1 2018-02-07 Richard Biener <rguenther@suse.de>
3 PR tree-optimization/84037
4 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
5 (cse_and_gimplify_to_preheader): Declare.
6 (vect_get_place_in_interleaving_chain): Likewise.
7 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
9 (_loop_vec_info::~_loop_vec_info): Delete it.
10 (cse_and_gimplify_to_preheader): New function.
11 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
12 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
13 (vectorizable_load): Likewise. For grouped stores always base
14 the IV on the first element.
15 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
16 condition before gimplifying.
18 2018-02-07 Jakub Jelinek <jakub@redhat.com>
20 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
21 *DIV_EXPR and *MOD_EXPR.
23 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
26 * config/i386/i386.c (ix86_option_override_internal): Mask out
27 the CF_SET bit when checking -fcf-protection.
29 2018-02-07 Tom de Vries <tom@codesourcery.com>
32 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
35 2018-02-07 Richard Biener <rguenther@suse.de>
37 PR tree-optimization/84204
38 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
41 PR tree-optimization/84205
42 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
43 special-case isl_ast_op_zdiv_r.
45 PR tree-optimization/84223
46 * graphite-scop-detection.c (gather_bbs::before_dom_children):
47 Only add conditions from within the region.
48 (gather_bbs::after_dom_children): Adjust.
50 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
53 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
54 * config/avr/avr.md: Only post-reload split REG-REG moves if
55 either register is REGERAL_REG_P.
57 2018-02-07 Jakub Jelinek <jakub@redhat.com>
59 PR tree-optimization/84235
60 * tree-ssa-scopedtables.c
61 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
62 if the subtraction is performed in floating point type where NaNs are
63 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
64 build 1. Formatting fix.
66 2018-02-06 Jakub Jelinek <jakub@redhat.com>
69 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
70 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
71 and skip it regardless of bb boundaries. Use CALL_P macro,
72 don't test INSN_P (insn) together with CALL_P or JUMP_P check
73 unnecessarily, formatting fix.
75 2018-02-06 Michael Collison <michael.collison@arm.com>
77 * config/arm/thumb2.md:
78 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
79 (*thumb_mov_notscc): Ditto.
81 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
84 * config/rs6000/rs6000.md (su code attribute): Use "u" for
85 unsigned_fix, not "s".
87 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
89 * configure.ac (gcc_fn_eh_frame_ro): New function.
90 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
91 correct .eh_frame permissions.
92 * configure: Regenerate.
94 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
96 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
99 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
101 * config/rs6000/rs6000.c (rs6000_option_override_internal):
102 Display warning message for -mno-speculate-indirect-jumps.
104 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
106 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
108 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
110 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
112 PR tree-optimization/84225
113 * tree-eh.c (find_trapping_overflow): Only call
114 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
116 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
119 * config/i386/i386.c: Reimplement the check of possible options
120 -mibt/-mshstk conbination. Change error messages.
121 * doc/invoke.texi: Fix a typo: remove extra '='.
123 2018-02-06 Marek Polacek <polacek@redhat.com>
125 PR tree-optimization/84228
126 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
128 2018-02-06 Tamar Christina <tamar.christina@arm.com>
131 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
132 emitted arch directives.
133 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
134 __ARM_FEATURE_COPROC before changing architectures.
136 2018-02-06 Richard Biener <rguenther@suse.de>
138 * config/i386/i386.c (print_reg): Fix typo.
139 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
141 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
143 * configure: Regenerate.
145 2018-02-05 Martin Sebor <msebor@redhat.com>
147 PR tree-optimization/83369
148 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
151 2018-02-05 Martin Liska <mliska@suse.cz>
153 * doc/invoke.texi: Cherry-pick upstream r323995.
155 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
157 * ira.c (ira_init_register_move_cost): Adjust comment.
159 2018-02-05 Martin Liska <mliska@suse.cz>
161 PR gcov-profile/84137
162 * doc/gcov.texi: Fix typo in documentation.
164 2018-02-05 Martin Liska <mliska@suse.cz>
166 PR gcov-profile/83879
167 * doc/gcov.texi: Document necessity of --dynamic-list-data when
168 using dlopen functionality.
170 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
172 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
173 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
174 _mm_maskz_range_ss, _mm_mask_range_round_ss,
175 _mm_maskz_range_round_ss): New intrinsics.
176 (__builtin_ia32_rangesd128_round)
177 (__builtin_ia32_rangess128_round): Remove.
178 (__builtin_ia32_rangesd128_mask_round,
179 __builtin_ia32_rangess128_mask_round): New builtins.
180 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
181 __builtin_ia32_rangess128_round): Remove.
182 (__builtin_ia32_rangesd128_mask_round,
183 __builtin_ia32_rangess128_mask_round): New builtins.
184 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
185 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
186 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
187 "<round_saeonly_constraint>")): Changed to ...
188 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
189 "<round_saeonly_scalar_constraint>")): ... this.
190 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
191 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
192 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
193 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
194 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
196 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
198 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
200 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
201 Remove all values except native, 8540 and 8548.
203 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
205 * config/i386/i386.c (ix86_output_function_return): Pass
206 INVALID_REGNUM, instead of -1, as invalid register number to
207 indirect_thunk_name and output_indirect_thunk.
209 2018-02-02 Julia Koval <julia.koval@intel.com>
211 * config.gcc: Add -march=icelake.
212 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
213 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
214 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
215 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
216 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
217 (processor_target_table): Add icelake.
218 (ix86_option_override_internal): Handle new PTAs.
219 (get_builtin_code_for_version): Handle icelake.
220 (M_INTEL_COREI7_ICELAKE): New.
221 (fold_builtin_cpu): Handle icelake.
222 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
223 * doc/invoke.texi: Add -march=icelake.
225 2018-02-02 Julia Koval <julia.koval@intel.com>
227 * config/i386/i386.c (ix86_option_override_internal): Change flags type
229 * wide-int-bitmask.h: New.
231 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
234 * config/i386/i386.md: Replace Pmode with word_mode in
235 builtin_setjmp_setup and builtin_longjmp to support x32.
237 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
241 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
243 (rs6000_supported_cpu_names): New static variable.
244 (linux_cpu_translation_table): Likewise.
245 (elf_platform) <cpu>: Define new static variable and use it.
246 Translate kernel AT_PLATFORM name to canonical name if needed.
247 Error if platform name is unknown.
249 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
252 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
254 2018-02-01 Jeff Law <law@redhat.com>
257 * config/i386/i386.c (release_scratch_register_on_entry): Add new
258 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
259 the scratch if RELEASE_VIA_POP is false.
260 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
261 If we have to save a temporary register, decrement SIZE appropriately.
262 Pass new arguments to release_scratch_register_on_entry.
263 (ix86_adjust_stack_and_probe): Likewise.
264 (ix86_emit_probe_stack_range): Pass new arguments to
265 release_scratch_register_on_entry.
267 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
269 PR rtl-optimization/84157
270 * combine.c (change_zero_ext): Use REG_P predicate in
271 front of HARD_REGISTER_P predicate.
273 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
275 * config/avr/avr.c (avr_option_override): Move disabling of
276 -fdelete-null-pointer-checks to...
277 * common/config/avr/avr-common.c (avr_option_optimization_table):
280 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
282 PR tree-optimization/81635
283 * tree-data-ref.c (split_constant_offset_1): For types that
284 wrap on overflow, try to use range info to prove that wrapping
287 2018-02-01 Renlin Li <renlin.li@arm.com>
290 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
292 (aarch64_register_move_cost): Likewise.
293 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
295 (REG_CLASS_NAMES): Likewise.
296 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
297 TAILCALL_ADDR_REGS. Remove IP registers.
298 * config/aarch64/aarch64.md (Ucs): Update register constraint.
300 2018-02-01 Richard Biener <rguenther@suse.de>
302 * domwalk.h (dom_walker::dom_walker): Add additional constructor
303 for specifying RPO order and allow NULL for that.
304 * domwalk.c (dom_walker::dom_walker): Likewise.
305 (dom_walker::walk): Handle NULL RPO order.
306 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
308 (rewrite_update_dom_walker): Likewise.
309 (mark_def_dom_walker): Likewise.
311 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
313 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
314 (aarch64_maybe_expand_sve_subreg_move): Declare.
315 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
316 * config/aarch64/predicates.md (aarch64_any_register_operand): New
318 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
319 that are semantically a reverse operation.
320 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
321 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
322 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
324 (aarch64_can_change_mode_class): For big-endian, forbid changes
325 between two SVE modes if they have different element sizes.
327 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
329 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
330 the TImode handling for big-endian targets.
332 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
334 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
335 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
337 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
338 Remove BSWAP handing for big-endian targets and use the form of
339 LD1RQ appropariate for the mode.
341 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
343 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
344 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
347 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
350 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
351 check for operands that need to go through aarch64_sve_reload_be.
353 2018-02-01 Jakub Jelinek <jakub@redhat.com>
355 PR tree-optimization/81661
356 PR tree-optimization/84117
357 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
358 * tree-eh.c: Include gimplify.h.
359 (find_trapping_overflow, replace_trapping_overflow,
360 rewrite_to_non_trapping_overflow): New functions.
361 * tree-vect-loop.c: Include tree-eh.h.
362 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
363 * tree-data-ref.c: Include tree-eh.h.
364 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
366 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
368 PR rtl-optimization/84123
369 * combine.c (change_zero_ext): Check if hard register satisfies
370 can_change_dest_mode before calling gen_lowpart_SUBREG.
372 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
375 * ira.c (ira_init_register_move_cost): Remove assert.
377 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
379 PR rtl-optimization/84071
380 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
381 * doc/tm.texi: Regenerate.
383 2018-01-31 Richard Biener <rguenther@suse.de>
385 PR tree-optimization/84132
386 * tree-data-ref.c (analyze_miv_subscript): Properly
387 check whether evolution_function_is_affine_multivariate_p
388 before calling gcd_of_steps_may_divide_p.
390 2018-01-31 Julia Koval <julia.koval@intel.com>
393 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
394 * config/i386/i386.md (rdpid_rex64) New.
395 (rdpid): Make 32bit only.
397 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
400 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
401 an IDENTIFIER_NODE for FUNCTION_TYPE's.
403 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
406 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
408 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
410 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
412 PR rtl-optimization/84071
413 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
414 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
416 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
418 * config/arc/arc.c (arc_handle_aux_attribute): New function.
419 (arc_attribute_table): Add 'aux' attribute.
420 (arc_in_small_data_p): Consider aux like variables.
421 (arc_is_aux_reg_p): New function.
422 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
423 (arc_get_aux_arg): New function.
424 (prepare_move_operands): Handle aux-register access.
425 (arc_handle_aux_attribute): New function.
426 * doc/extend.texi (ARC Variable attributes): Add subsection.
428 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
430 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
431 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
432 (arc_attribute_table): Add 'uncached' attribute.
433 (arc_print_operand): Print '.di' flag for uncached memory
435 (arc_in_small_data_p): Do not consider for small data the uncached
437 (arc_is_uncached_mem_p): New function.
438 * config/arc/predicates.md (compact_store_memory_operand): Check
439 for uncached memory accesses.
440 (nonvol_nonimm_operand): Likewise.
441 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
443 2018-01-31 Jakub Jelinek <jakub@redhat.com>
446 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
447 falign-loops=): Add Optimization flag.
449 2018-01-30 Jeff Law <law@redhat.com>
452 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
453 INT_REGISTERS_SAVED. Check it prior to calling
454 get_scratch_register_on_entry.
455 (ix86_adjust_stack_and_probe): Similarly.
456 (ix86_emit_probe_stack_range): Similarly.
457 (ix86_expand_prologue): Corresponding changes.
459 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
462 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
463 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
465 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
468 * lra-constraints.c (curr_insn_transform): Process AND in the
471 2018-01-30 Jakub Jelinek <jakub@redhat.com>
473 PR rtl-optimization/83986
474 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
475 dependence against last_pending_memory_flush in addition to
478 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
480 PR tree-optimization/81611
481 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
484 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
486 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
489 2018-01-30 Richard Biener <rguenther@suse.de>
490 Jakub Jelinek <jakub@redhat.com>
492 PR tree-optimization/84111
493 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
494 inner loops added during recursion, as they don't have up-to-date
497 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
500 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
501 (can_inline_edge_by_limits_p): ... here.
502 (can_early_inline_edge_p, check_callers,
503 update_caller_keys, update_callee_keys, recursive_inlining,
504 add_new_edges_to_heap, speculation_useful_p,
505 inline_small_functions,
506 inline_small_functions, flatten_function,
507 inline_to_all_callers_1): Update.
509 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
511 * profile-count.c (profile_count::combine_with_ipa_count): Handle
514 2018-01-30 Richard Biener <rguenther@suse.de>
516 PR tree-optimization/83008
517 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
518 invariant and constant vector uses in stmts when they need
521 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
524 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
525 * configure: Regenerate.
527 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
529 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
531 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
532 Use gen_rtx_REG rather than gen_lowpart.
534 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
536 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
537 rather than 0 when creating partial subregs.
539 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
541 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
544 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
547 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
548 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
549 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
550 flags. This restores the settings used before the 2017-07-24.
551 Turning off pre increment/decrement/modify allows IVOPTS to
552 optimize DF/SF loops where the index is an int.
554 2018-01-29 Richard Biener <rguenther@suse.de>
555 Kelvin Nilsen <kelvin@gcc.gnu.org>
558 * tree-vect-stmts.c (vectorizable_call): Don't call
559 targetm.vectorize_builtin_md_vectorized_function if callee is
562 2018-01-22 Carl Love <cel@us.ibm.com>
564 * doc/extend.tex: Fix typo in second arg in
565 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
567 2018-01-29 Richard Biener <rguenther@suse.de>
569 PR tree-optimization/84086
570 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
571 (flush_ssaname_freelist): When SSA names were released reset
574 2018-01-29 Richard Biener <rguenther@suse.de>
576 PR tree-optimization/84057
577 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
578 removed paths when removing edges.
580 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
582 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
583 -mfunction-return=@var{choice}.
585 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
588 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
590 (layout::print_source_line): Likewise.
591 (test_get_line_width_without_trailing_whitespace): Add test cases.
593 2018-01-27 Jakub Jelinek <jakub@redhat.com>
596 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
599 2018-01-26 Jim Wilson <jimw@sifive.com>
601 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
603 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
606 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
608 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
609 and CMP + SUB-immediate -> SUBS.
611 2018-01-26 Martin Sebor <msebor@redhat.com>
613 PR tree-optimization/83896
614 * tree-ssa-strlen.c (get_string_len): Rename...
615 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
616 Avoid assuming length is constant.
617 (handle_char_store): Use HOST_WIDE_INT for string length.
619 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
622 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
623 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
625 2018-01-26 Richard Biener <rguenther@suse.de>
627 PR rtl-optimization/84003
628 * dse.c (record_store): Only record redundant stores when
629 the earlier store aliases at least all accesses the later one does.
631 2018-01-26 Jakub Jelinek <jakub@redhat.com>
633 PR rtl-optimization/83985
634 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
635 REG_CFA_RESTORE insns.
636 (delete_unmarked_insns): Don't ignore separate shrink wrapping
637 REG_CFA_RESTORE insns here.
640 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
641 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
643 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
645 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
646 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
647 (arc_init): Likewise.
648 (arc_override_options): Likewise.
649 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
651 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
653 * config/arc/arc.h (TARGET_DBNZ): Define.
654 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
655 properly set the tune attribute.
656 (dbnz): Use TARGET_DBNZ guard.
657 * config/arc/arc.opt (mtune): Add core3 option.
659 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
661 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
662 recognize new pic like addresses.
663 (arc_delegitimize_address): Clean up.
665 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
667 * config/arc/arc-arches.def: Option mrf16 valid for all
669 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
670 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
671 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
672 * config/arc/arc-tables.opt: Regenerate.
673 * config/arc/arc.c (arc_conditional_register_usage): Handle
674 reduced register file case.
675 (arc_file_start): Set must have build attributes.
676 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
678 * config/arc/arc.opt (mrf16): Add new option.
679 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
680 * config/arc/genmultilib.awk: Handle new mrf16 option.
681 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
682 * config/arc/t-multilib: Regenerate.
683 * doc/invoke.texi (ARC Options): Document mrf16 option.
685 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
687 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
688 * config/arc/arc.c (arc_handle_secure_attribute): New function.
689 (arc_attribute_table): Add 'secure_call' attribute.
690 (arc_print_operand): Print secure call operand.
691 (arc_function_ok_for_sibcall): Don't optimize tail calls when
693 (arc_is_secure_call_p): New function. * config/arc/arc.md
694 (call_i): Add support for sjli instruction.
695 (call_value_i): Likewise.
696 * config/arc/constraints.md (Csc): New constraint.
698 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
699 John Eric Martin <John.Martin@emmicro-us.com>
701 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
702 * config/arc/arc.c (_arc_jli_section): New struct.
703 (arc_jli_section): New type.
704 (rc_jli_sections): New static variable.
705 (arc_handle_jli_attribute): New function.
706 (arc_attribute_table): Add jli_always and jli_fixed attribute.
707 (arc_file_end): New function.
708 (TARGET_ASM_FILE_END): Define.
709 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
710 (arc_add_jli_section): New function.
711 (jli_call_scan): Likewise.
712 (arc_reorg): Call jli_call_scan.
713 (arc_output_addsi): Remove 'S' from printing asm operand.
714 (arc_is_jli_call_p): New function.
715 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
717 (movhi_insn): Likewise.
718 (movsi_insn): Likewise.
719 (movsi_set_cc_insn): Likewise.
720 (loadqi_update): Likewise.
721 (load_zeroextendqisi_update): Likewise.
722 (load_signextendqisi_update): Likewise.
723 (loadhi_update): Likewise.
724 (load_zeroextendhisi_update): Likewise.
725 (load_signextendhisi_update): Likewise.
726 (loadsi_update): Likewise.
727 (loadsf_update): Likewise.
728 (movsicc_insn): Likewise.
729 (bset_insn): Likewise.
730 (bxor_insn): Likewise.
731 (bclr_insn): Likewise.
732 (bmsk_insn): Likewise.
733 (bicsi3_insn): Likewise.
734 (cmpsi_cc_c_insn): Likewise.
735 (movsi_ne): Likewise.
736 (movsi_cond_exec): Likewise.
737 (clrsbsi2): Likewise.
748 (call_i): Remove 'S' asm letter, add jli instruction.
749 (call_value_i): Likewise.
750 * config/arc/arc.op (mjli-always): New option.
751 * config/arc/constraints.md (Cji): New constraint.
752 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
754 (subsf3_fpx): Likewise.
755 (mulsf3_fpx): Likewise.
756 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
758 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
760 * doc/invoke.texi (ARC): Document mjli-always option.
762 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
764 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
765 and use incw and decw where possible
766 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
768 2018-01-26 Richard Biener <rguenther@suse.de>
770 PR tree-optimization/81082
771 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
772 association if it requires casting to unsigned.
773 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
774 from fold_plusminus_mult_expr to catch important cases late when
775 range info is available.
777 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
779 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
780 * configure.ac (hidden_linkonce): New test.
781 * configure: Regenerate.
782 * config.in: Regenerate.
784 2018-01-26 Julia Koval <julia.koval@intel.com>
786 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
787 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
788 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
789 _mm_mask_bitshuffle_epi64_mask): Fix type.
790 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
791 USI_FTYPE_V4DI_V4DI_USI): Remove.
792 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
793 __builtin_ia32_vpshufbitqmb256_mask,
794 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
795 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
796 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
798 2018-01-26 Alan Modra <amodra@gmail.com>
801 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
802 UNSPEC_VBPERMQ. Sort other unspecs.
804 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
806 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
808 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
811 * predict.c (drop_profile): Do not push/pop cfun; update also
813 (handle_missing_profiles): Fix logic looking for zero profiles.
815 2018-01-25 Jakub Jelinek <jakub@redhat.com>
818 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
819 on functions with #pragma omp declare simd or functions with simd
821 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
822 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
823 Remove trailing \n from warning_at calls.
825 2018-01-25 Tom de Vries <tom@codesourcery.com>
828 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
829 for neutered workers.
831 2018-01-24 Joseph Myers <joseph@codesourcery.com>
834 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
835 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
837 2018-01-24 Jeff Law <law@redhat.com>
840 * i386.c (get_probe_interval): Move to earlier point.
841 (ix86_compute_frame_layout): If -fstack-clash-protection and
842 the frame is larger than the probe interval, then use pushes
843 to save registers rather than reg->mem moves.
844 (ix86_expand_prologue): Remove conditional for int_registers_saved
847 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
850 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
851 min/max for never referenced object.
853 2018-01-24 Jakub Jelinek <jakub@redhat.com>
856 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
858 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
859 attributes from DECL_ATTRIBUTES (decl) without affecting
860 DECL_ATTRIBUTES (current_function_decl).
861 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
862 functions with non-NULL DECL_ABSTRACT_ORIGIN.
864 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
866 PR tree-optimization/83979
867 * fold-const.c (fold_comparison): Use constant_boolean_node
868 instead of boolean_{true,false}_node.
870 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
872 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
875 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
877 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
878 Simplify the clause that sets the length attribute.
879 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
880 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
881 clause that sets the length attribute.
882 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
884 2018-01-24 Tom de Vries <tom@codesourcery.com>
887 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
888 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
889 Add strict parameter.
890 (prevent_branch_around_nothing): Insert dummy insn between branch to
891 label and label with no ptx insn inbetween.
892 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
894 2018-01-24 Tom de Vries <tom@codesourcery.com>
897 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
898 for neutered threads in warp.
899 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
901 2018-01-24 Richard Biener <rguenther@suse.de>
903 PR tree-optimization/83176
904 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
907 2018-01-24 Richard Biener <rguenther@suse.de>
909 PR tree-optimization/82819
910 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
911 code generating pluses that are no-ops in the target precision.
913 2018-01-24 Richard Biener <rguenther@suse.de>
916 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
918 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
920 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
921 to merge probabilities.
