1 2018-02-07 Iain Sandoe <iain@codesourcery.com>
3 * config/rs6000/altivec.md (*restore_world): Remove LR use.
4 * config/rs6000/predicates.md (restore_world_operation): Adjust op
7 2018-02-07 Michael Meissner <meissner@linux.vnet.ibm.com>
9 * doc/install.texi (Configuration): Document the
10 --with-long-double-format={ibm,ieee} PowerPC configuration
14 * config/rs6000/rs6000.md (fix_trunc<SFDF:mode><QHI:mode>2):
15 Convert from define_expand to be define_insn_and_split. Rework
16 float/double/_Float128 conversions to QI/HI/SImode to work with
17 both ISA 2.07 (power8) or ISA 3.0 (power9). Fix regression where
18 conversions to QI/HImode types did a store and then a load to
19 truncate the value. For conversions to VSX registers, don't split
20 the insn, instead emit the code directly. Use the code iterator
21 any_fix to combine signed and unsigned conversions.
22 (fix<uns>_trunc<SFDF:mode>si2_p8): Likewise.
23 (fixuns_trunc<SFDF:mode><QHI:mode>2): Likewise.
24 (fix_trunc<IEEE128:mode><QHI:mode>2): Likewise.
25 (fix<uns>_trunc<SFDF:mode><QHI:mode>2): Likewise.
26 (fix_<mode>di2_hw): Likewise.
27 (fixuns_<mode>di2_hw): Likewise.
28 (fix_<mode>si2_hw): Likewise.
29 (fixuns_<mode>si2_hw): Likewise.
30 (fix<uns>_<IEEE128:mode><SDI:mode>2_hw): Likewise.
31 (fix<uns>_trunc<IEEE128:mode><QHI:mode>2): Likewise.
32 (fctiw<u>z_<mode>_smallint): Rename fctiw<u>z_<mode>_smallint to
33 fix<uns>_trunc<SFDF:mode>si2_p8.
34 (fix_trunc<SFDF:mode><QHI:mode>2_internal): Delete, no longer
36 (fixuns_trunc<SFDF:mode><QHI:mode>2_internal): Likewise.
37 (fix<uns>_<mode>_mem): Likewise.
38 (fctiw<u>z_<mode>_mem): Likewise.
39 (fix<uns>_<mode>_mem): Likewise.
40 (fix<uns>_trunc<SFDF:mode><QHSI:mode>2_mem): On ISA 3.0, prevent
41 the register allocator from doing a direct move to the GPRs to do
42 a store, and instead use the ISA 3.0 store byte/half-word from
43 vector register instruction. For IEEE 128-bit floating point,
44 also optimize stores of 32-bit ints.
45 (fix<uns>_trunc<IEEE128:mode><QHSI:mode>2_mem): Likewise.
47 2018-02-07 Alan Hayward <alan.hayward@arm.com>
49 * genextract.c (push_pathstr_operand): New function to support
51 (walk_rtx): Call push_pathstr_operand.
52 (print_path): Support [a-zA-Z].
54 2018-02-07 Richard Biener <rguenther@suse.de>
56 PR tree-optimization/84037
57 * tree-vectorizer.h (struct _loop_vec_info): Add ivexpr_map member.
58 (cse_and_gimplify_to_preheader): Declare.
59 (vect_get_place_in_interleaving_chain): Likewise.
60 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
62 (_loop_vec_info::~_loop_vec_info): Delete it.
63 (cse_and_gimplify_to_preheader): New function.
64 * tree-vect-slp.c (vect_get_place_in_interleaving_chain): Export.
65 * tree-vect-stmts.c (vectorizable_store): CSE base and steps.
66 (vectorizable_load): Likewise. For grouped stores always base
67 the IV on the first element.
68 * tree-vect-loop-manip.c (vect_loop_versioning): Unshare versioning
69 condition before gimplifying.
71 2018-02-07 Jakub Jelinek <jakub@redhat.com>
73 * tree-eh.c (operation_could_trap_helper_p): Ignore honor_trapv for
74 *DIV_EXPR and *MOD_EXPR.
76 2018-02-07 H.J. Lu <hongjiu.lu@intel.com>
79 * config/i386/i386.c (ix86_option_override_internal): Mask out
80 the CF_SET bit when checking -fcf-protection.
82 2018-02-07 Tom de Vries <tom@codesourcery.com>
85 * omp-expand.c (expand_oacc_collapse_init): Ensure diff_type is large
88 2018-02-07 Richard Biener <rguenther@suse.de>
90 PR tree-optimization/84204
91 * tree-chrec.c (chrec_fold_plus_1): Remove size limiting in
94 PR tree-optimization/84205
95 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Also
96 special-case isl_ast_op_zdiv_r.
98 PR tree-optimization/84223
99 * graphite-scop-detection.c (gather_bbs::before_dom_children):
100 Only add conditions from within the region.
101 (gather_bbs::after_dom_children): Adjust.
103 2018-02-07 Georg-Johann Lay <avr@gjlay.de>
106 * config/avr/avr.h (GENERAL_REGNO_P, GENERAL_REG_P): New macros.
107 * config/avr/avr.md: Only post-reload split REG-REG moves if
108 either register is REGERAL_REG_P.
110 2018-02-07 Jakub Jelinek <jakub@redhat.com>
112 PR tree-optimization/84235
113 * tree-ssa-scopedtables.c
114 (avail_exprs_stack::simplify_binary_operation): Fir MINUS_EXPR, punt
115 if the subtraction is performed in floating point type where NaNs are
116 honored. For *DIV_EXPR, punt for ALL_FRACT_MODE_Ps where we can't
117 build 1. Formatting fix.
119 2018-02-06 Jakub Jelinek <jakub@redhat.com>
122 * config/i386/i386.c (rest_of_insert_endbranch): Only skip
123 NOTE_INSN_CALL_ARG_LOCATION after a call, not anything else,
124 and skip it regardless of bb boundaries. Use CALL_P macro,
125 don't test INSN_P (insn) together with CALL_P or JUMP_P check
126 unnecessarily, formatting fix.
128 2018-02-06 Michael Collison <michael.collison@arm.com>
130 * config/arm/thumb2.md:
131 (*thumb2_mov_negscc): Split only if TARGET_THUMB2 && !arm_restrict_it.
132 (*thumb_mov_notscc): Ditto.
134 2018-02-06 Michael Meissner <meissner@linux.vnet.ibm.com>
137 * config/rs6000/rs6000.md (su code attribute): Use "u" for
138 unsigned_fix, not "s".
