| blob | 8787d37c6b737880cb4e71990ed1abb91200e915 |
1 2018-01-31 Jakub Jelinek <jakub@redhat.com>
3 PR c/84100
4 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
5 falign-loops=): Add Optimization flag.
7 2017-01-30 Jeff Law <law@redhat.com>
9 PR target/84064
10 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
11 INT_REGISTERS_SAVED. Check it prior to calling
12 get_scratch_register_on_entry.
13 (ix86_adjust_stack_and_probe): Similarly.
14 (ix86_emit_probe_stack_range): Similarly.
15 (ix86_expand_prologue): Corresponding changes.
17 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
19 PR target/40411
20 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
21 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
23 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
25 PR target/84112
26 * lra-constraints.c (curr_insn_transform): Process AND in the
27 address.
29 2018-01-30 Jakub Jelinek <jakub@redhat.com>
31 PR rtl-optimization/83986
32 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
33 dependence against last_pending_memory_flush in addition to
34 pending_jump_insns.
36 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
38 PR tree-optimization/81611
39 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
40 copies.
42 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
44 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
45 a reg rtx.
47 2018-01-30 Richard Biener <rguenther@suse.de>
48 Jakub Jelinek <jakub@redhat.com>
50 PR tree-optimization/84111
51 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
52 inner loops added during recursion, as they don't have up-to-date
53 SSA form.
55 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
57 PR ipa/81360
58 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
59 (can_inline_edge_by_limits_p): ... here.
60 (can_early_inline_edge_p, check_callers,
61 update_caller_keys, update_callee_keys, recursive_inlining,
62 add_new_edges_to_heap, speculation_useful_p,
63 inline_small_functions,
64 inline_small_functions, flatten_function,
65 inline_to_all_callers_1): Update.
67 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
69 * profile-count.c (profile_count::combine_with_ipa_count): Handle
70 zeros correctly.
72 2018-01-30 Richard Biener <rguenther@suse.de>
74 PR tree-optimization/83008
75 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
76 invariant and constant vector uses in stmts when they need
77 more than one stmt.
79 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
81 PR bootstrap/84017
82 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
83 * configure: Regenerate.
85 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
87 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
88 pattern.
89 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
90 Use gen_rtx_REG rather than gen_lowpart.
92 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
94 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
95 rather than 0 when creating partial subregs.
97 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
99 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
100 of usage.
102 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
104 PR target/81550
105 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
106 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
107 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
108 flags. This restores the settings used before the 2017-07-24.
109 Turning off pre increment/decrement/modify allows IVOPTS to
110 optimize DF/SF loops where the index is an int.
112 2018-01-29 Richard Biener <rguenther@suse.de>
113 Kelvin Nilsen <kelvin@gcc.gnu.org>
115 PR bootstrap/80867
116 * tree-vect-stmts.c (vectorizable_call): Don't call
117 targetm.vectorize_builtin_md_vectorized_function if callee is
118 NULL.
120 2018-01-22 Carl Love <cel@us.ibm.com>
122 * doc/extend.tex: Fix typo in second arg in
123 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
125 2018-01-29 Richard Biener <rguenther@suse.de>
127 PR tree-optimization/84086
128 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
129 (flush_ssaname_freelist): When SSA names were released reset
130 the SCEV hash table.
132 2018-01-29 Richard Biener <rguenther@suse.de>
134 PR tree-optimization/84057
135 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
136 removed paths when removing edges.
138 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
140 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
141 -mfunction-return=@var{choice}.
143 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
145 PR diagnostic/84034
146 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
147 Handle CR like TAB.
148 (layout::print_source_line): Likewise.
149 (test_get_line_width_without_trailing_whitespace): Add test cases.
151 2018-01-27 Jakub Jelinek <jakub@redhat.com>
153 PR middle-end/84040
154 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
155 debug insns.
157 2018-01-26 Jim Wilson <jimw@sifive.com>
159 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
161 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
162 specified.
164 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
166 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
167 and CMP + SUB-immediate -> SUBS.
169 2018-01-26 Martin Sebor <msebor@redhat.com>
171 PR tree-optimization/83896
172 * tree-ssa-strlen.c (get_string_len): Rename...
173 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
174 Avoid assuming length is constant.
175 (handle_char_store): Use HOST_WIDE_INT for string length.
177 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
179 PR target/81763
180 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
181 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
183 2018-01-26 Richard Biener <rguenther@suse.de>
185 PR rtl-optimization/84003
186 * dse.c (record_store): Only record redundant stores when
187 the earlier store aliases at least all accesses the later one does.
189 2018-01-26 Jakub Jelinek <jakub@redhat.com>
191 PR rtl-optimization/83985
192 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
193 REG_CFA_RESTORE insns.
194 (delete_unmarked_insns): Don't ignore separate shrink wrapping
195 REG_CFA_RESTORE insns here.
197 PR c/83989
198 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
199 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
201 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
203 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
204 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
205 (arc_init): Likewise.
206 (arc_override_options): Likewise.
207 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
208 value.
209 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
210 support.
211 * config/arc/arc.h (TARGET_DBNZ): Define.
212 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
213 properly set the tune attribute.
214 (dbnz): Use TARGET_DBNZ guard.
215 * config/arc/arc.opt (mtune): Add core3 option.
217 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
219 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
220 recognize new pic like addresses.
221 (arc_delegitimize_address): Clean up.
223 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
225 * config/arc/arc-arches.def: Option mrf16 valid for all
226 architectures.
227 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
228 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
229 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
230 * config/arc/arc-tables.opt: Regenerate.
231 * config/arc/arc.c (arc_conditional_register_usage): Handle
232 reduced register file case.
233 (arc_file_start): Set must have build attributes.
234 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
235 mrf16 option value.
236 * config/arc/arc.opt (mrf16): Add new option.
237 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
238 * config/arc/genmultilib.awk: Handle new mrf16 option.
239 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
240 * config/arc/t-multilib: Regenerate.
241 * doc/invoke.texi (ARC Options): Document mrf16 option.
243 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
245 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
246 * config/arc/arc.c (arc_handle_secure_attribute): New function.
247 (arc_attribute_table): Add 'secure_call' attribute.
248 (arc_print_operand): Print secure call operand.
249 (arc_function_ok_for_sibcall): Don't optimize tail calls when
250 secure.
251 (arc_is_secure_call_p): New function. * config/arc/arc.md
252 (call_i): Add support for sjli instruction.
253 (call_value_i): Likewise.
254 * config/arc/constraints.md (Csc): New constraint.
256 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
257 John Eric Martin <John.Martin@emmicro-us.com>
259 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
260 * config/arc/arc.c (_arc_jli_section): New struct.
261 (arc_jli_section): New type.
262 (rc_jli_sections): New static variable.
263 (arc_handle_jli_attribute): New function.
264 (arc_attribute_table): Add jli_always and jli_fixed attribute.
265 (arc_file_end): New function.
266 (TARGET_ASM_FILE_END): Define.
267 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
268 (arc_add_jli_section): New function.
269 (jli_call_scan): Likewise.
270 (arc_reorg): Call jli_call_scan.
271 (arc_output_addsi): Remove 'S' from printing asm operand.
272 (arc_is_jli_call_p): New function.
273 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
274 operand.
275 (movhi_insn): Likewise.
276 (movsi_insn): Likewise.
277 (movsi_set_cc_insn): Likewise.
278 (loadqi_update): Likewise.
279 (load_zeroextendqisi_update): Likewise.
280 (load_signextendqisi_update): Likewise.
281 (loadhi_update): Likewise.
282 (load_zeroextendhisi_update): Likewise.
283 (load_signextendhisi_update): Likewise.
284 (loadsi_update): Likewise.
285 (loadsf_update): Likewise.
286 (movsicc_insn): Likewise.
287 (bset_insn): Likewise.
288 (bxor_insn): Likewise.
289 (bclr_insn): Likewise.
290 (bmsk_insn): Likewise.
291 (bicsi3_insn): Likewise.
292 (cmpsi_cc_c_insn): Likewise.
293 (movsi_ne): Likewise.
294 (movsi_cond_exec): Likewise.
295 (clrsbsi2): Likewise.
296 (norm_f): Likewise.
297 (normw): Likewise.
298 (swap): Likewise.
299 (divaw): Likewise.
300 (flag): Likewise.
301 (sr): Likewise.
302 (kflag): Likewise.
303 (ffs): Likewise.
304 (ffs_f): Likewise.
305 (fls): Likewise.
306 (call_i): Remove 'S' asm letter, add jli instruction.
307 (call_value_i): Likewise.
308 * config/arc/arc.op (mjli-always): New option.
309 * config/arc/constraints.md (Cji): New constraint.
310 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
311 operand.
312 (subsf3_fpx): Likewise.
313 (mulsf3_fpx): Likewise.
314 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
315 asm operand.
316 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
317 function attrbutes.
318 * doc/invoke.texi (ARC): Document mjli-always option.
320 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
322 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
323 and use incw and decw where possible
324 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
326 2018-01-26 Richard Biener <rguenther@suse.de>
328 PR tree-optimization/81082
329 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
330 association if it requires casting to unsigned.
331 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
332 from fold_plusminus_mult_expr to catch important cases late when
333 range info is available.
335 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
337 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
338 * configure.ac (hidden_linkonce): New test.
339 * configure: Regenerate.
340 * config.in: Regenerate.
342 2018-01-26 Julia Koval <julia.koval@intel.com>
344 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
345 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
346 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
347 _mm_mask_bitshuffle_epi64_mask): Fix type.
348 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
349 USI_FTYPE_V4DI_V4DI_USI): Remove.
350 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
351 __builtin_ia32_vpshufbitqmb256_mask,
352 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
353 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
354 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
356 2018-01-26 Alan Modra <amodra@gmail.com>
358 PR target/84033
359 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
360 UNSPEC_VBPERMQ. Sort other unspecs.
362 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
364 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
366 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
368 PR middle-end/83055
369 * predict.c (drop_profile): Do not push/pop cfun; update also
370 node->count.
371 (handle_missing_profiles): Fix logic looking for zero profiles.
373 2018-01-25 Jakub Jelinek <jakub@redhat.com>
375 PR middle-end/83977
376 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
377 on functions with #pragma omp declare simd or functions with simd
378 attribute.
379 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
380 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
381 Remove trailing \n from warning_at calls.
383 2018-01-25 Tom de Vries <tom@codesourcery.com>
385 PR target/84028
386 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
387 for neutered workers.
389 2018-01-24 Joseph Myers <joseph@codesourcery.com>
391 PR target/68467
392 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
393 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
395 2017-01-24 Jeff Law <law@redhat.com>
397 PR target/83994
398 * i386.c (get_probe_interval): Move to earlier point.
399 (ix86_compute_frame_layout): If -fstack-clash-protection and
400 the frame is larger than the probe interval, then use pushes
401 to save registers rather than reg->mem moves.
402 (ix86_expand_prologue): Remove conditional for int_registers_saved
403 assertion.
405 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
407 PR target/84014
408 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
409 min/max for never referenced object.
411 2018-01-24 Jakub Jelinek <jakub@redhat.com>
413 PR middle-end/83977
414 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
415 here.
416 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
417 attributes from DECL_ATTRIBUTES (decl) without affecting
418 DECL_ATTRIBUTES (current_function_decl).
419 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
420 functions with non-NULL DECL_ABSTRACT_ORIGIN.
422 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
424 PR tree-optimization/83979
425 * fold-const.c (fold_comparison): Use constant_boolean_node
426 instead of boolean_{true,false}_node.
428 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
430 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
431 with zero counts.
433 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
435 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
436 Simplify the clause that sets the length attribute.
437 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
438 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
439 clause that sets the length attribute.
440 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
442 2018-01-24 Tom de Vries <tom@codesourcery.com>
444 PR target/83589
445 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
446 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
447 Add strict parameter.
448 (prevent_branch_around_nothing): Insert dummy insn between branch to
449 label and label with no ptx insn inbetween.
450 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
452 2018-01-24 Tom de Vries <tom@codesourcery.com>
454 PR target/81352
455 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
456 for neutered threads in warp.
457 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
459 2018-01-24 Richard Biener <rguenther@suse.de>
461 PR tree-optimization/83176
462 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
463 operands.
465 2018-01-24 Richard Biener <rguenther@suse.de>
467 PR tree-optimization/82819
468 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
469 code generating pluses that are no-ops in the target precision.
471 2018-01-24 Richard Biener <rguenther@suse.de>
473 PR middle-end/84000
474 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
476 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
478 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
479 to merge probabilities.
480 * predict.c (probably_never_executed): Also mark as cold functions
481 with global 0 profile and guessed local profile.
482 * profile-count.c (profile_probability::combine_with_count): New
483 member function.
484 * profile-count.h (profile_probability::operator*,
485 profile_probability::operator*=, profile_probability::operator/,
486 profile_probability::operator/=): Reduce precision to adjusted
487 and set value to guessed on contradictory divisions.
488 (profile_probability::combine_with_freq): Remove.
489 (profile_probability::combine_wiht_count): Declare.
490 (profile_count::force_nonzero):: Set to adjusted.
491 (profile_count::probability_in):: Set quality to adjusted.
492 * tree-ssa-tail-merge.c (replace_block_by): Use
493 combine_with_count.
495 2018-01-23 Andrew Waterman <andrew@sifive.com>
496 Jim Wilson <jimw@sifive.com>
498 * config/riscv/riscv.c (riscv_stack_boundary): New.
499 (riscv_option_override): Set riscv_stack_boundary. Handle
500 riscv_preferred_stack_boundary_arg.
501 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
502 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
503 (STACK_BOUNDARY): Set to riscv_stack_boundary.
504 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
505 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
506 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
508 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
510 PR target/83905
511 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
512 of struct ix86_frame.
513 (ix86_expand_epilogue): Likewise. Add a local variable for
514 the reg_save_offset field in struct ix86_frame.
516 2018-01-23 Bin Cheng <bin.cheng@arm.com>
518 PR tree-optimization/82604
519 * tree-loop-distribution.c (enum partition_kind): New enum item
520 PKIND_PARTIAL_MEMSET.
521 (partition_builtin_p): Support above new enum item.
522 (generate_code_for_partition): Ditto.
523 (compute_access_range): Differentiate cases that equality can be
524 proven at all loops, the innermost loops or no loops.
525 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
526 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
527 (finalize_partitions, distribute_loop): Don't fuse partition of
528 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
529 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
530 parloop is enabled.
532 2018-01-23 Martin Liska <mliska@suse.cz>
534 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
535 order to ignore the predictor.
536 (PRED_POLYMORPHIC_CALL): Likewise.
537 (PRED_RECURSIVE_CALL): Likewise.
539 2018-01-23 Martin Liska <mliska@suse.cz>
541 * tree-profile.c (tree_profiling): Print function header to
542 aware reader which function we are working on.
543 * value-prof.c (gimple_find_values_to_profile): Do not print
544 not interesting value histograms.
546 2018-01-23 Martin Liska <mliska@suse.cz>
548 * profile-count.h (enum profile_quality): Add
549 profile_uninitialized as the first value. Do not number values
550 as they are zero based.
551 (profile_count::verify): Update sanity check.
552 (profile_probability::verify): Likewise.
554 2018-01-23 Nathan Sidwell <nathan@acm.org>
556 * doc/invoke.texi (ffor-scope): Deprecate.
558 2018-01-23 David Malcolm <dmalcolm@redhat.com>
560 PR tree-optimization/83510
561 * domwalk.c (set_all_edges_as_executable): New function.
562 (dom_walker::dom_walker): Convert bool param
563 "skip_unreachable_blocks" to enum reachability. Move setup of
564 edge flags to set_all_edges_as_executable and only do it when
565 reachability is REACHABLE_BLOCKS.
566 * domwalk.h (enum dom_walker::reachability): New enum.
567 (dom_walker::dom_walker): Convert bool param
568 "skip_unreachable_blocks" to enum reachability.
569 (set_all_edges_as_executable): New decl.
570 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
571 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
572 "reachability".
573 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
574 but converting true to REACHABLE_BLOCKS.
575 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
576 * tree-vrp.c
577 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
578 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
579 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
580 REACHABLE_BLOCKS.
581 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
582 if check_all_array_refs will be called.
584 2018-01-23 David Malcolm <dmalcolm@redhat.com>
586 * tree.c (selftest::test_location_wrappers): Add more test
587 coverage.
589 2018-01-23 David Malcolm <dmalcolm@redhat.com>
591 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
592 (selftest::test_bit_in_range): Likewise.
594 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
596 PR testsuite/83888
597 * doc/sourcebuild.texi (vect_float): Say that the selector
598 only describes the situation when -funsafe-math-optimizations is on.
599 (vect_float_strict): Document.
601 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
603 PR tree-optimization/83965
604 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
605 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
606 instead of checking only for a reduction.
607 (vect_recog_widen_sum_pattern): Likewise.
609 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
611 * predict.c (probably_never_executed): Only use precise profile info.
612 (compute_function_frequency): Skip after inlining hack since we now
613 have quality checking.
615 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
617 * profile-count.h (profile_probability::very_unlikely,
618 profile_probability::unlikely, profile_probability::even): Set
619 precision to guessed.
621 2018-01-23 Richard Biener <rguenther@suse.de>
623 PR tree-optimization/83963
624 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
625 Properly terminate dominator walk when crossing the exit edge not
626 when visiting its source block.
628 2018-01-23 Jakub Jelinek <jakub@redhat.com>
630 PR c++/83918
631 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
632 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
634 2018-01-22 Jakub Jelinek <jakub@redhat.com>
636 PR tree-optimization/83957
637 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
638 semicolon after for body surrounded by braces.
640 PR tree-optimization/83081
641 * profile-count.h (profile_probability::split): New method.
642 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
643 Use profile_probability::split.
644 (do_compare_rtx_and_jump): Fix adjustment of probabilities
645 when splitting a single conditional jump into 2.
647 2018-01-22 David Malcolm <dmalcolm@redhat.com>
649 PR tree-optimization/69452
650 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
651 decl.
653 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
655 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
656 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
657 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
659 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
661 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
662 * config/rl78/rl78.md: New define_expand "movdi"
663 * config/rl78/rl78.c: New function definition rl78_split_movdi
665 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
667 PR target/83862
668 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
669 no longer used.
670 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
671 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
672 128-bit to produce an UNSPEC move to get the double word with the
673 signbit and then a shift directly to do signbit.
674 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
675 implementation with a new version that just does either a direct
676 move or a regular move. Move memory interface to separate insns.
677 Move insns so they are next to the expander.
678 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
679 with signbit move. Split big and little endian case.
680 (signbit<mode>2_dm_mem_le): Likewise.
681 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
682 (signbit<mode>2_dm2): Likewise.
684 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
686 * config/rl78/rl78.md: New define_expand "anddi3".
688 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
690 * config/rl78/rl78.md: New define_expand "umindi3".
692 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
694 * config/rl78/rl78.md: New define_expand "smindi3".
696 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
698 * config/rl78/rl78.md: New define_expand "smaxdi3".
700 2018-01-22 Carl Love <cel@us.ibm.com>
702 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
703 LVX_V1TI): Add macro expansion.
704 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
705 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
706 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
707 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
708 Change check to determine if the instruction is a byte reversing
709 entry. Fix typo in comment.
710 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
711 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
712 Add def_builtin calls for new builtins.
713 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
714 Add define_insn expansion.
716 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
718 * config/rl78/rl78.md: New define_expand "umaxdi3".
720 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
722 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
723 for non-QImode registers
725 2018-01-22 Richard Biener <rguenther@suse.de>
727 PR tree-optimization/83963
728 * graphite-scop-detection.c (scop_detection::get_sese): Delay
729 including the loop exit block.
730 (scop_detection::merge_sese): Likewise.
731 (scop_detection::add_scop): Do it here instead.
733 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
735 * doc/sourcebuild.texi (arm_softfloat): Document.
737 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
739 PR gcc/77734
740 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
741 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
742 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
744 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
745 David Edelsohn <dje.gcc@gmail.com>
747 PR target/83946
748 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
749 Change "crset eq" to "crset 2".
750 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
751 (*call_indirect_aix<mode>_nospec): Likewise.
752 (*call_value_indirect_aix<mode>_nospec): Likewise.
753 (*call_indirect_elfv2<mode>_nospec): Likewise.
754 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
755 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
756 change assembly output from . to $.
757 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
758 (indirect_jump<mode>_nospec): Change assembly output from . to $.
759 (*tablejump<mode>_internal1_nospec): Likewise.
761 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
763 PR target/80870
764 * config/sh/sh_optimize_sett_clrt.cc:
765 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
767 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
769 PR tree-optimization/83940
770 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
771 offset_dt to vect_constant_def rather than vect_unknown_def_type.
772 (vect_check_load_store_mask): Add a mask_dt_out parameter and
773 use it to pass back the definition type.
774 (vect_check_store_rhs): Likewise rhs_dt_out.
775 (vect_build_gather_load_calls): Add a mask_dt argument and use
776 it instead of a call to vect_is_simple_use.
777 (vectorizable_store): Update calls to vect_check_load_store_mask
778 and vect_check_store_rhs. Use the dt returned by the latter instead
779 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
780 instead of calls to vect_is_simple_use. Pass the scalar rather
781 than the vector operand to vect_is_simple_use when handling
782 second and subsequent copies of an rhs value.
783 (vectorizable_load): Update calls to vect_check_load_store_mask
784 and vect_build_gather_load_calls. Use the cached mask_dt and
785 gs_info.offset_dt instead of calls to vect_is_simple_use.
787 2018-01-20 Jakub Jelinek <jakub@redhat.com>
789 PR middle-end/83945
790 * tree-emutls.c: Include gimplify.h.
791 (lower_emutls_2): New function.
792 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
793 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
794 it before further processing.
796 PR target/83930
797 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
798 UINTVAL (trueop1) instead of INTVAL (op1).
800 2018-01-19 Jakub Jelinek <jakub@redhat.com>
802 PR debug/81570
803 PR debug/83728
804 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
805 INCOMING_FRAME_SP_OFFSET if not defined.
806 (scan_trace): Add ENTRY argument. If true and
807 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
808 emit a note to adjust the CFA offset.
