_perm_const_le.
5523 (rs6000_expand_vec_perm_const): Delete.
5524 (rs6000_vectorize_vec_perm_const_ok): Delete.
5525 (rs6000_vectorize_vec_perm_const): New function.
5526 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5527 an element count and rtx array.
5528 (rs6000_expand_extract_even): Update call accordingly.
5529 (rs6000_expand_interleave): Likewise.
5530 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5531 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5532 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5533 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5534 (rs6000_expand_vec_perm_const): Delete.
5535 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5536 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5537 (altivec_expand_vec_perm_const_le): Take each operand individually.
5538 Operate on constant selectors rather than rtxes.
5539 (altivec_expand_vec_perm_const): Likewise. Update call to
5540 altivec_expand_vec_perm_const_le.
5541 (rs6000_expand_vec_perm_const): Delete.
5542 (rs6000_vectorize_vec_perm_const_ok): Delete.
5543 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5544 reference to the SPE evmerge intructions.
5545 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5546 an element count and rtx array.
5547 (rs6000_expand_extract_even): Update call accordingly.
5548 (rs6000_expand_interleave): Likewise.
5549 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5550 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5551 new function.
5552 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5554 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5556 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5557 vector mode and that that mode matches the mode of the data
5558 being permuted.
5559 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5560 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5561 directly using expand_vec_perm_1 when forcing selectors into
5562 registers.
5563 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5565 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5567 * optabs-query.h (can_vec_perm_p): Delete.
5568 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5569 * optabs-query.c (can_vec_perm_p): Split into...
5570 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5571 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5572 particular selector is valid.
5573 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5574 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5575 (vect_grouped_load_supported): Likewise.
5576 (vect_shift_permute_load_chain): Likewise.
5577 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5578 (vect_transform_slp_perm_load): Likewise.
5579 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5580 (vectorizable_bswap): Likewise.
5581 (vect_gen_perm_mask_checked): Likewise.
5582 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5583 implementations of variable permutation vectors into account
5584 when deciding which selector to use.
5585 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5586 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5587 with a false third argument.
5588 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5589 to test whether the constant selector is valid and can_vec_perm_var_p
5590 to test whether a variable selector is valid.
5592 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5594 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5595 * optabs-query.c (can_vec_perm_p): Likewise.
5596 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5597 instead of vec_perm_indices.
5598 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5599 (vect_gen_perm_mask_checked): Likewise,
5600 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5601 (vect_gen_perm_mask_checked): Likewise,
5603 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5605 * optabs-query.h (qimode_for_vec_perm): Declare.
5606 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5607 (qimode_for_vec_perm): ...this new function.
5608 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5610 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5612 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5613 does not have a conditional at the top.
5615 2018-01-02 Richard Biener <rguenther@suse.de>
5617 * ipa-inline.c (big_speedup_p): Fix expression.
5619 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5621 PR target/81616
5622 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5623 for generic 4->6.
5625 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5627 PR target/81616
5628 Generic tuning.
5629 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5630 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5631 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5632 cond_taken_branch_cost 3->4.
5634 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5636 PR tree-optimization/83581
5637 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5638 TODO_cleanup_cfg if any changes have been made.
5640 PR middle-end/83608
5641 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5642 convert_modes if target mode has the right side, but different mode
5643 class.
5645 PR middle-end/83609
5646 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5647 last argument when extracting from CONCAT. If either from_real or
5648 from_imag is NULL, use expansion through memory. If result is not
5649 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5650 the parts directly to inner mode, if even that fails, use expansion
5651 through memory.
5653 PR middle-end/83623
5654 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5655 check for bswap in mode rather than HImode and use that in expand_unop
5656 too.
5657 \f
5658 Copyright (C) 2018 Free Software Foundation, Inc.
5660 Copying and distribution of this file, with or without modification,
5661 are permitted in any medium without royalty provided the copyright
5662 notice and this notice are preserved.