140 2018-02-06 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
142 * configure.ac (gcc_fn_eh_frame_ro): New function.
143 (gcc_cv_as_cfi_directive): Check both 32 and 64-bit assembler for
144 correct .eh_frame permissions.
145 * configure: Regenerate.
147 2018-02-06 Andrew Jenner <andrew@codeourcery.com>
149 * doc/invoke.texi: Add section for the PowerPC SPE backend. Remove
152 2018-02-06 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
154 * config/rs6000/rs6000.c (rs6000_option_override_internal):
155 Display warning message for -mno-speculate-indirect-jumps.
157 2018-02-06 Andrew Jenner <andrew@codesourcery.com>
159 * config/powerpcspe/powerpcspe.opt: (msimple-fpu, mfpu) Add
161 * config/powerpcspe/sysv4.opt (mbit-align): Likewise.
163 2018-02-06 Aldy Hernandez <aldyh@redhat.com>
165 PR tree-optimization/84225
166 * tree-eh.c (find_trapping_overflow): Only call
167 operation_no_trapping_overflow when ANY_INTEGRAL_TYPE_P.
169 2018-02-06 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
172 * config/i386/i386.c: Reimplement the check of possible options
173 -mibt/-mshstk conbination. Change error messages.
174 * doc/invoke.texi: Fix a typo: remove extra '='.
176 2018-02-06 Marek Polacek <polacek@redhat.com>
178 PR tree-optimization/84228
179 * tree-ssa-strlen.c (maybe_diag_stxncpy_trunc): Skip debug statements.
181 2018-02-06 Tamar Christina <tamar.christina@arm.com>
184 * config/arm/arm.c (arm_print_asm_arch_directives): Record already
185 emitted arch directives.
186 * config/arm/arm-c.c (arm_cpu_builtins): Undefine __ARM_ARCH and
187 __ARM_FEATURE_COPROC before changing architectures.
189 2018-02-06 Richard Biener <rguenther@suse.de>
191 * config/i386/i386.c (print_reg): Fix typo.
192 (ix86_loop_unroll_adjust): Do not unroll beyond the original nunroll.
194 2018-02-06 Eric Botcazou <ebotcazou@adacore.com>
196 * configure: Regenerate.
198 2018-02-05 Martin Sebor <msebor@redhat.com>
200 PR tree-optimization/83369
201 * tree-ssa-ccp.c (pass_post_ipa_warn::execute): Use %G to print
204 2018-02-05 Martin Liska <mliska@suse.cz>
206 * doc/invoke.texi: Cherry-pick upstream r323995.
208 2018-02-05 Richard Sandiford <richard.sandiford@linaro.org>
210 * ira.c (ira_init_register_move_cost): Adjust comment.
212 2018-02-05 Martin Liska <mliska@suse.cz>
214 PR gcov-profile/84137
215 * doc/gcov.texi: Fix typo in documentation.
217 2018-02-05 Martin Liska <mliska@suse.cz>
219 PR gcov-profile/83879
220 * doc/gcov.texi: Document necessity of --dynamic-list-data when
221 using dlopen functionality.
223 2018-02-05 Olga Makhotina <olga.makhotina@intel.com>
225 * config/i386/avx512dqintrin.h (_mm_mask_range_sd, _mm_maskz_range_sd,
226 _mm_mask_range_round_sd, _mm_maskz_range_round_sd, _mm_mask_range_ss,
227 _mm_maskz_range_ss, _mm_mask_range_round_ss,
228 _mm_maskz_range_round_ss): New intrinsics.
229 (__builtin_ia32_rangesd128_round)
230 (__builtin_ia32_rangess128_round): Remove.
231 (__builtin_ia32_rangesd128_mask_round,
232 __builtin_ia32_rangess128_mask_round): New builtins.
233 * config/i386/i386-builtin.def (__builtin_ia32_rangesd128_round,
234 __builtin_ia32_rangess128_round): Remove.
235 (__builtin_ia32_rangesd128_mask_round,
236 __builtin_ia32_rangess128_mask_round): New builtins.
237 * config/i386/sse.md (ranges<mode><round_saeonly_name>): Renamed to ...
238 (ranges<mode><mask_scalar_name><round_saeonly_scalar_name>): ... this.
239 ((match_operand:VF_128 2 "<round_saeonly_nimm_predicate>"
240 "<round_saeonly_constraint>")): Changed to ...
241 ((match_operand:VF_128 2 "<round_saeonly_scalar_nimm_predicate>"
242 "<round_saeonly_scalar_constraint>")): ... this.
243 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_op4>%2, %1, %0|
244 %0, %1, %2<round_saeonly_op4>, %3}"): Changed to ...
245 ("vrange<ssescalarmodesuffix>\t{%3, <round_saeonly_scalar_mask_op4>%2,
246 %1, %0<mask_scalar_operand4>|%0<mask_scalar_operand4>, %1,
247 %2<round_saeonly_scalar_mask_op4>, %3}"): ... this.
249 2018-02-02 Andrew Jenner <andrew@codesourcery.com>
251 * config/powerpcspe/powerpcspe.opt: Add Undocumented to irrelevant
253 * config/powerpcspe/powerpcspe-tables.opt (rs6000_cpu_opt_value):
254 Remove all values except native, 8540 and 8548.
256 2018-02-02 H.J. Lu <hongjiu.lu@intel.com>
258 * config/i386/i386.c (ix86_output_function_return): Pass
259 INVALID_REGNUM, instead of -1, as invalid register number to
260 indirect_thunk_name and output_indirect_thunk.
262 2018-02-02 Julia Koval <julia.koval@intel.com>
264 * config.gcc: Add -march=icelake.
265 * config/i386/driver-i386.c (host_detect_local_cpu): Detect icelake.
266 * config/i386/i386-c.c (ix86_target_macros_internal): Handle icelake.
267 * config/i386/i386.c (processor_costs): Add m_ICELAKE.
268 (PTA_ICELAKE, PTA_AVX512VNNI, PTA_GFNI, PTA_VAES, PTA_AVX512VBMI2,
269 PTA_VPCLMULQDQ, PTA_RDPID, PTA_AVX512BITALG): New.
270 (processor_target_table): Add icelake.
271 (ix86_option_override_internal): Handle new PTAs.
272 (get_builtin_code_for_version): Handle icelake.
273 (M_INTEL_COREI7_ICELAKE): New.
274 (fold_builtin_cpu): Handle icelake.