809 (create_cfi_notes): Adjust scan_trace callers.
810 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
811 INCOMING_FRAME_SP_OFFSET in the CIE.
812 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
813 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
814 Likewise.
815 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
816 * doc/tm.texi: Regenerated.
818 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
820 PR rtl-optimization/83147
821 * lra-constraints.c (remove_inheritance_pseudos): Use
822 lra_substitute_pseudo_within_insn.
824 2018-01-19 Tom de Vries <tom@codesourcery.com>
825 Cesar Philippidis <cesar@codesourcery.com>
827 PR target/83920
828 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
830 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
832 PR target/83790
833 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
834 spaces for function labels.
836 2018-01-19 Martin Liska <mliska@suse.cz>
838 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
839 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
840 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
841 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
842 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
843 (PRED_CONST_RETURN): Change from 69 to 65.
844 (PRED_NULL_RETURN): Change from 91 to 71.
845 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
846 (PRED_LOOP_GUARD): Change from 66 to 73.
848 2018-01-19 Martin Liska <mliska@suse.cz>
850 * predict.c (predict_insn_def): Add new assert.
851 (struct branch_predictor): Change type to signed integer.
852 (test_prediction_value_range): Amend test to cover
853 PROB_UNINITIALIZED.
854 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
855 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
856 (PRED_LOOP_ITERATIONS_MAX): Likewise.
857 (PRED_LOOP_IV_COMPARE): Likewise.
858 * predict.h (PROB_UNINITIALIZED): Define new constant.
860 2018-01-19 Martin Liska <mliska@suse.cz>
862 * predict.c (dump_prediction): Add new format for
863 analyze_brprob.py script which is enabled with -details
864 suboption.
865 * profile-count.h (precise_p): New function.
867 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
869 PR tree-optimization/83922
870 * tree-vect-loop.c (vect_verify_full_masking): Return false if
871 there are no statements that need masking.
872 (vect_active_double_reduction_p): New function.
873 (vect_analyze_loop_operations): Use it when handling phis that
874 are not in the loop header.
876 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
878 PR tree-optimization/83914
879 * tree-vect-loop.c (vectorizable_induction): Don't convert
880 init_expr or apply the peeling adjustment for inductions
881 that are nested within the vectorized loop.
883 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
885 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
886 instead of NEG.
888 2018-01-18 Jakub Jelinek <jakub@redhat.com>
890 PR sanitizer/81715
891 PR testsuite/83882
892 * function.h (gimplify_parameters): Add gimple_seq * argument.
893 * function.c: Include gimple.h and options.h.
894 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
895 for the added local temporaries if needed.
896 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
897 if there are any parameter cleanups, wrap whole body into a
898 try/finally with the cleanups.
900 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
902 PR target/82964
903 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
904 Use GET_MODE_CLASS for scalar floating point.
906 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
908 PR ipa/82256
909 patch by PaX Team
910 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
911 Fix call of call_cgraph_insertion_hooks.
913 2018-01-18 Martin Sebor <msebor@redhat.com>
915 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
917 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
919 PR ipa/83619
920 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
921 frequencies.
923 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
925 PR other/70268
926 * common.opt: (-ffile-prefix-map): New option.
927 * opts.c (common_handle_option): Defer it.
928 * opts-global.c (handle_common_deferred_options): Handle it.
929 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
930 * file-prefix-map.h: New file.
931 (remap_debug_filename, add_debug_prefix_map): ...here.
932 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
933 * final.c (debug_prefix_map, add_debug_prefix_map
934 remap_debug_filename): Move to...
935 * file-prefix-map.c: New file.
936 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
937 generalize, get rid of alloca(), use strrchr() instead of strchr().
938 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
939 Implement in terms of add_prefix_map().
940 (remap_macro_filename, remap_debug_filename): Implement in term of
941 remap_filename().
942 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
943 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
944 * dbxout.c: Include file-prefix-map.h.
945 * varasm.c: Likewise.
946 * vmsdbgout.c: Likewise.
947 * xcoffout.c: Likewise.
948 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
949 * doc/cppopts.texi (-fmacro-prefix-map): Document.
950 * doc/invoke.texi (-ffile-prefix-map): Document.
951 (-fdebug-prefix-map): Update description.
953 2018-01-18 Martin Liska <mliska@suse.cz>
955 * config/i386/i386.c (indirect_thunk_name): Document that also
956 lfence is emitted.
957 (output_indirect_thunk): Document why both instructions
958 (pause and lfence) are generated.
960 2018-01-18 Richard Biener <rguenther@suse.de>
962 PR tree-optimization/83887
963 * graphite-scop-detection.c
964 (scop_detection::get_nearest_dom_with_single_entry): Remove.
965 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
966 (scop_detection::merge_sese): Re-implement with a flood-fill
967 algorithm that properly finds a SESE region if it exists.
969 2018-01-18 Jakub Jelinek <jakub@redhat.com>
971 PR c/61240
972 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
973 pointer_diff optimizations use view_convert instead of convert.
975 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
977 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
978 Generate different code for -mno-speculate-indirect-jumps.
979 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
980 (*call_indirect_aix<mode>): Disable for
981 -mno-speculate-indirect-jumps.
982 (*call_indirect_aix<mode>_nospec): New define_insn.
983 (*call_value_indirect_aix<mode>): Disable for
984 -mno-speculate-indirect-jumps.
985 (*call_value_indirect_aix<mode>_nospec): New define_insn.
986 (*sibcall_nonlocal_sysv<mode>): Generate different code for
987 -mno-speculate-indirect-jumps.
988 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
990 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
992 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
993 long double type, set the flags for noting the default long double
994 type, even if we don't pass or return a long double type.
996 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
998 PR ipa/83051
999 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1000 failure.
1002 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1004 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1005 support for merge[hl].
1006 (fold_mergehl_helper): New helper function.
1007 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1008 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1009 (altivec_vmrglw_direct): Add xxmrglw insn.
1011 2018-01-17 Andrew Waterman <andrew@sifive.com>
1013 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1014 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1016 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1018 PR lto/83121
1019 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1020 call the lto_location_cache before reading the
1021 DECL_SOURCE_LOCATION of the types.
1023 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1024 Richard Sandiford <richard.sandiford@linaro.org>
1026 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1027 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1028 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1029 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1030 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1031 Add declaration.
1032 * config/aarch64/constraints.md (aarch64_movti_operand):
1033 Limit immediates.
1034 * config/aarch64/predicates.md (Uti): Add new constraint.
1036 2018-01-17 Carl Love <cel@us.ibm.com>
1037 * config/rs6000/vsx.md (define_expand xl_len_r,
1038 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1039 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1040 lxvll.
1041 (define_expand, define_insn): Move the shift left from the
1042 define_insn to the define_expand for lxvl and stxvl instructions.
1043 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1044 and XL_LEN_R definitions to PURE.
1046 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1048 * config/i386/i386.c (indirect_thunk_name): Declare regno
1049 as unsigned int. Compare regno with INVALID_REGNUM.
1050 (output_indirect_thunk): Ditto.
1051 (output_indirect_thunk_function): Ditto.
1052 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1053 in the call to output_indirect_thunk_function.
1055 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1057 PR middle-end/83884
1058 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1059 rather than the size of inner_type to determine the stack slot size
1060 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1062 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1064 PR target/83546
1065 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1066 to PTA_SILVERMONT.
1068 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1070 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1071 endian Linux systems to optionally enable multilibs for selecting
1072 the long double type if the user configured an explicit type.
1073 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1074 have no long double multilibs if not defined.
1075 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1076 warn if the user used -mabi={ieee,ibm}longdouble and we built
1077 multilibs for long double.
1078 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1079 appropriate multilib option.
1080 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1081 multilib options.
1082 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1083 for building long double multilibs.
1084 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1086 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1088 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1089 copies.
1091 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1092 64 bits.
1093 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1094 128 bits.
1096 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1097 variables.
1099 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1100 return value.
1102 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1104 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1105 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1107 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1109 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1110 different rtl trees depending on TARGET_64BIT.
1111 (rs6000_gen_lvx): Likewise.
1113 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1115 * config/visium/visium.md (nop): Tweak comment.
1116 (hazard_nop): Likewise.
1118 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1120 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1121 -mspeculate-indirect-jumps.
1122 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1123 for -mno-speculate-indirect-jumps.
1124 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1125 (*call_value_indirect_elfv2<mode>): Disable for
1126 -mno-speculate-indirect-jumps.
1127 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1128 (indirect_jump): Emit different RTL for
1129 -mno-speculate-indirect-jumps.
1130 (*indirect_jump<mode>): Disable for
1131 -mno-speculate-indirect-jumps.
1132 (*indirect_jump<mode>_nospec): New define_insn.
1133 (tablejump): Emit different RTL for
1134 -mno-speculate-indirect-jumps.
1135 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1136 (tablejumpsi_nospec): New define_expand.
1137 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1138 (tablejumpdi_nospec): New define_expand.
1139 (*tablejump<mode>_internal1): Disable for
1140 -mno-speculate-indirect-jumps.
1141 (*tablejump<mode>_internal1_nospec): New define_insn.
1142 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1143 option.
1145 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1147 * caller-save.c (insert_save): Drop unnecessary parameter. All
1148 callers updated.
1150 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1151 Richard Biener <rguenth@suse.de>
1153 PR libgomp/83590
1154 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1155 return early, inline manually is_gimple_sizepos. Make sure if we
1156 call gimplify_expr we don't end up with a gimple constant.
1157 * tree.c (variably_modified_type_p): Don't return true for
1158 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1159 * gimplify.h (is_gimple_sizepos): Remove.
1161 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1163 PR tree-optimization/83857
1164 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1165 vectorizable_live_operation for pure SLP statements.
1166 (vectorizable_live_operation): Handle PHIs.
1168 2018-01-16 Richard Biener <rguenther@suse.de>
1170 PR tree-optimization/83867
1171 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1172 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1174 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1176 PR c/83844
1177 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1178 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1179 If off is not INTEGER_CST, issue a may not be aligned warning
1180 rather than isn't aligned. Use isn%'t rather than isn't.
1181 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1182 into MULT_EXPR.
1183 <case MULT_EXPR>: Improve the case when bottom and one of the
1184 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1185 operand, in that case check if the other operand is multiple of
1186 bottom divided by the INTEGER_CST operand.
1188 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1190 PR target/83858
1191 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1192 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1193 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1194 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1195 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1196 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1197 (pa_function_arg_size): New function.
1199 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1201 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1202 in a separate statement.
1204 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1206 PR tree-optimization/83847
1207 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1208 group gathers and scatters.
1210 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1212 PR rtl-optimization/86620
1213 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1215 PR rtl-optimization/83213
1216 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1217 to last if both are JUMP_INSNs.
1219 PR tree-optimization/83843
1220 * gimple-ssa-store-merging.c
1221 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1222 store_immediate_info for bswap/nop orig_stores.
1224 2018-01-15 Andrew Waterman <andrew@sifive.com>
1226 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1227 !TARGET_MUL.
1228 <UDIV>: Increase cost if !TARGET_DIV.
1230 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1232 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1233 (define_attr "cr_logical_3op"): New.
1234 (cceq_ior_compare): Adjust.
1235 (cceq_ior_compare_complement): Adjust.
1236 (*cceq_rev_compare): Adjust.
1237 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1238 (is_cracked_insn): Adjust.
1239 (insn_must_be_first_in_group): Adjust.
1240 * config/rs6000/40x.md: Adjust.
1241 * config/rs6000/440.md: Adjust.
1242 * config/rs6000/476.md: Adjust.
1243 * config/rs6000/601.md: Adjust.
1244 * config/rs6000/603.md: Adjust.
1245 * config/rs6000/6xx.md: Adjust.
1246 * config/rs6000/7450.md: Adjust.
1247 * config/rs6000/7xx.md: Adjust.
1248 * config/rs6000/8540.md: Adjust.
1249 * config/rs6000/cell.md: Adjust.
1250 * config/rs6000/e300c2c3.md: Adjust.
1251 * config/rs6000/e500mc.md: Adjust.
1252 * config/rs6000/e500mc64.md: Adjust.
1253 * config/rs6000/e5500.md: Adjust.
1254 * config/rs6000/e6500.md: Adjust.
1255 * config/rs6000/mpc.md: Adjust.
1256 * config/rs6000/power4.md: Adjust.
1257 * config/rs6000/power5.md: Adjust.
1258 * config/rs6000/power6.md: Adjust.
1259 * config/rs6000/power7.md: Adjust.
1260 * config/rs6000/power8.md: Adjust.
1261 * config/rs6000/power9.md: Adjust.
1262 * config/rs6000/rs64.md: Adjust.
1263 * config/rs6000/titan.md: Adjust.
1265 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1267 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1268 ix86_indirect_branch_register logic.
1270 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1272 * config/i386/constraints.md (Bs): Update
1273 ix86_indirect_branch_register check. Don't check
1274 ix86_indirect_branch_register with GOT_memory_operand.
1275 (Bw): Likewise.
1276 * config/i386/predicates.md (GOT_memory_operand): Don't check
1277 ix86_indirect_branch_register here.
1278 (GOT32_symbol_operand): Likewise.
1280 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1282 * config/i386/predicates.md (constant_call_address_operand):
1283 Rewrite ix86_indirect_branch_register logic.
1284 (sibcall_insn_operand): Likewise.
1286 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1288 * config/i386/constraints.md (Bs): Replace
1289 ix86_indirect_branch_thunk_register with
1290 ix86_indirect_branch_register.
1291 (Bw): Likewise.
1292 * config/i386/i386.md (indirect_jump): Likewise.
1293 (tablejump): Likewise.
1294 (*sibcall_memory): Likewise.
1295 (*sibcall_value_memory): Likewise.
1296 Peepholes of indirect call and jump via memory: Likewise.
1297 * config/i386/i386.opt: Likewise.
1298 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1299 (GOT_memory_operand): Likewise.
1300 (call_insn_operand): Likewise.
1301 (sibcall_insn_operand): Likewise.
1302 (GOT32_symbol_operand): Likewise.
1304 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1306 PR middle-end/83837
1307 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1308 type rather than type addr's type points to.
1309 (expand_omp_atomic_mutex): Likewise.
1310 (expand_omp_atomic): Likewise.
1312 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1314 PR target/83839
1315 * config/i386/i386.c (output_indirect_thunk_function): Use
1316 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1317 for __x86_return_thunk.
1319 2018-01-15 Richard Biener <rguenther@suse.de>
1321 PR middle-end/83850
1322 * expmed.c (extract_bit_field_1): Fix typo.
1324 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1326 PR target/83687
1327 * config/arm/iterators.md (VF): New mode iterator.
1328 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1329 Remove integer-related logic from pattern.
1330 (neon_vabd<mode>_3): Likewise.
1332 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1334 PR middle-end/82694
1335 * common.opt (fstrict-overflow): No longer an alias.
1336 (fwrapv-pointer): New option.
1337 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1338 also for pointer types based on flag_wrapv_pointer.
1339 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1340 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1341 opts->x_flag_wrapv got set.
1342 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1343 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1344 POINTER_TYPE_OVERFLOW_UNDEFINED.
1345 * match.pd: Likewise in address comparison pattern.
1346 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1348 2018-01-15 Richard Biener <rguenther@suse.de>
1350 PR lto/83804
1351 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1352 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1353 Reset type names to their identifier if their TYPE_DECL doesn't
1354 have linkage (and thus is used for ODR and devirt).
1355 (save_debug_info_for_decl): Remove.
1356 (save_debug_info_for_type): Likewise.
1357 (add_tree_to_fld_list): Adjust.
1358 * tree-pretty-print.c (dump_generic_node): Make dumping of
1359 type names more robust.
1361 2018-01-15 Richard Biener <rguenther@suse.de>
1363 * BASE-VER: Bump to 8.0.1.
1365 2018-01-14 Martin Sebor <msebor@redhat.com>
1367 PR other/83508
1368 * builtins.c (check_access): Avoid warning when the no-warning bit
1369 is set.
1371 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1373 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1374 * ira-color (allocno_hard_regs_compare): Likewise.
1376 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1378 PR target/83013
1379 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1380 Use .pushsection/.popsection.
1382 2018-01-14 Martin Sebor <msebor@redhat.com>
1384 PR c++/81327
1385 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1387 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1389 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1390 entry from extra_headers.
1391 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1392 extra_headers, make the list bitwise identical to the i?86-*-* one.
1394 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1396 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1397 -mcmodel=large with -mindirect-branch=thunk,
1398 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1399 -mfunction-return=thunk-extern.
1400 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1401 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1402 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1404 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1406 * config/i386/i386.c (print_reg): Print the name of the full
1407 integer register without '%'.
1408 (ix86_print_operand): Handle 'V'.
1409 * doc/extend.texi: Document 'V' modifier.
1411 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1413 * config/i386/constraints.md (Bs): Disallow memory operand for
1414 -mindirect-branch-register.
1415 (Bw): Likewise.
1416 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1417 (GOT_memory_operand): Likewise.
1418 (call_insn_operand): Likewise.
1419 (sibcall_insn_operand): Likewise.
1420 (GOT32_symbol_operand): Likewise.
1421 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1422 for -mindirect-branch-register.
1423 (tablejump): Likewise.
1424 (*sibcall_memory): Likewise.
1425 (*sibcall_value_memory): Likewise.
1426 Disallow peepholes of indirect call and jump via memory for
1427 -mindirect-branch-register.
1428 (*call_pop): Replace m with Bw.
1429 (*call_value_pop): Likewise.
1430 (*sibcall_pop_memory): Replace m with Bs.
1431 * config/i386/i386.opt (mindirect-branch-register): New option.
1432 * doc/invoke.texi: Document -mindirect-branch-register option.
1434 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1436 * config/i386/i386-protos.h (ix86_output_function_return): New.
1437 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1438 set function_return_type.
1439 (indirect_thunk_name): Add ret_p to indicate thunk for function
1440 return.
1441 (output_indirect_thunk_function): Pass false to
1442 indirect_thunk_name.
1443 (ix86_output_indirect_branch_via_reg): Likewise.
1444 (ix86_output_indirect_branch_via_push): Likewise.
1445 (output_indirect_thunk_function): Create alias for function
1446 return thunk if regno < 0.
1447 (ix86_output_function_return): New function.
1448 (ix86_handle_fndecl_attribute): Handle function_return.
1449 (ix86_attribute_table): Add function_return.
1450 * config/i386/i386.h (machine_function): Add
1451 function_return_type.
1452 * config/i386/i386.md (simple_return_internal): Use
1453 ix86_output_function_return.
1454 (simple_return_internal_long): Likewise.
1455 * config/i386/i386.opt (mfunction-return=): New option.
1456 (indirect_branch): Mention -mfunction-return=.
1457 * doc/extend.texi: Document function_return function attribute.
1458 * doc/invoke.texi: Document -mfunction-return= option.
1460 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1462 * config/i386/i386-opts.h (indirect_branch): New.
1463 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1464 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1465 with local indirect jump when converting indirect call and jump.
1466 (ix86_set_indirect_branch_type): New.
1467 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1468 (indirectlabelno): New.
1469 (indirect_thunk_needed): Likewise.
1470 (indirect_thunk_bnd_needed): Likewise.
1471 (indirect_thunks_used): Likewise.
1472 (indirect_thunks_bnd_used): Likewise.
1473 (INDIRECT_LABEL): Likewise.
1474 (indirect_thunk_name): Likewise.
1475 (output_indirect_thunk): Likewise.
1476 (output_indirect_thunk_function): Likewise.
1477 (ix86_output_indirect_branch_via_reg): Likewise.
1478 (ix86_output_indirect_branch_via_push): Likewise.
1479 (ix86_output_indirect_branch): Likewise.
1480 (ix86_output_indirect_jmp): Likewise.
1481 (ix86_code_end): Call output_indirect_thunk_function if needed.
1482 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1483 needed.
1484 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1485 (ix86_attribute_table): Add indirect_branch.
1486 * config/i386/i386.h (machine_function): Add indirect_branch_type
1487 and has_local_indirect_jump.
1488 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1489 to true.
1490 (tablejump): Likewise.
1491 (*indirect_jump): Use ix86_output_indirect_jmp.
1492 (*tablejump_1): Likewise.
1493 (simple_return_indirect_internal): Likewise.
1494 * config/i386/i386.opt (mindirect-branch=): New option.
1495 (indirect_branch): New.
1496 (keep): Likewise.
1497 (thunk): Likewise.
1498 (thunk-inline): Likewise.
1499 (thunk-extern): Likewise.
1500 * doc/extend.texi: Document indirect_branch function attribute.
1501 * doc/invoke.texi: Document -mindirect-branch= option.
1503 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1505 PR ipa/83051
1506 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1508 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1510 * ipa-inline.c (want_inline_small_function_p): Return false if
1511 inlining has already failed with CIF_FINAL_ERROR.
1512 (update_caller_keys): Call want_inline_small_function_p before
1513 can_inline_edge_p.
1514 (update_callee_keys): Likewise.
1516 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1518 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1519 New function.
1520 (rs6000_quadword_masked_address_p): Likewise.
1521 (quad_aligned_load_p): Likewise.
1522 (quad_aligned_store_p): Likewise.
1523 (const_load_sequence_p): Add comment to describe the outer-most loop.
1524 (mimic_memory_attributes_and_flags): New function.
1525 (rs6000_gen_stvx): Likewise.
1526 (replace_swapped_aligned_store): Likewise.
1527 (rs6000_gen_lvx): Likewise.
1528 (replace_swapped_aligned_load): Likewise.
1529 (replace_swapped_load_constant): Capitalize argument name in
1530 comment describing this function.
1531 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1532 and stores that access quad-word aligned addresses and replace
1533 with stvx or lvx instructions when appropriate.
1534 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1535 New function prototype.
1536 (rs6000_quadword_masked_address_p): Likewise.
1537 (rs6000_gen_lvx): Likewise.
1538 (rs6000_gen_stvx): Likewise.
1539 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1540 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1541 when memory address is aligned.
1542 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1543 this split to select lvx instruction when memory address is aligned.
1544 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1545 instruction when memory address is aligned.