922 * predict.c (probably_never_executed): Also mark as cold functions
923 with global 0 profile and guessed local profile.
924 * profile-count.c (profile_probability::combine_with_count): New
926 * profile-count.h (profile_probability::operator*,
927 profile_probability::operator*=, profile_probability::operator/,
928 profile_probability::operator/=): Reduce precision to adjusted
929 and set value to guessed on contradictory divisions.
930 (profile_probability::combine_with_freq): Remove.
931 (profile_probability::combine_wiht_count): Declare.
932 (profile_count::force_nonzero):: Set to adjusted.
933 (profile_count::probability_in):: Set quality to adjusted.
934 * tree-ssa-tail-merge.c (replace_block_by): Use
937 2018-01-23 Andrew Waterman <andrew@sifive.com>
938 Jim Wilson <jimw@sifive.com>
940 * config/riscv/riscv.c (riscv_stack_boundary): New.
941 (riscv_option_override): Set riscv_stack_boundary. Handle
942 riscv_preferred_stack_boundary_arg.
943 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
944 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
945 (STACK_BOUNDARY): Set to riscv_stack_boundary.
946 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
947 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
948 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
950 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
953 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
954 of struct ix86_frame.
955 (ix86_expand_epilogue): Likewise. Add a local variable for
956 the reg_save_offset field in struct ix86_frame.
958 2018-01-23 Bin Cheng <bin.cheng@arm.com>
960 PR tree-optimization/82604
961 * tree-loop-distribution.c (enum partition_kind): New enum item
962 PKIND_PARTIAL_MEMSET.
963 (partition_builtin_p): Support above new enum item.
964 (generate_code_for_partition): Ditto.
965 (compute_access_range): Differentiate cases that equality can be
966 proven at all loops, the innermost loops or no loops.
967 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
968 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
969 (finalize_partitions, distribute_loop): Don't fuse partition of
970 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
971 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
974 2018-01-23 Martin Liska <mliska@suse.cz>
976 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
977 order to ignore the predictor.
978 (PRED_POLYMORPHIC_CALL): Likewise.
979 (PRED_RECURSIVE_CALL): Likewise.
981 2018-01-23 Martin Liska <mliska@suse.cz>
983 * tree-profile.c (tree_profiling): Print function header to
984 aware reader which function we are working on.
985 * value-prof.c (gimple_find_values_to_profile): Do not print
986 not interesting value histograms.
988 2018-01-23 Martin Liska <mliska@suse.cz>
990 * profile-count.h (enum profile_quality): Add
991 profile_uninitialized as the first value. Do not number values
992 as they are zero based.
993 (profile_count::verify): Update sanity check.
994 (profile_probability::verify): Likewise.
996 2018-01-23 Nathan Sidwell <nathan@acm.org>
998 * doc/invoke.texi (ffor-scope): Deprecate.
1000 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1002 PR tree-optimization/83510
1003 * domwalk.c (set_all_edges_as_executable): New function.
1004 (dom_walker::dom_walker): Convert bool param
1005 "skip_unreachable_blocks" to enum reachability. Move setup of
1006 edge flags to set_all_edges_as_executable and only do it when
1007 reachability is REACHABLE_BLOCKS.
1008 * domwalk.h (enum dom_walker::reachability): New enum.
1009 (dom_walker::dom_walker): Convert bool param
1010 "skip_unreachable_blocks" to enum reachability.
1011 (set_all_edges_as_executable): New decl.
1012 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1013 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1015 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1016 but converting true to REACHABLE_BLOCKS.
1017 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1019 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1020 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1021 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1023 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1024 if check_all_array_refs will be called.
1026 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1028 * tree.c (selftest::test_location_wrappers): Add more test
1031 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1033 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1034 (selftest::test_bit_in_range): Likewise.
1036 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1039 * doc/sourcebuild.texi (vect_float): Say that the selector
1040 only describes the situation when -funsafe-math-optimizations is on.
1041 (vect_float_strict): Document.
1043 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1045 PR tree-optimization/83965
1046 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1047 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1048 instead of checking only for a reduction.
1049 (vect_recog_widen_sum_pattern): Likewise.
1051 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1053 * predict.c (probably_never_executed): Only use precise profile info.
1054 (compute_function_frequency): Skip after inlining hack since we now
1055 have quality checking.
1057 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1059 * profile-count.h (profile_probability::very_unlikely,
1060 profile_probability::unlikely, profile_probability::even): Set
1061 precision to guessed.
1063 2018-01-23 Richard Biener <rguenther@suse.de>
1065 PR tree-optimization/83963
1066 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1067 Properly terminate dominator walk when crossing the exit edge not
1068 when visiting its source block.
1070 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1073 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1074 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1076 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1078 PR tree-optimization/83957
1079 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1080 semicolon after for body surrounded by braces.
1082 PR tree-optimization/83081
1083 * profile-count.h (profile_probability::split): New method.
1084 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1085 Use profile_probability::split.
1086 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1087 when splitting a single conditional jump into 2.
1089 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1091 PR tree-optimization/69452
1092 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1095 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1097 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1098 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1099 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1101 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1103 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1104 * config/rl78/rl78.md: New define_expand "movdi"
1105 * config/rl78/rl78.c: New function definition rl78_split_movdi
1107 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1110 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1112 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1113 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1114 128-bit to produce an UNSPEC move to get the double word with the
1115 signbit and then a shift directly to do signbit.
1116 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1117 implementation with a new version that just does either a direct
1118 move or a regular move. Move memory interface to separate insns.
1119 Move insns so they are next to the expander.
1120 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1121 with signbit move. Split big and little endian case.
1122 (signbit<mode>2_dm_mem_le): Likewise.
1123 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1124 (signbit<mode>2_dm2): Likewise.
1126 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1128 * config/rl78/rl78.md: New define_expand "anddi3".
1130 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1132 * config/rl78/rl78.md: New define_expand "umindi3".
1134 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1136 * config/rl78/rl78.md: New define_expand "smindi3".
1138 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1140 * config/rl78/rl78.md: New define_expand "smaxdi3".
1142 2018-01-22 Carl Love <cel@us.ibm.com>
1144 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1145 LVX_V1TI): Add macro expansion.
1146 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1147 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1148 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1149 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1150 Change check to determine if the instruction is a byte reversing
1151 entry. Fix typo in comment.
1152 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1153 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1154 Add def_builtin calls for new builtins.
1155 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1156 Add define_insn expansion.
1158 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1160 * config/rl78/rl78.md: New define_expand "umaxdi3".
1162 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1164 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1165 for non-QImode registers
1167 2018-01-22 Richard Biener <rguenther@suse.de>
1169 PR tree-optimization/83963
1170 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1171 including the loop exit block.
1172 (scop_detection::merge_sese): Likewise.
1173 (scop_detection::add_scop): Do it here instead.
1175 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1177 * doc/sourcebuild.texi (arm_softfloat): Document.
1179 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1182 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1183 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1184 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1186 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1187 David Edelsohn <dje.gcc@gmail.com>
1190 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1191 Change "crset eq" to "crset 2".
1192 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1193 (*call_indirect_aix<mode>_nospec): Likewise.
1194 (*call_value_indirect_aix<mode>_nospec): Likewise.
1195 (*call_indirect_elfv2<mode>_nospec): Likewise.
1196 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1197 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1198 change assembly output from . to $.
1199 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1200 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1201 (*tablejump<mode>_internal1_nospec): Likewise.
1203 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1206 * config/sh/sh_optimize_sett_clrt.cc:
1207 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1209 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1211 PR tree-optimization/83940
1212 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1213 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1214 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1215 use it to pass back the definition type.
1216 (vect_check_store_rhs): Likewise rhs_dt_out.
1217 (vect_build_gather_load_calls): Add a mask_dt argument and use
1218 it instead of a call to vect_is_simple_use.
1219 (vectorizable_store): Update calls to vect_check_load_store_mask
1220 and vect_check_store_rhs. Use the dt returned by the latter instead
1221 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1222 instead of calls to vect_is_simple_use. Pass the scalar rather
1223 than the vector operand to vect_is_simple_use when handling
1224 second and subsequent copies of an rhs value.
1225 (vectorizable_load): Update calls to vect_check_load_store_mask
1226 and vect_build_gather_load_calls. Use the cached mask_dt and
1227 gs_info.offset_dt instead of calls to vect_is_simple_use.
1229 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1232 * tree-emutls.c: Include gimplify.h.
1233 (lower_emutls_2): New function.
1234 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1235 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1236 it before further processing.
1239 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1240 UINTVAL (trueop1) instead of INTVAL (op1).
1242 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1246 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1247 INCOMING_FRAME_SP_OFFSET if not defined.
1248 (scan_trace): Add ENTRY argument. If true and
1249 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1250 emit a note to adjust the CFA offset.
1251 (create_cfi_notes): Adjust scan_trace callers.
1252 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1253 INCOMING_FRAME_SP_OFFSET in the CIE.
1254 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1255 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1257 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1258 * doc/tm.texi: Regenerated.
1260 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1262 PR rtl-optimization/83147
1263 * lra-constraints.c (remove_inheritance_pseudos): Use
1264 lra_substitute_pseudo_within_insn.
1266 2018-01-19 Tom de Vries <tom@codesourcery.com>
1267 Cesar Philippidis <cesar@codesourcery.com>
1270 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1272 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1275 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1276 spaces for function labels.
1278 2018-01-19 Martin Liska <mliska@suse.cz>
1280 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1281 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1282 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1283 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1284 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1285 (PRED_CONST_RETURN): Change from 69 to 65.
1286 (PRED_NULL_RETURN): Change from 91 to 71.
1287 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1288 (PRED_LOOP_GUARD): Change from 66 to 73.
1290 2018-01-19 Martin Liska <mliska@suse.cz>
1292 * predict.c (predict_insn_def): Add new assert.
1293 (struct branch_predictor): Change type to signed integer.
1294 (test_prediction_value_range): Amend test to cover
1296 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1297 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1298 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1299 (PRED_LOOP_IV_COMPARE): Likewise.
1300 * predict.h (PROB_UNINITIALIZED): Define new constant.
1302 2018-01-19 Martin Liska <mliska@suse.cz>
1304 * predict.c (dump_prediction): Add new format for
1305 analyze_brprob.py script which is enabled with -details
1307 * profile-count.h (precise_p): New function.
1309 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1311 PR tree-optimization/83922
1312 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1313 there are no statements that need masking.
1314 (vect_active_double_reduction_p): New function.
1315 (vect_analyze_loop_operations): Use it when handling phis that
1316 are not in the loop header.
1318 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1320 PR tree-optimization/83914
1321 * tree-vect-loop.c (vectorizable_induction): Don't convert
1322 init_expr or apply the peeling adjustment for inductions
1323 that are nested within the vectorized loop.
1325 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1327 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1330 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1334 * function.h (gimplify_parameters): Add gimple_seq * argument.
1335 * function.c: Include gimple.h and options.h.
1336 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1337 for the added local temporaries if needed.
1338 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1339 if there are any parameter cleanups, wrap whole body into a
1340 try/finally with the cleanups.
1342 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1345 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1346 Use GET_MODE_CLASS for scalar floating point.
1348 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1352 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1353 Fix call of call_cgraph_insertion_hooks.
1355 2018-01-18 Martin Sebor <msebor@redhat.com>
1357 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1359 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1362 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1365 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1368 * common.opt: (-ffile-prefix-map): New option.
1369 * opts.c (common_handle_option): Defer it.
1370 * opts-global.c (handle_common_deferred_options): Handle it.
1371 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1372 * file-prefix-map.h: New file.
1373 (remap_debug_filename, add_debug_prefix_map): ...here.
1374 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1375 * final.c (debug_prefix_map, add_debug_prefix_map
1376 remap_debug_filename): Move to...
1377 * file-prefix-map.c: New file.
1378 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1379 generalize, get rid of alloca(), use strrchr() instead of strchr().
1380 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1381 Implement in terms of add_prefix_map().
1382 (remap_macro_filename, remap_debug_filename): Implement in term of
1384 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1385 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1386 * dbxout.c: Include file-prefix-map.h.
1387 * varasm.c: Likewise.
1388 * vmsdbgout.c: Likewise.
1389 * xcoffout.c: Likewise.
1390 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1391 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1392 * doc/invoke.texi (-ffile-prefix-map): Document.
1393 (-fdebug-prefix-map): Update description.
1395 2018-01-18 Martin Liska <mliska@suse.cz>
1397 * config/i386/i386.c (indirect_thunk_name): Document that also
1399 (output_indirect_thunk): Document why both instructions
1400 (pause and lfence) are generated.
1402 2018-01-18 Richard Biener <rguenther@suse.de>
1404 PR tree-optimization/83887
1405 * graphite-scop-detection.c
1406 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1407 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1408 (scop_detection::merge_sese): Re-implement with a flood-fill
1409 algorithm that properly finds a SESE region if it exists.
1411 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1414 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1415 pointer_diff optimizations use view_convert instead of convert.
1417 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1419 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1420 Generate different code for -mno-speculate-indirect-jumps.
1421 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1422 (*call_indirect_aix<mode>): Disable for
1423 -mno-speculate-indirect-jumps.
1424 (*call_indirect_aix<mode>_nospec): New define_insn.
1425 (*call_value_indirect_aix<mode>): Disable for
1426 -mno-speculate-indirect-jumps.
1427 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1428 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1429 -mno-speculate-indirect-jumps.
1430 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1432 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1434 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1435 long double type, set the flags for noting the default long double
1436 type, even if we don't pass or return a long double type.
1438 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1441 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1444 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1446 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1447 support for merge[hl].
1448 (fold_mergehl_helper): New helper function.
1449 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1450 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1451 (altivec_vmrglw_direct): Add xxmrglw insn.
1453 2018-01-17 Andrew Waterman <andrew@sifive.com>
1455 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1456 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1458 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1461 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1462 call the lto_location_cache before reading the
1463 DECL_SOURCE_LOCATION of the types.
1465 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1466 Richard Sandiford <richard.sandiford@linaro.org>
1468 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1469 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1470 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1471 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1472 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1474 * config/aarch64/constraints.md (aarch64_movti_operand):
1476 * config/aarch64/predicates.md (Uti): Add new constraint.
1478 2018-01-17 Carl Love <cel@us.ibm.com>
1479 * config/rs6000/vsx.md (define_expand xl_len_r,
1480 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1481 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1483 (define_expand, define_insn): Move the shift left from the
1484 define_insn to the define_expand for lxvl and stxvl instructions.
1485 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1486 and XL_LEN_R definitions to PURE.
1488 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1490 * config/i386/i386.c (indirect_thunk_name): Declare regno
1491 as unsigned int. Compare regno with INVALID_REGNUM.
1492 (output_indirect_thunk): Ditto.
1493 (output_indirect_thunk_function): Ditto.
1494 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1495 in the call to output_indirect_thunk_function.
1497 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1500 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1501 rather than the size of inner_type to determine the stack slot size
1502 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1504 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1507 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1510 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1512 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1513 endian Linux systems to optionally enable multilibs for selecting
1514 the long double type if the user configured an explicit type.
1515 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1516 have no long double multilibs if not defined.
1517 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1518 warn if the user used -mabi={ieee,ibm}longdouble and we built
1519 multilibs for long double.
1520 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1521 appropriate multilib option.
1522 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1524 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1525 for building long double multilibs.
1526 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1528 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1530 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1533 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1535 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1538 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1541 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1544 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1546 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1547 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1549 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1551 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1552 different rtl trees depending on TARGET_64BIT.
1553 (rs6000_gen_lvx): Likewise.
1555 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1557 * config/visium/visium.md (nop): Tweak comment.
1558 (hazard_nop): Likewise.
1560 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1562 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1563 -mspeculate-indirect-jumps.
1564 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1565 for -mno-speculate-indirect-jumps.
1566 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1567 (*call_value_indirect_elfv2<mode>): Disable for
1568 -mno-speculate-indirect-jumps.
1569 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1570 (indirect_jump): Emit different RTL for
1571 -mno-speculate-indirect-jumps.
1572 (*indirect_jump<mode>): Disable for
1573 -mno-speculate-indirect-jumps.
1574 (*indirect_jump<mode>_nospec): New define_insn.
1575 (tablejump): Emit different RTL for
1576 -mno-speculate-indirect-jumps.
1577 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1578 (tablejumpsi_nospec): New define_expand.
1579 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1580 (tablejumpdi_nospec): New define_expand.
1581 (*tablejump<mode>_internal1): Disable for
1582 -mno-speculate-indirect-jumps.
1583 (*tablejump<mode>_internal1_nospec): New define_insn.
1584 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1587 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1589 * caller-save.c (insert_save): Drop unnecessary parameter. All
1592 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1593 Richard Biener <rguenth@suse.de>
1596 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1597 return early, inline manually is_gimple_sizepos. Make sure if we
1598 call gimplify_expr we don't end up with a gimple constant.
1599 * tree.c (variably_modified_type_p): Don't return true for
1600 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1601 * gimplify.h (is_gimple_sizepos): Remove.
1603 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1605 PR tree-optimization/83857
1606 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1607 vectorizable_live_operation for pure SLP statements.
1608 (vectorizable_live_operation): Handle PHIs.
1610 2018-01-16 Richard Biener <rguenther@suse.de>
1612 PR tree-optimization/83867
1613 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1614 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1616 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1619 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1620 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1621 If off is not INTEGER_CST, issue a may not be aligned warning
1622 rather than isn't aligned. Use isn%'t rather than isn't.
1623 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1625 <case MULT_EXPR>: Improve the case when bottom and one of the
1626 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1627 operand, in that case check if the other operand is multiple of
1628 bottom divided by the INTEGER_CST operand.
1630 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1633 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1634 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1635 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1636 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1637 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1638 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1639 (pa_function_arg_size): New function.
1641 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1643 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1644 in a separate statement.
1646 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1648 PR tree-optimization/83847
1649 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1650 group gathers and scatters.
1652 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1654 PR rtl-optimization/86620
1655 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1657 PR rtl-optimization/83213
1658 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1659 to last if both are JUMP_INSNs.
1661 PR tree-optimization/83843
1662 * gimple-ssa-store-merging.c
1663 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1664 store_immediate_info for bswap/nop orig_stores.
1666 2018-01-15 Andrew Waterman <andrew@sifive.com>
1668 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1670 <UDIV>: Increase cost if !TARGET_DIV.
1672 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1674 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1675 (define_attr "cr_logical_3op"): New.
1676 (cceq_ior_compare): Adjust.
1677 (cceq_ior_compare_complement): Adjust.
1678 (*cceq_rev_compare): Adjust.
1679 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1680 (is_cracked_insn): Adjust.
1681 (insn_must_be_first_in_group): Adjust.
1682 * config/rs6000/40x.md: Adjust.
1683 * config/rs6000/440.md: Adjust.
1684 * config/rs6000/476.md: Adjust.
1685 * config/rs6000/601.md: Adjust.
1686 * config/rs6000/603.md: Adjust.
1687 * config/rs6000/6xx.md: Adjust.
1688 * config/rs6000/7450.md: Adjust.
1689 * config/rs6000/7xx.md: Adjust.
1690 * config/rs6000/8540.md: Adjust.
1691 * config/rs6000/cell.md: Adjust.
1692 * config/rs6000/e300c2c3.md: Adjust.
1693 * config/rs6000/e500mc.md: Adjust.
1694 * config/rs6000/e500mc64.md: Adjust.
1695 * config/rs6000/e5500.md: Adjust.
1696 * config/rs6000/e6500.md: Adjust.
1697 * config/rs6000/mpc.md: Adjust.
1698 * config/rs6000/power4.md: Adjust.
1699 * config/rs6000/power5.md: Adjust.
1700 * config/rs6000/power6.md: Adjust.
1701 * config/rs6000/power7.md: Adjust.
1702 * config/rs6000/power8.md: Adjust.
1703 * config/rs6000/power9.md: Adjust.
1704 * config/rs6000/rs64.md: Adjust.
1705 * config/rs6000/titan.md: Adjust.
1707 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1709 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1710 ix86_indirect_branch_register logic.
1712 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1714 * config/i386/constraints.md (Bs): Update
1715 ix86_indirect_branch_register check. Don't check
1716 ix86_indirect_branch_register with GOT_memory_operand.
1718 * config/i386/predicates.md (GOT_memory_operand): Don't check
1719 ix86_indirect_branch_register here.
1720 (GOT32_symbol_operand): Likewise.
1722 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1724 * config/i386/predicates.md (constant_call_address_operand):
1725 Rewrite ix86_indirect_branch_register logic.
1726 (sibcall_insn_operand): Likewise.
1728 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1730 * config/i386/constraints.md (Bs): Replace
1731 ix86_indirect_branch_thunk_register with
1732 ix86_indirect_branch_register.
1734 * config/i386/i386.md (indirect_jump): Likewise.
1735 (tablejump): Likewise.
1736 (*sibcall_memory): Likewise.
1737 (*sibcall_value_memory): Likewise.
1738 Peepholes of indirect call and jump via memory: Likewise.
1739 * config/i386/i386.opt: Likewise.
1740 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1741 (GOT_memory_operand): Likewise.
1742 (call_insn_operand): Likewise.
1743 (sibcall_insn_operand): Likewise.
1744 (GOT32_symbol_operand): Likewise.
1746 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1749 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1750 type rather than type addr's type points to.
1751 (expand_omp_atomic_mutex): Likewise.
1752 (expand_omp_atomic): Likewise.
1754 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1757 * config/i386/i386.c (output_indirect_thunk_function): Use
1758 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1759 for __x86_return_thunk.
1761 2018-01-15 Richard Biener <rguenther@suse.de>
1764 * expmed.c (extract_bit_field_1): Fix typo.
1766 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1769 * config/arm/iterators.md (VF): New mode iterator.
1770 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1771 Remove integer-related logic from pattern.
1772 (neon_vabd<mode>_3): Likewise.
1774 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1777 * common.opt (fstrict-overflow): No longer an alias.
1778 (fwrapv-pointer): New option.