275 * config/i386/i386.h (TARGET_ICELAKE, PROCESSOR_ICELAKE): New.
276 * doc/invoke.texi: Add -march=icelake.
278 2018-02-02 Julia Koval <julia.koval@intel.com>
280 * config/i386/i386.c (ix86_option_override_internal): Change flags type
282 * wide-int-bitmask.h: New.
284 2018-02-02 Igor Tsimbalist <igor.v.tsimbalist@intel.com>
287 * config/i386/i386.md: Replace Pmode with word_mode in
288 builtin_setjmp_setup and builtin_longjmp to support x32.
290 2018-02-01 Peter Bergner <bergner@vnet.ibm.com>
294 * config/rs6000/driver-rs6000.c: #include "diagnostic.h".
296 (rs6000_supported_cpu_names): New static variable.
297 (linux_cpu_translation_table): Likewise.
298 (elf_platform) <cpu>: Define new static variable and use it.
299 Translate kernel AT_PLATFORM name to canonical name if needed.
300 Error if platform name is unknown.
302 2018-02-01 Aldy Hernandez <aldyh@redhat.com>
305 * config/pa/predicates.md (base14_operand): Handle E_VOIDmode.
307 2018-02-01 Jeff Law <law@redhat.com>
310 * config/i386/i386.c (release_scratch_register_on_entry): Add new
311 OFFSET and RELEASE_VIA_POP arguments. Use SP+OFFSET to restore
312 the scratch if RELEASE_VIA_POP is false.
313 (ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
314 If we have to save a temporary register, decrement SIZE appropriately.
315 Pass new arguments to release_scratch_register_on_entry.
316 (ix86_adjust_stack_and_probe): Likewise.
317 (ix86_emit_probe_stack_range): Pass new arguments to
318 release_scratch_register_on_entry.
320 2018-02-01 Uros Bizjak <ubizjak@gmail.com>
322 PR rtl-optimization/84157
323 * combine.c (change_zero_ext): Use REG_P predicate in
324 front of HARD_REGISTER_P predicate.
326 2018-02-01 Georg-Johann Lay <avr@gjlay.de>
328 * config/avr/avr.c (avr_option_override): Move disabling of
329 -fdelete-null-pointer-checks to...
330 * common/config/avr/avr-common.c (avr_option_optimization_table):
333 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
335 PR tree-optimization/81635
336 * tree-data-ref.c (split_constant_offset_1): For types that
337 wrap on overflow, try to use range info to prove that wrapping
340 2018-02-01 Renlin Li <renlin.li@arm.com>
343 * config/aarch64/aarch64.c (aarch64_class_max_nregs): Handle
345 (aarch64_register_move_cost): Likewise.
346 * config/aarch64/aarch64.h (reg_class): Rename CALLER_SAVE_REGS to
348 (REG_CLASS_NAMES): Likewise.
349 (REG_CLASS_CONTENTS): Rename CALLER_SAVE_REGS to
350 TAILCALL_ADDR_REGS. Remove IP registers.
351 * config/aarch64/aarch64.md (Ucs): Update register constraint.
353 2018-02-01 Richard Biener <rguenther@suse.de>
355 * domwalk.h (dom_walker::dom_walker): Add additional constructor
356 for specifying RPO order and allow NULL for that.
357 * domwalk.c (dom_walker::dom_walker): Likewise.
358 (dom_walker::walk): Handle NULL RPO order.
359 * tree-into-ssa.c (rewrite_dom_walker): Do not walk dom children
361 (rewrite_update_dom_walker): Likewise.
362 (mark_def_dom_walker): Likewise.
364 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
366 * config/aarch64/aarch64-protos.h (aarch64_split_sve_subreg_move)
367 (aarch64_maybe_expand_sve_subreg_move): Declare.
368 * config/aarch64/aarch64.md (UNSPEC_REV_SUBREG): New unspec.
369 * config/aarch64/predicates.md (aarch64_any_register_operand): New
371 * config/aarch64/aarch64-sve.md (mov<mode>): Optimize subreg moves
372 that are semantically a reverse operation.
373 (*aarch64_sve_mov<mode>_subreg_be): New pattern.
374 * config/aarch64/aarch64.c (aarch64_maybe_expand_sve_subreg_move):
375 (aarch64_replace_reg_mode, aarch64_split_sve_subreg_move): New
377 (aarch64_can_change_mode_class): For big-endian, forbid changes
378 between two SVE modes if they have different element sizes.
380 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
382 * config/aarch64/aarch64.c (aarch64_expand_sve_const_vector): Prefer
383 the TImode handling for big-endian targets.
385 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
387 * config/aarch64/aarch64-sve.md (sve_ld1rq): Replace with...
388 (*sve_ld1rq<Vesize>): ... this new pattern. Handle all element sizes,
390 * config/aarch64/aarch64.c (aarch64_expand_sve_widened_duplicate):
391 Remove BSWAP handing for big-endian targets and use the form of
392 LD1RQ appropariate for the mode.
394 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
396 * config/aarch64/aarch64.c (aarch64_simd_valid_immediate): Handle
397 all CONST_VECTOR_DUPLICATE_P vectors, not just those with a single
400 2018-02-01 Richard Sandiford <richard.sandiford@linaro.org>
403 * config/aarch64/aarch64.c (aarch64_secondary_reload): Tighten
404 check for operands that need to go through aarch64_sve_reload_be.
406 2018-02-01 Jakub Jelinek <jakub@redhat.com>
408 PR tree-optimization/81661
409 PR tree-optimization/84117
410 * tree-eh.h (rewrite_to_non_trapping_overflow): Declare.
411 * tree-eh.c: Include gimplify.h.
412 (find_trapping_overflow, replace_trapping_overflow,
413 rewrite_to_non_trapping_overflow): New functions.
414 * tree-vect-loop.c: Include tree-eh.h.
415 (vect_get_loop_niters): Use rewrite_to_non_trapping_overflow.
416 * tree-data-ref.c: Include tree-eh.h.
417 (get_segment_min_max): Use rewrite_to_non_trapping_overflow.
419 2018-01-31 Uros Bizjak <ubizjak@gmail.com>
421 PR rtl-optimization/84123
422 * combine.c (change_zero_ext): Check if hard register satisfies
423 can_change_dest_mode before calling gen_lowpart_SUBREG.
425 2018-01-31 Vladimir Makarov <vmakarov@redhat.com>
428 * ira.c (ira_init_register_move_cost): Remove assert.