1546 (*vsx_le_perm_load_v16qi): Likewise.
1547 (four unnamed splitters): Modify to select the stvx instruction
1548 when memory is aligned.
1550 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1552 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1553 which appears in the queue multiple times.
1555 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1556 Alan Hayward <alan.hayward@arm.com>
1557 David Sherwood <david.sherwood@arm.com>
1559 * tree-vectorizer.h (vec_lower_bound): New structure.
1560 (_loop_vec_info): Add check_nonzero and lower_bounds.
1561 (LOOP_VINFO_CHECK_NONZERO): New macro.
1562 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1563 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1564 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1565 fields. Make seg_len the distance travelled, not including the
1566 access size.
1567 (dr_direction_indicator): Declare.
1568 (dr_zero_step_indicator): Likewise.
1569 (dr_known_forward_stride_p): Likewise.
1570 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1571 tree-ssanames.h.
1572 (runtime_alias_check_p): Allow runtime alias checks with
1573 variable strides.
1574 (operator ==): Compare access_size and align.
1575 (prune_runtime_alias_test_list): Rework for new distinction between
1576 the access_size and seg_len.
1577 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1578 segment lengths.
1579 (get_segment_min_max): New function.
1580 (create_intersect_range_checks): Use it.
1581 (dr_step_indicator): New function.
1582 (dr_direction_indicator): Likewise.
1583 (dr_zero_step_indicator): Likewise.
1584 (dr_known_forward_stride_p): Likewise.
1585 * tree-loop-distribution.c (data_ref_segment_size): Return
1586 DR_STEP * (niters - 1).
1587 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1588 constructor.
1589 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1590 (vect_preserves_scalar_order_p): New function, split out from...
1591 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1592 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1593 (vect_vfa_access_size): New function.
1594 (vect_vfa_align): Likewise.
1595 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1596 (dump_lower_bound): New function.
1597 (vect_check_lower_bound): Likewise.
1598 (vect_small_gap_p): Likewise.
1599 (vectorizable_with_step_bound_p): Likewise.
1600 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1601 depencies if the vectorization factor is 1. Convert the checks
1602 for nonzero steps into checks on the bounds of DR_STEP. Try using
1603 a bunds check for variable steps if the minimum required step is
1604 relatively small. Update calls to the dr_with_seg_len
1605 constructor and to vect_compile_time_alias.
1606 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1607 function.
1608 (vect_loop_versioning): Call it.
1609 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1610 when retrying.
1611 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1613 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1614 Alan Hayward <alan.hayward@arm.com>
1615 David Sherwood <david.sherwood@arm.com>
1617 * doc/sourcebuild.texi (vect_scatter_store): Document.
1618 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1619 optabs.
1620 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1621 Document.
1622 * genopinit.c (main): Add supports_vec_scatter_store and
1623 supports_vec_scatter_store_cached to target_optabs.
1624 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1625 IFN_MASK_SCATTER_STORE.
1626 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1627 functions.
1628 * internal-fn.h (internal_store_fn_p): Declare.
1629 (internal_fn_stored_value_index): Likewise.
1630 * internal-fn.c (scatter_store_direct): New macro.
1631 (expand_scatter_store_optab_fn): New function.
1632 (direct_scatter_store_optab_supported_p): New macro.
1633 (internal_store_fn_p): New function.
1634 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1635 IFN_MASK_SCATTER_STORE.
1636 (internal_fn_mask_index): Likewise.
1637 (internal_fn_stored_value_index): New function.
1638 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1639 for scatter stores.
1640 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1641 * optabs-query.c (supports_vec_scatter_store_p): New function.
1642 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1643 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1644 true for scatter stores.
1645 (vect_gather_scatter_fn_p): Handle scatter stores too.
1646 (vect_check_gather_scatter): Consider using scatter stores if
1647 supports_vec_scatter_store_p.
1648 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1649 scatter stores too.
1650 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1651 internal_fn_stored_value_index.
1652 (check_load_store_masking): Handle scatter stores too.
1653 (vect_get_store_rhs): Make public.
1654 (vectorizable_call): Use internal_store_fn_p.
1655 (vectorizable_store): Handle scatter store internal functions.
1656 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1657 when deciding whether the end of the group has been reached.
1658 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1659 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1660 (mask_scatter_store<mode>): New insns.
1662 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1663 Alan Hayward <alan.hayward@arm.com>
1664 David Sherwood <david.sherwood@arm.com>
1666 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1667 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1668 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1669 function.
1670 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1671 Use vect_truncate_gather_scatter_offset if we can't treat the
1672 operation as a normal gather load or scatter store.
1673 (get_group_load_store_type): Take the gather_scatter_info
1674 as argument. Try using a gather load or scatter store for
1675 single-element groups.
1676 (get_load_store_type): Update calls to get_group_load_store_type
1677 and vect_use_strided_gather_scatters_p.
1679 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1680 Alan Hayward <alan.hayward@arm.com>
1681 David Sherwood <david.sherwood@arm.com>
1683 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1684 optional tree argument.
1685 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1686 null target hooks.
1687 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1688 but continue to use the current value as a fallback.
1689 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1690 to compare the updates.
1691 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1692 (get_load_store_type): Use it when handling a strided access.
1693 (vect_get_strided_load_store_ops): New function.
1694 (vect_get_data_ptr_increment): Likewise.
1695 (vectorizable_load): Handle strided gather loads. Always pass
1696 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1698 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1699 Alan Hayward <alan.hayward@arm.com>
1700 David Sherwood <david.sherwood@arm.com>
1702 * doc/md.texi (gather_load@var{m}): Document.
1703 (mask_gather_load@var{m}): Likewise.
1704 * genopinit.c (main): Add supports_vec_gather_load and
1705 supports_vec_gather_load_cached to target_optabs.
1706 * optabs-tree.c (init_tree_optimization_optabs): Use
1707 ggc_cleared_alloc to allocate target_optabs.
1708 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1709 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1710 functions.
1711 * internal-fn.h (internal_load_fn_p): Declare.
1712 (internal_gather_scatter_fn_p): Likewise.
1713 (internal_fn_mask_index): Likewise.
1714 (internal_gather_scatter_fn_supported_p): Likewise.
1715 * internal-fn.c (gather_load_direct): New macro.
1716 (expand_gather_load_optab_fn): New function.
1717 (direct_gather_load_optab_supported_p): New macro.
1718 (direct_internal_fn_optab): New function.
1719 (internal_load_fn_p): Likewise.
1720 (internal_gather_scatter_fn_p): Likewise.
1721 (internal_fn_mask_index): Likewise.
1722 (internal_gather_scatter_fn_supported_p): Likewise.
1723 * optabs-query.c (supports_at_least_one_mode_p): New function.
1724 (supports_vec_gather_load_p): Likewise.
1725 * optabs-query.h (supports_vec_gather_load_p): Declare.
1726 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1727 and memory_type field.
1728 (NUM_PATTERNS): Bump to 15.
1729 * tree-vect-data-refs.c: Include internal-fn.h.
1730 (vect_gather_scatter_fn_p): New function.
1731 (vect_describe_gather_scatter_call): Likewise.
1732 (vect_check_gather_scatter): Try using internal functions for
1733 gather loads. Recognize existing calls to a gather load function.
1734 (vect_analyze_data_refs): Consider using gather loads if
1735 supports_vec_gather_load_p.
1736 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1737 (vect_get_gather_scatter_offset_type): Likewise.
1738 (vect_convert_mask_for_vectype): Likewise.
1739 (vect_add_conversion_to_patterm): Likewise.
1740 (vect_try_gather_scatter_pattern): Likewise.
1741 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1742 (vect_vect_recog_func_ptrs): Add it.
1743 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1744 internal_fn_mask_index and internal_gather_scatter_fn_p.
1745 (check_load_store_masking): Take the gather_scatter_info as an
1746 argument and handle gather loads.
1747 (vect_get_gather_scatter_ops): New function.
1748 (vectorizable_call): Check internal_load_fn_p.
1749 (vectorizable_load): Likewise. Handle gather load internal
1750 functions.
1751 (vectorizable_store): Update call to check_load_store_masking.
1752 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1753 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1754 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1755 (aarch64_gather_scale_operand_d): New predicates.
1756 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1757 (mask_gather_load<mode>): New insns.
1759 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1760 Alan Hayward <alan.hayward@arm.com>
1761 David Sherwood <david.sherwood@arm.com>
1763 * optabs.def (fold_left_plus_optab): New optab.
1764 * doc/md.texi (fold_left_plus_@var{m}): Document.
1765 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1766 * internal-fn.c (fold_left_direct): Define.
1767 (expand_fold_left_optab_fn): Likewise.
1768 (direct_fold_left_optab_supported_p): Likewise.
1769 * fold-const-call.c (fold_const_fold_left): New function.
1770 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1771 * tree-parloops.c (valid_reduction_p): New function.
1772 (gather_scalar_reductions): Use it.
1773 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1774 (vect_finish_replace_stmt): Declare.
1775 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1776 (needs_fold_left_reduction_p): New function, split out from...
1777 (vect_is_simple_reduction): ...here. Accept reductions that
1778 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1779 (vect_force_simple_reduction): Also store the reduction type in
1780 the assignment's STMT_VINFO_REDUC_TYPE.
1781 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1782 (merge_with_identity): New function.
1783 (vect_expand_fold_left): Likewise.
1784 (vectorize_fold_left_reduction): Likewise.
1785 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1786 scalar phi in place for it. Check for target support and reject
1787 cases that would reassociate the operation. Defer the transform
1788 phase to vectorize_fold_left_reduction.
1789 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1790 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1791 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1793 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1795 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1796 call to ifc_temp_var.
1798 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1799 Alan Hayward <alan.hayward@arm.com>
1800 David Sherwood <david.sherwood@arm.com>
1802 * target.def (legitimize_address_displacement): Take the original
1803 offset as a poly_int.
1804 * targhooks.h (default_legitimize_address_displacement): Update
1805 accordingly.
1806 * targhooks.c (default_legitimize_address_displacement): Likewise.
1807 * doc/tm.texi: Regenerate.
1808 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1809 as an argument, moving assert of ad->disp == ad->disp_term to...
1810 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1811 Try calling targetm.legitimize_address_displacement before expanding
1812 the address rather than afterwards, and adjust for the new interface.
1813 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1814 Match the new hook interface. Handle SVE addresses.
1815 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1816 new hook interface.
1818 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1820 * Makefile.in (OBJS): Add early-remat.o.
1821 * target.def (select_early_remat_modes): New hook.
1822 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1823 * doc/tm.texi: Regenerate.
1824 * targhooks.h (default_select_early_remat_modes): Declare.
1825 * targhooks.c (default_select_early_remat_modes): New function.
1826 * timevar.def (TV_EARLY_REMAT): New timevar.
1827 * passes.def (pass_early_remat): New pass.
1828 * tree-pass.h (make_pass_early_remat): Declare.
1829 * early-remat.c: New file.
1830 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1831 function.
1832 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1834 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1835 Alan Hayward <alan.hayward@arm.com>
1836 David Sherwood <david.sherwood@arm.com>
1838 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1839 vfm1 with a bound_epilog parameter.
1840 (vect_do_peeling): Update calls accordingly, and move the prologue
1841 call earlier in the function. Treat the base bound_epilog as 0 for
1842 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1843 this base when peeling for gaps.
1844 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1845 with fully-masked loops.
1846 (vect_estimate_min_profitable_iters): Handle the single peeled
1847 iteration in that case.
1849 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1850 Alan Hayward <alan.hayward@arm.com>
1851 David Sherwood <david.sherwood@arm.com>
1853 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1854 single-element interleaving even if the size is not a power of 2.
1855 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1856 accesses for single-element interleaving if the group size is
1857 not a power of 2.
1859 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1860 Alan Hayward <alan.hayward@arm.com>
1861 David Sherwood <david.sherwood@arm.com>
1863 * doc/md.texi (fold_extract_last_@var{m}): Document.
1864 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1865 * optabs.def (fold_extract_last_optab): New optab.
1866 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1867 * internal-fn.c (fold_extract_direct): New macro.
1868 (expand_fold_extract_optab_fn): Likewise.
1869 (direct_fold_extract_optab_supported_p): Likewise.
1870 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1871 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1872 EXTRACT_LAST_REDUCTION.
1873 (get_initial_def_for_reduction): Do not create an initial vector
1874 for EXTRACT_LAST_REDUCTION reductions.
1875 (vectorizable_reduction): Leave the scalar phi in place for
1876 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1877 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1878 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1879 transform phase to vectorizable_condition.
1880 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1881 split out from...
1882 (vect_finish_stmt_generation): ...here.
1883 (vect_finish_replace_stmt): New function.
1884 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1885 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1886 pattern.
1887 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1889 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1890 Alan Hayward <alan.hayward@arm.com>
1891 David Sherwood <david.sherwood@arm.com>
1893 * doc/md.texi (extract_last_@var{m}): Document.
1894 * optabs.def (extract_last_optab): New optab.
1895 * internal-fn.def (EXTRACT_LAST): New internal function.
1896 * internal-fn.c (cond_unary_direct): New macro.
1897 (expand_cond_unary_optab_fn): Likewise.
1898 (direct_cond_unary_optab_supported_p): Likewise.
1899 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1900 loops using EXTRACT_LAST.
1901 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1902 (extract_last_<mode>): ...this optab.
1903 (vec_extract<mode><Vel>): Update accordingly.
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 * target.def (empty_mask_is_expensive): New hook.
1910 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
1911 * doc/tm.texi: Regenerate.
1912 * targhooks.h (default_empty_mask_is_expensive): Declare.
1913 * targhooks.c (default_empty_mask_is_expensive): New function.
1914 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
1915 if the target says that empty masks are expensive.
1916 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
1917 New function.
1918 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
1920 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1921 Alan Hayward <alan.hayward@arm.com>
1922 David Sherwood <david.sherwood@arm.com>
1924 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
1925 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
1926 (vect_use_loop_mask_for_alignment_p): New function.
1927 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
1928 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
1929 niters_skip argument. Make sure that the first niters_skip elements
1930 of the first iteration are inactive.
1931 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
1932 Update call to vect_set_loop_masks_directly.
1933 (get_misalign_in_elems): New function, split out from...
1934 (vect_gen_prolog_loop_niters): ...here.
1935 (vect_update_init_of_dr): Take a code argument that specifies whether
1936 the adjustment should be added or subtracted.
1937 (vect_update_init_of_drs): Likewise.
1938 (vect_prepare_for_masked_peels): New function.
1939 (vect_do_peeling): Skip prologue peeling if we're using a mask
1940 instead. Update call to vect_update_inits_of_drs.
1941 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1942 mask_skip_niters.
1943 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
1944 alignment. Do not include the number of peeled iterations in
1945 the minimum threshold in that case.
1946 (vectorizable_induction): Adjust the start value down by
1947 LOOP_VINFO_MASK_SKIP_NITERS iterations.
1948 (vect_transform_loop): Call vect_prepare_for_masked_peels.
1949 Take the number of skipped iterations into account when calculating
1950 the loop bounds.
1951 * tree-vect-stmts.c (vect_gen_while_not): New function.
1953 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1954 Alan Hayward <alan.hayward@arm.com>
1955 David Sherwood <david.sherwood@arm.com>
1957 * doc/sourcebuild.texi (vect_fully_masked): Document.
1958 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
1959 default value to 0.
1960 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
1961 split out from...
1962 (vect_analyze_loop_2): ...here. Don't check the vectorization
1963 factor against the number of loop iterations if the loop is
1964 fully-masked.
1966 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1967 Alan Hayward <alan.hayward@arm.com>
1968 David Sherwood <david.sherwood@arm.com>
1970 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
1971 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
1972 (dump_groups): Update accordingly.
1973 (iv_use::mem_type): New member variable.
1974 (address_p): New function.
1975 (record_use): Add a mem_type argument and initialize the new
1976 mem_type field.
1977 (record_group_use): Add a mem_type argument. Use address_p.
1978 Remove obsolete null checks of base_object. Update call to record_use.
1979 (find_interesting_uses_op): Update call to record_group_use.
1980 (find_interesting_uses_cond): Likewise.
1981 (find_interesting_uses_address): Likewise.
1982 (get_mem_type_for_internal_fn): New function.
1983 (find_address_like_use): Likewise.
1984 (find_interesting_uses_stmt): Try find_address_like_use before
1985 calling find_interesting_uses_op.
1986 (addr_offset_valid_p): Use the iv mem_type field as the type
1987 of the addressed memory.
1988 (add_autoinc_candidates): Likewise.
1989 (get_address_cost): Likewise.
1990 (split_small_address_groups_p): Use address_p.
1991 (split_address_groups): Likewise.
1992 (add_iv_candidate_for_use): Likewise.
1993 (autoinc_possible_for_pair): Likewise.
1994 (rewrite_groups): Likewise.
1995 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
1996 (determine_group_iv_cost): Update after split of USE_ADDRESS.
1997 (get_alias_ptr_type_for_ptr_address): New function.
1998 (rewrite_use_address): Rewrite address uses in calls that were
1999 identified by find_address_like_use.
2001 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2002 Alan Hayward <alan.hayward@arm.com>
2003 David Sherwood <david.sherwood@arm.com>
2005 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2006 TARGET_MEM_REFs.
2007 * gimple-expr.h (is_gimple_addressable: Likewise.
2008 * gimple-expr.c (is_gimple_address): Likewise.
2009 * internal-fn.c (expand_call_mem_ref): New function.
2010 (expand_mask_load_optab_fn): Use it.
2011 (expand_mask_store_optab_fn): Likewise.
2013 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2014 Alan Hayward <alan.hayward@arm.com>
2015 David Sherwood <david.sherwood@arm.com>
2017 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2018 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2019 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2020 (cond_umax@var{mode}): Document.
2021 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2022 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2023 (cond_umin_optab, cond_umax_optab): New optabs.
2024 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2025 (COND_IOR, COND_XOR): New internal functions.
2026 * internal-fn.h (get_conditional_internal_fn): Declare.
2027 * internal-fn.c (cond_binary_direct): New macro.
2028 (expand_cond_binary_optab_fn): Likewise.
2029 (direct_cond_binary_optab_supported_p): Likewise.
2030 (get_conditional_internal_fn): New function.
2031 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2032 Cope with reduction statements that are vectorized as calls rather
2033 than assignments.
2034 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2035 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2036 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2037 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2038 (UNSPEC_COND_EOR): New unspecs.
2039 (optab): Add mappings for them.
2040 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2041 (sve_int_op, sve_fp_op): New int attributes.
2043 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2044 Alan Hayward <alan.hayward@arm.com>
2045 David Sherwood <david.sherwood@arm.com>
2047 * optabs.def (while_ult_optab): New optab.
2048 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2049 * internal-fn.def (WHILE_ULT): New internal function.
2050 * internal-fn.h (direct_internal_fn_supported_p): New override
2051 that takes two types as argument.
2052 * internal-fn.c (while_direct): New macro.
2053 (expand_while_optab_fn): New function.
2054 (convert_optab_supported_p): Likewise.
2055 (direct_while_optab_supported_p): New macro.
2056 * wide-int.h (wi::udiv_ceil): New function.
2057 * tree-vectorizer.h (rgroup_masks): New structure.
2058 (vec_loop_masks): New typedef.
2059 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2060 and fully_masked_p.
2061 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2062 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2063 (vect_max_vf): New function.
2064 (slpeel_make_loop_iterate_ntimes): Delete.
2065 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2066 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2067 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2068 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2069 internal-fn.h, stor-layout.h and optabs-query.h.
2070 (vect_set_loop_mask): New function.
2071 (add_preheader_seq): Likewise.
2072 (add_header_seq): Likewise.
2073 (interleave_supported_p): Likewise.
2074 (vect_maybe_permute_loop_masks): Likewise.
2075 (vect_set_loop_masks_directly): Likewise.
2076 (vect_set_loop_condition_masked): Likewise.
2077 (vect_set_loop_condition_unmasked): New function, split out from
2078 slpeel_make_loop_iterate_ntimes.
2079 (slpeel_make_loop_iterate_ntimes): Rename to..
2080 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2081 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2082 (vect_do_peeling): Update call accordingly.
2083 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2084 loops.
2085 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2086 mask_compare_type, can_fully_mask_p and fully_masked_p.
2087 (release_vec_loop_masks): New function.
2088 (_loop_vec_info): Use it to free the loop masks.
2089 (can_produce_all_loop_masks_p): New function.
2090 (vect_get_max_nscalars_per_iter): Likewise.
2091 (vect_verify_full_masking): Likewise.
2092 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2093 retries, and free the mask rgroups before retrying. Check loop-wide
2094 reasons for disallowing fully-masked loops. Make the final decision
2095 about whether use a fully-masked loop or not.
2096 (vect_estimate_min_profitable_iters): Do not assume that peeling
2097 for the number of iterations will be needed for fully-masked loops.
2098 (vectorizable_reduction): Disable fully-masked loops.
2099 (vectorizable_live_operation): Likewise.
2100 (vect_halve_mask_nunits): New function.
2101 (vect_double_mask_nunits): Likewise.
2102 (vect_record_loop_mask): Likewise.
2103 (vect_get_loop_mask): Likewise.
2104 (vect_transform_loop): Handle the case in which the final loop
2105 iteration might handle a partial vector. Call vect_set_loop_condition
2106 instead of slpeel_make_loop_iterate_ntimes.
2107 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2108 (check_load_store_masking): New function.
2109 (prepare_load_store_mask): Likewise.
2110 (vectorizable_store): Handle fully-masked loops.
2111 (vectorizable_load): Likewise.
2112 (supportable_widening_operation): Use vect_halve_mask_nunits for
2113 booleans.
2114 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2115 (vect_gen_while): New function.
2116 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2117 (aarch64_uqdec<mode>): New insn.
2119 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2120 Alan Hayward <alan.hayward@arm.com>
2121 David Sherwood <david.sherwood@arm.com>
2123 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2124 (reduc_xor_scal_optab): New optabs.
2125 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2126 (reduc_xor_scal_@var{m}): Document.
2127 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2128 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2129 internal functions.
2130 * fold-const-call.c (fold_const_call): Handle them.