1779 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1780 also for pointer types based on flag_wrapv_pointer.
1781 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1782 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1783 opts->x_flag_wrapv got set.
1784 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1785 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1786 POINTER_TYPE_OVERFLOW_UNDEFINED.
1787 * match.pd: Likewise in address comparison pattern.
1788 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1790 2018-01-15 Richard Biener <rguenther@suse.de>
1793 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1794 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1795 Reset type names to their identifier if their TYPE_DECL doesn't
1796 have linkage (and thus is used for ODR and devirt).
1797 (save_debug_info_for_decl): Remove.
1798 (save_debug_info_for_type): Likewise.
1799 (add_tree_to_fld_list): Adjust.
1800 * tree-pretty-print.c (dump_generic_node): Make dumping of
1801 type names more robust.
1803 2018-01-15 Richard Biener <rguenther@suse.de>
1805 * BASE-VER: Bump to 8.0.1.
1807 2018-01-14 Martin Sebor <msebor@redhat.com>
1810 * builtins.c (check_access): Avoid warning when the no-warning bit
1813 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1815 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1816 * ira-color (allocno_hard_regs_compare): Likewise.
1818 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1821 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1822 Use .pushsection/.popsection.
1824 2018-01-14 Martin Sebor <msebor@redhat.com>
1827 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1829 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1831 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1832 entry from extra_headers.
1833 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1834 extra_headers, make the list bitwise identical to the i?86-*-* one.
1836 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1838 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1839 -mcmodel=large with -mindirect-branch=thunk,
1840 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1841 -mfunction-return=thunk-extern.
1842 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1843 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1844 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1846 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1848 * config/i386/i386.c (print_reg): Print the name of the full
1849 integer register without '%'.
1850 (ix86_print_operand): Handle 'V'.
1851 * doc/extend.texi: Document 'V' modifier.
1853 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1855 * config/i386/constraints.md (Bs): Disallow memory operand for
1856 -mindirect-branch-register.
1858 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1859 (GOT_memory_operand): Likewise.
1860 (call_insn_operand): Likewise.
1861 (sibcall_insn_operand): Likewise.
1862 (GOT32_symbol_operand): Likewise.
1863 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1864 for -mindirect-branch-register.
1865 (tablejump): Likewise.
1866 (*sibcall_memory): Likewise.
1867 (*sibcall_value_memory): Likewise.
1868 Disallow peepholes of indirect call and jump via memory for
1869 -mindirect-branch-register.
1870 (*call_pop): Replace m with Bw.
1871 (*call_value_pop): Likewise.
1872 (*sibcall_pop_memory): Replace m with Bs.
1873 * config/i386/i386.opt (mindirect-branch-register): New option.
1874 * doc/invoke.texi: Document -mindirect-branch-register option.
1876 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1878 * config/i386/i386-protos.h (ix86_output_function_return): New.
1879 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1880 set function_return_type.
1881 (indirect_thunk_name): Add ret_p to indicate thunk for function
1883 (output_indirect_thunk_function): Pass false to
1884 indirect_thunk_name.
1885 (ix86_output_indirect_branch_via_reg): Likewise.
1886 (ix86_output_indirect_branch_via_push): Likewise.
1887 (output_indirect_thunk_function): Create alias for function
1888 return thunk if regno < 0.
1889 (ix86_output_function_return): New function.
1890 (ix86_handle_fndecl_attribute): Handle function_return.
1891 (ix86_attribute_table): Add function_return.
1892 * config/i386/i386.h (machine_function): Add
1893 function_return_type.
1894 * config/i386/i386.md (simple_return_internal): Use
1895 ix86_output_function_return.
1896 (simple_return_internal_long): Likewise.
1897 * config/i386/i386.opt (mfunction-return=): New option.
1898 (indirect_branch): Mention -mfunction-return=.
1899 * doc/extend.texi: Document function_return function attribute.
1900 * doc/invoke.texi: Document -mfunction-return= option.
1902 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1904 * config/i386/i386-opts.h (indirect_branch): New.
1905 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1906 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1907 with local indirect jump when converting indirect call and jump.
1908 (ix86_set_indirect_branch_type): New.
1909 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1910 (indirectlabelno): New.
1911 (indirect_thunk_needed): Likewise.
1912 (indirect_thunk_bnd_needed): Likewise.
1913 (indirect_thunks_used): Likewise.
1914 (indirect_thunks_bnd_used): Likewise.
1915 (INDIRECT_LABEL): Likewise.
1916 (indirect_thunk_name): Likewise.
1917 (output_indirect_thunk): Likewise.
1918 (output_indirect_thunk_function): Likewise.
1919 (ix86_output_indirect_branch_via_reg): Likewise.
1920 (ix86_output_indirect_branch_via_push): Likewise.
1921 (ix86_output_indirect_branch): Likewise.
1922 (ix86_output_indirect_jmp): Likewise.
1923 (ix86_code_end): Call output_indirect_thunk_function if needed.
1924 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1926 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1927 (ix86_attribute_table): Add indirect_branch.
1928 * config/i386/i386.h (machine_function): Add indirect_branch_type
1929 and has_local_indirect_jump.
1930 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1932 (tablejump): Likewise.
1933 (*indirect_jump): Use ix86_output_indirect_jmp.
1934 (*tablejump_1): Likewise.
1935 (simple_return_indirect_internal): Likewise.
1936 * config/i386/i386.opt (mindirect-branch=): New option.
1937 (indirect_branch): New.
1940 (thunk-inline): Likewise.
1941 (thunk-extern): Likewise.
1942 * doc/extend.texi: Document indirect_branch function attribute.
1943 * doc/invoke.texi: Document -mindirect-branch= option.
1945 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1948 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1950 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1952 * ipa-inline.c (want_inline_small_function_p): Return false if
1953 inlining has already failed with CIF_FINAL_ERROR.
1954 (update_caller_keys): Call want_inline_small_function_p before
1956 (update_callee_keys): Likewise.
1958 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1960 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1962 (rs6000_quadword_masked_address_p): Likewise.
1963 (quad_aligned_load_p): Likewise.
1964 (quad_aligned_store_p): Likewise.
1965 (const_load_sequence_p): Add comment to describe the outer-most loop.
1966 (mimic_memory_attributes_and_flags): New function.
1967 (rs6000_gen_stvx): Likewise.
1968 (replace_swapped_aligned_store): Likewise.
1969 (rs6000_gen_lvx): Likewise.
1970 (replace_swapped_aligned_load): Likewise.
1971 (replace_swapped_load_constant): Capitalize argument name in
1972 comment describing this function.
1973 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1974 and stores that access quad-word aligned addresses and replace
1975 with stvx or lvx instructions when appropriate.
1976 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1977 New function prototype.
1978 (rs6000_quadword_masked_address_p): Likewise.
1979 (rs6000_gen_lvx): Likewise.
1980 (rs6000_gen_stvx): Likewise.
1981 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1982 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1983 when memory address is aligned.
1984 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1985 this split to select lvx instruction when memory address is aligned.
1986 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1987 instruction when memory address is aligned.
1988 (*vsx_le_perm_load_v16qi): Likewise.
1989 (four unnamed splitters): Modify to select the stvx instruction
1990 when memory is aligned.
1992 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1994 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1995 which appears in the queue multiple times.
1997 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1998 Alan Hayward <alan.hayward@arm.com>
1999 David Sherwood <david.sherwood@arm.com>
2001 * tree-vectorizer.h (vec_lower_bound): New structure.
2002 (_loop_vec_info): Add check_nonzero and lower_bounds.
2003 (LOOP_VINFO_CHECK_NONZERO): New macro.
2004 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2005 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2006 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2007 fields. Make seg_len the distance travelled, not including the
2009 (dr_direction_indicator): Declare.
2010 (dr_zero_step_indicator): Likewise.
2011 (dr_known_forward_stride_p): Likewise.
2012 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2014 (runtime_alias_check_p): Allow runtime alias checks with
2016 (operator ==): Compare access_size and align.
2017 (prune_runtime_alias_test_list): Rework for new distinction between
2018 the access_size and seg_len.
2019 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2021 (get_segment_min_max): New function.
2022 (create_intersect_range_checks): Use it.
2023 (dr_step_indicator): New function.
2024 (dr_direction_indicator): Likewise.
2025 (dr_zero_step_indicator): Likewise.
2026 (dr_known_forward_stride_p): Likewise.
2027 * tree-loop-distribution.c (data_ref_segment_size): Return
2028 DR_STEP * (niters - 1).
2029 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2031 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2032 (vect_preserves_scalar_order_p): New function, split out from...
2033 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2034 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2035 (vect_vfa_access_size): New function.
2036 (vect_vfa_align): Likewise.
2037 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2038 (dump_lower_bound): New function.
2039 (vect_check_lower_bound): Likewise.
2040 (vect_small_gap_p): Likewise.
2041 (vectorizable_with_step_bound_p): Likewise.
2042 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2043 depencies if the vectorization factor is 1. Convert the checks
2044 for nonzero steps into checks on the bounds of DR_STEP. Try using
2045 a bunds check for variable steps if the minimum required step is
2046 relatively small. Update calls to the dr_with_seg_len
2047 constructor and to vect_compile_time_alias.
2048 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2050 (vect_loop_versioning): Call it.
2051 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2053 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2055 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2056 Alan Hayward <alan.hayward@arm.com>
2057 David Sherwood <david.sherwood@arm.com>
2059 * doc/sourcebuild.texi (vect_scatter_store): Document.
2060 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2062 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2064 * genopinit.c (main): Add supports_vec_scatter_store and
2065 supports_vec_scatter_store_cached to target_optabs.
2066 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2067 IFN_MASK_SCATTER_STORE.
2068 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2070 * internal-fn.h (internal_store_fn_p): Declare.
2071 (internal_fn_stored_value_index): Likewise.
2072 * internal-fn.c (scatter_store_direct): New macro.
2073 (expand_scatter_store_optab_fn): New function.
2074 (direct_scatter_store_optab_supported_p): New macro.
2075 (internal_store_fn_p): New function.
2076 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2077 IFN_MASK_SCATTER_STORE.
2078 (internal_fn_mask_index): Likewise.
2079 (internal_fn_stored_value_index): New function.
2080 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2082 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2083 * optabs-query.c (supports_vec_scatter_store_p): New function.
2084 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2085 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2086 true for scatter stores.
2087 (vect_gather_scatter_fn_p): Handle scatter stores too.
2088 (vect_check_gather_scatter): Consider using scatter stores if
2089 supports_vec_scatter_store_p.
2090 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2092 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2093 internal_fn_stored_value_index.
2094 (check_load_store_masking): Handle scatter stores too.
2095 (vect_get_store_rhs): Make public.
2096 (vectorizable_call): Use internal_store_fn_p.
2097 (vectorizable_store): Handle scatter store internal functions.
2098 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2099 when deciding whether the end of the group has been reached.
2100 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2101 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2102 (mask_scatter_store<mode>): New insns.
2104 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2105 Alan Hayward <alan.hayward@arm.com>
2106 David Sherwood <david.sherwood@arm.com>
2108 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2109 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2110 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2112 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2113 Use vect_truncate_gather_scatter_offset if we can't treat the
2114 operation as a normal gather load or scatter store.
2115 (get_group_load_store_type): Take the gather_scatter_info
2116 as argument. Try using a gather load or scatter store for
2117 single-element groups.
2118 (get_load_store_type): Update calls to get_group_load_store_type
2119 and vect_use_strided_gather_scatters_p.
2121 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2122 Alan Hayward <alan.hayward@arm.com>
2123 David Sherwood <david.sherwood@arm.com>
2125 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2126 optional tree argument.
2127 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2129 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2130 but continue to use the current value as a fallback.
2131 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2132 to compare the updates.
2133 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2134 (get_load_store_type): Use it when handling a strided access.
2135 (vect_get_strided_load_store_ops): New function.
2136 (vect_get_data_ptr_increment): Likewise.
2137 (vectorizable_load): Handle strided gather loads. Always pass
2138 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2140 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2141 Alan Hayward <alan.hayward@arm.com>
2142 David Sherwood <david.sherwood@arm.com>
2144 * doc/md.texi (gather_load@var{m}): Document.
2145 (mask_gather_load@var{m}): Likewise.
2146 * genopinit.c (main): Add supports_vec_gather_load and
2147 supports_vec_gather_load_cached to target_optabs.
2148 * optabs-tree.c (init_tree_optimization_optabs): Use
2149 ggc_cleared_alloc to allocate target_optabs.
2150 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2151 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2153 * internal-fn.h (internal_load_fn_p): Declare.
2154 (internal_gather_scatter_fn_p): Likewise.
2155 (internal_fn_mask_index): Likewise.
2156 (internal_gather_scatter_fn_supported_p): Likewise.
2157 * internal-fn.c (gather_load_direct): New macro.
2158 (expand_gather_load_optab_fn): New function.
2159 (direct_gather_load_optab_supported_p): New macro.
2160 (direct_internal_fn_optab): New function.
2161 (internal_load_fn_p): Likewise.
2162 (internal_gather_scatter_fn_p): Likewise.
2163 (internal_fn_mask_index): Likewise.
2164 (internal_gather_scatter_fn_supported_p): Likewise.
2165 * optabs-query.c (supports_at_least_one_mode_p): New function.
2166 (supports_vec_gather_load_p): Likewise.
2167 * optabs-query.h (supports_vec_gather_load_p): Declare.
2168 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2169 and memory_type field.
2170 (NUM_PATTERNS): Bump to 15.
2171 * tree-vect-data-refs.c: Include internal-fn.h.
2172 (vect_gather_scatter_fn_p): New function.
2173 (vect_describe_gather_scatter_call): Likewise.
2174 (vect_check_gather_scatter): Try using internal functions for
2175 gather loads. Recognize existing calls to a gather load function.
2176 (vect_analyze_data_refs): Consider using gather loads if
2177 supports_vec_gather_load_p.
2178 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2179 (vect_get_gather_scatter_offset_type): Likewise.
2180 (vect_convert_mask_for_vectype): Likewise.
2181 (vect_add_conversion_to_patterm): Likewise.
2182 (vect_try_gather_scatter_pattern): Likewise.
2183 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2184 (vect_vect_recog_func_ptrs): Add it.
2185 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2186 internal_fn_mask_index and internal_gather_scatter_fn_p.
2187 (check_load_store_masking): Take the gather_scatter_info as an
2188 argument and handle gather loads.
2189 (vect_get_gather_scatter_ops): New function.
2190 (vectorizable_call): Check internal_load_fn_p.
2191 (vectorizable_load): Likewise. Handle gather load internal
2193 (vectorizable_store): Update call to check_load_store_masking.
2194 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2195 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2196 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2197 (aarch64_gather_scale_operand_d): New predicates.
2198 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2199 (mask_gather_load<mode>): New insns.
2201 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2202 Alan Hayward <alan.hayward@arm.com>
2203 David Sherwood <david.sherwood@arm.com>
2205 * optabs.def (fold_left_plus_optab): New optab.
2206 * doc/md.texi (fold_left_plus_@var{m}): Document.
2207 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2208 * internal-fn.c (fold_left_direct): Define.
2209 (expand_fold_left_optab_fn): Likewise.
2210 (direct_fold_left_optab_supported_p): Likewise.
2211 * fold-const-call.c (fold_const_fold_left): New function.
2212 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2213 * tree-parloops.c (valid_reduction_p): New function.
2214 (gather_scalar_reductions): Use it.
2215 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2216 (vect_finish_replace_stmt): Declare.
2217 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2218 (needs_fold_left_reduction_p): New function, split out from...
2219 (vect_is_simple_reduction): ...here. Accept reductions that
2220 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2221 (vect_force_simple_reduction): Also store the reduction type in
2222 the assignment's STMT_VINFO_REDUC_TYPE.
2223 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2224 (merge_with_identity): New function.
2225 (vect_expand_fold_left): Likewise.
2226 (vectorize_fold_left_reduction): Likewise.
2227 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2228 scalar phi in place for it. Check for target support and reject
2229 cases that would reassociate the operation. Defer the transform
2230 phase to vectorize_fold_left_reduction.
2231 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2232 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2233 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2235 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2237 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2238 call to ifc_temp_var.
2240 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2241 Alan Hayward <alan.hayward@arm.com>
2242 David Sherwood <david.sherwood@arm.com>
2244 * target.def (legitimize_address_displacement): Take the original
2245 offset as a poly_int.
2246 * targhooks.h (default_legitimize_address_displacement): Update
2248 * targhooks.c (default_legitimize_address_displacement): Likewise.
2249 * doc/tm.texi: Regenerate.
2250 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2251 as an argument, moving assert of ad->disp == ad->disp_term to...
2252 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2253 Try calling targetm.legitimize_address_displacement before expanding
2254 the address rather than afterwards, and adjust for the new interface.
2255 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2256 Match the new hook interface. Handle SVE addresses.
2257 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2260 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2262 * Makefile.in (OBJS): Add early-remat.o.
2263 * target.def (select_early_remat_modes): New hook.
2264 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2265 * doc/tm.texi: Regenerate.
2266 * targhooks.h (default_select_early_remat_modes): Declare.
2267 * targhooks.c (default_select_early_remat_modes): New function.
2268 * timevar.def (TV_EARLY_REMAT): New timevar.
2269 * passes.def (pass_early_remat): New pass.
2270 * tree-pass.h (make_pass_early_remat): Declare.
2271 * early-remat.c: New file.
2272 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2274 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2276 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2277 Alan Hayward <alan.hayward@arm.com>
2278 David Sherwood <david.sherwood@arm.com>
2280 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2281 vfm1 with a bound_epilog parameter.
2282 (vect_do_peeling): Update calls accordingly, and move the prologue
2283 call earlier in the function. Treat the base bound_epilog as 0 for
2284 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2285 this base when peeling for gaps.
2286 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2287 with fully-masked loops.
2288 (vect_estimate_min_profitable_iters): Handle the single peeled
2289 iteration in that case.
2291 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2292 Alan Hayward <alan.hayward@arm.com>
2293 David Sherwood <david.sherwood@arm.com>
2295 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2296 single-element interleaving even if the size is not a power of 2.
2297 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2298 accesses for single-element interleaving if the group size is
2301 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2302 Alan Hayward <alan.hayward@arm.com>
2303 David Sherwood <david.sherwood@arm.com>
2305 * doc/md.texi (fold_extract_last_@var{m}): Document.
2306 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2307 * optabs.def (fold_extract_last_optab): New optab.
2308 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2309 * internal-fn.c (fold_extract_direct): New macro.
2310 (expand_fold_extract_optab_fn): Likewise.
2311 (direct_fold_extract_optab_supported_p): Likewise.
2312 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2313 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2314 EXTRACT_LAST_REDUCTION.
2315 (get_initial_def_for_reduction): Do not create an initial vector
2316 for EXTRACT_LAST_REDUCTION reductions.
2317 (vectorizable_reduction): Leave the scalar phi in place for
2318 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2319 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2320 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2321 transform phase to vectorizable_condition.
2322 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2324 (vect_finish_stmt_generation): ...here.
2325 (vect_finish_replace_stmt): New function.
2326 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2327 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2329 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2331 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2332 Alan Hayward <alan.hayward@arm.com>
2333 David Sherwood <david.sherwood@arm.com>
2335 * doc/md.texi (extract_last_@var{m}): Document.
2336 * optabs.def (extract_last_optab): New optab.
2337 * internal-fn.def (EXTRACT_LAST): New internal function.
2338 * internal-fn.c (cond_unary_direct): New macro.
2339 (expand_cond_unary_optab_fn): Likewise.
2340 (direct_cond_unary_optab_supported_p): Likewise.
2341 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2342 loops using EXTRACT_LAST.
2343 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2344 (extract_last_<mode>): ...this optab.
2345 (vec_extract<mode><Vel>): Update accordingly.
2347 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2348 Alan Hayward <alan.hayward@arm.com>
2349 David Sherwood <david.sherwood@arm.com>
2351 * target.def (empty_mask_is_expensive): New hook.
2352 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2353 * doc/tm.texi: Regenerate.
2354 * targhooks.h (default_empty_mask_is_expensive): Declare.
2355 * targhooks.c (default_empty_mask_is_expensive): New function.
2356 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2357 if the target says that empty masks are expensive.
2358 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2360 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2362 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2363 Alan Hayward <alan.hayward@arm.com>
2364 David Sherwood <david.sherwood@arm.com>
2366 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2367 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2368 (vect_use_loop_mask_for_alignment_p): New function.
2369 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2370 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2371 niters_skip argument. Make sure that the first niters_skip elements
2372 of the first iteration are inactive.
2373 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2374 Update call to vect_set_loop_masks_directly.
2375 (get_misalign_in_elems): New function, split out from...
2376 (vect_gen_prolog_loop_niters): ...here.
2377 (vect_update_init_of_dr): Take a code argument that specifies whether
2378 the adjustment should be added or subtracted.
2379 (vect_update_init_of_drs): Likewise.
2380 (vect_prepare_for_masked_peels): New function.
2381 (vect_do_peeling): Skip prologue peeling if we're using a mask
2382 instead. Update call to vect_update_inits_of_drs.
2383 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2385 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2386 alignment. Do not include the number of peeled iterations in
2387 the minimum threshold in that case.
2388 (vectorizable_induction): Adjust the start value down by
2389 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2390 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2391 Take the number of skipped iterations into account when calculating
2393 * tree-vect-stmts.c (vect_gen_while_not): New function.
2395 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2396 Alan Hayward <alan.hayward@arm.com>
2397 David Sherwood <david.sherwood@arm.com>
2399 * doc/sourcebuild.texi (vect_fully_masked): Document.
2400 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2402 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2404 (vect_analyze_loop_2): ...here. Don't check the vectorization
2405 factor against the number of loop iterations if the loop is
2408 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2409 Alan Hayward <alan.hayward@arm.com>
2410 David Sherwood <david.sherwood@arm.com>
2412 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2413 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2414 (dump_groups): Update accordingly.
2415 (iv_use::mem_type): New member variable.