430 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
432 PR rtl-optimization/84071
433 * doc/tm.texi.in (WORD_REGISTER_OPERATIONS): Add explicit case.
434 * doc/tm.texi: Regenerate.
436 2018-01-31 Richard Biener <rguenther@suse.de>
438 PR tree-optimization/84132
439 * tree-data-ref.c (analyze_miv_subscript): Properly
440 check whether evolution_function_is_affine_multivariate_p
441 before calling gcd_of_steps_may_divide_p.
443 2018-01-31 Julia Koval <julia.koval@intel.com>
446 * config/i386/i386.c (ix86_expand_builtin): Handle IX86_BUILTIN_RDPID.
447 * config/i386/i386.md (rdpid_rex64) New.
448 (rdpid): Make 32bit only.
450 2018-01-29 Aldy Hernandez <aldyh@redhat.com>
453 * tree-pretty-print.c (dump_generic_node): Handle a TYPE_NAME with
454 an IDENTIFIER_NODE for FUNCTION_TYPE's.
456 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
459 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
461 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
463 2018-01-31 Eric Botcazou <ebotcazou@adacore.com>
465 PR rtl-optimization/84071
466 * combine.c (record_dead_and_set_regs_1): Record the source unmodified
467 for a paradoxical SUBREG on a WORD_REGISTER_OPERATIONS target.
469 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
471 * config/arc/arc.c (arc_handle_aux_attribute): New function.
472 (arc_attribute_table): Add 'aux' attribute.
473 (arc_in_small_data_p): Consider aux like variables.
474 (arc_is_aux_reg_p): New function.
475 (arc_asm_output_aligned_decl_local): Ignore 'aux' like variables.
476 (arc_get_aux_arg): New function.
477 (prepare_move_operands): Handle aux-register access.
478 (arc_handle_aux_attribute): New function.
479 * doc/extend.texi (ARC Variable attributes): Add subsection.
481 2018-01-31 Claudiu Zissulescu <claziss@synopsys.com>
483 * config/arc/arc-protos.h (arc_is_uncached_mem_p): Function proto.
484 * config/arc/arc.c (arc_handle_uncached_attribute): New function.
485 (arc_attribute_table): Add 'uncached' attribute.
486 (arc_print_operand): Print '.di' flag for uncached memory
488 (arc_in_small_data_p): Do not consider for small data the uncached
490 (arc_is_uncached_mem_p): New function.
491 * config/arc/predicates.md (compact_store_memory_operand): Check
492 for uncached memory accesses.
493 (nonvol_nonimm_operand): Likewise.
494 * gcc/doc/extend.texi (ARC Type Attribute): New subsection.
496 2018-01-31 Jakub Jelinek <jakub@redhat.com>
499 * common.opt (falign-functions=, falign-jumps=, falign-labels=,
500 falign-loops=): Add Optimization flag.
502 2018-01-30 Jeff Law <law@redhat.com>
505 * i386.c (ix86_adjust_stack_and_probe_stack_clash): New argument
506 INT_REGISTERS_SAVED. Check it prior to calling
507 get_scratch_register_on_entry.
508 (ix86_adjust_stack_and_probe): Similarly.
509 (ix86_emit_probe_stack_range): Similarly.
510 (ix86_expand_prologue): Corresponding changes.
512 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
515 * config/sol2.h (STARTFILE_ARCH_SPEC): Use -std=c*,
516 -std=iso9899:199409 instead of -pedantic to select values-Xc.o.
518 2018-01-30 Vladimir Makarov <vmakarov@redhat.com>
521 * lra-constraints.c (curr_insn_transform): Process AND in the
524 2018-01-30 Jakub Jelinek <jakub@redhat.com>
526 PR rtl-optimization/83986
527 * sched-deps.c (sched_analyze_insn): For frame related insns, add anti
528 dependence against last_pending_memory_flush in addition to
531 2018-01-30 Alexandre Oliva <aoliva@redhat.com>
533 PR tree-optimization/81611
534 * tree-ssa-dom.c (simple_iv_increment_p): Skip intervening
537 2018-01-30 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
539 * config/rs6000/rs6000.c (rs6000_internal_arg_pointer): Only return
542 2018-01-30 Richard Biener <rguenther@suse.de>
543 Jakub Jelinek <jakub@redhat.com>
545 PR tree-optimization/84111
546 * tree-ssa-loop-ivcanon.c (tree_unroll_loops_completely_1): Skip
547 inner loops added during recursion, as they don't have up-to-date
550 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
553 * ipa-inline.c (can_inline_edge_p): Break out late tests to...
554 (can_inline_edge_by_limits_p): ... here.
555 (can_early_inline_edge_p, check_callers,
556 update_caller_keys, update_callee_keys, recursive_inlining,
557 add_new_edges_to_heap, speculation_useful_p,
558 inline_small_functions,
559 inline_small_functions, flatten_function,
560 inline_to_all_callers_1): Update.
562 2018-01-30 Jan Hubicka <hubicka@ucw.cz>
564 * profile-count.c (profile_count::combine_with_ipa_count): Handle
567 2018-01-30 Richard Biener <rguenther@suse.de>
569 PR tree-optimization/83008
570 * tree-vect-slp.c (vect_analyze_slp_cost_1): Properly cost
571 invariant and constant vector uses in stmts when they need
574 2018-01-30 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
577 * configure.ac (gcc_cv_as_shf_merge): Disable on Solaris 10/x86.
578 * configure: Regenerate.
580 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
582 * config/aarch64/aarch64-sve.md (*vec_extract<mode><Vel>_0): New
584 (*vec_extract<mode><Vel>_v128): Require a nonzero lane number.
585 Use gen_rtx_REG rather than gen_lowpart.
587 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
589 * lra-constraints.c (match_reload): Use subreg_lowpart_offset
590 rather than 0 when creating partial subregs.
592 2018-01-30 Richard Sandiford <richard.sandiford@linaro.org>
594 * vec-perm-indices.c (vec_perm_indices::series_p): Give examples
597 2018-01-29 Michael Meissner <meissner@linux.vnet.ibm.com>
600 * config/rs6000/rs6000.c (rs6000_setup_reg_addr_masks): If DFmode
601 and SFmode can go in Altivec registers (-mcpu=power7 for DFmode,
602 -mcpu=power8 for SFmode) don't set the PRE_INCDEC or PRE_MODIFY
603 flags. This restores the settings used before the 2017-07-24.
604 Turning off pre increment/decrement/modify allows IVOPTS to
605 optimize DF/SF loops where the index is an int.