2131 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2132 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2133 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2134 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2135 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2136 (UNSPEC_XORV): New unspecs.
2137 (optab): Add entries for them.
2138 (BITWISEV): New int iterator.
2139 (bit_reduc_op): New int attributes.
2141 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2142 Alan Hayward <alan.hayward@arm.com>
2143 David Sherwood <david.sherwood@arm.com>
2145 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2146 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2147 * optabs.def (vec_shl_insert_optab): New optab.
2148 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2149 (duplicate_and_interleave): Likewise.
2150 * tree-vect-loop.c: Include internal-fn.h.
2151 (neutral_op_for_slp_reduction): New function, split out from
2152 get_initial_defs_for_reduction.
2153 (get_initial_def_for_reduction): Handle option 2 for variable-length
2154 vectors by loading the neutral value into a vector and then shifting
2155 the initial value into element 0.
2156 (get_initial_defs_for_reduction): Replace the code argument with
2157 the neutral value calculated by neutral_op_for_slp_reduction.
2158 Use gimple_build_vector for constant-length vectors.
2159 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2160 but the first group_size elements have a neutral value.
2161 Use duplicate_and_interleave otherwise.
2162 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2163 Update call to get_initial_defs_for_reduction. Handle SLP
2164 reductions for variable-length vectors by creating one vector
2165 result for each scalar result, with the elements associated
2166 with other scalar results stubbed out with the neutral value.
2167 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2168 Require IFN_VEC_SHL_INSERT for double reductions on
2169 variable-length vectors, or SLP reductions that have
2170 a neutral value. Require can_duplicate_and_interleave_p
2171 support for variable-length unchained SLP reductions if there
2172 is no neutral value, such as for MIN/MAX reductions. Also require
2173 the number of vector elements to be a multiple of the number of
2174 SLP statements when doing variable-length unchained SLP reductions.
2175 Update call to vect_create_epilog_for_reduction.
2176 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2177 and remove initial values.
2178 (duplicate_and_interleave): Make public.
2179 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2180 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
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 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2187 (can_duplicate_and_interleave_p): New function.
2188 (vect_get_and_check_slp_defs): Take the vector of statements
2189 rather than just the current one. Remove excess parentheses.
2190 Restriction rejectinon of vect_constant_def and vect_external_def
2191 for variable-length vectors to boolean types, or types for which
2192 can_duplicate_and_interleave_p is false.
2193 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2194 (duplicate_and_interleave): New function.
2195 (vect_get_constant_vectors): Use gimple_build_vector for
2196 constant-length vectors and suitable variable-length constant
2197 vectors. Use duplicate_and_interleave for other variable-length
2198 vectors. Don't defer the update when inserting new statements.
2200 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2201 Alan Hayward <alan.hayward@arm.com>
2202 David Sherwood <david.sherwood@arm.com>
2204 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2205 min_profitable_iters doesn't go negative.
2207 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2208 Alan Hayward <alan.hayward@arm.com>
2209 David Sherwood <david.sherwood@arm.com>
2211 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2212 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2213 * optabs.def (vec_mask_load_lanes_optab): New optab.
2214 (vec_mask_store_lanes_optab): Likewise.
2215 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2216 (MASK_STORE_LANES): Likewise.
2217 * internal-fn.c (mask_load_lanes_direct): New macro.
2218 (mask_store_lanes_direct): Likewise.
2219 (expand_mask_load_optab_fn): Handle masked operations.
2220 (expand_mask_load_lanes_optab_fn): New macro.
2221 (expand_mask_store_optab_fn): Handle masked operations.
2222 (expand_mask_store_lanes_optab_fn): New macro.
2223 (direct_mask_load_lanes_optab_supported_p): Likewise.
2224 (direct_mask_store_lanes_optab_supported_p): Likewise.
2225 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2226 parameter.
2227 (vect_load_lanes_supported): Likewise.
2228 * tree-vect-data-refs.c (strip_conversion): New function.
2229 (can_group_stmts_p): Likewise.
2230 (vect_analyze_data_ref_accesses): Use it instead of checking
2231 for a pair of assignments.
2232 (vect_store_lanes_supported): Take a masked_p parameter.
2233 (vect_load_lanes_supported): Likewise.
2234 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2235 vect_store_lanes_supported and vect_load_lanes_supported.
2236 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2237 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2238 parameter. Don't allow gaps for masked accesses.
2239 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2240 and vect_load_lanes_supported.
2241 (get_load_store_type): Take a masked_p parameter and update
2242 call to get_group_load_store_type.
2243 (vectorizable_store): Update call to get_load_store_type.
2244 Handle IFN_MASK_STORE_LANES.
2245 (vectorizable_load): Update call to get_load_store_type.
2246 Handle IFN_MASK_LOAD_LANES.
2248 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2249 Alan Hayward <alan.hayward@arm.com>
2250 David Sherwood <david.sherwood@arm.com>
2252 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2253 modes for SVE.
2254 * config/aarch64/aarch64-protos.h
2255 (aarch64_sve_struct_memory_operand_p): Declare.
2256 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2257 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2258 (VPRED, vpred): Handle SVE structure modes.
2259 * config/aarch64/constraints.md (Utx): New constraint.
2260 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2261 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2262 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2263 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2264 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2265 structure modes. Split into pieces after RA.
2266 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2267 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2268 New patterns.
2269 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2270 SVE structure modes.
2271 (aarch64_classify_address): Likewise.
2272 (sizetochar): Move earlier in file.
2273 (aarch64_print_operand): Handle SVE register lists.
2274 (aarch64_array_mode): New function.
2275 (aarch64_sve_struct_memory_operand_p): Likewise.
2276 (TARGET_ARRAY_MODE): Redefine.
2278 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2279 Alan Hayward <alan.hayward@arm.com>
2280 David Sherwood <david.sherwood@arm.com>
2282 * target.def (array_mode): New target hook.
2283 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2284 * doc/tm.texi: Regenerate.
2285 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2286 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2287 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2288 targetm.array_mode.
2289 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2290 type sizes.
2292 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2293 Alan Hayward <alan.hayward@arm.com>
2294 David Sherwood <david.sherwood@arm.com>
2296 * fold-const.c (fold_binary_loc): Check the argument types
2297 rather than the result type when testing for a vector operation.
2299 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2301 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2302 * doc/tm.texi: Regenerate.
2304 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2305 Alan Hayward <alan.hayward@arm.com>
2306 David Sherwood <david.sherwood@arm.com>
2308 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2309 (sve): Document new AArch64 extension.
2310 * doc/md.texi (w): Extend the description of the AArch64
2311 constraint to include SVE vectors.
2312 (Upl, Upa): Document new AArch64 predicate constraints.
2313 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2314 enum.
2315 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2316 (msve-vector-bits=): New option.
2317 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2318 SVE when these are disabled.
2319 (sve): New extension.
2320 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2321 modes. Adjust their number of units based on aarch64_sve_vg.
2322 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2323 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2324 aarch64_addr_query_type.
2325 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2326 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2327 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2328 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2329 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2330 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2331 (aarch64_simd_imm_zero_p): Delete.
2332 (aarch64_check_zero_based_sve_index_immediate): Declare.
2333 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2334 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2335 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2336 (aarch64_sve_float_mul_immediate_p): Likewise.
2337 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2338 rather than an rtx.
2339 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2340 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2341 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2342 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2343 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2344 (aarch64_regmode_natural_size): Likewise.
2345 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2346 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2347 left one place.
2348 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2349 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2350 for VG and the SVE predicate registers.
2351 (V_ALIASES): Add a "z"-prefixed alias.
2352 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2353 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2354 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2355 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2356 (REG_CLASS_NAMES): Add entries for them.
2357 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2358 and the predicate registers.
2359 (aarch64_sve_vg): Declare.
2360 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2361 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2362 (REGMODE_NATURAL_SIZE): Define.
2363 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2364 SVE macros.
2365 * config/aarch64/aarch64.c: Include cfgrtl.h.
2366 (simd_immediate_info): Add a constructor for series vectors,
2367 and an associated step field.
2368 (aarch64_sve_vg): New variable.
2369 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2370 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2371 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2372 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2373 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2374 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2375 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2376 (aarch64_get_mask_mode): New functions.
2377 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2378 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2379 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2380 predicate modes and predicate registers. Explicitly restrict
2381 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2382 to store a vector mode if it is recognized by
2383 aarch64_classify_vector_mode.
2384 (aarch64_regmode_natural_size): New function.
2385 (aarch64_hard_regno_caller_save_mode): Return the original mode
2386 for predicates.
2387 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2388 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2389 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2390 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2391 functions.
2392 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2393 does not overlap dest if the function is frame-related. Handle
2394 SVE constants.
2395 (aarch64_split_add_offset): New function.
2396 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2397 them aarch64_add_offset.
2398 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2399 and update call to aarch64_sub_sp.
2400 (aarch64_add_cfa_expression): New function.
2401 (aarch64_expand_prologue): Pass extra temporary registers to the
2402 functions above. Handle the case in which we need to emit new
2403 DW_CFA_expressions for registers that were originally saved
2404 relative to the stack pointer, but now have to be expressed
2405 relative to the frame pointer.
2406 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2407 functions above.
2408 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2409 IP0 and IP1 values for SVE frames.
2410 (aarch64_expand_vec_series): New function.
2411 (aarch64_expand_sve_widened_duplicate): Likewise.
2412 (aarch64_expand_sve_const_vector): Likewise.
2413 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2414 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2415 into the register, rather than emitting a SET directly.
2416 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2417 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2418 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2419 (offset_9bit_signed_scaled_p): New functions.
2420 (aarch64_replicate_bitmask_imm): New function.
2421 (aarch64_bitmask_imm): Use it.
2422 (aarch64_cannot_force_const_mem): Reject expressions involving
2423 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2424 (aarch64_classify_index): Handle SVE indices, by requiring
2425 a plain register index with a scale that matches the element size.
2426 (aarch64_classify_address): Handle SVE addresses. Assert that
2427 the mode of the address is VOIDmode or an integer mode.
2428 Update call to aarch64_classify_symbol.
2429 (aarch64_classify_symbolic_expression): Update call to
2430 aarch64_classify_symbol.
2431 (aarch64_const_vec_all_in_range_p): New function.
2432 (aarch64_print_vector_float_operand): Likewise.
2433 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2434 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2435 and the FP immediates 1.0 and 0.5.
2436 (aarch64_print_address_internal): Handle SVE addresses.
2437 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2438 (aarch64_regno_regclass): Handle predicate registers.
2439 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2440 data modes.
2441 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2442 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2443 (aarch64_convert_sve_vector_bits): New function.
2444 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2445 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2446 rather than an rtx.
2447 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2448 Handle SVE vector and predicate modes. Accept VL-based constants
2449 that need only one temporary register, and VL offsets that require
2450 no temporary registers.
2451 (aarch64_conditional_register_usage): Mark the predicate registers
2452 as fixed if SVE isn't available.
2453 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2454 Return true for SVE vector and predicate modes.
2455 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2456 rather than an unsigned int. Handle SVE modes.
2457 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2458 SVE modes.
2459 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2460 if SVE is enabled.
2461 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2462 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2463 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2464 (aarch64_sve_float_mul_immediate_p): New functions.
2465 (aarch64_sve_valid_immediate): New function.
2466 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2467 Explicitly reject structure modes. Check for INDEX constants.
2468 Handle PTRUE and PFALSE constants.
2469 (aarch64_check_zero_based_sve_index_immediate): New function.
2470 (aarch64_simd_imm_zero_p): Delete.
2471 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2472 vector modes. Accept constants in the range of CNT[BHWD].
2473 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2474 ask for an Advanced SIMD mode.
2475 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2476 (aarch64_simd_vector_alignment): Handle SVE predicates.
2477 (aarch64_vectorize_preferred_vector_alignment): New function.
2478 (aarch64_simd_vector_alignment_reachable): Use it instead of
2479 the vector size.
2480 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2481 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2482 functions.
2483 (MAX_VECT_LEN): Delete.
2484 (expand_vec_perm_d): Add a vec_flags field.
2485 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2486 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2487 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2488 for SVE modes.
2489 (aarch64_evpc_rev): Rename to...
2490 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2491 (aarch64_evpc_rev_global): New function.
2492 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2493 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2494 MAX_VECT_LEN.
2495 (aarch64_evpc_sve_tbl): New function.
2496 (aarch64_expand_vec_perm_const_1): Update after rename of
2497 aarch64_evpc_rev. Handle SVE permutes too, trying
2498 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2499 than aarch64_evpc_tbl.
2500 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2501 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2502 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2503 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2504 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2505 (aarch64_expand_sve_vcond): New functions.
2506 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2507 of aarch64_vector_mode_p.
2508 (aarch64_dwarf_poly_indeterminate_value): New function.
2509 (aarch64_compute_pressure_classes): Likewise.
2510 (aarch64_can_change_mode_class): Likewise.
2511 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2512 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2513 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2514 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2515 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2516 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2517 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2518 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2519 constraints.
2520 (Dn, Dl, Dr): Accept const as well as const_vector.
2521 (Dz): Likewise. Compare against CONST0_RTX.
2522 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2523 of "vector" where appropriate.
2524 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2525 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2526 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2527 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2528 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2529 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2530 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2531 (v_int_equiv): Extend to SVE modes.
2532 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2533 mode attributes.
2534 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2535 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2536 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2537 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2538 (SVE_COND_FP_CMP): New int iterators.
2539 (perm_hilo): Handle the new unpack unspecs.
2540 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2541 attributes.
2542 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2543 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2544 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2545 (aarch64_equality_operator, aarch64_constant_vector_operand)
2546 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2547 (aarch64_sve_nonimmediate_operand): Likewise.
2548 (aarch64_sve_general_operand): Likewise.
2549 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2550 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2551 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2552 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2553 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2554 (aarch64_sve_float_arith_immediate): Likewise.
2555 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2556 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2557 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2558 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2559 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2560 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2561 (aarch64_sve_float_arith_operand): Likewise.
2562 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2563 (aarch64_sve_float_mul_operand): Likewise.
2564 (aarch64_sve_vec_perm_operand): Likewise.
2565 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2566 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2567 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2568 as well as const_vector.
2569 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2570 in file. Use CONST0_RTX and CONSTM1_RTX.
2571 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2572 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2573 Use aarch64_simd_imm_zero.
2574 * config/aarch64/aarch64-sve.md: New file.
2575 * config/aarch64/aarch64.md: Include it.
2576 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2577 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2578 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2579 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2580 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2581 (sve): New attribute.
2582 (enabled): Disable instructions with the sve attribute unless
2583 TARGET_SVE.
2584 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2585 aarch64_expand_mov_immediate.
2586 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2587 CNT[BHSD] immediates.
2588 (movti): Split CONST_POLY_INT moves into two halves.
2589 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2590 Split additions that need a temporary here if the destination
2591 is the stack pointer.
2592 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2593 (*add<mode>3_poly_1): New instruction.
2594 (set_clobber_cc): New expander.
2596 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2598 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2599 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2600 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2601 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2602 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2603 Change innermode from fixed_mode_size to machine_mode.
2604 (simplify_subreg): Update call accordingly. Handle a constant-sized
2605 subreg of a variable-length CONST_VECTOR.
2607 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2608 Alan Hayward <alan.hayward@arm.com>
2609 David Sherwood <david.sherwood@arm.com>
2611 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2612 (add_offset_to_base): New function, split out from...
2613 (create_mem_ref): ...here. When handling a scale other than 1,
2614 check first whether the address is valid without the offset.
2615 Add it into the base if so, leaving the index and scale as-is.
2617 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2619 PR c++/83778
2620 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2621 fold_for_warn before checking if arg2 is INTEGER_CST.
2623 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2625 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2626 (store_multiple_operation): Delete.
2627 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2628 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2629 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2630 guarded by TARGET_STRING.
2631 (rs6000_output_load_multiple): Delete.
2632 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2633 OPTION_MASK_STRING / TARGET_STRING handling.
2634 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2635 (const rs6000_opt_masks) <"string">: Change mask to 0.
2636 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2637 (MASK_STRING): Delete.
2638 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2639 parts. Simplify.
2640 (load_multiple): Delete.
2641 (*ldmsi8): Delete.
2642 (*ldmsi7): Delete.
2643 (*ldmsi6): Delete.
2644 (*ldmsi5): Delete.
2645 (*ldmsi4): Delete.
2646 (*ldmsi3): Delete.
2647 (store_multiple): Delete.
2648 (*stmsi8): Delete.
2649 (*stmsi7): Delete.
2650 (*stmsi6): Delete.
2651 (*stmsi5): Delete.
2652 (*stmsi4): Delete.
2653 (*stmsi3): Delete.
2654 (movmemsi_8reg): Delete.
2655 (corresponding unnamed define_insn): Delete.
2656 (movmemsi_6reg): Delete.
2657 (corresponding unnamed define_insn): Delete.
2658 (movmemsi_4reg): Delete.
2659 (corresponding unnamed define_insn): Delete.
2660 (movmemsi_2reg): Delete.
2661 (corresponding unnamed define_insn): Delete.
2662 (movmemsi_1reg): Delete.
2663 (corresponding unnamed define_insn): Delete.
2664 * config/rs6000/rs6000.opt (mno-string): New.
2665 (mstring): Replace by deprecation warning stub.
2666 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2668 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2670 * regrename.c (regrename_do_replace): If replacing the same
2671 reg multiple times, try to reuse last created gen_raw_REG.
2673 PR debug/81155
2674 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2675 main to workaround a bug in GDB.
2677 2018-01-12 Tom de Vries <tom@codesourcery.com>
2679 PR target/83737
2680 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2682 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2684 PR rtl-optimization/80481
2685 * ira-color.c (get_cap_member): New function.
2686 (allocnos_conflict_by_live_ranges_p): Use it.
2687 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2688 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2690 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2692 PR target/83628
2693 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2694 (*saddl_se_1): Ditto.
2695 (*ssubsi_1): Ditto.
2696 (*ssubl_se_1): Ditto.
2698 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2700 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2701 rather than wi::to_widest for DR_INITs.
2702 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2703 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2704 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2705 INTEGER_CSTs.
2706 (vect_analyze_group_access_1): Note that here.
2708 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2710 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2711 polynomial type sizes.
2713 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2715 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2716 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2717 (gimple_add_tmp_var): Likewise.
2719 2018-01-12 Martin Liska <mliska@suse.cz>
2721 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2722 (gimple_alloc_sizes): Likewise.
2723 (dump_gimple_statistics): Use PRIu64 in printf format.
2724 * gimple.h: Change uint64_t to int.
2726 2018-01-12 Martin Liska <mliska@suse.cz>
2728 * tree-core.h: Use uint64_t instead of int.
2729 * tree.c (tree_node_counts): Likewise.
2730 (tree_node_sizes): Likewise.
2731 (dump_tree_statistics): Use PRIu64 in printf format.
2733 2018-01-12 Martin Liska <mliska@suse.cz>
2735 * Makefile.in: As qsort_chk is implemented in vec.c, add
2736 vec.o to linkage of gencfn-macros.
2737 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2738 passing the info to record_node_allocation_statistics.
2739 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2740 and pass the info.
2741 * ggc-common.c (struct ggc_usage): Add operator== and use
2742 it in operator< and compare function.
2743 * mem-stats.h (struct mem_usage): Likewise.
2744 * vec.c (struct vec_usage): Remove operator< and compare
2745 function. Can be simply inherited.
2747 2018-01-12 Martin Jambor <mjambor@suse.cz>
2749 PR target/81616
2750 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2751 * tree-ssa-math-opts.c: Include domwalk.h.
2752 (convert_mult_to_fma_1): New function.
2753 (fma_transformation_info): New type.
2754 (fma_deferring_state): Likewise.
2755 (cancel_fma_deferring): New function.
2756 (result_of_phi): Likewise.
2757 (last_fma_candidate_feeds_initial_phi): Likewise.
2758 (convert_mult_to_fma): Added deferring logic, split actual
2759 transformation to convert_mult_to_fma_1.
2760 (math_opts_dom_walker): New type.
2761 (math_opts_dom_walker::after_dom_children): New method, body moved
2762 here from pass_optimize_widening_mul::execute, added deferring logic
2763 bits.
2764 (pass_optimize_widening_mul::execute): Moved most of code to
2765 math_opts_dom_walker::after_dom_children.
2766 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2767 * config/i386/i386.c (ix86_option_override_internal): Added
2768 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2770 2018-01-12 Richard Biener <rguenther@suse.de>
2772 PR debug/83157
2773 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2774 inline instance vars.
2776 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2778 PR target/81819
2779 * config/rx/rx.c (rx_is_restricted_memory_address):
2780 Handle SUBREG case.
2782 2018-01-12 Richard Biener <rguenther@suse.de>
2784 PR tree-optimization/80846
2785 * target.def (split_reduction): New target hook.
2786 * targhooks.c (default_split_reduction): New function.
2787 * targhooks.h (default_split_reduction): Declare.
2788 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2789 target requests first reduce vectors by combining low and high
2790 parts.
2791 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2792 (get_vectype_for_scalar_type_and_size): Export.
2793 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2794 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2795 * doc/tm.texi: Regenerate.
2796 * config/i386/i386.c (ix86_split_reduction): Implement
2797 TARGET_VECTORIZE_SPLIT_REDUCTION.
2799 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2801 PR target/83368
2802 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2803 in PIC mode except for TARGET_VXWORKS_RTP.
2804 * config/sparc/sparc.c: Include cfgrtl.h.
2805 (TARGET_INIT_PIC_REG): Define.
2806 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2807 (sparc_pic_register_p): New predicate.
2808 (sparc_legitimate_address_p): Use it.
2809 (sparc_legitimize_pic_address): Likewise.
2810 (sparc_delegitimize_address): Likewise.
2811 (sparc_mode_dependent_address_p): Likewise.
2812 (gen_load_pcrel_sym): Remove 4th parameter.
2813 (load_got_register): Adjust call to above. Remove obsolete stuff.
2814 (sparc_expand_prologue): Do not call load_got_register here.
2815 (sparc_flat_expand_prologue): Likewise.