2416 (address_p): New function.
2417 (record_use): Add a mem_type argument and initialize the new
2419 (record_group_use): Add a mem_type argument. Use address_p.
2420 Remove obsolete null checks of base_object. Update call to record_use.
2421 (find_interesting_uses_op): Update call to record_group_use.
2422 (find_interesting_uses_cond): Likewise.
2423 (find_interesting_uses_address): Likewise.
2424 (get_mem_type_for_internal_fn): New function.
2425 (find_address_like_use): Likewise.
2426 (find_interesting_uses_stmt): Try find_address_like_use before
2427 calling find_interesting_uses_op.
2428 (addr_offset_valid_p): Use the iv mem_type field as the type
2429 of the addressed memory.
2430 (add_autoinc_candidates): Likewise.
2431 (get_address_cost): Likewise.
2432 (split_small_address_groups_p): Use address_p.
2433 (split_address_groups): Likewise.
2434 (add_iv_candidate_for_use): Likewise.
2435 (autoinc_possible_for_pair): Likewise.
2436 (rewrite_groups): Likewise.
2437 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2438 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2439 (get_alias_ptr_type_for_ptr_address): New function.
2440 (rewrite_use_address): Rewrite address uses in calls that were
2441 identified by find_address_like_use.
2443 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2444 Alan Hayward <alan.hayward@arm.com>
2445 David Sherwood <david.sherwood@arm.com>
2447 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2449 * gimple-expr.h (is_gimple_addressable: Likewise.
2450 * gimple-expr.c (is_gimple_address): Likewise.
2451 * internal-fn.c (expand_call_mem_ref): New function.
2452 (expand_mask_load_optab_fn): Use it.
2453 (expand_mask_store_optab_fn): Likewise.
2455 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2456 Alan Hayward <alan.hayward@arm.com>
2457 David Sherwood <david.sherwood@arm.com>
2459 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2460 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2461 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2462 (cond_umax@var{mode}): Document.
2463 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2464 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2465 (cond_umin_optab, cond_umax_optab): New optabs.
2466 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2467 (COND_IOR, COND_XOR): New internal functions.
2468 * internal-fn.h (get_conditional_internal_fn): Declare.
2469 * internal-fn.c (cond_binary_direct): New macro.
2470 (expand_cond_binary_optab_fn): Likewise.
2471 (direct_cond_binary_optab_supported_p): Likewise.
2472 (get_conditional_internal_fn): New function.
2473 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2474 Cope with reduction statements that are vectorized as calls rather
2476 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2477 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2478 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2479 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2480 (UNSPEC_COND_EOR): New unspecs.
2481 (optab): Add mappings for them.
2482 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2483 (sve_int_op, sve_fp_op): New int attributes.
2485 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2486 Alan Hayward <alan.hayward@arm.com>
2487 David Sherwood <david.sherwood@arm.com>
2489 * optabs.def (while_ult_optab): New optab.
2490 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2491 * internal-fn.def (WHILE_ULT): New internal function.
2492 * internal-fn.h (direct_internal_fn_supported_p): New override
2493 that takes two types as argument.
2494 * internal-fn.c (while_direct): New macro.
2495 (expand_while_optab_fn): New function.
2496 (convert_optab_supported_p): Likewise.
2497 (direct_while_optab_supported_p): New macro.
2498 * wide-int.h (wi::udiv_ceil): New function.
2499 * tree-vectorizer.h (rgroup_masks): New structure.
2500 (vec_loop_masks): New typedef.
2501 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2503 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2504 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2505 (vect_max_vf): New function.
2506 (slpeel_make_loop_iterate_ntimes): Delete.
2507 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2508 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2509 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2510 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2511 internal-fn.h, stor-layout.h and optabs-query.h.
2512 (vect_set_loop_mask): New function.
2513 (add_preheader_seq): Likewise.
2514 (add_header_seq): Likewise.
2515 (interleave_supported_p): Likewise.
2516 (vect_maybe_permute_loop_masks): Likewise.
2517 (vect_set_loop_masks_directly): Likewise.
2518 (vect_set_loop_condition_masked): Likewise.
2519 (vect_set_loop_condition_unmasked): New function, split out from
2520 slpeel_make_loop_iterate_ntimes.
2521 (slpeel_make_loop_iterate_ntimes): Rename to..
2522 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2523 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2524 (vect_do_peeling): Update call accordingly.
2525 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2527 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2528 mask_compare_type, can_fully_mask_p and fully_masked_p.
2529 (release_vec_loop_masks): New function.
2530 (_loop_vec_info): Use it to free the loop masks.
2531 (can_produce_all_loop_masks_p): New function.
2532 (vect_get_max_nscalars_per_iter): Likewise.
2533 (vect_verify_full_masking): Likewise.
2534 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2535 retries, and free the mask rgroups before retrying. Check loop-wide
2536 reasons for disallowing fully-masked loops. Make the final decision
2537 about whether use a fully-masked loop or not.
2538 (vect_estimate_min_profitable_iters): Do not assume that peeling
2539 for the number of iterations will be needed for fully-masked loops.
2540 (vectorizable_reduction): Disable fully-masked loops.
2541 (vectorizable_live_operation): Likewise.
2542 (vect_halve_mask_nunits): New function.
2543 (vect_double_mask_nunits): Likewise.
2544 (vect_record_loop_mask): Likewise.
2545 (vect_get_loop_mask): Likewise.
2546 (vect_transform_loop): Handle the case in which the final loop
2547 iteration might handle a partial vector. Call vect_set_loop_condition
2548 instead of slpeel_make_loop_iterate_ntimes.
2549 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2550 (check_load_store_masking): New function.
2551 (prepare_load_store_mask): Likewise.
2552 (vectorizable_store): Handle fully-masked loops.
2553 (vectorizable_load): Likewise.
2554 (supportable_widening_operation): Use vect_halve_mask_nunits for
2556 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2557 (vect_gen_while): New function.
2558 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2559 (aarch64_uqdec<mode>): New insn.
2561 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2562 Alan Hayward <alan.hayward@arm.com>
2563 David Sherwood <david.sherwood@arm.com>
2565 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2566 (reduc_xor_scal_optab): New optabs.
2567 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2568 (reduc_xor_scal_@var{m}): Document.
2569 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2570 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2572 * fold-const-call.c (fold_const_call): Handle them.
2573 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2574 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2575 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2576 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2577 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2578 (UNSPEC_XORV): New unspecs.
2579 (optab): Add entries for them.
2580 (BITWISEV): New int iterator.
2581 (bit_reduc_op): New int attributes.
2583 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2584 Alan Hayward <alan.hayward@arm.com>
2585 David Sherwood <david.sherwood@arm.com>
2587 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2588 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2589 * optabs.def (vec_shl_insert_optab): New optab.
2590 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2591 (duplicate_and_interleave): Likewise.
2592 * tree-vect-loop.c: Include internal-fn.h.
2593 (neutral_op_for_slp_reduction): New function, split out from
2594 get_initial_defs_for_reduction.
2595 (get_initial_def_for_reduction): Handle option 2 for variable-length
2596 vectors by loading the neutral value into a vector and then shifting
2597 the initial value into element 0.
2598 (get_initial_defs_for_reduction): Replace the code argument with
2599 the neutral value calculated by neutral_op_for_slp_reduction.
2600 Use gimple_build_vector for constant-length vectors.
2601 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2602 but the first group_size elements have a neutral value.
2603 Use duplicate_and_interleave otherwise.
2604 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2605 Update call to get_initial_defs_for_reduction. Handle SLP
2606 reductions for variable-length vectors by creating one vector
2607 result for each scalar result, with the elements associated
2608 with other scalar results stubbed out with the neutral value.
2609 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2610 Require IFN_VEC_SHL_INSERT for double reductions on
2611 variable-length vectors, or SLP reductions that have
2612 a neutral value. Require can_duplicate_and_interleave_p
2613 support for variable-length unchained SLP reductions if there
2614 is no neutral value, such as for MIN/MAX reductions. Also require
2615 the number of vector elements to be a multiple of the number of
2616 SLP statements when doing variable-length unchained SLP reductions.
2617 Update call to vect_create_epilog_for_reduction.
2618 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2619 and remove initial values.
2620 (duplicate_and_interleave): Make public.
2621 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2622 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2624 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2625 Alan Hayward <alan.hayward@arm.com>
2626 David Sherwood <david.sherwood@arm.com>
2628 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2629 (can_duplicate_and_interleave_p): New function.
2630 (vect_get_and_check_slp_defs): Take the vector of statements
2631 rather than just the current one. Remove excess parentheses.
2632 Restriction rejectinon of vect_constant_def and vect_external_def
2633 for variable-length vectors to boolean types, or types for which
2634 can_duplicate_and_interleave_p is false.
2635 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2636 (duplicate_and_interleave): New function.
2637 (vect_get_constant_vectors): Use gimple_build_vector for
2638 constant-length vectors and suitable variable-length constant
2639 vectors. Use duplicate_and_interleave for other variable-length
2640 vectors. Don't defer the update when inserting new statements.
2642 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2643 Alan Hayward <alan.hayward@arm.com>
2644 David Sherwood <david.sherwood@arm.com>
2646 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2647 min_profitable_iters doesn't go negative.
2649 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2650 Alan Hayward <alan.hayward@arm.com>
2651 David Sherwood <david.sherwood@arm.com>
2653 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2654 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2655 * optabs.def (vec_mask_load_lanes_optab): New optab.
2656 (vec_mask_store_lanes_optab): Likewise.
2657 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2658 (MASK_STORE_LANES): Likewise.
2659 * internal-fn.c (mask_load_lanes_direct): New macro.
2660 (mask_store_lanes_direct): Likewise.
2661 (expand_mask_load_optab_fn): Handle masked operations.
2662 (expand_mask_load_lanes_optab_fn): New macro.
2663 (expand_mask_store_optab_fn): Handle masked operations.
2664 (expand_mask_store_lanes_optab_fn): New macro.
2665 (direct_mask_load_lanes_optab_supported_p): Likewise.
2666 (direct_mask_store_lanes_optab_supported_p): Likewise.
2667 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2669 (vect_load_lanes_supported): Likewise.
2670 * tree-vect-data-refs.c (strip_conversion): New function.
2671 (can_group_stmts_p): Likewise.
2672 (vect_analyze_data_ref_accesses): Use it instead of checking
2673 for a pair of assignments.
2674 (vect_store_lanes_supported): Take a masked_p parameter.
2675 (vect_load_lanes_supported): Likewise.
2676 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2677 vect_store_lanes_supported and vect_load_lanes_supported.
2678 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2679 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2680 parameter. Don't allow gaps for masked accesses.
2681 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2682 and vect_load_lanes_supported.
2683 (get_load_store_type): Take a masked_p parameter and update
2684 call to get_group_load_store_type.
2685 (vectorizable_store): Update call to get_load_store_type.
2686 Handle IFN_MASK_STORE_LANES.
2687 (vectorizable_load): Update call to get_load_store_type.
2688 Handle IFN_MASK_LOAD_LANES.
2690 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2691 Alan Hayward <alan.hayward@arm.com>
2692 David Sherwood <david.sherwood@arm.com>
2694 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2696 * config/aarch64/aarch64-protos.h
2697 (aarch64_sve_struct_memory_operand_p): Declare.
2698 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2699 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2700 (VPRED, vpred): Handle SVE structure modes.
2701 * config/aarch64/constraints.md (Utx): New constraint.
2702 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2703 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2704 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2705 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2706 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2707 structure modes. Split into pieces after RA.
2708 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2709 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2711 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2712 SVE structure modes.
2713 (aarch64_classify_address): Likewise.
2714 (sizetochar): Move earlier in file.
2715 (aarch64_print_operand): Handle SVE register lists.
2716 (aarch64_array_mode): New function.
2717 (aarch64_sve_struct_memory_operand_p): Likewise.
2718 (TARGET_ARRAY_MODE): Redefine.
2720 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2721 Alan Hayward <alan.hayward@arm.com>
2722 David Sherwood <david.sherwood@arm.com>
2724 * target.def (array_mode): New target hook.
2725 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2726 * doc/tm.texi: Regenerate.
2727 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2728 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2729 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2731 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2734 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2735 Alan Hayward <alan.hayward@arm.com>
2736 David Sherwood <david.sherwood@arm.com>
2738 * fold-const.c (fold_binary_loc): Check the argument types
2739 rather than the result type when testing for a vector operation.
2741 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2743 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2744 * doc/tm.texi: Regenerate.
2746 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2747 Alan Hayward <alan.hayward@arm.com>
2748 David Sherwood <david.sherwood@arm.com>
2750 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2751 (sve): Document new AArch64 extension.
2752 * doc/md.texi (w): Extend the description of the AArch64
2753 constraint to include SVE vectors.
2754 (Upl, Upa): Document new AArch64 predicate constraints.
2755 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2757 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2758 (msve-vector-bits=): New option.
2759 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2760 SVE when these are disabled.
2761 (sve): New extension.
2762 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2763 modes. Adjust their number of units based on aarch64_sve_vg.
2764 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2765 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2766 aarch64_addr_query_type.
2767 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2768 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2769 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2770 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2771 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2772 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2773 (aarch64_simd_imm_zero_p): Delete.
2774 (aarch64_check_zero_based_sve_index_immediate): Declare.
2775 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2776 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2777 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2778 (aarch64_sve_float_mul_immediate_p): Likewise.
2779 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2781 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2782 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2783 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2784 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2785 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2786 (aarch64_regmode_natural_size): Likewise.
2787 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2788 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2790 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2791 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2792 for VG and the SVE predicate registers.
2793 (V_ALIASES): Add a "z"-prefixed alias.
2794 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2795 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2796 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2797 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2798 (REG_CLASS_NAMES): Add entries for them.
2799 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2800 and the predicate registers.
2801 (aarch64_sve_vg): Declare.
2802 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2803 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2804 (REGMODE_NATURAL_SIZE): Define.
2805 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2807 * config/aarch64/aarch64.c: Include cfgrtl.h.
2808 (simd_immediate_info): Add a constructor for series vectors,
2809 and an associated step field.
2810 (aarch64_sve_vg): New variable.
2811 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2812 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2813 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2814 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2815 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2816 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2817 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2818 (aarch64_get_mask_mode): New functions.
2819 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2820 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2821 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2822 predicate modes and predicate registers. Explicitly restrict
2823 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2824 to store a vector mode if it is recognized by
2825 aarch64_classify_vector_mode.
2826 (aarch64_regmode_natural_size): New function.
2827 (aarch64_hard_regno_caller_save_mode): Return the original mode
2829 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2830 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2831 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2832 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2834 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2835 does not overlap dest if the function is frame-related. Handle
2837 (aarch64_split_add_offset): New function.
2838 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2839 them aarch64_add_offset.
2840 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2841 and update call to aarch64_sub_sp.
2842 (aarch64_add_cfa_expression): New function.
2843 (aarch64_expand_prologue): Pass extra temporary registers to the
2844 functions above. Handle the case in which we need to emit new
2845 DW_CFA_expressions for registers that were originally saved
2846 relative to the stack pointer, but now have to be expressed
2847 relative to the frame pointer.
2848 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2850 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2851 IP0 and IP1 values for SVE frames.
2852 (aarch64_expand_vec_series): New function.
2853 (aarch64_expand_sve_widened_duplicate): Likewise.
2854 (aarch64_expand_sve_const_vector): Likewise.
2855 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2856 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2857 into the register, rather than emitting a SET directly.
2858 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2859 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2860 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2861 (offset_9bit_signed_scaled_p): New functions.
2862 (aarch64_replicate_bitmask_imm): New function.
2863 (aarch64_bitmask_imm): Use it.
2864 (aarch64_cannot_force_const_mem): Reject expressions involving
2865 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2866 (aarch64_classify_index): Handle SVE indices, by requiring
2867 a plain register index with a scale that matches the element size.
2868 (aarch64_classify_address): Handle SVE addresses. Assert that
2869 the mode of the address is VOIDmode or an integer mode.
2870 Update call to aarch64_classify_symbol.
2871 (aarch64_classify_symbolic_expression): Update call to
2872 aarch64_classify_symbol.
2873 (aarch64_const_vec_all_in_range_p): New function.
2874 (aarch64_print_vector_float_operand): Likewise.
2875 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2876 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2877 and the FP immediates 1.0 and 0.5.
2878 (aarch64_print_address_internal): Handle SVE addresses.
2879 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2880 (aarch64_regno_regclass): Handle predicate registers.
2881 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2883 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2884 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2885 (aarch64_convert_sve_vector_bits): New function.
2886 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2887 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2889 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2890 Handle SVE vector and predicate modes. Accept VL-based constants
2891 that need only one temporary register, and VL offsets that require
2892 no temporary registers.
2893 (aarch64_conditional_register_usage): Mark the predicate registers
2894 as fixed if SVE isn't available.
2895 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2896 Return true for SVE vector and predicate modes.
2897 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2898 rather than an unsigned int. Handle SVE modes.
2899 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2901 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2903 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2904 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2905 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2906 (aarch64_sve_float_mul_immediate_p): New functions.
2907 (aarch64_sve_valid_immediate): New function.
2908 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2909 Explicitly reject structure modes. Check for INDEX constants.
2910 Handle PTRUE and PFALSE constants.
2911 (aarch64_check_zero_based_sve_index_immediate): New function.
2912 (aarch64_simd_imm_zero_p): Delete.
2913 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2914 vector modes. Accept constants in the range of CNT[BHWD].
2915 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2916 ask for an Advanced SIMD mode.
2917 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2918 (aarch64_simd_vector_alignment): Handle SVE predicates.
2919 (aarch64_vectorize_preferred_vector_alignment): New function.
2920 (aarch64_simd_vector_alignment_reachable): Use it instead of
2922 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2923 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2925 (MAX_VECT_LEN): Delete.
2926 (expand_vec_perm_d): Add a vec_flags field.
2927 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2928 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2929 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2931 (aarch64_evpc_rev): Rename to...
2932 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2933 (aarch64_evpc_rev_global): New function.
2934 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2935 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2937 (aarch64_evpc_sve_tbl): New function.
2938 (aarch64_expand_vec_perm_const_1): Update after rename of
2939 aarch64_evpc_rev. Handle SVE permutes too, trying
2940 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2941 than aarch64_evpc_tbl.
2942 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2943 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2944 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2945 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2946 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2947 (aarch64_expand_sve_vcond): New functions.
2948 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2949 of aarch64_vector_mode_p.
2950 (aarch64_dwarf_poly_indeterminate_value): New function.
2951 (aarch64_compute_pressure_classes): Likewise.
2952 (aarch64_can_change_mode_class): Likewise.
2953 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2954 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2955 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2956 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2957 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2958 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2959 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2960 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2962 (Dn, Dl, Dr): Accept const as well as const_vector.
2963 (Dz): Likewise. Compare against CONST0_RTX.
2964 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2965 of "vector" where appropriate.
2966 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2967 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2968 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2969 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2970 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2971 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2972 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2973 (v_int_equiv): Extend to SVE modes.
2974 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2976 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2977 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2978 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2979 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2980 (SVE_COND_FP_CMP): New int iterators.
2981 (perm_hilo): Handle the new unpack unspecs.
2982 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2984 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2985 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2986 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2987 (aarch64_equality_operator, aarch64_constant_vector_operand)
2988 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2989 (aarch64_sve_nonimmediate_operand): Likewise.
2990 (aarch64_sve_general_operand): Likewise.
2991 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2992 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2993 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2994 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2995 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2996 (aarch64_sve_float_arith_immediate): Likewise.
2997 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2998 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2999 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3000 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3001 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3002 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3003 (aarch64_sve_float_arith_operand): Likewise.
3004 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3005 (aarch64_sve_float_mul_operand): Likewise.
3006 (aarch64_sve_vec_perm_operand): Likewise.
3007 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3008 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3009 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3010 as well as const_vector.
3011 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3012 in file. Use CONST0_RTX and CONSTM1_RTX.
3013 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3014 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3015 Use aarch64_simd_imm_zero.
3016 * config/aarch64/aarch64-sve.md: New file.
3017 * config/aarch64/aarch64.md: Include it.
3018 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3019 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3020 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3021 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3022 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3023 (sve): New attribute.
3024 (enabled): Disable instructions with the sve attribute unless
3026 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3027 aarch64_expand_mov_immediate.
3028 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3029 CNT[BHSD] immediates.
3030 (movti): Split CONST_POLY_INT moves into two halves.
3031 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3032 Split additions that need a temporary here if the destination
3033 is the stack pointer.
3034 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3035 (*add<mode>3_poly_1): New instruction.
3036 (set_clobber_cc): New expander.
3038 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3040 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3041 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3042 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3043 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3044 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3045 Change innermode from fixed_mode_size to machine_mode.
3046 (simplify_subreg): Update call accordingly. Handle a constant-sized
3047 subreg of a variable-length CONST_VECTOR.
3049 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3050 Alan Hayward <alan.hayward@arm.com>
3051 David Sherwood <david.sherwood@arm.com>
3053 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3054 (add_offset_to_base): New function, split out from...
3055 (create_mem_ref): ...here. When handling a scale other than 1,
3056 check first whether the address is valid without the offset.
3057 Add it into the base if so, leaving the index and scale as-is.
3059 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3062 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3063 fold_for_warn before checking if arg2 is INTEGER_CST.
3065 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3067 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3068 (store_multiple_operation): Delete.
3069 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3070 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3071 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3072 guarded by TARGET_STRING.
3073 (rs6000_output_load_multiple): Delete.
3074 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3075 OPTION_MASK_STRING / TARGET_STRING handling.
3076 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3077 (const rs6000_opt_masks) <"string">: Change mask to 0.
3078 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3079 (MASK_STRING): Delete.
3080 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3082 (load_multiple): Delete.
3089 (store_multiple): Delete.
3096 (movmemsi_8reg): Delete.
3097 (corresponding unnamed define_insn): Delete.
3098 (movmemsi_6reg): Delete.
3099 (corresponding unnamed define_insn): Delete.