607 2018-01-29 Richard Biener <rguenther@suse.de>
608 Kelvin Nilsen <kelvin@gcc.gnu.org>
611 * tree-vect-stmts.c (vectorizable_call): Don't call
612 targetm.vectorize_builtin_md_vectorized_function if callee is
615 2018-01-22 Carl Love <cel@us.ibm.com>
617 * doc/extend.tex: Fix typo in second arg in
618 __builtin_bcdadd_{lt|eq|gt|ov} and __builtin_bcdsub_{lt|eq|gt|ov}.
620 2018-01-29 Richard Biener <rguenther@suse.de>
622 PR tree-optimization/84086
623 * tree-ssanames.c: Include cfgloop.h and tree-scalar-evolution.h.
624 (flush_ssaname_freelist): When SSA names were released reset
627 2018-01-29 Richard Biener <rguenther@suse.de>
629 PR tree-optimization/84057
630 * tree-ssa-loop-ivcanon.c (unloop_loops): Deal with already
631 removed paths when removing edges.
633 2018-01-27 H.J. Lu <hongjiu.lu@intel.com>
635 * doc/invoke.texi: Replace -mfunction-return==@var{choice} with
636 -mfunction-return=@var{choice}.
638 2018-01-27 Bernd Edlinger <bernd.edlinger@hotmail.de>
641 * diagnostic-show-locus.c (get_line_width_without_trailing_whitespace):
643 (layout::print_source_line): Likewise.
644 (test_get_line_width_without_trailing_whitespace): Add test cases.
646 2018-01-27 Jakub Jelinek <jakub@redhat.com>
649 * sched-deps.c (sched_macro_fuse_insns): Return immediately for
652 2018-01-26 Jim Wilson <jimw@sifive.com>
654 * config/riscv/riscv.h (MAX_FIXED_MODE_SIZE): New.
656 * config/riscv/elf.h (LIB_SPEC): Don't include -lgloss when nosys.specs
659 2018-01-26 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
661 * config/aarch64/aarch64.md: Add peepholes for CMP + SUB -> SUBS
662 and CMP + SUB-immediate -> SUBS.
664 2018-01-26 Martin Sebor <msebor@redhat.com>
666 PR tree-optimization/83896
667 * tree-ssa-strlen.c (get_string_len): Rename...
668 (get_string_cst_length): ...to this. Return HOST_WIDE_INT.
669 Avoid assuming length is constant.
670 (handle_char_store): Use HOST_WIDE_INT for string length.
672 2018-01-26 Uros Bizjak <ubizjak@gmail.com>
675 * config/i386/i386.md (*andndi3_doubleword): Add earlyclobber
676 to (=&r,r,rm) alternative. Add (=r,0,rm) and (=r,r,0) alternatives.
678 2018-01-26 Richard Biener <rguenther@suse.de>
680 PR rtl-optimization/84003
681 * dse.c (record_store): Only record redundant stores when
682 the earlier store aliases at least all accesses the later one does.
684 2018-01-26 Jakub Jelinek <jakub@redhat.com>
686 PR rtl-optimization/83985
687 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
688 REG_CFA_RESTORE insns.
689 (delete_unmarked_insns): Don't ignore separate shrink wrapping
690 REG_CFA_RESTORE insns here.
693 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
694 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
696 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
698 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
699 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
700 (arc_init): Likewise.
701 (arc_override_options): Likewise.
702 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
704 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
706 * config/arc/arc.h (TARGET_DBNZ): Define.
707 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
708 properly set the tune attribute.
709 (dbnz): Use TARGET_DBNZ guard.
710 * config/arc/arc.opt (mtune): Add core3 option.
712 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
714 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
715 recognize new pic like addresses.
716 (arc_delegitimize_address): Clean up.
718 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
720 * config/arc/arc-arches.def: Option mrf16 valid for all
722 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
723 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
724 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
725 * config/arc/arc-tables.opt: Regenerate.
726 * config/arc/arc.c (arc_conditional_register_usage): Handle
727 reduced register file case.
728 (arc_file_start): Set must have build attributes.
729 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
731 * config/arc/arc.opt (mrf16): Add new option.
732 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
733 * config/arc/genmultilib.awk: Handle new mrf16 option.
734 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
735 * config/arc/t-multilib: Regenerate.
736 * doc/invoke.texi (ARC Options): Document mrf16 option.
738 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
740 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
741 * config/arc/arc.c (arc_handle_secure_attribute): New function.
742 (arc_attribute_table): Add 'secure_call' attribute.
743 (arc_print_operand): Print secure call operand.
744 (arc_function_ok_for_sibcall): Don't optimize tail calls when
746 (arc_is_secure_call_p): New function. * config/arc/arc.md
747 (call_i): Add support for sjli instruction.
748 (call_value_i): Likewise.
749 * config/arc/constraints.md (Csc): New constraint.
751 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
752 John Eric Martin <John.Martin@emmicro-us.com>
754 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
755 * config/arc/arc.c (_arc_jli_section): New struct.
756 (arc_jli_section): New type.
757 (rc_jli_sections): New static variable.
758 (arc_handle_jli_attribute): New function.
759 (arc_attribute_table): Add jli_always and jli_fixed attribute.
760 (arc_file_end): New function.
761 (TARGET_ASM_FILE_END): Define.
762 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
763 (arc_add_jli_section): New function.
764 (jli_call_scan): Likewise.
765 (arc_reorg): Call jli_call_scan.
766 (arc_output_addsi): Remove 'S' from printing asm operand.
767 (arc_is_jli_call_p): New function.
768 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
770 (movhi_insn): Likewise.
771 (movsi_insn): Likewise.
772 (movsi_set_cc_insn): Likewise.
773 (loadqi_update): Likewise.
774 (load_zeroextendqisi_update): Likewise.
775 (load_signextendqisi_update): Likewise.
776 (loadhi_update): Likewise.
777 (load_zeroextendhisi_update): Likewise.
778 (load_signextendhisi_update): Likewise.
779 (loadsi_update): Likewise.
780 (loadsf_update): Likewise.
781 (movsicc_insn): Likewise.
782 (bset_insn): Likewise.
783 (bxor_insn): Likewise.
784 (bclr_insn): Likewise.
785 (bmsk_insn): Likewise.
786 (bicsi3_insn): Likewise.
787 (cmpsi_cc_c_insn): Likewise.
788 (movsi_ne): Likewise.
789 (movsi_cond_exec): Likewise.