2816 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2817 (sparc_use_pseudo_pic_reg): New function.
2818 (sparc_init_pic_reg): Likewise.
2819 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2820 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2822 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2824 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2825 Add item for branch_cost.
2827 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2829 PR rtl-optimization/83565
2830 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2831 not extend the result to a larger mode for rotate operations.
2832 (num_sign_bit_copies1): Likewise.
2834 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2836 PR target/40411
2837 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2838 -symbolic.
2839 Use values-Xc.o for -pedantic.
2840 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2842 2018-01-12 Martin Liska <mliska@suse.cz>
2844 PR ipa/83054
2845 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2846 New function.
2847 (possible_polymorphic_call_targets): Use it.
2848 (ipa_devirt): Likewise.
2850 2018-01-12 Martin Liska <mliska@suse.cz>
2852 * profile-count.h (enum profile_quality): Use 0 as invalid
2853 enum value of profile_quality.
2855 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2857 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2858 -mext-string options.
2860 2018-01-12 Richard Biener <rguenther@suse.de>
2862 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2863 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2864 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2865 Likewise.
2866 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2868 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2870 * configure.ac (--with-long-double-format): Add support for the
2871 configuration option to change the default long double format on
2872 PowerPC systems.
2873 * config.gcc (powerpc*-linux*-*): Likewise.
2874 * configure: Regenerate.
2875 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2876 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2877 used without modification.
2879 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2881 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2882 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2883 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2884 MISC_BUILTIN_SPEC_BARRIER.
2885 (rs6000_init_builtins): Likewise.
2886 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2887 enum value.
2888 (speculation_barrier): New define_insn.
2889 * doc/extend.texi: Document __builtin_speculation_barrier.
2891 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2893 PR target/83203
2894 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2895 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2896 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2897 iterators.
2898 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2899 integral modes instead of "ss" and "sd".
2900 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2901 vectors with 32-bit and 64-bit elements.
2902 (vecdupssescalarmodesuffix): New mode attribute.
2903 (vec_dup<mode>): Use it.
2905 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2907 PR target/83330
2908 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2909 frame if argument is passed on stack.
2911 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2913 PR target/82682
2914 * ree.c (combine_reaching_defs): Optimize also
2915 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
2916 reg2=any_extend(exp); reg1=reg2;, formatting fix.
2918 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2920 PR middle-end/83189
2921 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
2923 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2925 PR middle-end/83718
2926 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
2927 after they are computed.
2929 2018-01-11 Bin Cheng <bin.cheng@arm.com>
2931 PR tree-optimization/83695
2932 * gimple-loop-linterchange.cc
2933 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
2934 reset cached scev information after interchange.
2935 (pass_linterchange::execute): Remove call to scev_reset_htab.
2937 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2939 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
2940 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
2941 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
2942 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
2943 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
2944 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
2945 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
2946 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
2947 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
2948 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
2949 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
2950 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
2951 (V_lane_reg): Likewise.
2952 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
2953 New define_expand.
2954 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
2955 (vfmal_lane_low<mode>_intrinsic,
2956 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
2957 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
2958 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
2959 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
2960 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
2961 vfmsl_lane_high<mode>_intrinsic): New define_insns.
2963 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2965 * config/arm/arm-cpus.in (fp16fml): New feature.
2966 (ALL_SIMD): Add fp16fml.
2967 (armv8.2-a): Add fp16fml as an option.
2968 (armv8.3-a): Likewise.
2969 (armv8.4-a): Add fp16fml as part of fp16.
2970 * config/arm/arm.h (TARGET_FP16FML): Define.
2971 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
2972 when appropriate.
2973 * config/arm/arm-modes.def (V2HF): Define.
2974 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
2975 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
2976 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
2977 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
2978 vfmsl_low, vfmsl_high): New set of builtins.
2979 * config/arm/iterators.md (PLUSMINUS): New code iterator.
2980 (vfml_op): New code attribute.
2981 (VFMLHALVES): New int iterator.
2982 (VFML, VFMLSEL): New mode attributes.
2983 (V_reg): Define mapping for V2HF.
2984 (V_hi, V_lo): New mode attributes.
2985 (VF_constraint): Likewise.
2986 (vfml_half, vfml_half_selector): New int attributes.
2987 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
2988 define_expand.
2989 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
2990 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
2991 New define_insn.
2992 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
2993 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
2994 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
2995 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
2996 documentation.
2997 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
2998 Document new effective target and option set.
3000 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3002 * config/arm/arm-cpus.in (armv8_4): New feature.
3003 (ARMv8_4a): New fgroup.
3004 (armv8.4-a): New arch.
3005 * config/arm/arm-tables.opt: Regenerate.
3006 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3007 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3008 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3009 Add matching rules for -march=armv8.4-a and extensions.
3010 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3012 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3014 PR target/81821
3015 * config/rx/rx.md (BW): New mode attribute.
3016 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3018 2018-01-11 Richard Biener <rguenther@suse.de>
3020 PR tree-optimization/83435
3021 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3022 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3023 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3025 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3026 Alan Hayward <alan.hayward@arm.com>
3027 David Sherwood <david.sherwood@arm.com>
3029 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3030 field.
3031 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3032 (aarch64_print_address_internal): Use it to check for a zero offset.
3034 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3035 Alan Hayward <alan.hayward@arm.com>
3036 David Sherwood <david.sherwood@arm.com>
3038 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3039 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3040 Return a poly_int64 rather than a HOST_WIDE_INT.
3041 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3042 rather than a HOST_WIDE_INT.
3043 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3044 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3045 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3046 final_offset from HOST_WIDE_INT to poly_int64.
3047 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3048 to_constant when getting the number of units in an Advanced SIMD
3049 mode.
3050 (aarch64_builtin_vectorized_function): Check for a constant number
3051 of units.
3052 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3053 GET_MODE_SIZE.
3054 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3055 attribute instead of GET_MODE_NUNITS.
3056 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3057 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3058 GET_MODE_SIZE for fixed-size registers.
3059 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3060 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3061 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3062 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3063 (aarch64_print_operand, aarch64_print_address_internal)
3064 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3065 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3066 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3067 Handle polynomial GET_MODE_SIZE.
3068 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3069 wider than SImode without modification.
3070 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3071 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3072 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3073 passing and returning SVE modes.
3074 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3075 rather than GEN_INT.
3076 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3077 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3078 (aarch64_allocate_and_probe_stack_space): Likewise.
3079 (aarch64_layout_frame): Cope with polynomial offsets.
3080 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3081 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3082 polynomial offsets.
3083 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3084 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3085 poly_int64 rather than a HOST_WIDE_INT.
3086 (aarch64_get_separate_components, aarch64_process_components)
3087 (aarch64_expand_prologue, aarch64_expand_epilogue)
3088 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3089 (aarch64_anchor_offset): New function, split out from...
3090 (aarch64_legitimize_address): ...here.
3091 (aarch64_builtin_vectorization_cost): Handle polynomial
3092 TYPE_VECTOR_SUBPARTS.
3093 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3094 GET_MODE_NUNITS.
3095 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3096 number of elements from the PARALLEL rather than the mode.
3097 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3098 rather than GET_MODE_BITSIZE.
3099 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3100 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3101 (aarch64_expand_vec_perm_const_1): Handle polynomial
3102 d->perm.length () and d->perm elements.
3103 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3104 Apply to_constant to d->perm elements.
3105 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3106 polynomial CONST_VECTOR_NUNITS.
3107 (aarch64_move_pointer): Take amount as a poly_int64 rather
3108 than an int.
3109 (aarch64_progress_pointer): Avoid temporary variable.
3110 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3111 the mode attribute instead of GET_MODE.
3113 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3114 Alan Hayward <alan.hayward@arm.com>
3115 David Sherwood <david.sherwood@arm.com>
3117 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3118 x exists before using it.
3119 (aarch64_add_constant_internal): Rename to...
3120 (aarch64_add_offset_1): ...this. Replace regnum with separate
3121 src and dest rtxes. Handle the case in which they're different,
3122 including when the offset is zero. Replace scratchreg with an rtx.
3123 Use 2 additions if there is no spare register into which we can
3124 move a 16-bit constant.
3125 (aarch64_add_constant): Delete.
3126 (aarch64_add_offset): Replace reg with separate src and dest
3127 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3128 Use aarch64_add_offset_1.
3129 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3130 an rtx rather than an int. Take the delta as a poly_int64
3131 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3132 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3133 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3134 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3135 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3136 and aarch64_add_sp.
3137 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3138 aarch64_add_constant.
3140 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3142 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3143 Use scalar_float_mode.
3145 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3147 * config/aarch64/aarch64-simd.md
3148 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3149 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3150 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3151 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3152 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3153 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3154 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3155 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3156 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3157 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3159 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3161 PR target/83514
3162 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3163 targ_options->x_arm_arch_string is non NULL.
3165 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3167 * config/aarch64/aarch64.h
3168 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3170 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3172 PR target/82096
3173 * expmed.c (emit_store_flag_force): Swap if const op0
3174 and change VOIDmode to mode of op0.
3176 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3178 PR rtl-optimization/83761
3179 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3180 than bytes to mode_for_size.
3182 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3184 PR middle-end/83189
3185 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3186 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3187 profile.
3189 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3191 PR middle-end/83575
3192 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3193 when in layout mode.
3194 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3195 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3196 partition fixup.
3198 2018-01-10 Michael Collison <michael.collison@arm.com>
3200 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3201 * config/aarch64/aarch64-option-extension.def: Add
3202 AARCH64_OPT_EXTENSION of 'fp16fml'.
3203 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3204 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3205 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3206 * config/aarch64/constraints.md (Ui7): New constraint.
3207 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3208 (VFMLA_SEL_W): Ditto.
3209 (f16quad): Ditto.
3210 (f16mac1): Ditto.
3211 (VFMLA16_LOW): New int iterator.
3212 (VFMLA16_HIGH): Ditto.
3213 (UNSPEC_FMLAL): New unspec.
3214 (UNSPEC_FMLSL): Ditto.
3215 (UNSPEC_FMLAL2): Ditto.
3216 (UNSPEC_FMLSL2): Ditto.
3217 (f16mac): New code attribute.
3218 * config/aarch64/aarch64-simd-builtins.def
3219 (aarch64_fmlal_lowv2sf): Ditto.
3220 (aarch64_fmlsl_lowv2sf): Ditto.
3221 (aarch64_fmlalq_lowv4sf): Ditto.
3222 (aarch64_fmlslq_lowv4sf): Ditto.
3223 (aarch64_fmlal_highv2sf): Ditto.
3224 (aarch64_fmlsl_highv2sf): Ditto.
3225 (aarch64_fmlalq_highv4sf): Ditto.
3226 (aarch64_fmlslq_highv4sf): Ditto.
3227 (aarch64_fmlal_lane_lowv2sf): Ditto.
3228 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3229 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3230 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3231 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3232 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3233 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3234 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3235 (aarch64_fmlal_lane_highv2sf): Ditto.
3236 (aarch64_fmlsl_lane_highv2sf): Ditto.
3237 (aarch64_fmlal_laneq_highv2sf): Ditto.
3238 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3239 (aarch64_fmlalq_lane_highv4sf): Ditto.
3240 (aarch64_fmlsl_lane_highv4sf): Ditto.
3241 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3242 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3243 * config/aarch64/aarch64-simd.md:
3244 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3245 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3246 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3247 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3248 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3249 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3250 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3251 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3252 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3253 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3254 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3255 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3256 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3257 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3258 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3259 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3260 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3261 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3262 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3263 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3264 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3265 (vfmlsl_low_u32): Ditto.
3266 (vfmlalq_low_u32): Ditto.
3267 (vfmlslq_low_u32): Ditto.
3268 (vfmlal_high_u32): Ditto.
3269 (vfmlsl_high_u32): Ditto.
3270 (vfmlalq_high_u32): Ditto.
3271 (vfmlslq_high_u32): Ditto.
3272 (vfmlal_lane_low_u32): Ditto.
3273 (vfmlsl_lane_low_u32): Ditto.
3274 (vfmlal_laneq_low_u32): Ditto.
3275 (vfmlsl_laneq_low_u32): Ditto.
3276 (vfmlalq_lane_low_u32): Ditto.
3277 (vfmlslq_lane_low_u32): Ditto.
3278 (vfmlalq_laneq_low_u32): Ditto.
3279 (vfmlslq_laneq_low_u32): Ditto.
3280 (vfmlal_lane_high_u32): Ditto.
3281 (vfmlsl_lane_high_u32): Ditto.
3282 (vfmlal_laneq_high_u32): Ditto.
3283 (vfmlsl_laneq_high_u32): Ditto.
3284 (vfmlalq_lane_high_u32): Ditto.
3285 (vfmlslq_lane_high_u32): Ditto.
3286 (vfmlalq_laneq_high_u32): Ditto.
3287 (vfmlslq_laneq_high_u32): Ditto.
3288 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3289 (AARCH64_FL_FOR_ARCH8_4): New.
3290 (AARCH64_ISA_F16FML): New ISA flag.
3291 (TARGET_F16FML): New feature flag for fp16fml.
3292 (doc/invoke.texi): Document new fp16fml option.
3294 2018-01-10 Michael Collison <michael.collison@arm.com>
3296 * config/aarch64/aarch64-builtins.c:
3297 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3298 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3299 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3300 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3301 (AARCH64_ISA_SHA3): New ISA flag.
3302 (TARGET_SHA3): New feature flag for sha3.
3303 * config/aarch64/iterators.md (sha512_op): New int attribute.
3304 (CRYPTO_SHA512): New int iterator.
3305 (UNSPEC_SHA512H): New unspec.
3306 (UNSPEC_SHA512H2): Ditto.
3307 (UNSPEC_SHA512SU0): Ditto.
3308 (UNSPEC_SHA512SU1): Ditto.
3309 * config/aarch64/aarch64-simd-builtins.def
3310 (aarch64_crypto_sha512hqv2di): New builtin.
3311 (aarch64_crypto_sha512h2qv2di): Ditto.
3312 (aarch64_crypto_sha512su0qv2di): Ditto.
3313 (aarch64_crypto_sha512su1qv2di): Ditto.
3314 (aarch64_eor3qv8hi): Ditto.
3315 (aarch64_rax1qv2di): Ditto.
3316 (aarch64_xarqv2di): Ditto.
3317 (aarch64_bcaxqv8hi): Ditto.
3318 * config/aarch64/aarch64-simd.md:
3319 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3320 (aarch64_crypto_sha512su0qv2di): Ditto.
3321 (aarch64_crypto_sha512su1qv2di): Ditto.
3322 (aarch64_eor3qv8hi): Ditto.
3323 (aarch64_rax1qv2di): Ditto.
3324 (aarch64_xarqv2di): Ditto.
3325 (aarch64_bcaxqv8hi): Ditto.
3326 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3327 (vsha512h2q_u64): Ditto.
3328 (vsha512su0q_u64): Ditto.
3329 (vsha512su1q_u64): Ditto.
3330 (veor3q_u16): Ditto.
3331 (vrax1q_u64): Ditto.
3332 (vxarq_u64): Ditto.
3333 (vbcaxq_u16): Ditto.
3334 * config/arm/types.md (crypto_sha512): New type attribute.
3335 (crypto_sha3): Ditto.
3336 (doc/invoke.texi): Document new sha3 option.
3338 2018-01-10 Michael Collison <michael.collison@arm.com>
3340 * config/aarch64/aarch64-builtins.c:
3341 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3342 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3343 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3344 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3345 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3346 (AARCH64_ISA_SM4): New ISA flag.
3347 (TARGET_SM4): New feature flag for sm4.
3348 * config/aarch64/aarch64-simd-builtins.def
3349 (aarch64_sm3ss1qv4si): Ditto.
3350 (aarch64_sm3tt1aq4si): Ditto.
3351 (aarch64_sm3tt1bq4si): Ditto.
3352 (aarch64_sm3tt2aq4si): Ditto.
3353 (aarch64_sm3tt2bq4si): Ditto.
3354 (aarch64_sm3partw1qv4si): Ditto.
3355 (aarch64_sm3partw2qv4si): Ditto.
3356 (aarch64_sm4eqv4si): Ditto.
3357 (aarch64_sm4ekeyqv4si): Ditto.
3358 * config/aarch64/aarch64-simd.md:
3359 (aarch64_sm3ss1qv4si): Ditto.
3360 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3361 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3362 (aarch64_sm4eqv4si): Ditto.
3363 (aarch64_sm4ekeyqv4si): Ditto.
3364 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3365 (sm3part_op): Ditto.
3366 (CRYPTO_SM3TT): Ditto.
3367 (CRYPTO_SM3PART): Ditto.
3368 (UNSPEC_SM3SS1): New unspec.
3369 (UNSPEC_SM3TT1A): Ditto.
3370 (UNSPEC_SM3TT1B): Ditto.
3371 (UNSPEC_SM3TT2A): Ditto.
3372 (UNSPEC_SM3TT2B): Ditto.
3373 (UNSPEC_SM3PARTW1): Ditto.
3374 (UNSPEC_SM3PARTW2): Ditto.
3375 (UNSPEC_SM4E): Ditto.
3376 (UNSPEC_SM4EKEY): Ditto.
3377 * config/aarch64/constraints.md (Ui2): New constraint.
3378 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3379 * config/arm/types.md (crypto_sm3): New type attribute.
3380 (crypto_sm4): Ditto.
3381 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3382 (vsm3tt1aq_u32): Ditto.
3383 (vsm3tt1bq_u32): Ditto.
3384 (vsm3tt2aq_u32): Ditto.
3385 (vsm3tt2bq_u32): Ditto.
3386 (vsm3partw1q_u32): Ditto.
3387 (vsm3partw2q_u32): Ditto.
3388 (vsm4eq_u32): Ditto.
3389 (vsm4ekeyq_u32): Ditto.
3390 (doc/invoke.texi): Document new sm4 option.
3392 2018-01-10 Michael Collison <michael.collison@arm.com>
3394 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3395 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3396 (AARCH64_FL_FOR_ARCH8_4): New.
3397 (AARCH64_FL_V8_4): New flag.
3398 (doc/invoke.texi): Document new armv8.4-a option.
3400 2018-01-10 Michael Collison <michael.collison@arm.com>
3402 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3403 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3404 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3405 * config/aarch64/aarch64-option-extension.def: Add
3406 AARCH64_OPT_EXTENSION of 'sha2'.
3407 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3408 (crypto): Disable sha2 and aes if crypto disabled.
3409 (crypto): Enable aes and sha2 if enabled.
3410 (simd): Disable sha2 and aes if simd disabled.
3411 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3412 New flags.
3413 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3414 (TARGET_SHA2): New feature flag for sha2.
3415 (TARGET_AES): New feature flag for aes.
3416 * config/aarch64/aarch64-simd.md:
3417 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3418 conditional on TARGET_AES.
3419 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3420 (aarch64_crypto_sha1hsi): Make pattern conditional
3421 on TARGET_SHA2.
3422 (aarch64_crypto_sha1hv4si): Ditto.
3423 (aarch64_be_crypto_sha1hv4si): Ditto.
3424 (aarch64_crypto_sha1su1v4si): Ditto.
3425 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3426 (aarch64_crypto_sha1su0v4si): Ditto.
3427 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3428 (aarch64_crypto_sha256su0v4si): Ditto.
3429 (aarch64_crypto_sha256su1v4si): Ditto.
3430 (doc/invoke.texi): Document new aes and sha2 options.
3432 2018-01-10 Martin Sebor <msebor@redhat.com>
3434 PR tree-optimization/83781
3435 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3436 as string arrays.
3438 2018-01-11 Martin Sebor <msebor@gmail.com>
3439 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3441 PR tree-optimization/83501
3442 PR tree-optimization/81703
3444 * tree-ssa-strlen.c (get_string_cst): Rename...
3445 (get_string_len): ...to this. Handle global constants.
3446 (handle_char_store): Adjust.
3448 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3449 Jim Wilson <jimw@sifive.com>
3451 * config/riscv/riscv-protos.h (riscv_output_return): New.
3452 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3453 (riscv_attribute_table, riscv_output_return),
3454 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3455 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3456 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3457 function.
3458 (riscv_expand_prologue): Add early return for naked function.
3459 (riscv_expand_epilogue): Likewise.
3460 (riscv_function_ok_for_sibcall): Return false for naked function.
3461 (riscv_set_current_function): New.
3462 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3463 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3464 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3465 * doc/extend.texi (RISC-V Function Attributes): New.
3467 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3469 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3470 check for 128-bit long double before checking TCmode.
3471 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3472 128-bit long doubles before checking TFmode or TCmode.
3473 (FLOAT128_IBM_P): Likewise.
3475 2018-01-10 Martin Sebor <msebor@redhat.com>
3477 PR tree-optimization/83671
3478 * builtins.c (c_strlen): Unconditionally return zero for the empty
3479 string.
3480 Use -Warray-bounds for warnings.
3481 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3482 for non-constant array indices with COMPONENT_REF, arrays of
3483 arrays, and pointers to arrays.
3484 (gimple_fold_builtin_strlen): Determine and set length range for
3485 non-constant character arrays.
3487 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3489 PR middle-end/81897
3490 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3491 empty blocks.
3493 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3495 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3497 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3499 PR target/83399
3500 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3501 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3502 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3503 indexed_or_indirect_operand predicate.
3504 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3505 (*vsx_le_perm_load_v8hi): Likewise.
3506 (*vsx_le_perm_load_v16qi): Likewise.
3507 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3508 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3509 (*vsx_le_perm_store_v8hi): Likewise.
3510 (*vsx_le_perm_store_v16qi): Likewise.
3511 (eight unnamed splitters): Likewise.
3513 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3515 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3516 * config/rs6000/emmintrin.h: Likewise.
3517 * config/rs6000/mmintrin.h: Likewise.
3518 * config/rs6000/xmmintrin.h: Likewise.
3520 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3522 PR c++/43486
3523 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3524 "public_flag".
3525 * tree.c (tree_nop_conversion): Return true for location wrapper
3526 nodes.
3527 (maybe_wrap_with_location): New function.
3528 (selftest::check_strip_nops): New function.
3529 (selftest::test_location_wrappers): New function.