3100 (movmemsi_4reg): Delete.
3101 (corresponding unnamed define_insn): Delete.
3102 (movmemsi_2reg): Delete.
3103 (corresponding unnamed define_insn): Delete.
3104 (movmemsi_1reg): Delete.
3105 (corresponding unnamed define_insn): Delete.
3106 * config/rs6000/rs6000.opt (mno-string): New.
3107 (mstring): Replace by deprecation warning stub.
3108 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3110 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3112 * regrename.c (regrename_do_replace): If replacing the same
3113 reg multiple times, try to reuse last created gen_raw_REG.
3116 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3117 main to workaround a bug in GDB.
3119 2018-01-12 Tom de Vries <tom@codesourcery.com>
3122 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3124 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3126 PR rtl-optimization/80481
3127 * ira-color.c (get_cap_member): New function.
3128 (allocnos_conflict_by_live_ranges_p): Use it.
3129 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3130 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3132 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3135 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3136 (*saddl_se_1): Ditto.
3138 (*ssubl_se_1): Ditto.
3140 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3142 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3143 rather than wi::to_widest for DR_INITs.
3144 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3145 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3146 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3148 (vect_analyze_group_access_1): Note that here.
3150 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3152 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3153 polynomial type sizes.
3155 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3157 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3158 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3159 (gimple_add_tmp_var): Likewise.
3161 2018-01-12 Martin Liska <mliska@suse.cz>
3163 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3164 (gimple_alloc_sizes): Likewise.
3165 (dump_gimple_statistics): Use PRIu64 in printf format.
3166 * gimple.h: Change uint64_t to int.
3168 2018-01-12 Martin Liska <mliska@suse.cz>
3170 * tree-core.h: Use uint64_t instead of int.
3171 * tree.c (tree_node_counts): Likewise.
3172 (tree_node_sizes): Likewise.
3173 (dump_tree_statistics): Use PRIu64 in printf format.
3175 2018-01-12 Martin Liska <mliska@suse.cz>
3177 * Makefile.in: As qsort_chk is implemented in vec.c, add
3178 vec.o to linkage of gencfn-macros.
3179 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3180 passing the info to record_node_allocation_statistics.
3181 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3183 * ggc-common.c (struct ggc_usage): Add operator== and use
3184 it in operator< and compare function.
3185 * mem-stats.h (struct mem_usage): Likewise.
3186 * vec.c (struct vec_usage): Remove operator< and compare
3187 function. Can be simply inherited.
3189 2018-01-12 Martin Jambor <mjambor@suse.cz>
3192 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3193 * tree-ssa-math-opts.c: Include domwalk.h.
3194 (convert_mult_to_fma_1): New function.
3195 (fma_transformation_info): New type.
3196 (fma_deferring_state): Likewise.
3197 (cancel_fma_deferring): New function.
3198 (result_of_phi): Likewise.
3199 (last_fma_candidate_feeds_initial_phi): Likewise.
3200 (convert_mult_to_fma): Added deferring logic, split actual
3201 transformation to convert_mult_to_fma_1.
3202 (math_opts_dom_walker): New type.
3203 (math_opts_dom_walker::after_dom_children): New method, body moved
3204 here from pass_optimize_widening_mul::execute, added deferring logic
3206 (pass_optimize_widening_mul::execute): Moved most of code to
3207 math_opts_dom_walker::after_dom_children.
3208 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3209 * config/i386/i386.c (ix86_option_override_internal): Added
3210 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3212 2018-01-12 Richard Biener <rguenther@suse.de>
3215 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3216 inline instance vars.
3218 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3221 * config/rx/rx.c (rx_is_restricted_memory_address):
3224 2018-01-12 Richard Biener <rguenther@suse.de>
3226 PR tree-optimization/80846
3227 * target.def (split_reduction): New target hook.
3228 * targhooks.c (default_split_reduction): New function.
3229 * targhooks.h (default_split_reduction): Declare.
3230 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3231 target requests first reduce vectors by combining low and high
3233 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3234 (get_vectype_for_scalar_type_and_size): Export.
3235 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3236 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3237 * doc/tm.texi: Regenerate.
3238 * config/i386/i386.c (ix86_split_reduction): Implement
3239 TARGET_VECTORIZE_SPLIT_REDUCTION.
3241 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3244 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3245 in PIC mode except for TARGET_VXWORKS_RTP.
3246 * config/sparc/sparc.c: Include cfgrtl.h.
3247 (TARGET_INIT_PIC_REG): Define.
3248 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3249 (sparc_pic_register_p): New predicate.
3250 (sparc_legitimate_address_p): Use it.
3251 (sparc_legitimize_pic_address): Likewise.
3252 (sparc_delegitimize_address): Likewise.
3253 (sparc_mode_dependent_address_p): Likewise.
3254 (gen_load_pcrel_sym): Remove 4th parameter.
3255 (load_got_register): Adjust call to above. Remove obsolete stuff.
3256 (sparc_expand_prologue): Do not call load_got_register here.
3257 (sparc_flat_expand_prologue): Likewise.
3258 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3259 (sparc_use_pseudo_pic_reg): New function.
3260 (sparc_init_pic_reg): Likewise.
3261 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3262 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3264 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3266 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3267 Add item for branch_cost.
3269 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3271 PR rtl-optimization/83565
3272 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3273 not extend the result to a larger mode for rotate operations.
3274 (num_sign_bit_copies1): Likewise.
3276 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3279 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3281 Use values-Xc.o for -pedantic.
3282 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3284 2018-01-12 Martin Liska <mliska@suse.cz>
3287 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3289 (possible_polymorphic_call_targets): Use it.
3290 (ipa_devirt): Likewise.
3292 2018-01-12 Martin Liska <mliska@suse.cz>
3294 * profile-count.h (enum profile_quality): Use 0 as invalid
3295 enum value of profile_quality.
3297 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3299 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3300 -mext-string options.
3302 2018-01-12 Richard Biener <rguenther@suse.de>
3304 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3305 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3306 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3308 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3310 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3312 * configure.ac (--with-long-double-format): Add support for the
3313 configuration option to change the default long double format on
3315 * config.gcc (powerpc*-linux*-*): Likewise.
3316 * configure: Regenerate.
3317 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3318 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3319 used without modification.
3321 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3323 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3324 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3325 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3326 MISC_BUILTIN_SPEC_BARRIER.
3327 (rs6000_init_builtins): Likewise.
3328 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3330 (speculation_barrier): New define_insn.
3331 * doc/extend.texi: Document __builtin_speculation_barrier.
3333 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3336 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3337 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3338 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3340 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3341 integral modes instead of "ss" and "sd".
3342 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3343 vectors with 32-bit and 64-bit elements.
3344 (vecdupssescalarmodesuffix): New mode attribute.
3345 (vec_dup<mode>): Use it.
3347 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3350 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3351 frame if argument is passed on stack.
3353 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3356 * ree.c (combine_reaching_defs): Optimize also
3357 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3358 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3360 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3363 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3365 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3368 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3369 after they are computed.
3371 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3373 PR tree-optimization/83695
3374 * gimple-loop-linterchange.cc
3375 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3376 reset cached scev information after interchange.
3377 (pass_linterchange::execute): Remove call to scev_reset_htab.
3379 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3381 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3382 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3383 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3384 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3385 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3386 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3387 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3388 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3389 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3390 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3391 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3392 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3393 (V_lane_reg): Likewise.
3394 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3396 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3397 (vfmal_lane_low<mode>_intrinsic,
3398 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3399 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3400 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3401 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3402 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3403 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3405 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3407 * config/arm/arm-cpus.in (fp16fml): New feature.
3408 (ALL_SIMD): Add fp16fml.
3409 (armv8.2-a): Add fp16fml as an option.
3410 (armv8.3-a): Likewise.
3411 (armv8.4-a): Add fp16fml as part of fp16.
3412 * config/arm/arm.h (TARGET_FP16FML): Define.
3413 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3415 * config/arm/arm-modes.def (V2HF): Define.
3416 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3417 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3418 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3419 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3420 vfmsl_low, vfmsl_high): New set of builtins.
3421 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3422 (vfml_op): New code attribute.
3423 (VFMLHALVES): New int iterator.
3424 (VFML, VFMLSEL): New mode attributes.
3425 (V_reg): Define mapping for V2HF.
3426 (V_hi, V_lo): New mode attributes.
3427 (VF_constraint): Likewise.
3428 (vfml_half, vfml_half_selector): New int attributes.
3429 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3431 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3432 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3434 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3435 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3436 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3437 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3439 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3440 Document new effective target and option set.
3442 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3444 * config/arm/arm-cpus.in (armv8_4): New feature.
3445 (ARMv8_4a): New fgroup.
3446 (armv8.4-a): New arch.
3447 * config/arm/arm-tables.opt: Regenerate.
3448 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3449 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3450 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3451 Add matching rules for -march=armv8.4-a and extensions.
3452 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3454 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3457 * config/rx/rx.md (BW): New mode attribute.
3458 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3460 2018-01-11 Richard Biener <rguenther@suse.de>
3462 PR tree-optimization/83435
3463 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3464 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3465 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3467 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3468 Alan Hayward <alan.hayward@arm.com>
3469 David Sherwood <david.sherwood@arm.com>
3471 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3473 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3474 (aarch64_print_address_internal): Use it to check for a zero offset.
3476 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3477 Alan Hayward <alan.hayward@arm.com>
3478 David Sherwood <david.sherwood@arm.com>
3480 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3481 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3482 Return a poly_int64 rather than a HOST_WIDE_INT.
3483 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3484 rather than a HOST_WIDE_INT.
3485 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3486 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3487 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3488 final_offset from HOST_WIDE_INT to poly_int64.
3489 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3490 to_constant when getting the number of units in an Advanced SIMD
3492 (aarch64_builtin_vectorized_function): Check for a constant number
3494 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3496 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3497 attribute instead of GET_MODE_NUNITS.
3498 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3499 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3500 GET_MODE_SIZE for fixed-size registers.
3501 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3502 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3503 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3504 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3505 (aarch64_print_operand, aarch64_print_address_internal)
3506 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3507 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3508 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3509 Handle polynomial GET_MODE_SIZE.
3510 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3511 wider than SImode without modification.
3512 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3513 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3514 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3515 passing and returning SVE modes.
3516 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3517 rather than GEN_INT.
3518 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3519 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3520 (aarch64_allocate_and_probe_stack_space): Likewise.
3521 (aarch64_layout_frame): Cope with polynomial offsets.
3522 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3523 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3525 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3526 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3527 poly_int64 rather than a HOST_WIDE_INT.
3528 (aarch64_get_separate_components, aarch64_process_components)
3529 (aarch64_expand_prologue, aarch64_expand_epilogue)
3530 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3531 (aarch64_anchor_offset): New function, split out from...
3532 (aarch64_legitimize_address): ...here.
3533 (aarch64_builtin_vectorization_cost): Handle polynomial
3534 TYPE_VECTOR_SUBPARTS.
3535 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3537 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3538 number of elements from the PARALLEL rather than the mode.
3539 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3540 rather than GET_MODE_BITSIZE.
3541 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3542 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3543 (aarch64_expand_vec_perm_const_1): Handle polynomial
3544 d->perm.length () and d->perm elements.
3545 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3546 Apply to_constant to d->perm elements.
3547 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3548 polynomial CONST_VECTOR_NUNITS.
3549 (aarch64_move_pointer): Take amount as a poly_int64 rather
3551 (aarch64_progress_pointer): Avoid temporary variable.
3552 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3553 the mode attribute instead of GET_MODE.
3555 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3556 Alan Hayward <alan.hayward@arm.com>
3557 David Sherwood <david.sherwood@arm.com>
3559 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3560 x exists before using it.
3561 (aarch64_add_constant_internal): Rename to...
3562 (aarch64_add_offset_1): ...this. Replace regnum with separate
3563 src and dest rtxes. Handle the case in which they're different,
3564 including when the offset is zero. Replace scratchreg with an rtx.
3565 Use 2 additions if there is no spare register into which we can
3566 move a 16-bit constant.
3567 (aarch64_add_constant): Delete.
3568 (aarch64_add_offset): Replace reg with separate src and dest
3569 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3570 Use aarch64_add_offset_1.
3571 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3572 an rtx rather than an int. Take the delta as a poly_int64
3573 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3574 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3575 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3576 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3577 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3579 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3580 aarch64_add_constant.
3582 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3584 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3585 Use scalar_float_mode.
3587 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3589 * config/aarch64/aarch64-simd.md
3590 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3591 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3592 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3593 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3594 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3595 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3596 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3597 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3598 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3599 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3601 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3604 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3605 targ_options->x_arm_arch_string is non NULL.
3607 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3609 * config/aarch64/aarch64.h
3610 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3612 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3615 * expmed.c (emit_store_flag_force): Swap if const op0
3616 and change VOIDmode to mode of op0.
3618 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3620 PR rtl-optimization/83761
3621 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3622 than bytes to mode_for_size.
3624 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3627 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3628 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3631 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3634 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3635 when in layout mode.
3636 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3637 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3640 2018-01-10 Michael Collison <michael.collison@arm.com>
3642 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3643 * config/aarch64/aarch64-option-extension.def: Add
3644 AARCH64_OPT_EXTENSION of 'fp16fml'.
3645 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3646 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3647 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3648 * config/aarch64/constraints.md (Ui7): New constraint.
3649 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3650 (VFMLA_SEL_W): Ditto.
3653 (VFMLA16_LOW): New int iterator.
3654 (VFMLA16_HIGH): Ditto.
3655 (UNSPEC_FMLAL): New unspec.
3656 (UNSPEC_FMLSL): Ditto.
3657 (UNSPEC_FMLAL2): Ditto.
3658 (UNSPEC_FMLSL2): Ditto.
3659 (f16mac): New code attribute.
3660 * config/aarch64/aarch64-simd-builtins.def
3661 (aarch64_fmlal_lowv2sf): Ditto.
3662 (aarch64_fmlsl_lowv2sf): Ditto.
3663 (aarch64_fmlalq_lowv4sf): Ditto.
3664 (aarch64_fmlslq_lowv4sf): Ditto.
3665 (aarch64_fmlal_highv2sf): Ditto.
3666 (aarch64_fmlsl_highv2sf): Ditto.
3667 (aarch64_fmlalq_highv4sf): Ditto.
3668 (aarch64_fmlslq_highv4sf): Ditto.
3669 (aarch64_fmlal_lane_lowv2sf): Ditto.
3670 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3671 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3672 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3673 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3674 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3675 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3676 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3677 (aarch64_fmlal_lane_highv2sf): Ditto.
3678 (aarch64_fmlsl_lane_highv2sf): Ditto.
3679 (aarch64_fmlal_laneq_highv2sf): Ditto.
3680 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3681 (aarch64_fmlalq_lane_highv4sf): Ditto.
3682 (aarch64_fmlsl_lane_highv4sf): Ditto.
3683 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3684 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3685 * config/aarch64/aarch64-simd.md:
3686 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3687 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3688 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3689 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3690 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3691 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3692 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3693 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3694 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3695 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3696 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3697 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3698 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3699 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3700 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3701 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3702 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3703 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3704 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3705 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3706 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3707 (vfmlsl_low_u32): Ditto.
3708 (vfmlalq_low_u32): Ditto.
3709 (vfmlslq_low_u32): Ditto.
3710 (vfmlal_high_u32): Ditto.
3711 (vfmlsl_high_u32): Ditto.
3712 (vfmlalq_high_u32): Ditto.
3713 (vfmlslq_high_u32): Ditto.
3714 (vfmlal_lane_low_u32): Ditto.
3715 (vfmlsl_lane_low_u32): Ditto.
3716 (vfmlal_laneq_low_u32): Ditto.
3717 (vfmlsl_laneq_low_u32): Ditto.
3718 (vfmlalq_lane_low_u32): Ditto.
3719 (vfmlslq_lane_low_u32): Ditto.
3720 (vfmlalq_laneq_low_u32): Ditto.
3721 (vfmlslq_laneq_low_u32): Ditto.
3722 (vfmlal_lane_high_u32): Ditto.
3723 (vfmlsl_lane_high_u32): Ditto.
3724 (vfmlal_laneq_high_u32): Ditto.
3725 (vfmlsl_laneq_high_u32): Ditto.
3726 (vfmlalq_lane_high_u32): Ditto.
3727 (vfmlslq_lane_high_u32): Ditto.
3728 (vfmlalq_laneq_high_u32): Ditto.
3729 (vfmlslq_laneq_high_u32): Ditto.
3730 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3731 (AARCH64_FL_FOR_ARCH8_4): New.
3732 (AARCH64_ISA_F16FML): New ISA flag.
3733 (TARGET_F16FML): New feature flag for fp16fml.
3734 (doc/invoke.texi): Document new fp16fml option.
3736 2018-01-10 Michael Collison <michael.collison@arm.com>
3738 * config/aarch64/aarch64-builtins.c:
3739 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3740 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3741 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3742 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3743 (AARCH64_ISA_SHA3): New ISA flag.
3744 (TARGET_SHA3): New feature flag for sha3.
3745 * config/aarch64/iterators.md (sha512_op): New int attribute.
3746 (CRYPTO_SHA512): New int iterator.
3747 (UNSPEC_SHA512H): New unspec.
3748 (UNSPEC_SHA512H2): Ditto.
3749 (UNSPEC_SHA512SU0): Ditto.
3750 (UNSPEC_SHA512SU1): Ditto.
3751 * config/aarch64/aarch64-simd-builtins.def
3752 (aarch64_crypto_sha512hqv2di): New builtin.
3753 (aarch64_crypto_sha512h2qv2di): Ditto.
3754 (aarch64_crypto_sha512su0qv2di): Ditto.
3755 (aarch64_crypto_sha512su1qv2di): Ditto.
3756 (aarch64_eor3qv8hi): Ditto.
3757 (aarch64_rax1qv2di): Ditto.
3758 (aarch64_xarqv2di): Ditto.
3759 (aarch64_bcaxqv8hi): Ditto.
3760 * config/aarch64/aarch64-simd.md:
3761 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3762 (aarch64_crypto_sha512su0qv2di): Ditto.
3763 (aarch64_crypto_sha512su1qv2di): Ditto.
3764 (aarch64_eor3qv8hi): Ditto.
3765 (aarch64_rax1qv2di): Ditto.
3766 (aarch64_xarqv2di): Ditto.
3767 (aarch64_bcaxqv8hi): Ditto.
3768 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3769 (vsha512h2q_u64): Ditto.
3770 (vsha512su0q_u64): Ditto.
3771 (vsha512su1q_u64): Ditto.
3772 (veor3q_u16): Ditto.
3773 (vrax1q_u64): Ditto.
3775 (vbcaxq_u16): Ditto.
3776 * config/arm/types.md (crypto_sha512): New type attribute.
3777 (crypto_sha3): Ditto.
3778 (doc/invoke.texi): Document new sha3 option.
3780 2018-01-10 Michael Collison <michael.collison@arm.com>
3782 * config/aarch64/aarch64-builtins.c:
3783 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3784 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3785 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3786 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3787 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3788 (AARCH64_ISA_SM4): New ISA flag.
3789 (TARGET_SM4): New feature flag for sm4.
3790 * config/aarch64/aarch64-simd-builtins.def
3791 (aarch64_sm3ss1qv4si): Ditto.
3792 (aarch64_sm3tt1aq4si): Ditto.
3793 (aarch64_sm3tt1bq4si): Ditto.
3794 (aarch64_sm3tt2aq4si): Ditto.
3795 (aarch64_sm3tt2bq4si): Ditto.
3796 (aarch64_sm3partw1qv4si): Ditto.
3797 (aarch64_sm3partw2qv4si): Ditto.
3798 (aarch64_sm4eqv4si): Ditto.
3799 (aarch64_sm4ekeyqv4si): Ditto.
3800 * config/aarch64/aarch64-simd.md:
3801 (aarch64_sm3ss1qv4si): Ditto.
3802 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3803 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3804 (aarch64_sm4eqv4si): Ditto.
3805 (aarch64_sm4ekeyqv4si): Ditto.
3806 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3807 (sm3part_op): Ditto.
3808 (CRYPTO_SM3TT): Ditto.
3809 (CRYPTO_SM3PART): Ditto.
3810 (UNSPEC_SM3SS1): New unspec.
3811 (UNSPEC_SM3TT1A): Ditto.
3812 (UNSPEC_SM3TT1B): Ditto.
3813 (UNSPEC_SM3TT2A): Ditto.
3814 (UNSPEC_SM3TT2B): Ditto.
3815 (UNSPEC_SM3PARTW1): Ditto.
3816 (UNSPEC_SM3PARTW2): Ditto.
3817 (UNSPEC_SM4E): Ditto.
3818 (UNSPEC_SM4EKEY): Ditto.
3819 * config/aarch64/constraints.md (Ui2): New constraint.
3820 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3821 * config/arm/types.md (crypto_sm3): New type attribute.
3822 (crypto_sm4): Ditto.
3823 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3824 (vsm3tt1aq_u32): Ditto.
3825 (vsm3tt1bq_u32): Ditto.
3826 (vsm3tt2aq_u32): Ditto.
3827 (vsm3tt2bq_u32): Ditto.
3828 (vsm3partw1q_u32): Ditto.
3829 (vsm3partw2q_u32): Ditto.
3830 (vsm4eq_u32): Ditto.
3831 (vsm4ekeyq_u32): Ditto.
3832 (doc/invoke.texi): Document new sm4 option.
3834 2018-01-10 Michael Collison <michael.collison@arm.com>
3836 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3837 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3838 (AARCH64_FL_FOR_ARCH8_4): New.
3839 (AARCH64_FL_V8_4): New flag.
3840 (doc/invoke.texi): Document new armv8.4-a option.
3842 2018-01-10 Michael Collison <michael.collison@arm.com>
3844 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3845 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3846 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3847 * config/aarch64/aarch64-option-extension.def: Add
3848 AARCH64_OPT_EXTENSION of 'sha2'.
3849 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3850 (crypto): Disable sha2 and aes if crypto disabled.