790 (clrsbsi2): Likewise.
801 (call_i): Remove 'S' asm letter, add jli instruction.
802 (call_value_i): Likewise.
803 * config/arc/arc.op (mjli-always): New option.
804 * config/arc/constraints.md (Cji): New constraint.
805 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
807 (subsf3_fpx): Likewise.
808 (mulsf3_fpx): Likewise.
809 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
811 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
813 * doc/invoke.texi (ARC): Document mjli-always option.
815 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
817 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
818 and use incw and decw where possible
819 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
821 2018-01-26 Richard Biener <rguenther@suse.de>
823 PR tree-optimization/81082
824 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
825 association if it requires casting to unsigned.
826 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
827 from fold_plusminus_mult_expr to catch important cases late when
828 range info is available.
830 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
832 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
833 * configure.ac (hidden_linkonce): New test.
834 * configure: Regenerate.
835 * config.in: Regenerate.
837 2018-01-26 Julia Koval <julia.koval@intel.com>
839 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
840 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
841 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
842 _mm_mask_bitshuffle_epi64_mask): Fix type.
843 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
844 USI_FTYPE_V4DI_V4DI_USI): Remove.
845 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
846 __builtin_ia32_vpshufbitqmb256_mask,
847 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
848 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
849 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
851 2018-01-26 Alan Modra <amodra@gmail.com>
854 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
855 UNSPEC_VBPERMQ. Sort other unspecs.
857 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
859 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
861 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
864 * predict.c (drop_profile): Do not push/pop cfun; update also
866 (handle_missing_profiles): Fix logic looking for zero profiles.
868 2018-01-25 Jakub Jelinek <jakub@redhat.com>
871 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
872 on functions with #pragma omp declare simd or functions with simd
874 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
875 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
876 Remove trailing \n from warning_at calls.
878 2018-01-25 Tom de Vries <tom@codesourcery.com>
881 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
882 for neutered workers.
884 2018-01-24 Joseph Myers <joseph@codesourcery.com>
887 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
888 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
890 2018-01-24 Jeff Law <law@redhat.com>
893 * i386.c (get_probe_interval): Move to earlier point.
894 (ix86_compute_frame_layout): If -fstack-clash-protection and
895 the frame is larger than the probe interval, then use pushes
896 to save registers rather than reg->mem moves.
897 (ix86_expand_prologue): Remove conditional for int_registers_saved
900 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
903 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
904 min/max for never referenced object.
906 2018-01-24 Jakub Jelinek <jakub@redhat.com>
909 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
911 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
912 attributes from DECL_ATTRIBUTES (decl) without affecting
913 DECL_ATTRIBUTES (current_function_decl).
914 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
915 functions with non-NULL DECL_ABSTRACT_ORIGIN.
917 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
919 PR tree-optimization/83979
920 * fold-const.c (fold_comparison): Use constant_boolean_node
921 instead of boolean_{true,false}_node.
923 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
925 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
928 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
930 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
931 Simplify the clause that sets the length attribute.
932 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
933 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
934 clause that sets the length attribute.
935 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
937 2018-01-24 Tom de Vries <tom@codesourcery.com>
940 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
941 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
942 Add strict parameter.
943 (prevent_branch_around_nothing): Insert dummy insn between branch to
944 label and label with no ptx insn inbetween.
945 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
947 2018-01-24 Tom de Vries <tom@codesourcery.com>
950 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
951 for neutered threads in warp.
952 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
954 2018-01-24 Richard Biener <rguenther@suse.de>
956 PR tree-optimization/83176
957 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
960 2018-01-24 Richard Biener <rguenther@suse.de>
962 PR tree-optimization/82819
963 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
964 code generating pluses that are no-ops in the target precision.
966 2018-01-24 Richard Biener <rguenther@suse.de>
969 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
971 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
973 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
974 to merge probabilities.
975 * predict.c (probably_never_executed): Also mark as cold functions
976 with global 0 profile and guessed local profile.
977 * profile-count.c (profile_probability::combine_with_count): New
979 * profile-count.h (profile_probability::operator*,
980 profile_probability::operator*=, profile_probability::operator/,
981 profile_probability::operator/=): Reduce precision to adjusted
982 and set value to guessed on contradictory divisions.
983 (profile_probability::combine_with_freq): Remove.
984 (profile_probability::combine_wiht_count): Declare.
985 (profile_count::force_nonzero):: Set to adjusted.
986 (profile_count::probability_in):: Set quality to adjusted.
987 * tree-ssa-tail-merge.c (replace_block_by): Use
990 2018-01-23 Andrew Waterman <andrew@sifive.com>
991 Jim Wilson <jimw@sifive.com>
993 * config/riscv/riscv.c (riscv_stack_boundary): New.
994 (riscv_option_override): Set riscv_stack_boundary. Handle
995 riscv_preferred_stack_boundary_arg.
996 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
997 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
998 (STACK_BOUNDARY): Set to riscv_stack_boundary.
999 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
1000 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
1001 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
1003 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
1006 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
1007 of struct ix86_frame.
1008 (ix86_expand_epilogue): Likewise. Add a local variable for
1009 the reg_save_offset field in struct ix86_frame.
1011 2018-01-23 Bin Cheng <bin.cheng@arm.com>
1013 PR tree-optimization/82604
1014 * tree-loop-distribution.c (enum partition_kind): New enum item
1015 PKIND_PARTIAL_MEMSET.
1016 (partition_builtin_p): Support above new enum item.
1017 (generate_code_for_partition): Ditto.
1018 (compute_access_range): Differentiate cases that equality can be
1019 proven at all loops, the innermost loops or no loops.
1020 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
1021 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
1022 (finalize_partitions, distribute_loop): Don't fuse partition of
1023 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
1024 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
1027 2018-01-23 Martin Liska <mliska@suse.cz>
1029 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
1030 order to ignore the predictor.
1031 (PRED_POLYMORPHIC_CALL): Likewise.
1032 (PRED_RECURSIVE_CALL): Likewise.
1034 2018-01-23 Martin Liska <mliska@suse.cz>
1036 * tree-profile.c (tree_profiling): Print function header to
1037 aware reader which function we are working on.
1038 * value-prof.c (gimple_find_values_to_profile): Do not print
1039 not interesting value histograms.
1041 2018-01-23 Martin Liska <mliska@suse.cz>
1043 * profile-count.h (enum profile_quality): Add
1044 profile_uninitialized as the first value. Do not number values
1045 as they are zero based.
1046 (profile_count::verify): Update sanity check.