3530 (selftest::tree_c_tests): Call it.
3531 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3532 (maybe_wrap_with_location): New decl.
3533 (EXPR_LOCATION_WRAPPER_P): New macro.
3534 (location_wrapper_p): New inline function.
3535 (tree_strip_any_location_wrapper): New inline function.
3537 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3539 PR target/83735
3540 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3541 stack_realign_offset for the largest alignment of stack slot
3542 actually used.
3543 (ix86_find_max_used_stack_alignment): New function.
3544 (ix86_finalize_stack_frame_flags): Use it. Set
3545 max_used_stack_alignment if we don't realign stack.
3546 * config/i386/i386.h (machine_function): Add
3547 max_used_stack_alignment.
3549 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3551 * config/arm/arm.opt (-mbranch-cost): New option.
3552 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3553 account.
3555 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3557 PR target/83629
3558 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3559 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3561 2018-01-10 Richard Biener <rguenther@suse.de>
3563 PR debug/83765
3564 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3565 early out so it also covers the case where we have a non-NULL
3566 origin.
3568 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3570 PR tree-optimization/83753
3571 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3572 for non-strided grouped accesses if the number of elements is 1.
3574 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3576 PR target/81616
3577 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3578 * i386.h (TARGET_USE_GATHER): Define.
3579 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3581 2018-01-10 Martin Liska <mliska@suse.cz>
3583 PR bootstrap/82831
3584 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3585 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3586 partitioning.
3587 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3588 CLEANUP_NO_PARTITIONING is not set.
3590 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3592 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3593 for vectors, as a partial revert of r254296.
3594 * rtl.h (const_vec_p): Delete.
3595 (const_vec_duplicate_p): Don't test for vector CONSTs.
3596 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3597 * expmed.c (make_tree): Likewise.
3599 Revert:
3600 * common.md (E, F): Use CONSTANT_P instead of checking for
3601 CONST_VECTOR.
3602 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3603 checking for CONST_VECTOR.
3605 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3607 PR middle-end/83575
3608 * predict.c (force_edge_cold): Handle in more sane way edges
3609 with no prediction.
3611 2018-01-09 Carl Love <cel@us.ibm.com>
3613 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3614 V4SI, V4SF types.
3615 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3616 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3617 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3618 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3619 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3620 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3621 * config/rs6000/rs6000-protos.h: Add extern defition for
3622 rs6000_generate_float2_double_code.
3623 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3624 function.
3625 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3626 (float2_v2df): Add define_expand.
3628 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3630 PR target/83628
3631 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3632 op_mode in the force_to_mode call.
3634 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3636 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3637 instead of checking each element individually.
3638 (aarch64_evpc_uzp): Likewise.
3639 (aarch64_evpc_zip): Likewise.
3640 (aarch64_evpc_ext): Likewise.
3641 (aarch64_evpc_rev): Likewise.
3642 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3643 instead of checking each element individually. Return true without
3644 generating rtl if
3645 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3646 whether all selected elements come from the same input, instead of
3647 checking each element individually. Remove calls to gen_rtx_REG,
3648 start_sequence and end_sequence and instead assert that no rtl is
3649 generated.
3651 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3653 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3654 order of HIGH and CONST checks.
3656 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3658 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3659 if the destination isn't an SSA_NAME.
3661 2018-01-09 Richard Biener <rguenther@suse.de>
3663 PR tree-optimization/83668
3664 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3665 move prologue...
3666 (canonicalize_loop_form): ... here, renamed from ...
3667 (canonicalize_loop_closed_ssa_form): ... this and amended to
3668 swap successor edges for loop exit blocks to make us use
3669 the RPO order we need for initial schedule generation.
3671 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3673 PR tree-optimization/64811
3674 * match.pd: When optimizing comparisons with Inf, avoid
3675 introducing or losing exceptions from comparisons with NaN.
3677 2018-01-09 Martin Liska <mliska@suse.cz>
3679 PR sanitizer/82517
3680 * asan.c (shadow_mem_size): Add gcc_assert.
3682 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3684 Don't save registers in main().
3686 PR target/83738
3687 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3688 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3689 * config/avr/avr.c (avr_set_current_function): Don't error if
3690 naked, OS_task or OS_main are specified at the same time.
3691 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3692 OS_main.
3693 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3694 attribute.
3695 * common/config/avr/avr-common.c (avr_option_optimization_table):
3696 Switch on -mmain-is-OS_task for optimizing compilations.
3698 2018-01-09 Richard Biener <rguenther@suse.de>
3700 PR tree-optimization/83572
3701 * graphite.c: Include cfganal.h.
3702 (graphite_transform_loops): Connect infinite loops to exit
3703 and remove fake edges at the end.
3705 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3707 * ipa-inline.c (edge_badness): Revert accidental checkin.
3709 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3711 PR ipa/80763
3712 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3713 symbols; not inline clones.
3715 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3717 PR target/83507
3718 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3719 hard registers. Formatting fixes.
3721 PR preprocessor/83722
3722 * gcc.c (try_generate_repro): Pass
3723 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3724 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3725 do_report_bug.
3727 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3728 Kito Cheng <kito.cheng@gmail.com>
3730 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3731 (riscv_leaf_function_p): Delete.
3732 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3734 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3736 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3737 function.
3738 (do_ifelse): New function.
3739 (do_isel): New function.
3740 (do_sub3): New function.
3741 (do_add3): New function.
3742 (do_load_mask_compare): New function.
3743 (do_overlap_load_compare): New function.
3744 (expand_compare_loop): New function.
3745 (expand_block_compare): Call expand_compare_loop() when appropriate.
3746 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3747 option description.
3748 (-mblock-compare-inline-loop-limit): New option.
3750 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3752 PR target/83677
3753 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3754 Reverse order of second and third operands in first alternative.
3755 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3756 of first and second elements in UNSPEC_VPERMR vector.
3757 (altivec_expand_vec_perm_le): Likewise.
3759 2017-01-08 Jeff Law <law@redhat.com>
3761 PR rtl-optimizatin/81308
3762 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3763 (process_switch): If group_case_labels makes a change, then set
3764 cfg_altered.
3765 (pass_convert_switch::execute): If a switch is converted, then
3766 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3768 PR rtl-optimization/81308
3769 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3770 splitting insns.
3772 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3774 PR target/83663 - Revert r255946
3775 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3776 generation for cases where splatting a value is not useful.
3777 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3778 across a vec_duplicate and a paradoxical subreg forming a vector
3779 mode to a vec_concat.
3781 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3783 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3784 -march=armv8.3-a variants.
3785 * config/arm/t-multilib: Likewise.
3786 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3788 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3790 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3791 to generate rtl.
3792 (cceq_ior_compare_complement): Give it a name so I can use it, and
3793 change boolean_or_operator predicate to boolean_operator so it can
3794 be used to generate a crand.
3795 (eqne): New code iterator.
3796 (bd/bd_neg): New code_attrs.
3797 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3798 a single define_insn.
3799 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3800 decrement (bdnzt/bdnzf/bdzt/bdzf).
3801 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3802 with the new names of the branch decrement patterns, and added the
3803 names of the branch decrement conditional patterns.
3805 2018-01-08 Richard Biener <rguenther@suse.de>
3807 PR tree-optimization/83563
3808 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3809 cache.
3811 2018-01-08 Richard Biener <rguenther@suse.de>
3813 PR middle-end/83713
3814 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3816 2018-01-08 Richard Biener <rguenther@suse.de>
3818 PR tree-optimization/83685
3819 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3820 references to abnormals.
3822 2018-01-08 Richard Biener <rguenther@suse.de>
3824 PR lto/83719
3825 * dwarf2out.c (output_indirect_strings): Handle empty
3826 skeleton_debug_str_hash.
3827 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3829 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3831 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3832 (emit_store_direct): Likewise.
3833 (arc_trampoline_adjust_address): Likewise.
3834 (arc_asm_trampoline_template): New function.
3835 (arc_initialize_trampoline): Use asm_trampoline_template.
3836 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3837 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3838 * config/arc/arc.md (flush_icache): Delete pattern.
3840 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3842 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3843 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3844 munaligned-access.
3846 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3848 PR target/83681
3849 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3850 by not USED_FOR_TARGET.
3851 (make_pass_resolve_sw_modes): Likewise.
3853 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3855 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3856 USED_FOR_TARGET.
3858 2018-01-08 Richard Biener <rguenther@suse.de>
3860 PR middle-end/83580
3861 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3863 2018-01-08 Richard Biener <rguenther@suse.de>
3865 PR middle-end/83517
3866 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3868 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3870 PR middle-end/81897
3871 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3872 basic blocks with a small number of successors.
3873 (convert_control_dep_chain_into_preds): Improve handling of
3874 forwarder blocks.
3875 (dump_predicates): Split apart into...
3876 (dump_pred_chain): ...here...
3877 (dump_pred_info): ...and here.
3878 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3879 (can_chain_union_be_invalidated_p): Improve check for invalidation
3880 of paths.
3881 (uninit_uses_cannot_happen): Avoid unnecessary if
3882 convert_control_dep_chain_into_preds yielded nothing.
3884 2018-01-06 Martin Sebor <msebor@redhat.com>
3886 PR tree-optimization/83640
3887 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3888 subtracting negative offset from size.
3889 (builtin_access::overlap): Adjust offset bounds of the access to fall
3890 within the size of the object if possible.
3892 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3894 PR rtl-optimization/83699
3895 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3896 extract_bit_field_as_subreg to cases in which the extracted
3897 value is also a vector.
3899 * lra-constraints.c (process_alt_operands): Test for the equivalence
3900 substitutions when detecting a possible reload cycle.
3902 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3904 PR debug/83480
3905 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3906 by default if flag_selective_schedling{,2}. Formatting fixes.
3908 PR rtl-optimization/83682
3909 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
3910 if it has non-VECTOR_MODE element mode.
3911 (vec_duplicate_p): Likewise.
3913 PR middle-end/83694
3914 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
3915 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
3917 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3919 PR target/83604
3920 * config/i386/i386-builtin.def
3921 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
3922 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
3923 Require also OPTION_MASK_ISA_AVX512F in addition to
3924 OPTION_MASK_ISA_GFNI.
3925 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
3926 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
3927 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
3928 to OPTION_MASK_ISA_GFNI.
3929 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
3930 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
3931 OPTION_MASK_ISA_AVX512BW.
3932 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
3933 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
3934 addition to OPTION_MASK_ISA_GFNI.
3935 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
3936 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
3937 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
3938 to OPTION_MASK_ISA_GFNI.
3939 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
3940 a requirement for all ISAs rather than any of them with a few
3941 exceptions.
3942 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
3943 processing.
3944 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
3945 bitmasks to be enabled with 3 exceptions, instead of requiring any
3946 enabled ISA with lots of exceptions.
3947 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
3948 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
3949 Change avx512bw in isa attribute to avx512f.
3950 * config/i386/sgxintrin.h: Add license boilerplate.
3951 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
3952 to __AVX512F__ and __AVX512VL to __AVX512VL__.
3953 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
3954 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
3955 defined.
3956 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
3957 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
3958 temporarily sse2 rather than sse if not enabled already.
3960 PR target/83604
3961 * config/i386/sse.md (VI248_VLBW): Rename to ...
3962 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
3963 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
3964 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
3965 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
3966 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
3967 mode iterator instead of VI248_VLBW.
3969 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
3971 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
3972 (record_modified): Skip clobbers; add debug output.
3973 (param_change_prob): Use sreal frequencies.
3975 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3977 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
3978 punt for user-aligned variables.
3980 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3982 * tree-chrec.c (chrec_contains_symbols): Return true for
3983 POLY_INT_CST.
3985 2018-01-05 Sudakshina Das <sudi.das@arm.com>
3987 PR target/82439
3988 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
3989 of (x|y) == x for BICS pattern.
3991 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3993 PR tree-optimization/83605
3994 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
3995 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
3996 can throw.
3998 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4000 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4001 * config/epiphany/rtems.h: New file.
4003 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4004 Uros Bizjak <ubizjak@gmail.com>
4006 PR target/83554
4007 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4008 QIreg_operand instead of register_operand predicate.
4009 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4010 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4011 comments instead of -fmitigate[-_]rop.
4013 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4015 PR bootstrap/81926
4016 * cgraphunit.c (symbol_table::compile): Switch to text_section
4017 before calling assembly_start debug hook.
4018 * run-rtl-passes.c (run_rtl_passes): Likewise.
4019 Include output.h.
4021 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4023 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4024 range_int_cst_p rather than !symbolic_range_p before calling
4025 extract_range_from_multiplicative_op_1.
4027 2017-01-04 Jeff Law <law@redhat.com>
4029 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4030 redundant test in assertion.
4032 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4034 * doc/rtl.texi: Document machine_mode wrapper classes.
4036 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4038 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4039 using tree_to_uhwi.
4041 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4043 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4044 the VEC_PERM_EXPR fold to fail.
4046 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4048 PR debug/83585
4049 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4050 to switched_sections.
4052 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4054 PR target/83680
4055 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4056 test for d.testing.
4058 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4060 PR target/83387
4061 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4062 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4064 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4066 PR debug/83666
4067 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4068 is BLKmode and bitpos not zero or mode change is needed.
4070 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4072 PR target/83675
4073 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4074 TARGET_VIS2.
4076 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4078 PR target/83628
4079 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4080 instead of MULT rtx. Update all corresponding splitters.
4081 (*saddl_se): Ditto.
4082 (*ssub<modesuffix>): Ditto.
4083 (*ssubl_se): Ditto.
4084 (*cmp_sadd_di): Update split patterns.
4085 (*cmp_sadd_si): Ditto.
4086 (*cmp_sadd_sidi): Ditto.
4087 (*cmp_ssub_di): Ditto.
4088 (*cmp_ssub_si): Ditto.
4089 (*cmp_ssub_sidi): Ditto.
4090 * config/alpha/predicates.md (const23_operand): New predicate.
4091 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4092 Look for ASHIFT, not MULT inner operand.
4093 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4095 2018-01-04 Martin Liska <mliska@suse.cz>
4097 PR gcov-profile/83669
4098 * gcov.c (output_intermediate_file): Add version to intermediate
4099 gcov file.
4100 * doc/gcov.texi: Document new field 'version' in intermediate
4101 file format. Fix location of '-k' option of gcov command.
4103 2018-01-04 Martin Liska <mliska@suse.cz>
4105 PR ipa/82352
4106 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4108 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4110 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4112 2018-01-03 Martin Sebor <msebor@redhat.com>
4114 PR tree-optimization/83655
4115 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4116 checking calls with invalid arguments.
4118 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4120 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4121 (vectorizable_mask_load_store): Delete.
4122 (vectorizable_call): Return false for masked loads and stores.
4123 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4124 instead of gimple_assign_rhs1.
4125 (vectorizable_load): Handle IFN_MASK_LOAD.
4126 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4128 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4130 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4131 split out from..,
4132 (vectorizable_mask_load_store): ...here.
4133 (vectorizable_load): ...and here.
4135 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4137 * tree-vect-stmts.c (vect_build_all_ones_mask)
4138 (vect_build_zero_merge_argument): New functions, split out from...
4139 (vectorizable_load): ...here.
4141 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4143 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4144 split out from...
4145 (vectorizable_mask_load_store): ...here.
4146 (vectorizable_store): ...and here.
4148 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4150 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4151 split out from...
4152 (vectorizable_mask_load_store): ...here.
4154 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4156 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4157 (vect_model_store_cost): Take a vec_load_store_type instead of a
4158 vect_def_type.
4159 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4160 (vect_model_store_cost): Take a vec_load_store_type instead of a
4161 vect_def_type.
4162 (vectorizable_mask_load_store): Update accordingly.
4163 (vectorizable_store): Likewise.
4164 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4166 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4168 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4169 IFN_MASK_LOAD calls here rather than...
4170 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4172 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4173 Alan Hayward <alan.hayward@arm.com>
4174 David Sherwood <david.sherwood@arm.com>
4176 * expmed.c (extract_bit_field_1): For vector extracts,
4177 fall back to extract_bit_field_as_subreg if vec_extract
4178 isn't available.
4180 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4181 Alan Hayward <alan.hayward@arm.com>
4182 David Sherwood <david.sherwood@arm.com>
4184 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4185 they are variable or constant sized.
4186 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4187 slots for constant-sized data.
4189 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4190 Alan Hayward <alan.hayward@arm.com>
4191 David Sherwood <david.sherwood@arm.com>
4193 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4194 handling COND_EXPRs with boolean comparisons, try to find a better
4195 basis for the mask type than the boolean itself.
4197 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4199 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4200 is calculated and how it can be overridden.
4201 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4202 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4203 if defined.
4204 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4205 if nonzero.
4207 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4208 Alan Hayward <alan.hayward@arm.com>
4209 David Sherwood <david.sherwood@arm.com>
4211 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4212 Remove the mode argument.
4213 (aarch64_simd_valid_immediate): Remove the mode and inverse
4214 arguments.
4215 * config/aarch64/iterators.md (bitsize): New iterator.
4216 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4217 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4218 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4219 aarch64_simd_valid_immediate.
4220 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4221 (aarch64_reg_or_bic_imm): Likewise.
4222 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4223 with an insn_type enum and msl with a modifier_type enum.
4224 Replace element_width with a scalar_mode. Change the shift
4225 to unsigned int. Add constructors for scalar_float_mode and
4226 scalar_int_mode elements.
4227 (aarch64_vect_float_const_representable_p): Delete.
4228 (aarch64_can_const_movi_rtx_p)
4229 (aarch64_simd_scalar_immediate_valid_for_move)
4230 (aarch64_simd_make_constant): Update call to
4231 aarch64_simd_valid_immediate.
4232 (aarch64_advsimd_valid_immediate_hs): New function.
4233 (aarch64_advsimd_valid_immediate): Likewise.
4234 (aarch64_simd_valid_immediate): Remove mode and inverse
4235 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4236 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4237 and aarch64_float_const_representable_p on the result.
4238 (aarch64_output_simd_mov_immediate): Remove mode argument.
4239 Update call to aarch64_simd_valid_immediate and use of
4240 simd_immediate_info.
4241 (aarch64_output_scalar_simd_mov_immediate): Update call
4242 accordingly.
4244 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4245 Alan Hayward <alan.hayward@arm.com>
4246 David Sherwood <david.sherwood@arm.com>
4248 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4249 (mode_nunits): Likewise CONST_MODE_NUNITS.
4250 * machmode.def (ADJUST_NUNITS): Document.
4251 * genmodes.c (mode_data::need_nunits_adj): New field.
4252 (blank_mode): Update accordingly.
4253 (adj_nunits): New variable.
4254 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4255 parameter.
4256 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4257 listed in adj_nunits.
4258 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4259 listed in adj_nunits. Don't emit case statements for such modes.
4260 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4261 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4262 nothing if adj_nunits is nonnull.
4263 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4264 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4265 (emit_mode_fbit): Update use of print_maybe_const_decl.
4266 (emit_move_size): Likewise. Treat the array as non-const
4267 if adj_nunits.
4268 (emit_mode_adjustments): Handle adj_nunits.
4270 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4272 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4273 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4274 (VECTOR_MODES): Use it.
4275 (make_vector_modes): Take the prefix as an argument.
4277 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4278 Alan Hayward <alan.hayward@arm.com>
4279 David Sherwood <david.sherwood@arm.com>
4281 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4282 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4283 for MODE_VECTOR_BOOL.
4284 * machmode.def (VECTOR_BOOL_MODE): Document.
4285 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4286 (make_vector_bool_mode): New function.
4287 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4288 MODE_VECTOR_BOOL.
4289 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4290 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4291 Likewise.
4292 * stor-layout.c (int_mode_for_mode): Likewise.
4293 * tree.c (build_vector_type_for_mode): Likewise.
4294 * varasm.c (output_constant_pool_2): Likewise.
4295 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4296 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4297 for MODE_VECTOR_BOOL.
4298 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4299 of mode class checks.
4300 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4301 instead of a list of mode class checks.
4302 (expand_vector_scalar_condition): Likewise.
4303 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4305 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4306 Alan Hayward <alan.hayward@arm.com>
4307 David Sherwood <david.sherwood@arm.com>
4309 * machmode.h (mode_size): Change from unsigned short to
4310 poly_uint16_pod.
4311 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4312 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4313 or if measurement_type is not polynomial.
4314 (fixed_size_mode::includes_p): Check for constant-sized modes.
4315 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4316 return a poly_uint16 rather than an unsigned short.
4317 (emit_mode_size): Change the type of mode_size from unsigned short
4318 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4319 (emit_mode_adjustments): Cope with polynomial vector sizes.
4320 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4321 for GET_MODE_SIZE.
4322 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4323 for GET_MODE_SIZE.
4324 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4325 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4326 * caller-save.c (setup_save_areas): Likewise.
4327 (replace_reg_with_saved_mem): Likewise.
4328 * calls.c (emit_library_call_value_1): Likewise.
4329 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4330 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4331 (gen_lowpart_for_combine): Likewise.
4332 * convert.c (convert_to_integer_1): Likewise.
4333 * cse.c (equiv_constant, cse_insn): Likewise.
4334 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4335 (cselib_subst_to_values): Likewise.
4336 * dce.c (word_dce_process_block): Likewise.
4337 * df-problems.c (df_word_lr_mark_ref): Likewise.
4338 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4339 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4340 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4341 (rtl_for_decl_location): Likewise.
4342 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4343 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4344 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4345 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4346 (expand_expr_real_1): Likewise.
4347 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4348 (pad_below): Likewise.
4349 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4350 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4351 * ira.c (get_subreg_tracking_sizes): Likewise.
4352 * ira-build.c (ira_create_allocno_objects): Likewise.
4353 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4354 (ira_sort_regnos_for_alter_reg): Likewise.
4355 * ira-costs.c (record_operand_costs): Likewise.
4356 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4357 (resolve_simple_move): Likewise.
4358 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4359 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4360 (lra_constraints): Likewise.
4361 (CONST_POOL_OK_P): Reject variable-sized modes.
4362 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4363 (add_pseudo_to_slot, lra_spill): Likewise.