3851 (crypto): Enable aes and sha2 if enabled.
3852 (simd): Disable sha2 and aes if simd disabled.
3853 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3855 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3856 (TARGET_SHA2): New feature flag for sha2.
3857 (TARGET_AES): New feature flag for aes.
3858 * config/aarch64/aarch64-simd.md:
3859 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3860 conditional on TARGET_AES.
3861 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3862 (aarch64_crypto_sha1hsi): Make pattern conditional
3864 (aarch64_crypto_sha1hv4si): Ditto.
3865 (aarch64_be_crypto_sha1hv4si): Ditto.
3866 (aarch64_crypto_sha1su1v4si): Ditto.
3867 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3868 (aarch64_crypto_sha1su0v4si): Ditto.
3869 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3870 (aarch64_crypto_sha256su0v4si): Ditto.
3871 (aarch64_crypto_sha256su1v4si): Ditto.
3872 (doc/invoke.texi): Document new aes and sha2 options.
3874 2018-01-10 Martin Sebor <msebor@redhat.com>
3876 PR tree-optimization/83781
3877 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3880 2018-01-11 Martin Sebor <msebor@gmail.com>
3881 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3883 PR tree-optimization/83501
3884 PR tree-optimization/81703
3886 * tree-ssa-strlen.c (get_string_cst): Rename...
3887 (get_string_len): ...to this. Handle global constants.
3888 (handle_char_store): Adjust.
3890 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3891 Jim Wilson <jimw@sifive.com>
3893 * config/riscv/riscv-protos.h (riscv_output_return): New.
3894 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3895 (riscv_attribute_table, riscv_output_return),
3896 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3897 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3898 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3900 (riscv_expand_prologue): Add early return for naked function.
3901 (riscv_expand_epilogue): Likewise.
3902 (riscv_function_ok_for_sibcall): Return false for naked function.
3903 (riscv_set_current_function): New.
3904 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3905 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3906 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3907 * doc/extend.texi (RISC-V Function Attributes): New.
3909 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3911 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3912 check for 128-bit long double before checking TCmode.
3913 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3914 128-bit long doubles before checking TFmode or TCmode.
3915 (FLOAT128_IBM_P): Likewise.
3917 2018-01-10 Martin Sebor <msebor@redhat.com>
3919 PR tree-optimization/83671
3920 * builtins.c (c_strlen): Unconditionally return zero for the empty
3922 Use -Warray-bounds for warnings.
3923 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3924 for non-constant array indices with COMPONENT_REF, arrays of
3925 arrays, and pointers to arrays.
3926 (gimple_fold_builtin_strlen): Determine and set length range for
3927 non-constant character arrays.
3929 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3932 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3935 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3937 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3939 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3942 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3943 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3944 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3945 indexed_or_indirect_operand predicate.
3946 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3947 (*vsx_le_perm_load_v8hi): Likewise.
3948 (*vsx_le_perm_load_v16qi): Likewise.
3949 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3950 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3951 (*vsx_le_perm_store_v8hi): Likewise.
3952 (*vsx_le_perm_store_v16qi): Likewise.
3953 (eight unnamed splitters): Likewise.
3955 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3957 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3958 * config/rs6000/emmintrin.h: Likewise.
3959 * config/rs6000/mmintrin.h: Likewise.
3960 * config/rs6000/xmmintrin.h: Likewise.
3962 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3965 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3967 * tree.c (tree_nop_conversion): Return true for location wrapper
3969 (maybe_wrap_with_location): New function.
3970 (selftest::check_strip_nops): New function.
3971 (selftest::test_location_wrappers): New function.
3972 (selftest::tree_c_tests): Call it.
3973 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3974 (maybe_wrap_with_location): New decl.
3975 (EXPR_LOCATION_WRAPPER_P): New macro.
3976 (location_wrapper_p): New inline function.
3977 (tree_strip_any_location_wrapper): New inline function.
3979 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3982 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3983 stack_realign_offset for the largest alignment of stack slot
3985 (ix86_find_max_used_stack_alignment): New function.
3986 (ix86_finalize_stack_frame_flags): Use it. Set
3987 max_used_stack_alignment if we don't realign stack.
3988 * config/i386/i386.h (machine_function): Add
3989 max_used_stack_alignment.
3991 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3993 * config/arm/arm.opt (-mbranch-cost): New option.
3994 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3997 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4000 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4001 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4003 2018-01-10 Richard Biener <rguenther@suse.de>
4006 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4007 early out so it also covers the case where we have a non-NULL
4010 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4012 PR tree-optimization/83753
4013 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4014 for non-strided grouped accesses if the number of elements is 1.
4016 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4019 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4020 * i386.h (TARGET_USE_GATHER): Define.
4021 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4023 2018-01-10 Martin Liska <mliska@suse.cz>
4026 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4027 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4029 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4030 CLEANUP_NO_PARTITIONING is not set.
4032 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4034 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4035 for vectors, as a partial revert of r254296.
4036 * rtl.h (const_vec_p): Delete.
4037 (const_vec_duplicate_p): Don't test for vector CONSTs.
4038 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4039 * expmed.c (make_tree): Likewise.
4042 * common.md (E, F): Use CONSTANT_P instead of checking for
4044 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4045 checking for CONST_VECTOR.
4047 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4050 * predict.c (force_edge_cold): Handle in more sane way edges
4053 2018-01-09 Carl Love <cel@us.ibm.com>
4055 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4057 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4058 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4059 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4060 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4061 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4062 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4063 * config/rs6000/rs6000-protos.h: Add extern defition for
4064 rs6000_generate_float2_double_code.
4065 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4067 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4068 (float2_v2df): Add define_expand.
4070 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4073 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4074 op_mode in the force_to_mode call.
4076 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4078 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4079 instead of checking each element individually.
4080 (aarch64_evpc_uzp): Likewise.
4081 (aarch64_evpc_zip): Likewise.
4082 (aarch64_evpc_ext): Likewise.
4083 (aarch64_evpc_rev): Likewise.
4084 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4085 instead of checking each element individually. Return true without
4087 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4088 whether all selected elements come from the same input, instead of
4089 checking each element individually. Remove calls to gen_rtx_REG,
4090 start_sequence and end_sequence and instead assert that no rtl is
4093 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4095 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4096 order of HIGH and CONST checks.
4098 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4100 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4101 if the destination isn't an SSA_NAME.
4103 2018-01-09 Richard Biener <rguenther@suse.de>
4105 PR tree-optimization/83668
4106 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4108 (canonicalize_loop_form): ... here, renamed from ...
4109 (canonicalize_loop_closed_ssa_form): ... this and amended to
4110 swap successor edges for loop exit blocks to make us use
4111 the RPO order we need for initial schedule generation.
4113 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4115 PR tree-optimization/64811
4116 * match.pd: When optimizing comparisons with Inf, avoid
4117 introducing or losing exceptions from comparisons with NaN.
4119 2018-01-09 Martin Liska <mliska@suse.cz>
4122 * asan.c (shadow_mem_size): Add gcc_assert.
4124 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4126 Don't save registers in main().
4129 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4130 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4131 * config/avr/avr.c (avr_set_current_function): Don't error if
4132 naked, OS_task or OS_main are specified at the same time.
4133 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4135 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4137 * common/config/avr/avr-common.c (avr_option_optimization_table):
4138 Switch on -mmain-is-OS_task for optimizing compilations.
4140 2018-01-09 Richard Biener <rguenther@suse.de>
4142 PR tree-optimization/83572
4143 * graphite.c: Include cfganal.h.
4144 (graphite_transform_loops): Connect infinite loops to exit
4145 and remove fake edges at the end.
4147 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4149 * ipa-inline.c (edge_badness): Revert accidental checkin.
4151 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4154 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4155 symbols; not inline clones.
4157 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4160 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4161 hard registers. Formatting fixes.
4163 PR preprocessor/83722
4164 * gcc.c (try_generate_repro): Pass
4165 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4166 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4169 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4170 Kito Cheng <kito.cheng@gmail.com>
4172 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4173 (riscv_leaf_function_p): Delete.
4174 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4176 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4178 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4180 (do_ifelse): New function.
4181 (do_isel): New function.
4182 (do_sub3): New function.
4183 (do_add3): New function.
4184 (do_load_mask_compare): New function.
4185 (do_overlap_load_compare): New function.
4186 (expand_compare_loop): New function.
4187 (expand_block_compare): Call expand_compare_loop() when appropriate.
4188 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4190 (-mblock-compare-inline-loop-limit): New option.
4192 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4195 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4196 Reverse order of second and third operands in first alternative.
4197 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4198 of first and second elements in UNSPEC_VPERMR vector.
4199 (altivec_expand_vec_perm_le): Likewise.
4201 2018-01-08 Jeff Law <law@redhat.com>
4203 PR rtl-optimizatin/81308
4204 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4205 (process_switch): If group_case_labels makes a change, then set
4207 (pass_convert_switch::execute): If a switch is converted, then
4208 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4210 PR rtl-optimization/81308
4211 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4214 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4216 PR target/83663 - Revert r255946
4217 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4218 generation for cases where splatting a value is not useful.
4219 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4220 across a vec_duplicate and a paradoxical subreg forming a vector
4221 mode to a vec_concat.
4223 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4225 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4226 -march=armv8.3-a variants.
4227 * config/arm/t-multilib: Likewise.
4228 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4230 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4232 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4234 (cceq_ior_compare_complement): Give it a name so I can use it, and
4235 change boolean_or_operator predicate to boolean_operator so it can
4236 be used to generate a crand.
4237 (eqne): New code iterator.
4238 (bd/bd_neg): New code_attrs.
4239 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4240 a single define_insn.
4241 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4242 decrement (bdnzt/bdnzf/bdzt/bdzf).
4243 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4244 with the new names of the branch decrement patterns, and added the
4245 names of the branch decrement conditional patterns.
4247 2018-01-08 Richard Biener <rguenther@suse.de>
4249 PR tree-optimization/83563
4250 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4253 2018-01-08 Richard Biener <rguenther@suse.de>
4256 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4258 2018-01-08 Richard Biener <rguenther@suse.de>
4260 PR tree-optimization/83685
4261 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4262 references to abnormals.
4264 2018-01-08 Richard Biener <rguenther@suse.de>
4267 * dwarf2out.c (output_indirect_strings): Handle empty
4268 skeleton_debug_str_hash.
4269 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4271 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4273 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4274 (emit_store_direct): Likewise.
4275 (arc_trampoline_adjust_address): Likewise.
4276 (arc_asm_trampoline_template): New function.
4277 (arc_initialize_trampoline): Use asm_trampoline_template.
4278 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4279 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4280 * config/arc/arc.md (flush_icache): Delete pattern.
4282 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4284 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4285 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4288 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4291 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4292 by not USED_FOR_TARGET.
4293 (make_pass_resolve_sw_modes): Likewise.
4295 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4297 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4300 2018-01-08 Richard Biener <rguenther@suse.de>
4303 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4305 2018-01-08 Richard Biener <rguenther@suse.de>
4308 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4310 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4313 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4314 basic blocks with a small number of successors.
4315 (convert_control_dep_chain_into_preds): Improve handling of
4317 (dump_predicates): Split apart into...
4318 (dump_pred_chain): ...here...
4319 (dump_pred_info): ...and here.
4320 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4321 (can_chain_union_be_invalidated_p): Improve check for invalidation
4323 (uninit_uses_cannot_happen): Avoid unnecessary if
4324 convert_control_dep_chain_into_preds yielded nothing.
4326 2018-01-06 Martin Sebor <msebor@redhat.com>
4328 PR tree-optimization/83640
4329 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4330 subtracting negative offset from size.
4331 (builtin_access::overlap): Adjust offset bounds of the access to fall
4332 within the size of the object if possible.
4334 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4336 PR rtl-optimization/83699
4337 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4338 extract_bit_field_as_subreg to cases in which the extracted
4339 value is also a vector.
4341 * lra-constraints.c (process_alt_operands): Test for the equivalence
4342 substitutions when detecting a possible reload cycle.
4344 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4347 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4348 by default if flag_selective_schedling{,2}. Formatting fixes.
4350 PR rtl-optimization/83682
4351 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4352 if it has non-VECTOR_MODE element mode.
4353 (vec_duplicate_p): Likewise.
4356 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4357 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4359 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4362 * config/i386/i386-builtin.def
4363 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4364 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4365 Require also OPTION_MASK_ISA_AVX512F in addition to
4366 OPTION_MASK_ISA_GFNI.
4367 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4368 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4369 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4370 to OPTION_MASK_ISA_GFNI.
4371 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4372 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4373 OPTION_MASK_ISA_AVX512BW.
4374 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4375 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4376 addition to OPTION_MASK_ISA_GFNI.
4377 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4378 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4379 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4380 to OPTION_MASK_ISA_GFNI.
4381 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4382 a requirement for all ISAs rather than any of them with a few
4384 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4386 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4387 bitmasks to be enabled with 3 exceptions, instead of requiring any
4388 enabled ISA with lots of exceptions.
4389 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4390 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4391 Change avx512bw in isa attribute to avx512f.
4392 * config/i386/sgxintrin.h: Add license boilerplate.
4393 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4394 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4395 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4396 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4398 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4399 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4400 temporarily sse2 rather than sse if not enabled already.
4403 * config/i386/sse.md (VI248_VLBW): Rename to ...
4404 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4405 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4406 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4407 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4408 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4409 mode iterator instead of VI248_VLBW.
4411 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4413 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4414 (record_modified): Skip clobbers; add debug output.
4415 (param_change_prob): Use sreal frequencies.
4417 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4419 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4420 punt for user-aligned variables.
4422 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4424 * tree-chrec.c (chrec_contains_symbols): Return true for
4427 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4430 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4431 of (x|y) == x for BICS pattern.
4433 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4435 PR tree-optimization/83605
4436 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4437 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4440 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4442 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4443 * config/epiphany/rtems.h: New file.
4445 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4446 Uros Bizjak <ubizjak@gmail.com>
4449 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4450 QIreg_operand instead of register_operand predicate.
4451 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4452 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4453 comments instead of -fmitigate[-_]rop.
4455 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4458 * cgraphunit.c (symbol_table::compile): Switch to text_section
4459 before calling assembly_start debug hook.
4460 * run-rtl-passes.c (run_rtl_passes): Likewise.
4463 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4465 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4466 range_int_cst_p rather than !symbolic_range_p before calling
4467 extract_range_from_multiplicative_op_1.
4469 2018-01-04 Jeff Law <law@redhat.com>
4471 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4472 redundant test in assertion.
4474 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4476 * doc/rtl.texi: Document machine_mode wrapper classes.
4478 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4480 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4483 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4485 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4486 the VEC_PERM_EXPR fold to fail.
4488 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4491 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4492 to switched_sections.
4494 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4497 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4500 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4503 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4504 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4506 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4509 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4510 is BLKmode and bitpos not zero or mode change is needed.
4512 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4515 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4518 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4521 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4522 instead of MULT rtx. Update all corresponding splitters.
4524 (*ssub<modesuffix>): Ditto.
4526 (*cmp_sadd_di): Update split patterns.
4527 (*cmp_sadd_si): Ditto.
4528 (*cmp_sadd_sidi): Ditto.
4529 (*cmp_ssub_di): Ditto.
4530 (*cmp_ssub_si): Ditto.
4531 (*cmp_ssub_sidi): Ditto.
4532 * config/alpha/predicates.md (const23_operand): New predicate.
4533 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4534 Look for ASHIFT, not MULT inner operand.
4535 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4537 2018-01-04 Martin Liska <mliska@suse.cz>
4539 PR gcov-profile/83669
4540 * gcov.c (output_intermediate_file): Add version to intermediate
4542 * doc/gcov.texi: Document new field 'version' in intermediate
4543 file format. Fix location of '-k' option of gcov command.
4545 2018-01-04 Martin Liska <mliska@suse.cz>
4548 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4550 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4552 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4554 2018-01-03 Martin Sebor <msebor@redhat.com>
4556 PR tree-optimization/83655
4557 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4558 checking calls with invalid arguments.
4560 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4562 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4563 (vectorizable_mask_load_store): Delete.
4564 (vectorizable_call): Return false for masked loads and stores.
4565 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4566 instead of gimple_assign_rhs1.
4567 (vectorizable_load): Handle IFN_MASK_LOAD.
4568 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4570 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4572 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4574 (vectorizable_mask_load_store): ...here.
4575 (vectorizable_load): ...and here.
4577 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4579 * tree-vect-stmts.c (vect_build_all_ones_mask)
4580 (vect_build_zero_merge_argument): New functions, split out from...
4581 (vectorizable_load): ...here.
4583 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4585 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4587 (vectorizable_mask_load_store): ...here.
4588 (vectorizable_store): ...and here.
4590 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4592 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4594 (vectorizable_mask_load_store): ...here.
4596 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4598 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4599 (vect_model_store_cost): Take a vec_load_store_type instead of a
4601 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4602 (vect_model_store_cost): Take a vec_load_store_type instead of a
4604 (vectorizable_mask_load_store): Update accordingly.
4605 (vectorizable_store): Likewise.
4606 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4608 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4610 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4611 IFN_MASK_LOAD calls here rather than...
4612 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4614 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4615 Alan Hayward <alan.hayward@arm.com>
4616 David Sherwood <david.sherwood@arm.com>
4618 * expmed.c (extract_bit_field_1): For vector extracts,
4619 fall back to extract_bit_field_as_subreg if vec_extract
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 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4627 they are variable or constant sized.
4628 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4629 slots for constant-sized data.
4631 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4632 Alan Hayward <alan.hayward@arm.com>
4633 David Sherwood <david.sherwood@arm.com>
4635 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4636 handling COND_EXPRs with boolean comparisons, try to find a better
4637 basis for the mask type than the boolean itself.
4639 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4641 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4642 is calculated and how it can be overridden.
4643 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4644 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4646 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4649 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4650 Alan Hayward <alan.hayward@arm.com>
4651 David Sherwood <david.sherwood@arm.com>
4653 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4654 Remove the mode argument.
4655 (aarch64_simd_valid_immediate): Remove the mode and inverse
4657 * config/aarch64/iterators.md (bitsize): New iterator.
4658 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4659 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4660 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4661 aarch64_simd_valid_immediate.
4662 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4663 (aarch64_reg_or_bic_imm): Likewise.
4664 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4665 with an insn_type enum and msl with a modifier_type enum.
4666 Replace element_width with a scalar_mode. Change the shift
4667 to unsigned int. Add constructors for scalar_float_mode and
4668 scalar_int_mode elements.
4669 (aarch64_vect_float_const_representable_p): Delete.
4670 (aarch64_can_const_movi_rtx_p)
4671 (aarch64_simd_scalar_immediate_valid_for_move)
4672 (aarch64_simd_make_constant): Update call to
4673 aarch64_simd_valid_immediate.
4674 (aarch64_advsimd_valid_immediate_hs): New function.
4675 (aarch64_advsimd_valid_immediate): Likewise.
4676 (aarch64_simd_valid_immediate): Remove mode and inverse
4677 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4678 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4679 and aarch64_float_const_representable_p on the result.
4680 (aarch64_output_simd_mov_immediate): Remove mode argument.
4681 Update call to aarch64_simd_valid_immediate and use of
4682 simd_immediate_info.
4683 (aarch64_output_scalar_simd_mov_immediate): Update call
4686 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4687 Alan Hayward <alan.hayward@arm.com>
4688 David Sherwood <david.sherwood@arm.com>
4690 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4691 (mode_nunits): Likewise CONST_MODE_NUNITS.
4692 * machmode.def (ADJUST_NUNITS): Document.
4693 * genmodes.c (mode_data::need_nunits_adj): New field.
4694 (blank_mode): Update accordingly.
4695 (adj_nunits): New variable.
4696 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4698 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4699 listed in adj_nunits.
4700 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4701 listed in adj_nunits. Don't emit case statements for such modes.
4702 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4703 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4704 nothing if adj_nunits is nonnull.
4705 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4706 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4707 (emit_mode_fbit): Update use of print_maybe_const_decl.
4708 (emit_move_size): Likewise. Treat the array as non-const
4710 (emit_mode_adjustments): Handle adj_nunits.
4712 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4714 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4715 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4716 (VECTOR_MODES): Use it.
4717 (make_vector_modes): Take the prefix as an argument.
4719 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4720 Alan Hayward <alan.hayward@arm.com>
4721 David Sherwood <david.sherwood@arm.com>
4723 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4724 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4725 for MODE_VECTOR_BOOL.
4726 * machmode.def (VECTOR_BOOL_MODE): Document.
4727 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4728 (make_vector_bool_mode): New function.
4729 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4731 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4732 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4734 * stor-layout.c (int_mode_for_mode): Likewise.
4735 * tree.c (build_vector_type_for_mode): Likewise.
4736 * varasm.c (output_constant_pool_2): Likewise.
4737 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4738 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4739 for MODE_VECTOR_BOOL.
4740 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4741 of mode class checks.
4742 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4743 instead of a list of mode class checks.
4744 (expand_vector_scalar_condition): Likewise.
4745 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4747 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4748 Alan Hayward <alan.hayward@arm.com>
4749 David Sherwood <david.sherwood@arm.com>
4751 * machmode.h (mode_size): Change from unsigned short to
4753 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4754 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4755 or if measurement_type is not polynomial.
4756 (fixed_size_mode::includes_p): Check for constant-sized modes.
4757 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4758 return a poly_uint16 rather than an unsigned short.
4759 (emit_mode_size): Change the type of mode_size from unsigned short
4760 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4761 (emit_mode_adjustments): Cope with polynomial vector sizes.
4762 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4764 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4766 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4767 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4768 * caller-save.c (setup_save_areas): Likewise.
4769 (replace_reg_with_saved_mem): Likewise.
4770 * calls.c (emit_library_call_value_1): Likewise.
4771 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4772 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4773 (gen_lowpart_for_combine): Likewise.
4774 * convert.c (convert_to_integer_1): Likewise.