1047 (profile_probability::verify): Likewise.
1049 2018-01-23 Nathan Sidwell <nathan@acm.org>
1051 * doc/invoke.texi (ffor-scope): Deprecate.
1053 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1055 PR tree-optimization/83510
1056 * domwalk.c (set_all_edges_as_executable): New function.
1057 (dom_walker::dom_walker): Convert bool param
1058 "skip_unreachable_blocks" to enum reachability. Move setup of
1059 edge flags to set_all_edges_as_executable and only do it when
1060 reachability is REACHABLE_BLOCKS.
1061 * domwalk.h (enum dom_walker::reachability): New enum.
1062 (dom_walker::dom_walker): Convert bool param
1063 "skip_unreachable_blocks" to enum reachability.
1064 (set_all_edges_as_executable): New decl.
1065 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
1066 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
1068 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
1069 but converting true to REACHABLE_BLOCKS.
1070 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
1072 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
1073 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
1074 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
1076 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
1077 if check_all_array_refs will be called.
1079 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1081 * tree.c (selftest::test_location_wrappers): Add more test
1084 2018-01-23 David Malcolm <dmalcolm@redhat.com>
1086 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
1087 (selftest::test_bit_in_range): Likewise.
1089 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1092 * doc/sourcebuild.texi (vect_float): Say that the selector
1093 only describes the situation when -funsafe-math-optimizations is on.
1094 (vect_float_strict): Document.
1096 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
1098 PR tree-optimization/83965
1099 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
1100 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
1101 instead of checking only for a reduction.
1102 (vect_recog_widen_sum_pattern): Likewise.
1104 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1106 * predict.c (probably_never_executed): Only use precise profile info.
1107 (compute_function_frequency): Skip after inlining hack since we now
1108 have quality checking.
1110 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
1112 * profile-count.h (profile_probability::very_unlikely,
1113 profile_probability::unlikely, profile_probability::even): Set
1114 precision to guessed.
1116 2018-01-23 Richard Biener <rguenther@suse.de>
1118 PR tree-optimization/83963
1119 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
1120 Properly terminate dominator walk when crossing the exit edge not
1121 when visiting its source block.
1123 2018-01-23 Jakub Jelinek <jakub@redhat.com>
1126 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
1127 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
1129 2018-01-22 Jakub Jelinek <jakub@redhat.com>
1131 PR tree-optimization/83957
1132 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
1133 semicolon after for body surrounded by braces.
1135 PR tree-optimization/83081
1136 * profile-count.h (profile_probability::split): New method.
1137 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
1138 Use profile_probability::split.
1139 (do_compare_rtx_and_jump): Fix adjustment of probabilities
1140 when splitting a single conditional jump into 2.
1142 2018-01-22 David Malcolm <dmalcolm@redhat.com>
1144 PR tree-optimization/69452
1145 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
1148 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1150 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
1151 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
1152 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
1154 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1156 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
1157 * config/rl78/rl78.md: New define_expand "movdi"
1158 * config/rl78/rl78.c: New function definition rl78_split_movdi
1160 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
1163 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
1165 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
1166 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
1167 128-bit to produce an UNSPEC move to get the double word with the
1168 signbit and then a shift directly to do signbit.
1169 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
1170 implementation with a new version that just does either a direct
1171 move or a regular move. Move memory interface to separate insns.
1172 Move insns so they are next to the expander.
1173 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
1174 with signbit move. Split big and little endian case.
1175 (signbit<mode>2_dm_mem_le): Likewise.
1176 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
1177 (signbit<mode>2_dm2): Likewise.
1179 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1181 * config/rl78/rl78.md: New define_expand "anddi3".
1183 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1185 * config/rl78/rl78.md: New define_expand "umindi3".
1187 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1189 * config/rl78/rl78.md: New define_expand "smindi3".
1191 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1193 * config/rl78/rl78.md: New define_expand "smaxdi3".
1195 2018-01-22 Carl Love <cel@us.ibm.com>
1197 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
1198 LVX_V1TI): Add macro expansion.
1199 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
1200 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
1201 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
1202 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
1203 Change check to determine if the instruction is a byte reversing
1204 entry. Fix typo in comment.
1205 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
1206 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
1207 Add def_builtin calls for new builtins.
1208 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
1209 Add define_insn expansion.
1211 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1213 * config/rl78/rl78.md: New define_expand "umaxdi3".
1215 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
1217 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
1218 for non-QImode registers
1220 2018-01-22 Richard Biener <rguenther@suse.de>
1222 PR tree-optimization/83963
1223 * graphite-scop-detection.c (scop_detection::get_sese): Delay
1224 including the loop exit block.
1225 (scop_detection::merge_sese): Likewise.
1226 (scop_detection::add_scop): Do it here instead.
1228 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1230 * doc/sourcebuild.texi (arm_softfloat): Document.
1232 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
1235 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
1236 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
1237 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
1239 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1240 David Edelsohn <dje.gcc@gmail.com>
1243 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1244 Change "crset eq" to "crset 2".
1245 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1246 (*call_indirect_aix<mode>_nospec): Likewise.
1247 (*call_value_indirect_aix<mode>_nospec): Likewise.
1248 (*call_indirect_elfv2<mode>_nospec): Likewise.
1249 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
1250 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
1251 change assembly output from . to $.
1252 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1253 (indirect_jump<mode>_nospec): Change assembly output from . to $.
1254 (*tablejump<mode>_internal1_nospec): Likewise.
1256 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
1259 * config/sh/sh_optimize_sett_clrt.cc:
1260 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
1262 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
1264 PR tree-optimization/83940
1265 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
1266 offset_dt to vect_constant_def rather than vect_unknown_def_type.
1267 (vect_check_load_store_mask): Add a mask_dt_out parameter and
1268 use it to pass back the definition type.
1269 (vect_check_store_rhs): Likewise rhs_dt_out.
1270 (vect_build_gather_load_calls): Add a mask_dt argument and use
1271 it instead of a call to vect_is_simple_use.
1272 (vectorizable_store): Update calls to vect_check_load_store_mask
1273 and vect_check_store_rhs. Use the dt returned by the latter instead
1274 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
1275 instead of calls to vect_is_simple_use. Pass the scalar rather
1276 than the vector operand to vect_is_simple_use when handling
1277 second and subsequent copies of an rhs value.
1278 (vectorizable_load): Update calls to vect_check_load_store_mask
1279 and vect_build_gather_load_calls. Use the cached mask_dt and
1280 gs_info.offset_dt instead of calls to vect_is_simple_use.