4364 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4365 * optabs-query.c (get_best_extraction_insn): Likewise.
4366 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4367 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4368 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4369 * recog.c (offsettable_address_addr_space_p): Likewise.
4370 * regcprop.c (maybe_mode_change): Likewise.
4371 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4372 * regrename.c (build_def_use): Likewise.
4373 * regstat.c (dump_reg_info): Likewise.
4374 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4375 (find_reloads, find_reloads_subreg_address): Likewise.
4376 * reload1.c (eliminate_regs_1): Likewise.
4377 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4378 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4379 (simplify_binary_operation_1, simplify_subreg): Likewise.
4380 * targhooks.c (default_function_arg_padding): Likewise.
4381 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4382 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4383 (verify_gimple_assign_ternary): Likewise.
4384 * tree-inline.c (estimate_move_cost): Likewise.
4385 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4386 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4387 (get_address_cost_ainc): Likewise.
4388 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4389 (vect_supportable_dr_alignment): Likewise.
4390 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4391 (vectorizable_reduction): Likewise.
4392 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4393 (vectorizable_operation, vectorizable_load): Likewise.
4394 * tree.c (build_same_sized_truth_vector_type): Likewise.
4395 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4396 * var-tracking.c (emit_note_insn_var_location): Likewise.
4397 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4398 (ADDR_VEC_ALIGN): Likewise.
4400 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4401 Alan Hayward <alan.hayward@arm.com>
4402 David Sherwood <david.sherwood@arm.com>
4404 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4405 unsigned short.
4406 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4407 or if measurement_type is polynomial.
4408 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4409 * combine.c (make_extraction): Likewise.
4410 * dse.c (find_shift_sequence): Likewise.
4411 * dwarf2out.c (mem_loc_descriptor): Likewise.
4412 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4413 (extract_bit_field, extract_low_bits): Likewise.
4414 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4415 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4416 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4417 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4418 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4419 * reload.c (find_reloads): Likewise.
4420 * reload1.c (alter_reg): Likewise.
4421 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4422 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4423 * tree-if-conv.c (predicate_mem_writes): Likewise.
4424 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4425 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4426 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4427 * valtrack.c (dead_debug_insert_temp): Likewise.
4428 * varasm.c (mergeable_constant_section): Likewise.
4429 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4431 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4432 Alan Hayward <alan.hayward@arm.com>
4433 David Sherwood <david.sherwood@arm.com>
4435 * expr.c (expand_assignment): Cope with polynomial mode sizes
4436 when assigning to a CONCAT.
4438 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4439 Alan Hayward <alan.hayward@arm.com>
4440 David Sherwood <david.sherwood@arm.com>
4442 * machmode.h (mode_precision): Change from unsigned short to
4443 poly_uint16_pod.
4444 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4445 short.
4446 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4447 or if measurement_type is not polynomial.
4448 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4449 in which the mode is already known to be a scalar_int_mode.
4450 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4451 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4452 initializer.
4453 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4454 for GET_MODE_PRECISION.
4455 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4456 for GET_MODE_PRECISION.
4457 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4458 as polynomial.
4459 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4460 (expand_field_assignment, make_extraction): Likewise.
4461 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4462 (get_last_value): Likewise.
4463 * convert.c (convert_to_integer_1): Likewise.
4464 * cse.c (cse_insn): Likewise.
4465 * expr.c (expand_expr_real_1): Likewise.
4466 * lra-constraints.c (simplify_operand_subreg): Likewise.
4467 * optabs-query.c (can_atomic_load_p): Likewise.
4468 * optabs.c (expand_atomic_load): Likewise.
4469 (expand_atomic_store): Likewise.
4470 * ree.c (combine_reaching_defs): Likewise.
4471 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4472 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4473 * tree.h (type_has_mode_precision_p): Likewise.
4474 * ubsan.c (instrument_si_overflow): Likewise.
4476 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4477 Alan Hayward <alan.hayward@arm.com>
4478 David Sherwood <david.sherwood@arm.com>
4480 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4481 polynomial numbers of units.
4482 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4483 (valid_vector_subparts_p): New function.
4484 (build_vector_type): Remove temporary shim and take the number
4485 of units as a poly_uint64 rather than an int.
4486 (build_opaque_vector_type): Take the number of units as a
4487 poly_uint64 rather than an int.
4488 * tree.c (build_vector_from_ctor): Handle polynomial
4489 TYPE_VECTOR_SUBPARTS.
4490 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4491 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4492 (build_vector_from_val): If the number of units is variable,
4493 use build_vec_duplicate_cst for constant operands and
4494 VEC_DUPLICATE_EXPR otherwise.
4495 (make_vector_type): Remove temporary is_constant ().
4496 (build_vector_type, build_opaque_vector_type): Take the number of
4497 units as a poly_uint64 rather than an int.
4498 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4499 VECTOR_CST_NELTS.
4500 * cfgexpand.c (expand_debug_expr): Likewise.
4501 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4502 (store_constructor, expand_expr_real_1): Likewise.
4503 (const_scalar_mask_from_tree): Likewise.
4504 * fold-const-call.c (fold_const_reduction): Likewise.
4505 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4506 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4507 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4508 (fold_relational_const): Likewise.
4509 (native_interpret_vector): Likewise. Change the size from an
4510 int to an unsigned int.
4511 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4512 TYPE_VECTOR_SUBPARTS.
4513 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4514 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4515 duplicating a non-constant operand into a variable-length vector.
4516 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4517 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4518 * ipa-icf.c (sem_variable::equals): Likewise.
4519 * match.pd: Likewise.
4520 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4521 * print-tree.c (print_node): Likewise.
4522 * stor-layout.c (layout_type): Likewise.
4523 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4524 * tree-cfg.c (verify_gimple_comparison): Likewise.
4525 (verify_gimple_assign_binary): Likewise.
4526 (verify_gimple_assign_ternary): Likewise.
4527 (verify_gimple_assign_single): Likewise.
4528 * tree-pretty-print.c (dump_generic_node): Likewise.
4529 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4530 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4531 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4532 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4533 (vect_shift_permute_load_chain): Likewise.
4534 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4535 (expand_vector_condition, optimize_vector_constructor): Likewise.
4536 (lower_vec_perm, get_compute_type): Likewise.
4537 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4538 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4539 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4540 (vect_recog_mask_conversion_pattern): Likewise.
4541 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4542 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4543 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4544 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4545 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4546 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4547 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4548 (supportable_widening_operation): Likewise.
4549 (supportable_narrowing_operation): Likewise.
4550 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4551 Likewise.
4552 * varasm.c (output_constant): Likewise.
4554 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4555 Alan Hayward <alan.hayward@arm.com>
4556 David Sherwood <david.sherwood@arm.com>
4558 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4559 so that both the length == 3 and length != 3 cases set up their
4560 own permute vectors. Add comments explaining why we know the
4561 number of elements is constant.
4562 (vect_permute_load_chain): Likewise.
4564 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4565 Alan Hayward <alan.hayward@arm.com>
4566 David Sherwood <david.sherwood@arm.com>
4568 * machmode.h (mode_nunits): Change from unsigned char to
4569 poly_uint16_pod.
4570 (ONLY_FIXED_SIZE_MODES): New macro.
4571 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4572 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4573 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4574 New typedefs.
4575 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4576 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4577 or if measurement_type is not polynomial.
4578 * genmodes.c (ZERO_COEFFS): New macro.
4579 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4580 poly_uint16.
4581 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4582 Use ZERO_COEFFS when emitting initializers.
4583 * data-streamer.h (bp_pack_poly_value): New function.
4584 (bp_unpack_poly_value): Likewise.
4585 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4586 for GET_MODE_NUNITS.
4587 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4588 for GET_MODE_NUNITS.
4589 * tree.c (make_vector_type): Remove temporary shim and make
4590 the real function take the number of units as a poly_uint64
4591 rather than an int.
4592 (build_vector_type_for_mode): Handle polynomial nunits.
4593 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4594 * emit-rtl.c (const_vec_series_p_1): Likewise.
4595 (gen_rtx_CONST_VECTOR): Likewise.
4596 * fold-const.c (test_vec_duplicate_folding): Likewise.
4597 * genrecog.c (validate_pattern): Likewise.
4598 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4599 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4600 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4601 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4602 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4603 * rtlanal.c (subreg_get_info): Likewise.
4604 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4605 (vect_grouped_load_supported): Likewise.
4606 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4607 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4608 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4609 (simplify_const_unary_operation, simplify_binary_operation_1)
4610 (simplify_const_binary_operation, simplify_ternary_operation)
4611 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4612 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4613 instead of CONST_VECTOR_NUNITS.
4614 * varasm.c (output_constant_pool_2): Likewise.
4615 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4616 explicit-encoded elements in the XVEC for variable-length vectors.
4618 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4620 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4622 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4623 Alan Hayward <alan.hayward@arm.com>
4624 David Sherwood <david.sherwood@arm.com>
4626 * coretypes.h (fixed_size_mode): Declare.
4627 (fixed_size_mode_pod): New typedef.
4628 * builtins.h (target_builtins::x_apply_args_mode)
4629 (target_builtins::x_apply_result_mode): Change type to
4630 fixed_size_mode_pod.
4631 * builtins.c (apply_args_size, apply_result_size, result_vector)
4632 (expand_builtin_apply_args_1, expand_builtin_apply)
4633 (expand_builtin_return): Update accordingly.
4635 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4637 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4638 * cselib.c (cselib_hash_rtx): Likewise.
4639 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4640 CONST_VECTOR encoding.
4642 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4643 Jeff Law <law@redhat.com>
4645 PR target/83641
4646 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4647 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4648 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4649 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4651 PR target/83641
4652 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4653 explicitly probe *sp in a noreturn function if there were any callee
4654 register saves or frame pointer is needed.
4656 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4658 PR debug/83621
4659 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4660 BLKmode for ternary, binary or unary expressions.
4662 PR debug/83645
4663 * var-tracking.c (delete_vta_debug_insn): New inline function.
4664 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4665 insns from get_insns () to NULL instead of each bb separately.
4666 Use delete_vta_debug_insn. No longer static.
4667 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4668 delete_vta_debug_insns callers.
4669 * rtl.h (delete_vta_debug_insns): Declare.
4670 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4671 instead of variable_tracking_main.
4673 2018-01-03 Martin Sebor <msebor@redhat.com>
4675 PR tree-optimization/83603
4676 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4677 arguments past the endof the argument list in functions declared
4678 without a prototype.
4679 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4680 Avoid checking when arguments are null.
4682 2018-01-03 Martin Sebor <msebor@redhat.com>
4684 PR c/83559
4685 * doc/extend.texi (attribute const): Fix a typo.
4686 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4687 issuing -Wsuggest-attribute for void functions.
4689 2018-01-03 Martin Sebor <msebor@redhat.com>
4691 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4692 offset_int::from instead of wide_int::to_shwi.
4693 (maybe_diag_overlap): Remove assertion.
4694 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4695 * gimple-ssa-sprintf.c (format_directive): Same.
4696 (parse_directive): Same.
4697 (sprintf_dom_walker::compute_format_length): Same.
4698 (try_substitute_return_value): Same.
4700 2017-01-03 Jeff Law <law@redhat.com>
4702 PR middle-end/83654
4703 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4704 non-constant residual for zero at runtime and avoid probing in
4705 that case. Reorganize code for trailing problem to mirror handling
4706 of the residual.
4708 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4710 PR tree-optimization/83501
4711 * tree-ssa-strlen.c (get_string_cst): New.
4712 (handle_char_store): Call get_string_cst.
4714 2018-01-03 Martin Liska <mliska@suse.cz>
4716 PR tree-optimization/83593
4717 * tree-ssa-strlen.c: Include tree-cfg.h.
4718 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4719 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4720 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4721 to false.
4722 (strlen_dom_walker::before_dom_children): Call
4723 gimple_purge_dead_eh_edges. Dump tranformation with details
4724 dump flags.
4725 (strlen_dom_walker::before_dom_children): Update call by adding
4726 new argument cleanup_eh.
4727 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4729 2018-01-03 Martin Liska <mliska@suse.cz>
4731 PR ipa/83549
4732 * cif-code.def (VARIADIC_THUNK): New enum value.
4733 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4734 thunks.
4736 2018-01-03 Jan Beulich <jbeulich@suse.com>
4738 * sse.md (mov<mode>_internal): Tighten condition for when to use
4739 vmovdqu<ssescalarsize> for TI and OI modes.
4741 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4743 Update copyright years.
4745 2018-01-03 Martin Liska <mliska@suse.cz>
4747 PR ipa/83594
4748 * ipa-visibility.c (function_and_variable_visibility): Skip
4749 functions with noipa attribure.
4751 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4753 * gcc.c (process_command): Update copyright notice dates.
4754 * gcov-dump.c (print_version): Ditto.
4755 * gcov.c (print_version): Ditto.
4756 * gcov-tool.c (print_version): Ditto.
4757 * gengtype.c (create_file): Ditto.
4758 * doc/cpp.texi: Bump @copying's copyright year.
4759 * doc/cppinternals.texi: Ditto.
4760 * doc/gcc.texi: Ditto.
4761 * doc/gccint.texi: Ditto.
4762 * doc/gcov.texi: Ditto.
4763 * doc/install.texi: Ditto.
4764 * doc/invoke.texi: Ditto.
4766 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4768 * vector-builder.h (vector_builder::m_full_nelts): Change from
4769 unsigned int to poly_uint64.
4770 (vector_builder::full_nelts): Update prototype accordingly.
4771 (vector_builder::new_vector): Likewise.
4772 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4773 (vector_builder::operator ==): Likewise.
4774 (vector_builder::finalize): Likewise.
4775 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4776 Take the number of elements as a poly_uint64 rather than an
4777 unsigned int.
4778 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4779 from unsigned int to poly_uint64.
4780 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4781 (vec_perm_indices::new_vector): Likewise.
4782 (vec_perm_indices::length): Likewise.
4783 (vec_perm_indices::nelts_per_input): Likewise.
4784 (vec_perm_indices::input_nelts): Likewise.
4785 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4786 number of elements per input as a poly_uint64 rather than an
4787 unsigned int. Use the original encoding for variable-length
4788 vectors, rather than clamping each individual element.
4789 For the second and subsequent elements in each pattern,
4790 clamp the step and base before clamping their sum.
4791 (vec_perm_indices::series_p): Handle polynomial element counts.
4792 (vec_perm_indices::all_in_range_p): Likewise.
4793 (vec_perm_indices_to_tree): Likewise.
4794 (vec_perm_indices_to_rtx): Likewise.
4795 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4796 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4797 (tree_vector_builder::new_binary_operation): Handle polynomial
4798 element counts. Return false if we need to know the number
4799 of elements at compile time.
4800 * fold-const.c (fold_vec_perm): Punt if the number of elements
4801 isn't known at compile time.
4803 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4805 * vec-perm-indices.h (vec_perm_builder): Change element type
4806 from HOST_WIDE_INT to poly_int64.
4807 (vec_perm_indices::element_type): Update accordingly.
4808 (vec_perm_indices::clamp): Handle polynomial element_types.
4809 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4810 (vec_perm_indices::all_in_range_p): Likewise.
4811 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4812 than shwi trees.
4813 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4814 polynomial vec_perm_indices element types.
4815 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4816 * fold-const.c (fold_vec_perm): Likewise.
4817 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4818 * tree-vect-generic.c (lower_vec_perm): Likewise.
4819 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4820 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4821 element type to HOST_WIDE_INT.
4823 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4824 Alan Hayward <alan.hayward@arm.com>
4825 David Sherwood <david.sherwood@arm.com>
4827 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4828 rather than an int. Use plus_constant.
4829 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4830 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4832 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4833 Alan Hayward <alan.hayward@arm.com>
4834 David Sherwood <david.sherwood@arm.com>
4836 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4837 a HOST_WIDE_INT to a poly_int64.
4839 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4840 Alan Hayward <alan.hayward@arm.com>
4841 David Sherwood <david.sherwood@arm.com>
4843 * calls.c (load_register_parameters): Cope with polynomial
4844 mode sizes. Require a constant size for BLKmode parameters
4845 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4846 forces a parameter to be padded at the lsb end in order to
4847 fill a complete number of words, require the parameter size
4848 to be ordered wrt UNITS_PER_WORD.
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 * reload1.c (spill_stack_slot_width): Change element type
4855 from unsigned int to poly_uint64_pod.
4856 (alter_reg): Treat mode sizes as polynomial.
4858 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4859 Alan Hayward <alan.hayward@arm.com>
4860 David Sherwood <david.sherwood@arm.com>
4862 * reload.c (complex_word_subreg_p): New function.
4863 (reload_inner_reg_of_subreg, push_reload): Use it.
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 * lra-constraints.c (process_alt_operands): Reject matched
4870 operands whose sizes aren't ordered.
4871 (match_reload): Refer to this check here.
4873 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4874 Alan Hayward <alan.hayward@arm.com>
4875 David Sherwood <david.sherwood@arm.com>
4877 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4878 that the mode size is in the set {1, 2, 4, 8, 16}.
4880 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4881 Alan Hayward <alan.hayward@arm.com>
4882 David Sherwood <david.sherwood@arm.com>
4884 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4885 Use plus_constant instead of gen_rtx_PLUS.
4887 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4888 Alan Hayward <alan.hayward@arm.com>
4889 David Sherwood <david.sherwood@arm.com>
4891 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4892 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4893 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4894 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4895 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4896 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4897 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4898 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4899 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4900 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4901 a poly_int64.
4902 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4903 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4904 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4905 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4906 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4907 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4908 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4909 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
4910 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
4911 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
4912 function.
4913 * expr.c (emit_move_resolve_push): Treat the input and result
4914 of PUSH_ROUNDING as a poly_int64.
4915 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
4916 (emit_push_insn): Likewise.
4917 * lra-eliminations.c (mark_not_eliminable): Likewise.
4918 * recog.c (push_operand): Likewise.
4919 * reload1.c (elimination_effects): Likewise.
4920 * rtlanal.c (nonzero_bits1): Likewise.
4921 * calls.c (store_one_arg): Likewise. Require the padding to be
4922 known at compile time.
4924 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4925 Alan Hayward <alan.hayward@arm.com>
4926 David Sherwood <david.sherwood@arm.com>
4928 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
4929 Use plus_constant instead of gen_rtx_PLUS.
4931 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4932 Alan Hayward <alan.hayward@arm.com>
4933 David Sherwood <david.sherwood@arm.com>
4935 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
4936 rather than an int.
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 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
4943 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
4944 via stack temporaries. Treat the mode size as polynomial too.
4946 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4947 Alan Hayward <alan.hayward@arm.com>
4948 David Sherwood <david.sherwood@arm.com>
4950 * expr.c (expand_expr_real_2): When handling conversions involving
4951 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
4952 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
4953 as a poly_uint64 too.
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 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
4961 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4962 Alan Hayward <alan.hayward@arm.com>
4963 David Sherwood <david.sherwood@arm.com>
4965 * combine.c (can_change_dest_mode): Handle polynomial
4966 REGMODE_NATURAL_SIZE.
4967 * expmed.c (store_bit_field_1): Likewise.
4968 * expr.c (store_constructor): Likewise.
4969 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
4970 and polynomial REGMODE_NATURAL_SIZE.
4971 (gen_lowpart_common): Likewise.
4972 * reginfo.c (record_subregs_of_mode): Likewise.
4973 * rtlanal.c (read_modify_subreg_p): Likewise.
4975 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4976 Alan Hayward <alan.hayward@arm.com>
4977 David Sherwood <david.sherwood@arm.com>
4979 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
4980 numbers of elements.
4982 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4983 Alan Hayward <alan.hayward@arm.com>
4984 David Sherwood <david.sherwood@arm.com>
4986 * match.pd: Cope with polynomial numbers of vector elements.
4988 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4989 Alan Hayward <alan.hayward@arm.com>
4990 David Sherwood <david.sherwood@arm.com>
4992 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
4993 in a POINTER_PLUS_EXPR.
4995 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4996 Alan Hayward <alan.hayward@arm.com>
4997 David Sherwood <david.sherwood@arm.com>
4999 * omp-simd-clone.c (simd_clone_subparts): New function.
5000 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5001 (ipa_simd_modify_function_body): Likewise.
5003 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5004 Alan Hayward <alan.hayward@arm.com>
5005 David Sherwood <david.sherwood@arm.com>
5007 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5008 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5009 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5010 (expand_vector_condition, vector_element): Likewise.
5011 (subparts_gt): New function.
5012 (get_compute_type): Use subparts_gt.
5013 (count_type_subparts): Delete.
5014 (expand_vector_operations_1): Use subparts_gt instead of
5015 count_type_subparts.
5017 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5018 Alan Hayward <alan.hayward@arm.com>
5019 David Sherwood <david.sherwood@arm.com>
5021 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5022 (vect_compile_time_alias): ...this new function. Do the calculation
5023 on poly_ints rather than trees.
5024 (vect_prune_runtime_alias_test_list): Update call accordingly.
5026 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5027 Alan Hayward <alan.hayward@arm.com>
5028 David Sherwood <david.sherwood@arm.com>
5030 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5031 numbers of units.
5032 (vect_schedule_slp_instance): Likewise.
5034 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5035 Alan Hayward <alan.hayward@arm.com>
5036 David Sherwood <david.sherwood@arm.com>
5038 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5039 constant and extern definitions for variable-length vectors.
5040 (vect_get_constant_vectors): Note that the number of units
5041 is known to be constant.
5043 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5044 Alan Hayward <alan.hayward@arm.com>
5045 David Sherwood <david.sherwood@arm.com>
5047 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5048 of units as polynomial. Choose between WIDE and NARROW based
5049 on multiple_p.
5051 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5052 Alan Hayward <alan.hayward@arm.com>
5053 David Sherwood <david.sherwood@arm.com>
5055 * tree-vect-stmts.c (simd_clone_subparts): New function.
5056 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
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 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5063 vectors as polynomial. Use build_index_vector for
5064 IFN_GOMP_SIMD_LANE.
5066 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5067 Alan Hayward <alan.hayward@arm.com>
5068 David Sherwood <david.sherwood@arm.com>
5070 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5071 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5072 for variable-length vectors.