4775 * cse.c (equiv_constant, cse_insn): Likewise.
4776 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4777 (cselib_subst_to_values): Likewise.
4778 * dce.c (word_dce_process_block): Likewise.
4779 * df-problems.c (df_word_lr_mark_ref): Likewise.
4780 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4781 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4782 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4783 (rtl_for_decl_location): Likewise.
4784 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4785 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4786 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4787 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4788 (expand_expr_real_1): Likewise.
4789 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4790 (pad_below): Likewise.
4791 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4792 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4793 * ira.c (get_subreg_tracking_sizes): Likewise.
4794 * ira-build.c (ira_create_allocno_objects): Likewise.
4795 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4796 (ira_sort_regnos_for_alter_reg): Likewise.
4797 * ira-costs.c (record_operand_costs): Likewise.
4798 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4799 (resolve_simple_move): Likewise.
4800 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4801 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4802 (lra_constraints): Likewise.
4803 (CONST_POOL_OK_P): Reject variable-sized modes.
4804 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4805 (add_pseudo_to_slot, lra_spill): Likewise.
4806 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4807 * optabs-query.c (get_best_extraction_insn): Likewise.
4808 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4809 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4810 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4811 * recog.c (offsettable_address_addr_space_p): Likewise.
4812 * regcprop.c (maybe_mode_change): Likewise.
4813 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4814 * regrename.c (build_def_use): Likewise.
4815 * regstat.c (dump_reg_info): Likewise.
4816 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4817 (find_reloads, find_reloads_subreg_address): Likewise.
4818 * reload1.c (eliminate_regs_1): Likewise.
4819 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4820 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4821 (simplify_binary_operation_1, simplify_subreg): Likewise.
4822 * targhooks.c (default_function_arg_padding): Likewise.
4823 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4824 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4825 (verify_gimple_assign_ternary): Likewise.
4826 * tree-inline.c (estimate_move_cost): Likewise.
4827 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4828 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4829 (get_address_cost_ainc): Likewise.
4830 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4831 (vect_supportable_dr_alignment): Likewise.
4832 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4833 (vectorizable_reduction): Likewise.
4834 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4835 (vectorizable_operation, vectorizable_load): Likewise.
4836 * tree.c (build_same_sized_truth_vector_type): Likewise.
4837 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4838 * var-tracking.c (emit_note_insn_var_location): Likewise.
4839 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4840 (ADDR_VEC_ALIGN): Likewise.
4842 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4843 Alan Hayward <alan.hayward@arm.com>
4844 David Sherwood <david.sherwood@arm.com>
4846 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4848 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4849 or if measurement_type is polynomial.
4850 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4851 * combine.c (make_extraction): Likewise.
4852 * dse.c (find_shift_sequence): Likewise.
4853 * dwarf2out.c (mem_loc_descriptor): Likewise.
4854 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4855 (extract_bit_field, extract_low_bits): Likewise.
4856 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4857 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4858 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4859 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4860 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4861 * reload.c (find_reloads): Likewise.
4862 * reload1.c (alter_reg): Likewise.
4863 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4864 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4865 * tree-if-conv.c (predicate_mem_writes): Likewise.
4866 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4867 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4868 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4869 * valtrack.c (dead_debug_insert_temp): Likewise.
4870 * varasm.c (mergeable_constant_section): Likewise.
4871 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
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 * expr.c (expand_assignment): Cope with polynomial mode sizes
4878 when assigning to a CONCAT.
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 * machmode.h (mode_precision): Change from unsigned short to
4886 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4888 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4889 or if measurement_type is not polynomial.
4890 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4891 in which the mode is already known to be a scalar_int_mode.
4892 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4893 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4895 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4896 for GET_MODE_PRECISION.
4897 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4898 for GET_MODE_PRECISION.
4899 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4901 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4902 (expand_field_assignment, make_extraction): Likewise.
4903 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4904 (get_last_value): Likewise.
4905 * convert.c (convert_to_integer_1): Likewise.
4906 * cse.c (cse_insn): Likewise.
4907 * expr.c (expand_expr_real_1): Likewise.
4908 * lra-constraints.c (simplify_operand_subreg): Likewise.
4909 * optabs-query.c (can_atomic_load_p): Likewise.
4910 * optabs.c (expand_atomic_load): Likewise.
4911 (expand_atomic_store): Likewise.
4912 * ree.c (combine_reaching_defs): Likewise.
4913 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4914 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4915 * tree.h (type_has_mode_precision_p): Likewise.
4916 * ubsan.c (instrument_si_overflow): Likewise.
4918 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4919 Alan Hayward <alan.hayward@arm.com>
4920 David Sherwood <david.sherwood@arm.com>
4922 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4923 polynomial numbers of units.
4924 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4925 (valid_vector_subparts_p): New function.
4926 (build_vector_type): Remove temporary shim and take the number
4927 of units as a poly_uint64 rather than an int.
4928 (build_opaque_vector_type): Take the number of units as a
4929 poly_uint64 rather than an int.
4930 * tree.c (build_vector_from_ctor): Handle polynomial
4931 TYPE_VECTOR_SUBPARTS.
4932 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4933 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4934 (build_vector_from_val): If the number of units is variable,
4935 use build_vec_duplicate_cst for constant operands and
4936 VEC_DUPLICATE_EXPR otherwise.
4937 (make_vector_type): Remove temporary is_constant ().
4938 (build_vector_type, build_opaque_vector_type): Take the number of
4939 units as a poly_uint64 rather than an int.
4940 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4942 * cfgexpand.c (expand_debug_expr): Likewise.
4943 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4944 (store_constructor, expand_expr_real_1): Likewise.
4945 (const_scalar_mask_from_tree): Likewise.
4946 * fold-const-call.c (fold_const_reduction): Likewise.
4947 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4948 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4949 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4950 (fold_relational_const): Likewise.
4951 (native_interpret_vector): Likewise. Change the size from an
4952 int to an unsigned int.
4953 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4954 TYPE_VECTOR_SUBPARTS.
4955 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4956 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4957 duplicating a non-constant operand into a variable-length vector.
4958 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4959 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4960 * ipa-icf.c (sem_variable::equals): Likewise.
4961 * match.pd: Likewise.
4962 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4963 * print-tree.c (print_node): Likewise.
4964 * stor-layout.c (layout_type): Likewise.
4965 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4966 * tree-cfg.c (verify_gimple_comparison): Likewise.
4967 (verify_gimple_assign_binary): Likewise.
4968 (verify_gimple_assign_ternary): Likewise.
4969 (verify_gimple_assign_single): Likewise.
4970 * tree-pretty-print.c (dump_generic_node): Likewise.
4971 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4972 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4973 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4974 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4975 (vect_shift_permute_load_chain): Likewise.
4976 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4977 (expand_vector_condition, optimize_vector_constructor): Likewise.
4978 (lower_vec_perm, get_compute_type): Likewise.
4979 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4980 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4981 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4982 (vect_recog_mask_conversion_pattern): Likewise.
4983 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4984 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4985 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4986 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4987 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4988 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4989 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4990 (supportable_widening_operation): Likewise.
4991 (supportable_narrowing_operation): Likewise.
4992 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4994 * varasm.c (output_constant): Likewise.
4996 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4997 Alan Hayward <alan.hayward@arm.com>
4998 David Sherwood <david.sherwood@arm.com>
5000 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5001 so that both the length == 3 and length != 3 cases set up their
5002 own permute vectors. Add comments explaining why we know the
5003 number of elements is constant.
5004 (vect_permute_load_chain): Likewise.
5006 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5007 Alan Hayward <alan.hayward@arm.com>
5008 David Sherwood <david.sherwood@arm.com>
5010 * machmode.h (mode_nunits): Change from unsigned char to
5012 (ONLY_FIXED_SIZE_MODES): New macro.
5013 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5014 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5015 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5017 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5018 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5019 or if measurement_type is not polynomial.
5020 * genmodes.c (ZERO_COEFFS): New macro.
5021 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5023 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5024 Use ZERO_COEFFS when emitting initializers.
5025 * data-streamer.h (bp_pack_poly_value): New function.
5026 (bp_unpack_poly_value): Likewise.
5027 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5028 for GET_MODE_NUNITS.
5029 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5030 for GET_MODE_NUNITS.
5031 * tree.c (make_vector_type): Remove temporary shim and make
5032 the real function take the number of units as a poly_uint64
5034 (build_vector_type_for_mode): Handle polynomial nunits.
5035 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5036 * emit-rtl.c (const_vec_series_p_1): Likewise.
5037 (gen_rtx_CONST_VECTOR): Likewise.
5038 * fold-const.c (test_vec_duplicate_folding): Likewise.
5039 * genrecog.c (validate_pattern): Likewise.
5040 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5041 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5042 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5043 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5044 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5045 * rtlanal.c (subreg_get_info): Likewise.
5046 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5047 (vect_grouped_load_supported): Likewise.
5048 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5049 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5050 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5051 (simplify_const_unary_operation, simplify_binary_operation_1)
5052 (simplify_const_binary_operation, simplify_ternary_operation)
5053 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5054 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5055 instead of CONST_VECTOR_NUNITS.
5056 * varasm.c (output_constant_pool_2): Likewise.
5057 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5058 explicit-encoded elements in the XVEC for variable-length vectors.
5060 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5062 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5064 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5065 Alan Hayward <alan.hayward@arm.com>
5066 David Sherwood <david.sherwood@arm.com>
5068 * coretypes.h (fixed_size_mode): Declare.
5069 (fixed_size_mode_pod): New typedef.
5070 * builtins.h (target_builtins::x_apply_args_mode)
5071 (target_builtins::x_apply_result_mode): Change type to
5072 fixed_size_mode_pod.
5073 * builtins.c (apply_args_size, apply_result_size, result_vector)
5074 (expand_builtin_apply_args_1, expand_builtin_apply)
5075 (expand_builtin_return): Update accordingly.
5077 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5079 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5080 * cselib.c (cselib_hash_rtx): Likewise.
5081 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5082 CONST_VECTOR encoding.
5084 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5085 Jeff Law <law@redhat.com>
5088 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5089 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5090 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5091 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5094 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5095 explicitly probe *sp in a noreturn function if there were any callee
5096 register saves or frame pointer is needed.
5098 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5101 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5102 BLKmode for ternary, binary or unary expressions.
5105 * var-tracking.c (delete_vta_debug_insn): New inline function.
5106 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5107 insns from get_insns () to NULL instead of each bb separately.
5108 Use delete_vta_debug_insn. No longer static.
5109 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5110 delete_vta_debug_insns callers.
5111 * rtl.h (delete_vta_debug_insns): Declare.
5112 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5113 instead of variable_tracking_main.
5115 2018-01-03 Martin Sebor <msebor@redhat.com>
5117 PR tree-optimization/83603
5118 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5119 arguments past the endof the argument list in functions declared
5120 without a prototype.
5121 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5122 Avoid checking when arguments are null.
5124 2018-01-03 Martin Sebor <msebor@redhat.com>
5127 * doc/extend.texi (attribute const): Fix a typo.
5128 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5129 issuing -Wsuggest-attribute for void functions.
5131 2018-01-03 Martin Sebor <msebor@redhat.com>
5133 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5134 offset_int::from instead of wide_int::to_shwi.
5135 (maybe_diag_overlap): Remove assertion.
5136 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5137 * gimple-ssa-sprintf.c (format_directive): Same.
5138 (parse_directive): Same.
5139 (sprintf_dom_walker::compute_format_length): Same.
5140 (try_substitute_return_value): Same.
5142 2018-01-03 Jeff Law <law@redhat.com>
5145 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5146 non-constant residual for zero at runtime and avoid probing in
5147 that case. Reorganize code for trailing problem to mirror handling
5150 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5152 PR tree-optimization/83501
5153 * tree-ssa-strlen.c (get_string_cst): New.
5154 (handle_char_store): Call get_string_cst.
5156 2018-01-03 Martin Liska <mliska@suse.cz>
5158 PR tree-optimization/83593
5159 * tree-ssa-strlen.c: Include tree-cfg.h.
5160 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5161 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5162 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5164 (strlen_dom_walker::before_dom_children): Call
5165 gimple_purge_dead_eh_edges. Dump tranformation with details
5167 (strlen_dom_walker::before_dom_children): Update call by adding
5168 new argument cleanup_eh.
5169 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5171 2018-01-03 Martin Liska <mliska@suse.cz>
5174 * cif-code.def (VARIADIC_THUNK): New enum value.
5175 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5178 2018-01-03 Jan Beulich <jbeulich@suse.com>
5180 * sse.md (mov<mode>_internal): Tighten condition for when to use
5181 vmovdqu<ssescalarsize> for TI and OI modes.
5183 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5185 Update copyright years.
5187 2018-01-03 Martin Liska <mliska@suse.cz>
5190 * ipa-visibility.c (function_and_variable_visibility): Skip
5191 functions with noipa attribure.
5193 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5195 * gcc.c (process_command): Update copyright notice dates.
5196 * gcov-dump.c (print_version): Ditto.
5197 * gcov.c (print_version): Ditto.
5198 * gcov-tool.c (print_version): Ditto.
5199 * gengtype.c (create_file): Ditto.
5200 * doc/cpp.texi: Bump @copying's copyright year.
5201 * doc/cppinternals.texi: Ditto.
5202 * doc/gcc.texi: Ditto.
5203 * doc/gccint.texi: Ditto.
5204 * doc/gcov.texi: Ditto.
5205 * doc/install.texi: Ditto.
5206 * doc/invoke.texi: Ditto.
5208 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5210 * vector-builder.h (vector_builder::m_full_nelts): Change from
5211 unsigned int to poly_uint64.
5212 (vector_builder::full_nelts): Update prototype accordingly.
5213 (vector_builder::new_vector): Likewise.
5214 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5215 (vector_builder::operator ==): Likewise.
5216 (vector_builder::finalize): Likewise.
5217 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5218 Take the number of elements as a poly_uint64 rather than an
5220 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5221 from unsigned int to poly_uint64.
5222 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5223 (vec_perm_indices::new_vector): Likewise.
5224 (vec_perm_indices::length): Likewise.
5225 (vec_perm_indices::nelts_per_input): Likewise.
5226 (vec_perm_indices::input_nelts): Likewise.
5227 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5228 number of elements per input as a poly_uint64 rather than an
5229 unsigned int. Use the original encoding for variable-length
5230 vectors, rather than clamping each individual element.
5231 For the second and subsequent elements in each pattern,
5232 clamp the step and base before clamping their sum.
5233 (vec_perm_indices::series_p): Handle polynomial element counts.
5234 (vec_perm_indices::all_in_range_p): Likewise.
5235 (vec_perm_indices_to_tree): Likewise.
5236 (vec_perm_indices_to_rtx): Likewise.
5237 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5238 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5239 (tree_vector_builder::new_binary_operation): Handle polynomial
5240 element counts. Return false if we need to know the number
5241 of elements at compile time.
5242 * fold-const.c (fold_vec_perm): Punt if the number of elements
5243 isn't known at compile time.
5245 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5247 * vec-perm-indices.h (vec_perm_builder): Change element type
5248 from HOST_WIDE_INT to poly_int64.
5249 (vec_perm_indices::element_type): Update accordingly.
5250 (vec_perm_indices::clamp): Handle polynomial element_types.
5251 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5252 (vec_perm_indices::all_in_range_p): Likewise.
5253 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5255 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5256 polynomial vec_perm_indices element types.
5257 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5258 * fold-const.c (fold_vec_perm): Likewise.
5259 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5260 * tree-vect-generic.c (lower_vec_perm): Likewise.
5261 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5262 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5263 element type to HOST_WIDE_INT.
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 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5270 rather than an int. Use plus_constant.
5271 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5272 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5274 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5275 Alan Hayward <alan.hayward@arm.com>
5276 David Sherwood <david.sherwood@arm.com>
5278 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5279 a HOST_WIDE_INT to a poly_int64.
5281 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5282 Alan Hayward <alan.hayward@arm.com>
5283 David Sherwood <david.sherwood@arm.com>
5285 * calls.c (load_register_parameters): Cope with polynomial
5286 mode sizes. Require a constant size for BLKmode parameters
5287 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5288 forces a parameter to be padded at the lsb end in order to
5289 fill a complete number of words, require the parameter size
5290 to be ordered wrt UNITS_PER_WORD.
5292 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5293 Alan Hayward <alan.hayward@arm.com>
5294 David Sherwood <david.sherwood@arm.com>
5296 * reload1.c (spill_stack_slot_width): Change element type
5297 from unsigned int to poly_uint64_pod.
5298 (alter_reg): Treat mode sizes as polynomial.
5300 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5301 Alan Hayward <alan.hayward@arm.com>
5302 David Sherwood <david.sherwood@arm.com>
5304 * reload.c (complex_word_subreg_p): New function.
5305 (reload_inner_reg_of_subreg, push_reload): Use it.
5307 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5308 Alan Hayward <alan.hayward@arm.com>
5309 David Sherwood <david.sherwood@arm.com>
5311 * lra-constraints.c (process_alt_operands): Reject matched
5312 operands whose sizes aren't ordered.
5313 (match_reload): Refer to this check here.
5315 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5316 Alan Hayward <alan.hayward@arm.com>
5317 David Sherwood <david.sherwood@arm.com>
5319 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5320 that the mode size is in the set {1, 2, 4, 8, 16}.
5322 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5323 Alan Hayward <alan.hayward@arm.com>
5324 David Sherwood <david.sherwood@arm.com>
5326 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5327 Use plus_constant instead of gen_rtx_PLUS.
5329 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5330 Alan Hayward <alan.hayward@arm.com>
5331 David Sherwood <david.sherwood@arm.com>
5333 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5334 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5335 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5336 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5337 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5338 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5339 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5340 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5341 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5342 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5344 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5345 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5346 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5347 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5348 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5349 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5350 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5351 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5352 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5353 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5355 * expr.c (emit_move_resolve_push): Treat the input and result
5356 of PUSH_ROUNDING as a poly_int64.
5357 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5358 (emit_push_insn): Likewise.
5359 * lra-eliminations.c (mark_not_eliminable): Likewise.
5360 * recog.c (push_operand): Likewise.
5361 * reload1.c (elimination_effects): Likewise.
5362 * rtlanal.c (nonzero_bits1): Likewise.
5363 * calls.c (store_one_arg): Likewise. Require the padding to be
5364 known at compile time.
5366 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5367 Alan Hayward <alan.hayward@arm.com>
5368 David Sherwood <david.sherwood@arm.com>
5370 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5371 Use plus_constant instead of gen_rtx_PLUS.
5373 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5374 Alan Hayward <alan.hayward@arm.com>
5375 David Sherwood <david.sherwood@arm.com>
5377 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5380 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5381 Alan Hayward <alan.hayward@arm.com>
5382 David Sherwood <david.sherwood@arm.com>
5384 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5385 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5386 via stack temporaries. Treat the mode size as polynomial too.
5388 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5389 Alan Hayward <alan.hayward@arm.com>
5390 David Sherwood <david.sherwood@arm.com>
5392 * expr.c (expand_expr_real_2): When handling conversions involving
5393 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5394 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5395 as a poly_uint64 too.
5397 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5398 Alan Hayward <alan.hayward@arm.com>
5399 David Sherwood <david.sherwood@arm.com>
5401 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5403 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5404 Alan Hayward <alan.hayward@arm.com>
5405 David Sherwood <david.sherwood@arm.com>
5407 * combine.c (can_change_dest_mode): Handle polynomial
5408 REGMODE_NATURAL_SIZE.
5409 * expmed.c (store_bit_field_1): Likewise.
5410 * expr.c (store_constructor): Likewise.
5411 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5412 and polynomial REGMODE_NATURAL_SIZE.
5413 (gen_lowpart_common): Likewise.
5414 * reginfo.c (record_subregs_of_mode): Likewise.
5415 * rtlanal.c (read_modify_subreg_p): Likewise.
5417 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5418 Alan Hayward <alan.hayward@arm.com>
5419 David Sherwood <david.sherwood@arm.com>
5421 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5422 numbers of elements.
5424 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5425 Alan Hayward <alan.hayward@arm.com>
5426 David Sherwood <david.sherwood@arm.com>
5428 * match.pd: Cope with polynomial numbers of vector elements.
5430 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5431 Alan Hayward <alan.hayward@arm.com>
5432 David Sherwood <david.sherwood@arm.com>
5434 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5435 in a POINTER_PLUS_EXPR.
5437 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5438 Alan Hayward <alan.hayward@arm.com>
5439 David Sherwood <david.sherwood@arm.com>
5441 * omp-simd-clone.c (simd_clone_subparts): New function.
5442 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5443 (ipa_simd_modify_function_body): Likewise.
5445 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5446 Alan Hayward <alan.hayward@arm.com>
5447 David Sherwood <david.sherwood@arm.com>
5449 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5450 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5451 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5452 (expand_vector_condition, vector_element): Likewise.
5453 (subparts_gt): New function.
5454 (get_compute_type): Use subparts_gt.
5455 (count_type_subparts): Delete.
5456 (expand_vector_operations_1): Use subparts_gt instead of
5457 count_type_subparts.
5459 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5460 Alan Hayward <alan.hayward@arm.com>
5461 David Sherwood <david.sherwood@arm.com>
5463 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5464 (vect_compile_time_alias): ...this new function. Do the calculation
5465 on poly_ints rather than trees.
5466 (vect_prune_runtime_alias_test_list): Update call accordingly.
5468 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5469 Alan Hayward <alan.hayward@arm.com>
5470 David Sherwood <david.sherwood@arm.com>
5472 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5474 (vect_schedule_slp_instance): Likewise.
5476 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5477 Alan Hayward <alan.hayward@arm.com>
5478 David Sherwood <david.sherwood@arm.com>
5480 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5481 constant and extern definitions for variable-length vectors.
5482 (vect_get_constant_vectors): Note that the number of units
5483 is known to be constant.