5073 (vectorizable_mask_load_store): Treat the number of units as
5074 polynomial, asserting that it is constant if the condition has
5075 already been enforced.
5076 (vectorizable_store, vectorizable_load): Likewise.
5078 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5079 Alan Hayward <alan.hayward@arm.com>
5080 David Sherwood <david.sherwood@arm.com>
5082 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5083 of units as polynomial. Punt if we can't tell at compile time
5084 which vector contains the final result.
5086 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5087 Alan Hayward <alan.hayward@arm.com>
5088 David Sherwood <david.sherwood@arm.com>
5090 * tree-vect-loop.c (vectorizable_induction): Treat the number
5091 of units as polynomial. Punt on SLP inductions. Use an integer
5092 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5093 cast of such a series for variable-length floating-point
5094 reductions.
5096 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5097 Alan Hayward <alan.hayward@arm.com>
5098 David Sherwood <david.sherwood@arm.com>
5100 * tree.h (build_index_vector): Declare.
5101 * tree.c (build_index_vector): New function.
5102 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5103 of units as polynomial, forcibly converting it to a constant if
5104 vectorizable_reduction has already enforced the condition.
5105 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5106 to create a {1,2,3,...} vector.
5107 (vectorizable_reduction): Treat the number of units as polynomial.
5108 Choose vectype_in based on the largest scalar element size rather
5109 than the smallest number of units. Enforce the restrictions
5110 relied on above.
5112 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5113 Alan Hayward <alan.hayward@arm.com>
5114 David Sherwood <david.sherwood@arm.com>
5116 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5117 number of units as polynomial.
5119 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5120 Alan Hayward <alan.hayward@arm.com>
5121 David Sherwood <david.sherwood@arm.com>
5123 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5124 * target.def (autovectorize_vector_sizes): Return the vector sizes
5125 by pointer, using vector_sizes rather than a bitmask.
5126 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5127 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5128 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5129 Likewise.
5130 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5131 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5132 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5133 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5134 * omp-general.c (omp_max_vf): Likewise.
5135 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5136 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5137 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5138 * tree-vect-slp.c (vect_slp_bb): Likewise.
5139 * doc/tm.texi: Regenerate.
5140 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5141 to a poly_uint64.
5142 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5143 the vector size as a poly_uint64 rather than an unsigned int.
5144 (current_vector_size): Change from an unsigned int to a poly_uint64.
5145 (get_vectype_for_scalar_type): Update accordingly.
5146 * tree.h (build_truth_vector_type): Take the size and number of
5147 units as a poly_uint64 rather than an unsigned int.
5148 (build_vector_type): Add a temporary overload that takes
5149 the number of units as a poly_uint64 rather than an unsigned int.
5150 * tree.c (make_vector_type): Likewise.
5151 (build_truth_vector_type): Take the number of units as a poly_uint64
5152 rather than an unsigned int.
5154 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5155 Alan Hayward <alan.hayward@arm.com>
5156 David Sherwood <david.sherwood@arm.com>
5158 * target.def (get_mask_mode): Take the number of units and length
5159 as poly_uint64s rather than unsigned ints.
5160 * targhooks.h (default_get_mask_mode): Update accordingly.
5161 * targhooks.c (default_get_mask_mode): Likewise.
5162 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5163 * doc/tm.texi: Regenerate.
5165 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5166 Alan Hayward <alan.hayward@arm.com>
5167 David Sherwood <david.sherwood@arm.com>
5169 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5170 * omp-general.c (omp_max_vf): Likewise.
5171 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5172 (expand_omp_simd): Handle polynomial safelen.
5173 * omp-low.c (omplow_simd_context): Add a default constructor.
5174 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5175 (lower_rec_simd_input_clauses): Update accordingly.
5176 (lower_rec_input_clauses): Likewise.
5178 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5179 Alan Hayward <alan.hayward@arm.com>
5180 David Sherwood <david.sherwood@arm.com>
5182 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5183 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5184 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5185 (vect_analyze_slp_cost): Likewise.
5186 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5187 (vect_model_load_cost): Likewise.
5189 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5190 Alan Hayward <alan.hayward@arm.com>
5191 David Sherwood <david.sherwood@arm.com>
5193 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5194 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5195 from an unsigned int * to a poly_uint64_pod *.
5196 (calculate_unrolling_factor): New function.
5197 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5199 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5200 Alan Hayward <alan.hayward@arm.com>
5201 David Sherwood <david.sherwood@arm.com>
5203 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5204 from an unsigned int to a poly_uint64.
5205 (_loop_vec_info::slp_unrolling_factor): Likewise.
5206 (_loop_vec_info::vectorization_factor): Change from an int
5207 to a poly_uint64.
5208 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5209 (vect_get_num_vectors): New function.
5210 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5211 (vect_get_num_copies): Use vect_get_num_vectors.
5212 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5213 to an unsigned int *.
5214 (vect_analyze_data_refs): Change min_vf from an int * to a
5215 poly_uint64 *.
5216 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5217 than an unsigned HOST_WIDE_INT.
5218 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5219 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5220 to an unsigned int *.
5221 (vect_analyze_data_ref_dependences): Likewise.
5222 (vect_compute_data_ref_alignment): Handle polynomial vf.
5223 (vect_enhance_data_refs_alignment): Likewise.
5224 (vect_prune_runtime_alias_test_list): Likewise.
5225 (vect_shift_permute_load_chain): Likewise.
5226 (vect_supportable_dr_alignment): Likewise.
5227 (dependence_distance_ge_vf): Take the vectorization factor as a
5228 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5229 (vect_analyze_data_refs): Change min_vf from an int * to a
5230 poly_uint64 *.
5231 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5232 vfm1 as a poly_uint64 rather than an int. Make the same change
5233 for the returned bound_scalar.
5234 (vect_gen_vector_loop_niters): Handle polynomial vf.
5235 (vect_do_peeling): Likewise. Update call to
5236 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5237 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5238 be constant.
5239 * tree-vect-loop.c (vect_determine_vectorization_factor)
5240 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5241 (vect_get_known_peeling_cost): Likewise.
5242 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5243 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5244 (vect_transform_loop): Likewise. Use the lowest possible VF when
5245 updating the upper bounds of the loop.
5246 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5247 rather than an int.
5248 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5249 polynomial unroll factors.
5250 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5251 (vect_make_slp_decision): Likewise.
5252 (vect_supported_load_permutation_p): Likewise, and polynomial
5253 vf too.
5254 (vect_analyze_slp_cost): Handle polynomial vf.
5255 (vect_slp_analyze_node_operations): Likewise.
5256 (vect_slp_analyze_bb_1): Likewise.
5257 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5258 than an unsigned HOST_WIDE_INT.
5259 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5260 (vectorizable_load): Handle polynomial vf.
5261 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5262 a poly_uint64.
5263 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5265 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5266 Alan Hayward <alan.hayward@arm.com>
5267 David Sherwood <david.sherwood@arm.com>
5269 * match.pd: Handle bit operations involving three constants
5270 and try to fold one pair.
5272 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5274 * tree-vect-loop-manip.c: Include gimple-fold.h.
5275 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5276 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5277 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5278 Add a path that uses a step of VF instead of 1, but disable it
5279 for now.
5280 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5281 and niters_no_overflow parameters. Update calls to
5282 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5283 Create a new SSA name if the latter choses to use a ste other
5284 than zero, and return it via niters_vector_mult_vf_var.
5285 * tree-vect-loop.c (vect_transform_loop): Update calls to
5286 vect_do_peeling, vect_gen_vector_loop_niters and
5287 slpeel_make_loop_iterate_ntimes.
5288 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5289 (vect_gen_vector_loop_niters): Update declarations after above changes.
5291 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5293 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5294 128-bit round to integer instructions.
5295 (ceil<mode>2): Likewise.
5296 (btrunc<mode>2): Likewise.
5297 (round<mode>2): Likewise.
5299 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5301 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5302 unaligned VSX load/store on P8/P9.
5303 (expand_block_clear): Allow the use of unaligned VSX
5304 load/store on P8/P9.
5306 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5308 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5309 New function.
5310 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5311 swap associated with both a load and a store.
5313 2018-01-02 Andrew Waterman <andrew@sifive.com>
5315 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5316 * config/riscv/riscv.md (clear_cache): Use it.
5318 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5320 * web.c: Remove out-of-date comment.
5322 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5324 * expr.c (fixup_args_size_notes): Check that any existing
5325 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5326 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5327 (emit_single_push_insn): ...here.
5329 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5331 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5332 (const_vector_encoded_nelts): New function.
5333 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5334 (const_vector_int_elt, const_vector_elt): Declare.
5335 * emit-rtl.c (const_vector_int_elt_1): New function.
5336 (const_vector_elt): Likewise.
5337 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5338 of CONST_VECTOR_ELT.
5340 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5342 * expr.c: Include rtx-vector-builder.h.
5343 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5344 directly on the tree encoding.
5345 (const_vector_from_tree): Likewise.
5346 * optabs.c: Include rtx-vector-builder.h.
5347 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5348 sequence of "u" values.
5349 * vec-perm-indices.c: Include rtx-vector-builder.h.
5350 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5351 directly on the vec_perm_indices encoding.
5353 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5355 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5356 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5357 * rtx-vector-builder.h: New file.
5358 * rtx-vector-builder.c: Likewise.
5359 * rtl.h (rtx_def::u2): Add a const_vector field.
5360 (CONST_VECTOR_NPATTERNS): New macro.
5361 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5362 (CONST_VECTOR_DUPLICATE_P): Likewise.
5363 (CONST_VECTOR_STEPPED_P): Likewise.
5364 (CONST_VECTOR_ENCODED_ELT): Likewise.
5365 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5366 (unwrap_const_vec_duplicate): Likewise.
5367 (const_vec_series_p): Check for a non-duplicated vector encoding.
5368 Say that the function only returns true for integer vectors.
5369 * emit-rtl.c: Include rtx-vector-builder.h.
5370 (gen_const_vec_duplicate_1): Delete.
5371 (gen_const_vector): Call gen_const_vec_duplicate instead of
5372 gen_const_vec_duplicate_1.
5373 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5374 (gen_const_vec_duplicate): Use rtx_vector_builder.
5375 (gen_const_vec_series): Likewise.
5376 (gen_rtx_CONST_VECTOR): Likewise.
5377 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5378 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5379 Build a new vector rather than modifying a CONST_VECTOR in-place.
5380 (handle_special_swappables): Update call accordingly.
5381 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5382 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5383 Build a new vector rather than modifying a CONST_VECTOR in-place.
5384 (handle_special_swappables): Update call accordingly.
5386 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5388 * simplify-rtx.c (simplify_const_binary_operation): Use
5389 CONST_VECTOR_ELT instead of XVECEXP.
5391 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5393 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5394 the selector elements to be different from the data elements
5395 if the selector is a VECTOR_CST.
5396 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5397 ssizetype for the selector.
5399 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5401 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5402 before testing each element individually.
5403 * tree-vect-generic.c (lower_vec_perm): Likewise.
5405 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5407 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5408 * selftest-run-tests.c (selftest::run_tests): Call it.
5409 * vector-builder.h (vector_builder::operator ==): New function.
5410 (vector_builder::operator !=): Likewise.
5411 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5412 (vec_perm_indices::all_from_input_p): New function.
5413 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5414 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5415 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5416 instead of reading the VECTOR_CST directly. Detect whether both
5417 vector inputs are the same before constructing the vec_perm_indices,
5418 and update the number of inputs argument accordingly. Use the
5419 utility functions added above. Only construct sel2 if we need to.
5421 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5423 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5424 the broadcast of the low byte.
5425 (expand_mult_highpart): Use an explicit encoding for the permutes.
5426 * optabs-query.c (can_mult_highpart_p): Likewise.
5427 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5428 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5429 (vectorizable_bswap): Likewise.
5430 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5431 explicit encoding for the power-of-2 permutes.
5432 (vect_permute_store_chain): Likewise.
5433 (vect_grouped_load_supported): Likewise.
5434 (vect_permute_load_chain): Likewise.
5436 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5438 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5439 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5440 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5441 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5442 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5443 (vect_gen_perm_mask_any): Likewise.
5445 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5447 * int-vector-builder.h: New file.
5448 * vec-perm-indices.h: Include int-vector-builder.h.
5449 (vec_perm_indices): Redefine as an int_vector_builder.
5450 (auto_vec_perm_indices): Delete.
5451 (vec_perm_builder): Redefine as a stand-alone class.
5452 (vec_perm_indices::vec_perm_indices): New function.
5453 (vec_perm_indices::clamp): Likewise.
5454 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5455 (vec_perm_indices::new_vector): New function.
5456 (vec_perm_indices::new_expanded_vector): Update for new
5457 vec_perm_indices class.
5458 (vec_perm_indices::rotate_inputs): New function.
5459 (vec_perm_indices::all_in_range_p): Operate directly on the
5460 encoded form, without computing elided elements.
5461 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5462 encoding. Update for new vec_perm_indices class.
5463 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5464 the given vec_perm_builder.
5465 (expand_vec_perm_var): Update vec_perm_builder constructor.
5466 (expand_mult_highpart): Use vec_perm_builder instead of
5467 auto_vec_perm_indices.
5468 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5469 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5470 or double series encoding as appropriate.
5471 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5472 vec_perm_indices instead of auto_vec_perm_indices.
5473 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5474 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5475 (vect_permute_store_chain): Likewise.
5476 (vect_grouped_load_supported): Likewise.
5477 (vect_permute_load_chain): Likewise.
5478 (vect_shift_permute_load_chain): Likewise.
5479 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5480 (vect_transform_slp_perm_load): Likewise.
5481 (vect_schedule_slp_instance): Likewise.
5482 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5483 (vectorizable_mask_load_store): Likewise.
5484 (vectorizable_bswap): Likewise.
5485 (vectorizable_store): Likewise.
5486 (vectorizable_load): Likewise.
5487 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5488 vec_perm_indices instead of auto_vec_perm_indices. Use
5489 tree_to_vec_perm_builder to read the vector from a tree.
5490 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5491 vec_perm_builder instead of a vec_perm_indices.
5492 (have_whole_vector_shift): Use vec_perm_builder and
5493 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5494 truncation to calc_vec_perm_mask_for_shift.
5495 (vect_create_epilog_for_reduction): Likewise.
5496 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5497 from auto_vec_perm_indices to vec_perm_indices.
5498 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5499 instead of changing individual elements.
5500 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5501 the vector in d.perm.
5502 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5503 from auto_vec_perm_indices to vec_perm_indices.
5504 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5505 instead of changing individual elements.
5506 (arm_vectorize_vec_perm_const): Use new_vector to install
5507 the vector in d.perm.
5508 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5509 Update vec_perm_builder constructor.
5510 (rs6000_expand_interleave): Likewise.
5511 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5512 (rs6000_expand_interleave): Likewise.
5514 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5516 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5517 to qimode could truncate the indices.
5518 * optabs.c (expand_vec_perm_var): Likewise.
5520 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5522 * Makefile.in (OBJS): Add vec-perm-indices.o.
5523 * vec-perm-indices.h: New file.
5524 * vec-perm-indices.c: Likewise.
5525 * target.h (vec_perm_indices): Replace with a forward class
5526 declaration.
5527 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5528 * optabs.h: Include vec-perm-indices.h.
5529 (expand_vec_perm): Delete.
5530 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5531 (expand_vec_perm_const): Declare.
5532 * target.def (vec_perm_const_ok): Replace with...
5533 (vec_perm_const): ...this new hook.
5534 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5535 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5536 * doc/tm.texi: Regenerate.
5537 * optabs.def (vec_perm_const): Delete.
5538 * doc/md.texi (vec_perm_const): Likewise.
5539 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5540 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5541 expand_vec_perm for constant permutation vectors. Assert that
5542 the mode of variable permutation vectors is the integer equivalent
5543 of the mode that is being permuted.
5544 * optabs-query.h (selector_fits_mode_p): Declare.
5545 * optabs-query.c: Include vec-perm-indices.h.
5546 (selector_fits_mode_p): New function.
5547 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5548 is defined, instead of checking whether the vec_perm_const_optab
5549 exists. Use targetm.vectorize.vec_perm_const instead of
5550 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5551 fit in the vector mode before using a variable permute.
5552 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5553 vec_perm_indices instead of an rtx.
5554 (expand_vec_perm): Replace with...
5555 (expand_vec_perm_const): ...this new function. Take the selector
5556 as a vec_perm_indices rather than an rtx. Also take the mode of
5557 the selector. Update call to shift_amt_for_vec_perm_mask.
5558 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5559 Use vec_perm_indices::new_expanded_vector to expand the original
5560 selector into bytes. Check whether the indices fit in the vector
5561 mode before using a variable permute.
5562 (expand_vec_perm_var): Make global.
5563 (expand_mult_highpart): Use expand_vec_perm_const.
5564 * fold-const.c: Includes vec-perm-indices.h.
5565 * tree-ssa-forwprop.c: Likewise.
5566 * tree-vect-data-refs.c: Likewise.
5567 * tree-vect-generic.c: Likewise.
5568 * tree-vect-loop.c: Likewise.
5569 * tree-vect-slp.c: Likewise.
5570 * tree-vect-stmts.c: Likewise.
5571 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5572 Delete.
5573 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5574 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5575 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5576 (aarch64_vectorize_vec_perm_const): ...this new function.
5577 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5578 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5579 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5580 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5581 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5582 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5583 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5584 into...
5585 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5586 check for NEON modes.
5587 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5588 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5589 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5590 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5591 into...
5592 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5593 the old VEC_PERM_CONST conditions.
5594 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5595 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5596 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5597 (ia64_vectorize_vec_perm_const_ok): Merge into...
5598 (ia64_vectorize_vec_perm_const): ...this new function.
5599 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5600 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5601 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5602 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5603 * config/mips/mips.c (mips_expand_vec_perm_const)
5604 (mips_vectorize_vec_perm_const_ok): Merge into...
5605 (mips_vectorize_vec_perm_const): ...this new function.
5606 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5607 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5608 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5609 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5610 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5611 (rs6000_expand_vec_perm_const): Delete.
5612 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5613 Delete.
5614 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5615 (altivec_expand_vec_perm_const_le): Take each operand individually.
5616 Operate on constant selectors rather than rtxes.
5617 (altivec_expand_vec_perm_const): Likewise. Update call to
5618 altivec_expand_vec_perm_const_le.
5619 (rs6000_expand_vec_perm_const): Delete.
5620 (rs6000_vectorize_vec_perm_const_ok): Delete.
5621 (rs6000_vectorize_vec_perm_const): New function.
5622 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5623 an element count and rtx array.
5624 (rs6000_expand_extract_even): Update call accordingly.
5625 (rs6000_expand_interleave): Likewise.
5626 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5627 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5628 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5629 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5630 (rs6000_expand_vec_perm_const): Delete.
5631 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5632 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5633 (altivec_expand_vec_perm_const_le): Take each operand individually.
5634 Operate on constant selectors rather than rtxes.
5635 (altivec_expand_vec_perm_const): Likewise. Update call to
5636 altivec_expand_vec_perm_const_le.
5637 (rs6000_expand_vec_perm_const): Delete.
5638 (rs6000_vectorize_vec_perm_const_ok): Delete.
5639 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5640 reference to the SPE evmerge intructions.
5641 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5642 an element count and rtx array.
5643 (rs6000_expand_extract_even): Update call accordingly.
5644 (rs6000_expand_interleave): Likewise.
5645 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5646 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5647 new function.
5648 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5650 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5652 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5653 vector mode and that that mode matches the mode of the data
5654 being permuted.
5655 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5656 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5657 directly using expand_vec_perm_1 when forcing selectors into
5658 registers.
5659 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5661 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5663 * optabs-query.h (can_vec_perm_p): Delete.
5664 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5665 * optabs-query.c (can_vec_perm_p): Split into...
5666 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5667 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5668 particular selector is valid.
5669 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5670 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5671 (vect_grouped_load_supported): Likewise.
5672 (vect_shift_permute_load_chain): Likewise.
5673 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5674 (vect_transform_slp_perm_load): Likewise.
5675 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5676 (vectorizable_bswap): Likewise.
5677 (vect_gen_perm_mask_checked): Likewise.
5678 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5679 implementations of variable permutation vectors into account
5680 when deciding which selector to use.
5681 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5682 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5683 with a false third argument.
5684 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5685 to test whether the constant selector is valid and can_vec_perm_var_p
5686 to test whether a variable selector is valid.
5688 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5690 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5691 * optabs-query.c (can_vec_perm_p): Likewise.
5692 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5693 instead of vec_perm_indices.
5694 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5695 (vect_gen_perm_mask_checked): Likewise,
5696 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5697 (vect_gen_perm_mask_checked): Likewise,
5699 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5701 * optabs-query.h (qimode_for_vec_perm): Declare.
5702 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5703 (qimode_for_vec_perm): ...this new function.
5704 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5706 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5708 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5709 does not have a conditional at the top.
5711 2018-01-02 Richard Biener <rguenther@suse.de>
5713 * ipa-inline.c (big_speedup_p): Fix expression.
5715 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5717 PR target/81616
5718 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5719 for generic 4->6.
5721 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5723 PR target/81616
5724 Generic tuning.
5725 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5726 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5727 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5728 cond_taken_branch_cost 3->4.
5730 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5732 PR tree-optimization/83581
5733 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5734 TODO_cleanup_cfg if any changes have been made.
5736 PR middle-end/83608
5737 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5738 convert_modes if target mode has the right side, but different mode
5739 class.
5741 PR middle-end/83609
5742 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5743 last argument when extracting from CONCAT. If either from_real or
5744 from_imag is NULL, use expansion through memory. If result is not
5745 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5746 the parts directly to inner mode, if even that fails, use expansion
5747 through memory.
5749 PR middle-end/83623
5750 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5751 check for bswap in mode rather than HImode and use that in expand_unop
5752 too.
5753 \f
5754 Copyright (C) 2018 Free Software Foundation, Inc.
5756 Copying and distribution of this file, with or without modification,
5757 are permitted in any medium without royalty provided the copyright
5758 notice and this notice are preserved.