5485 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5486 Alan Hayward <alan.hayward@arm.com>
5487 David Sherwood <david.sherwood@arm.com>
5489 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
5490 of units as polynomial. Choose between WIDE and NARROW based
5493 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5494 Alan Hayward <alan.hayward@arm.com>
5495 David Sherwood <david.sherwood@arm.com>
5497 * tree-vect-stmts.c (simd_clone_subparts): New function.
5498 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5500 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5501 Alan Hayward <alan.hayward@arm.com>
5502 David Sherwood <david.sherwood@arm.com>
5504 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5505 vectors as polynomial. Use build_index_vector for
5508 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5509 Alan Hayward <alan.hayward@arm.com>
5510 David Sherwood <david.sherwood@arm.com>
5512 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5513 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5514 for variable-length vectors.
5515 (vectorizable_mask_load_store): Treat the number of units as
5516 polynomial, asserting that it is constant if the condition has
5517 already been enforced.
5518 (vectorizable_store, vectorizable_load): Likewise.
5520 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5521 Alan Hayward <alan.hayward@arm.com>
5522 David Sherwood <david.sherwood@arm.com>
5524 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5525 of units as polynomial. Punt if we can't tell at compile time
5526 which vector contains the final result.
5528 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5529 Alan Hayward <alan.hayward@arm.com>
5530 David Sherwood <david.sherwood@arm.com>
5532 * tree-vect-loop.c (vectorizable_induction): Treat the number
5533 of units as polynomial. Punt on SLP inductions. Use an integer
5534 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5535 cast of such a series for variable-length floating-point
5538 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5539 Alan Hayward <alan.hayward@arm.com>
5540 David Sherwood <david.sherwood@arm.com>
5542 * tree.h (build_index_vector): Declare.
5543 * tree.c (build_index_vector): New function.
5544 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5545 of units as polynomial, forcibly converting it to a constant if
5546 vectorizable_reduction has already enforced the condition.
5547 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5548 to create a {1,2,3,...} vector.
5549 (vectorizable_reduction): Treat the number of units as polynomial.
5550 Choose vectype_in based on the largest scalar element size rather
5551 than the smallest number of units. Enforce the restrictions
5554 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5555 Alan Hayward <alan.hayward@arm.com>
5556 David Sherwood <david.sherwood@arm.com>
5558 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5559 number of units as polynomial.
5561 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5562 Alan Hayward <alan.hayward@arm.com>
5563 David Sherwood <david.sherwood@arm.com>
5565 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5566 * target.def (autovectorize_vector_sizes): Return the vector sizes
5567 by pointer, using vector_sizes rather than a bitmask.
5568 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5569 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5570 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5572 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5573 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5574 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5575 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5576 * omp-general.c (omp_max_vf): Likewise.
5577 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5578 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5579 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5580 * tree-vect-slp.c (vect_slp_bb): Likewise.
5581 * doc/tm.texi: Regenerate.
5582 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5584 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5585 the vector size as a poly_uint64 rather than an unsigned int.
5586 (current_vector_size): Change from an unsigned int to a poly_uint64.
5587 (get_vectype_for_scalar_type): Update accordingly.
5588 * tree.h (build_truth_vector_type): Take the size and number of
5589 units as a poly_uint64 rather than an unsigned int.
5590 (build_vector_type): Add a temporary overload that takes
5591 the number of units as a poly_uint64 rather than an unsigned int.
5592 * tree.c (make_vector_type): Likewise.
5593 (build_truth_vector_type): Take the number of units as a poly_uint64
5594 rather than an unsigned int.
5596 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5597 Alan Hayward <alan.hayward@arm.com>
5598 David Sherwood <david.sherwood@arm.com>
5600 * target.def (get_mask_mode): Take the number of units and length
5601 as poly_uint64s rather than unsigned ints.
5602 * targhooks.h (default_get_mask_mode): Update accordingly.
5603 * targhooks.c (default_get_mask_mode): Likewise.
5604 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5605 * doc/tm.texi: Regenerate.
5607 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5608 Alan Hayward <alan.hayward@arm.com>
5609 David Sherwood <david.sherwood@arm.com>
5611 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5612 * omp-general.c (omp_max_vf): Likewise.
5613 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5614 (expand_omp_simd): Handle polynomial safelen.
5615 * omp-low.c (omplow_simd_context): Add a default constructor.
5616 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5617 (lower_rec_simd_input_clauses): Update accordingly.
5618 (lower_rec_input_clauses): Likewise.
5620 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5621 Alan Hayward <alan.hayward@arm.com>
5622 David Sherwood <david.sherwood@arm.com>
5624 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5625 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5626 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5627 (vect_analyze_slp_cost): Likewise.
5628 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5629 (vect_model_load_cost): Likewise.
5631 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5632 Alan Hayward <alan.hayward@arm.com>
5633 David Sherwood <david.sherwood@arm.com>
5635 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5636 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5637 from an unsigned int * to a poly_uint64_pod *.
5638 (calculate_unrolling_factor): New function.
5639 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5641 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5642 Alan Hayward <alan.hayward@arm.com>
5643 David Sherwood <david.sherwood@arm.com>
5645 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5646 from an unsigned int to a poly_uint64.
5647 (_loop_vec_info::slp_unrolling_factor): Likewise.
5648 (_loop_vec_info::vectorization_factor): Change from an int
5650 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5651 (vect_get_num_vectors): New function.
5652 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5653 (vect_get_num_copies): Use vect_get_num_vectors.
5654 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5655 to an unsigned int *.
5656 (vect_analyze_data_refs): Change min_vf from an int * to a
5658 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5659 than an unsigned HOST_WIDE_INT.
5660 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5661 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5662 to an unsigned int *.
5663 (vect_analyze_data_ref_dependences): Likewise.
5664 (vect_compute_data_ref_alignment): Handle polynomial vf.
5665 (vect_enhance_data_refs_alignment): Likewise.
5666 (vect_prune_runtime_alias_test_list): Likewise.
5667 (vect_shift_permute_load_chain): Likewise.
5668 (vect_supportable_dr_alignment): Likewise.
5669 (dependence_distance_ge_vf): Take the vectorization factor as a
5670 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5671 (vect_analyze_data_refs): Change min_vf from an int * to a
5673 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5674 vfm1 as a poly_uint64 rather than an int. Make the same change
5675 for the returned bound_scalar.
5676 (vect_gen_vector_loop_niters): Handle polynomial vf.
5677 (vect_do_peeling): Likewise. Update call to
5678 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5679 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5681 * tree-vect-loop.c (vect_determine_vectorization_factor)
5682 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5683 (vect_get_known_peeling_cost): Likewise.
5684 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5685 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5686 (vect_transform_loop): Likewise. Use the lowest possible VF when
5687 updating the upper bounds of the loop.
5688 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5690 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5691 polynomial unroll factors.
5692 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5693 (vect_make_slp_decision): Likewise.
5694 (vect_supported_load_permutation_p): Likewise, and polynomial
5696 (vect_analyze_slp_cost): Handle polynomial vf.
5697 (vect_slp_analyze_node_operations): Likewise.
5698 (vect_slp_analyze_bb_1): Likewise.
5699 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5700 than an unsigned HOST_WIDE_INT.
5701 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5702 (vectorizable_load): Handle polynomial vf.
5703 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5705 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5707 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5708 Alan Hayward <alan.hayward@arm.com>
5709 David Sherwood <david.sherwood@arm.com>
5711 * match.pd: Handle bit operations involving three constants
5712 and try to fold one pair.
5714 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5716 * tree-vect-loop-manip.c: Include gimple-fold.h.
5717 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5718 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5719 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5720 Add a path that uses a step of VF instead of 1, but disable it
5722 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5723 and niters_no_overflow parameters. Update calls to
5724 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5725 Create a new SSA name if the latter choses to use a ste other
5726 than zero, and return it via niters_vector_mult_vf_var.
5727 * tree-vect-loop.c (vect_transform_loop): Update calls to
5728 vect_do_peeling, vect_gen_vector_loop_niters and
5729 slpeel_make_loop_iterate_ntimes.
5730 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5731 (vect_gen_vector_loop_niters): Update declarations after above changes.
5733 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5735 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5736 128-bit round to integer instructions.
5737 (ceil<mode>2): Likewise.
5738 (btrunc<mode>2): Likewise.
5739 (round<mode>2): Likewise.
5741 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5743 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5744 unaligned VSX load/store on P8/P9.
5745 (expand_block_clear): Allow the use of unaligned VSX
5746 load/store on P8/P9.
5748 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5750 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5752 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5753 swap associated with both a load and a store.
5755 2018-01-02 Andrew Waterman <andrew@sifive.com>
5757 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5758 * config/riscv/riscv.md (clear_cache): Use it.
5760 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5762 * web.c: Remove out-of-date comment.
5764 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5766 * expr.c (fixup_args_size_notes): Check that any existing
5767 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5768 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5769 (emit_single_push_insn): ...here.
5771 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5773 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5774 (const_vector_encoded_nelts): New function.
5775 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5776 (const_vector_int_elt, const_vector_elt): Declare.
5777 * emit-rtl.c (const_vector_int_elt_1): New function.
5778 (const_vector_elt): Likewise.
5779 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5780 of CONST_VECTOR_ELT.
5782 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5784 * expr.c: Include rtx-vector-builder.h.
5785 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5786 directly on the tree encoding.
5787 (const_vector_from_tree): Likewise.
5788 * optabs.c: Include rtx-vector-builder.h.
5789 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5790 sequence of "u" values.
5791 * vec-perm-indices.c: Include rtx-vector-builder.h.
5792 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5793 directly on the vec_perm_indices encoding.
5795 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5797 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5798 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5799 * rtx-vector-builder.h: New file.
5800 * rtx-vector-builder.c: Likewise.
5801 * rtl.h (rtx_def::u2): Add a const_vector field.
5802 (CONST_VECTOR_NPATTERNS): New macro.
5803 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5804 (CONST_VECTOR_DUPLICATE_P): Likewise.
5805 (CONST_VECTOR_STEPPED_P): Likewise.
5806 (CONST_VECTOR_ENCODED_ELT): Likewise.
5807 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5808 (unwrap_const_vec_duplicate): Likewise.
5809 (const_vec_series_p): Check for a non-duplicated vector encoding.
5810 Say that the function only returns true for integer vectors.
5811 * emit-rtl.c: Include rtx-vector-builder.h.
5812 (gen_const_vec_duplicate_1): Delete.
5813 (gen_const_vector): Call gen_const_vec_duplicate instead of
5814 gen_const_vec_duplicate_1.
5815 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5816 (gen_const_vec_duplicate): Use rtx_vector_builder.
5817 (gen_const_vec_series): Likewise.
5818 (gen_rtx_CONST_VECTOR): Likewise.
5819 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5820 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5821 Build a new vector rather than modifying a CONST_VECTOR in-place.
5822 (handle_special_swappables): Update call accordingly.
5823 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5824 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5825 Build a new vector rather than modifying a CONST_VECTOR in-place.
5826 (handle_special_swappables): Update call accordingly.
5828 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5830 * simplify-rtx.c (simplify_const_binary_operation): Use
5831 CONST_VECTOR_ELT instead of XVECEXP.
5833 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5835 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5836 the selector elements to be different from the data elements
5837 if the selector is a VECTOR_CST.
5838 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5839 ssizetype for the selector.
5841 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5843 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5844 before testing each element individually.
5845 * tree-vect-generic.c (lower_vec_perm): Likewise.
5847 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5849 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5850 * selftest-run-tests.c (selftest::run_tests): Call it.
5851 * vector-builder.h (vector_builder::operator ==): New function.
5852 (vector_builder::operator !=): Likewise.
5853 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5854 (vec_perm_indices::all_from_input_p): New function.
5855 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5856 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5857 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5858 instead of reading the VECTOR_CST directly. Detect whether both
5859 vector inputs are the same before constructing the vec_perm_indices,
5860 and update the number of inputs argument accordingly. Use the
5861 utility functions added above. Only construct sel2 if we need to.
5863 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5865 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5866 the broadcast of the low byte.
5867 (expand_mult_highpart): Use an explicit encoding for the permutes.
5868 * optabs-query.c (can_mult_highpart_p): Likewise.
5869 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5870 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5871 (vectorizable_bswap): Likewise.
5872 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5873 explicit encoding for the power-of-2 permutes.
5874 (vect_permute_store_chain): Likewise.
5875 (vect_grouped_load_supported): Likewise.
5876 (vect_permute_load_chain): Likewise.
5878 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5880 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5881 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5882 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5883 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5884 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5885 (vect_gen_perm_mask_any): Likewise.
5887 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5889 * int-vector-builder.h: New file.
5890 * vec-perm-indices.h: Include int-vector-builder.h.
5891 (vec_perm_indices): Redefine as an int_vector_builder.
5892 (auto_vec_perm_indices): Delete.
5893 (vec_perm_builder): Redefine as a stand-alone class.
5894 (vec_perm_indices::vec_perm_indices): New function.
5895 (vec_perm_indices::clamp): Likewise.
5896 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5897 (vec_perm_indices::new_vector): New function.
5898 (vec_perm_indices::new_expanded_vector): Update for new
5899 vec_perm_indices class.
5900 (vec_perm_indices::rotate_inputs): New function.
5901 (vec_perm_indices::all_in_range_p): Operate directly on the
5902 encoded form, without computing elided elements.
5903 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5904 encoding. Update for new vec_perm_indices class.
5905 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5906 the given vec_perm_builder.
5907 (expand_vec_perm_var): Update vec_perm_builder constructor.
5908 (expand_mult_highpart): Use vec_perm_builder instead of
5909 auto_vec_perm_indices.
5910 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5911 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5912 or double series encoding as appropriate.
5913 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5914 vec_perm_indices instead of auto_vec_perm_indices.
5915 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5916 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5917 (vect_permute_store_chain): Likewise.
5918 (vect_grouped_load_supported): Likewise.
5919 (vect_permute_load_chain): Likewise.
5920 (vect_shift_permute_load_chain): Likewise.
5921 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5922 (vect_transform_slp_perm_load): Likewise.
5923 (vect_schedule_slp_instance): Likewise.
5924 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5925 (vectorizable_mask_load_store): Likewise.
5926 (vectorizable_bswap): Likewise.
5927 (vectorizable_store): Likewise.
5928 (vectorizable_load): Likewise.
5929 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5930 vec_perm_indices instead of auto_vec_perm_indices. Use
5931 tree_to_vec_perm_builder to read the vector from a tree.
5932 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5933 vec_perm_builder instead of a vec_perm_indices.
5934 (have_whole_vector_shift): Use vec_perm_builder and
5935 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5936 truncation to calc_vec_perm_mask_for_shift.
5937 (vect_create_epilog_for_reduction): Likewise.
5938 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5939 from auto_vec_perm_indices to vec_perm_indices.
5940 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5941 instead of changing individual elements.
5942 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5943 the vector in d.perm.
5944 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5945 from auto_vec_perm_indices to vec_perm_indices.
5946 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5947 instead of changing individual elements.
5948 (arm_vectorize_vec_perm_const): Use new_vector to install
5949 the vector in d.perm.
5950 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5951 Update vec_perm_builder constructor.
5952 (rs6000_expand_interleave): Likewise.
5953 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5954 (rs6000_expand_interleave): Likewise.
5956 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5958 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5959 to qimode could truncate the indices.
5960 * optabs.c (expand_vec_perm_var): Likewise.
5962 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5964 * Makefile.in (OBJS): Add vec-perm-indices.o.
5965 * vec-perm-indices.h: New file.
5966 * vec-perm-indices.c: Likewise.
5967 * target.h (vec_perm_indices): Replace with a forward class
5969 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5970 * optabs.h: Include vec-perm-indices.h.
5971 (expand_vec_perm): Delete.
5972 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5973 (expand_vec_perm_const): Declare.
5974 * target.def (vec_perm_const_ok): Replace with...
5975 (vec_perm_const): ...this new hook.
5976 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5977 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5978 * doc/tm.texi: Regenerate.
5979 * optabs.def (vec_perm_const): Delete.
5980 * doc/md.texi (vec_perm_const): Likewise.
5981 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5982 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5983 expand_vec_perm for constant permutation vectors. Assert that
5984 the mode of variable permutation vectors is the integer equivalent
5985 of the mode that is being permuted.
5986 * optabs-query.h (selector_fits_mode_p): Declare.
5987 * optabs-query.c: Include vec-perm-indices.h.
5988 (selector_fits_mode_p): New function.
5989 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5990 is defined, instead of checking whether the vec_perm_const_optab
5991 exists. Use targetm.vectorize.vec_perm_const instead of
5992 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5993 fit in the vector mode before using a variable permute.
5994 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5995 vec_perm_indices instead of an rtx.
5996 (expand_vec_perm): Replace with...
5997 (expand_vec_perm_const): ...this new function. Take the selector
5998 as a vec_perm_indices rather than an rtx. Also take the mode of
5999 the selector. Update call to shift_amt_for_vec_perm_mask.
6000 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6001 Use vec_perm_indices::new_expanded_vector to expand the original
6002 selector into bytes. Check whether the indices fit in the vector
6003 mode before using a variable permute.
6004 (expand_vec_perm_var): Make global.
6005 (expand_mult_highpart): Use expand_vec_perm_const.
6006 * fold-const.c: Includes vec-perm-indices.h.
6007 * tree-ssa-forwprop.c: Likewise.
6008 * tree-vect-data-refs.c: Likewise.
6009 * tree-vect-generic.c: Likewise.
6010 * tree-vect-loop.c: Likewise.
6011 * tree-vect-slp.c: Likewise.
6012 * tree-vect-stmts.c: Likewise.
6013 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6015 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6016 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6017 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6018 (aarch64_vectorize_vec_perm_const): ...this new function.
6019 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6020 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6021 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6022 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6023 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6024 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6025 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6027 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6028 check for NEON modes.
6029 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6030 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6031 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6032 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6034 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6035 the old VEC_PERM_CONST conditions.
6036 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6037 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6038 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6039 (ia64_vectorize_vec_perm_const_ok): Merge into...
6040 (ia64_vectorize_vec_perm_const): ...this new function.
6041 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6042 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6043 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6044 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6045 * config/mips/mips.c (mips_expand_vec_perm_const)
6046 (mips_vectorize_vec_perm_const_ok): Merge into...
6047 (mips_vectorize_vec_perm_const): ...this new function.
6048 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6049 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6050 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6051 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6052 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6053 (rs6000_expand_vec_perm_const): Delete.
6054 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6056 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6057 (altivec_expand_vec_perm_const_le): Take each operand individually.
6058 Operate on constant selectors rather than rtxes.
6059 (altivec_expand_vec_perm_const): Likewise. Update call to
6060 altivec_expand_vec_perm_const_le.
6061 (rs6000_expand_vec_perm_const): Delete.
6062 (rs6000_vectorize_vec_perm_const_ok): Delete.
6063 (rs6000_vectorize_vec_perm_const): New function.
6064 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6065 an element count and rtx array.
6066 (rs6000_expand_extract_even): Update call accordingly.
6067 (rs6000_expand_interleave): Likewise.
6068 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6069 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6070 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6071 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6072 (rs6000_expand_vec_perm_const): Delete.
6073 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6074 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6075 (altivec_expand_vec_perm_const_le): Take each operand individually.
6076 Operate on constant selectors rather than rtxes.
6077 (altivec_expand_vec_perm_const): Likewise. Update call to
6078 altivec_expand_vec_perm_const_le.
6079 (rs6000_expand_vec_perm_const): Delete.
6080 (rs6000_vectorize_vec_perm_const_ok): Delete.
6081 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6082 reference to the SPE evmerge intructions.
6083 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6084 an element count and rtx array.
6085 (rs6000_expand_extract_even): Update call accordingly.
6086 (rs6000_expand_interleave): Likewise.
6087 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6088 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6090 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6092 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6094 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6095 vector mode and that that mode matches the mode of the data
6097 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6098 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6099 directly using expand_vec_perm_1 when forcing selectors into
6101 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6103 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6105 * optabs-query.h (can_vec_perm_p): Delete.
6106 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6107 * optabs-query.c (can_vec_perm_p): Split into...
6108 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6109 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6110 particular selector is valid.
6111 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6112 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6113 (vect_grouped_load_supported): Likewise.
6114 (vect_shift_permute_load_chain): Likewise.
6115 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6116 (vect_transform_slp_perm_load): Likewise.
6117 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6118 (vectorizable_bswap): Likewise.
6119 (vect_gen_perm_mask_checked): Likewise.
6120 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6121 implementations of variable permutation vectors into account
6122 when deciding which selector to use.
6123 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6124 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6125 with a false third argument.
6126 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6127 to test whether the constant selector is valid and can_vec_perm_var_p
6128 to test whether a variable selector is valid.
6130 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6132 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6133 * optabs-query.c (can_vec_perm_p): Likewise.
6134 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6135 instead of vec_perm_indices.
6136 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6137 (vect_gen_perm_mask_checked): Likewise,
6138 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6139 (vect_gen_perm_mask_checked): Likewise,
6141 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6143 * optabs-query.h (qimode_for_vec_perm): Declare.
6144 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6145 (qimode_for_vec_perm): ...this new function.
6146 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6148 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6150 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6151 does not have a conditional at the top.
6153 2018-01-02 Richard Biener <rguenther@suse.de>
6155 * ipa-inline.c (big_speedup_p): Fix expression.
6157 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6160 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6163 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6167 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6168 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6169 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6170 cond_taken_branch_cost 3->4.
6172 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6174 PR tree-optimization/83581
6175 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6176 TODO_cleanup_cfg if any changes have been made.
6179 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6180 convert_modes if target mode has the right side, but different mode
6184 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6185 last argument when extracting from CONCAT. If either from_real or
6186 from_imag is NULL, use expansion through memory. If result is not
6187 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6188 the parts directly to inner mode, if even that fails, use expansion
6192 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6193 check for bswap in mode rather than HImode and use that in expand_unop
6196 Copyright (C) 2018 Free Software Foundation, Inc.
6198 Copying and distribution of this file, with or without modification,
6199 are permitted in any medium without royalty provided the copyright
6200 notice and this notice are preserved.