1282 2018-01-20 Jakub Jelinek <jakub@redhat.com>
1285 * tree-emutls.c: Include gimplify.h.
1286 (lower_emutls_2): New function.
1287 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
1288 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
1289 it before further processing.
1292 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
1293 UINTVAL (trueop1) instead of INTVAL (op1).
1295 2018-01-19 Jakub Jelinek <jakub@redhat.com>
1299 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
1300 INCOMING_FRAME_SP_OFFSET if not defined.
1301 (scan_trace): Add ENTRY argument. If true and
1302 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
1303 emit a note to adjust the CFA offset.
1304 (create_cfi_notes): Adjust scan_trace callers.
1305 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
1306 INCOMING_FRAME_SP_OFFSET in the CIE.
1307 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
1308 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
1310 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
1311 * doc/tm.texi: Regenerated.
1313 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
1315 PR rtl-optimization/83147
1316 * lra-constraints.c (remove_inheritance_pseudos): Use
1317 lra_substitute_pseudo_within_insn.
1319 2018-01-19 Tom de Vries <tom@codesourcery.com>
1320 Cesar Philippidis <cesar@codesourcery.com>
1323 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
1325 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
1328 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
1329 spaces for function labels.
1331 2018-01-19 Martin Liska <mliska@suse.cz>
1333 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
1334 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
1335 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
1336 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
1337 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
1338 (PRED_CONST_RETURN): Change from 69 to 65.
1339 (PRED_NULL_RETURN): Change from 91 to 71.
1340 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
1341 (PRED_LOOP_GUARD): Change from 66 to 73.
1343 2018-01-19 Martin Liska <mliska@suse.cz>
1345 * predict.c (predict_insn_def): Add new assert.
1346 (struct branch_predictor): Change type to signed integer.
1347 (test_prediction_value_range): Amend test to cover
1349 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
1350 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
1351 (PRED_LOOP_ITERATIONS_MAX): Likewise.
1352 (PRED_LOOP_IV_COMPARE): Likewise.
1353 * predict.h (PROB_UNINITIALIZED): Define new constant.
1355 2018-01-19 Martin Liska <mliska@suse.cz>
1357 * predict.c (dump_prediction): Add new format for
1358 analyze_brprob.py script which is enabled with -details
1360 * profile-count.h (precise_p): New function.
1362 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1364 PR tree-optimization/83922
1365 * tree-vect-loop.c (vect_verify_full_masking): Return false if
1366 there are no statements that need masking.
1367 (vect_active_double_reduction_p): New function.
1368 (vect_analyze_loop_operations): Use it when handling phis that
1369 are not in the loop header.
1371 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
1373 PR tree-optimization/83914
1374 * tree-vect-loop.c (vectorizable_induction): Don't convert
1375 init_expr or apply the peeling adjustment for inductions
1376 that are nested within the vectorized loop.
1378 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1380 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
1383 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1387 * function.h (gimplify_parameters): Add gimple_seq * argument.
1388 * function.c: Include gimple.h and options.h.
1389 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
1390 for the added local temporaries if needed.
1391 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
1392 if there are any parameter cleanups, wrap whole body into a
1393 try/finally with the cleanups.
1395 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
1398 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
1399 Use GET_MODE_CLASS for scalar floating point.
1401 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1405 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
1406 Fix call of call_cgraph_insertion_hooks.
1408 2018-01-18 Martin Sebor <msebor@redhat.com>
1410 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
1412 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
1415 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
1418 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
1421 * common.opt: (-ffile-prefix-map): New option.
1422 * opts.c (common_handle_option): Defer it.
1423 * opts-global.c (handle_common_deferred_options): Handle it.
1424 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
1425 * file-prefix-map.h: New file.
1426 (remap_debug_filename, add_debug_prefix_map): ...here.
1427 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
1428 * final.c (debug_prefix_map, add_debug_prefix_map
1429 remap_debug_filename): Move to...
1430 * file-prefix-map.c: New file.
1431 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
1432 generalize, get rid of alloca(), use strrchr() instead of strchr().
1433 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
1434 Implement in terms of add_prefix_map().
1435 (remap_macro_filename, remap_debug_filename): Implement in term of
1437 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
1438 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
1439 * dbxout.c: Include file-prefix-map.h.
1440 * varasm.c: Likewise.
1441 * vmsdbgout.c: Likewise.
1442 * xcoffout.c: Likewise.
1443 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
1444 * doc/cppopts.texi (-fmacro-prefix-map): Document.
1445 * doc/invoke.texi (-ffile-prefix-map): Document.
1446 (-fdebug-prefix-map): Update description.
1448 2018-01-18 Martin Liska <mliska@suse.cz>
1450 * config/i386/i386.c (indirect_thunk_name): Document that also
1452 (output_indirect_thunk): Document why both instructions
1453 (pause and lfence) are generated.
1455 2018-01-18 Richard Biener <rguenther@suse.de>
1457 PR tree-optimization/83887
1458 * graphite-scop-detection.c
1459 (scop_detection::get_nearest_dom_with_single_entry): Remove.
1460 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
1461 (scop_detection::merge_sese): Re-implement with a flood-fill
1462 algorithm that properly finds a SESE region if it exists.
1464 2018-01-18 Jakub Jelinek <jakub@redhat.com>
1467 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
1468 pointer_diff optimizations use view_convert instead of convert.
1470 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1472 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
1473 Generate different code for -mno-speculate-indirect-jumps.
1474 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
1475 (*call_indirect_aix<mode>): Disable for
1476 -mno-speculate-indirect-jumps.
1477 (*call_indirect_aix<mode>_nospec): New define_insn.
1478 (*call_value_indirect_aix<mode>): Disable for
1479 -mno-speculate-indirect-jumps.
1480 (*call_value_indirect_aix<mode>_nospec): New define_insn.
1481 (*sibcall_nonlocal_sysv<mode>): Generate different code for
1482 -mno-speculate-indirect-jumps.
1483 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
1485 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
1487 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
1488 long double type, set the flags for noting the default long double
1489 type, even if we don't pass or return a long double type.
1491 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
1494 * ipa-inline.c (flatten_function): Do not overwrite final inlining
1497 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
1499 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
1500 support for merge[hl].
1501 (fold_mergehl_helper): New helper function.
1502 (tree-vector-builder.h): New #include for tree_vector_builder usage.
1503 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
1504 (altivec_vmrglw_direct): Add xxmrglw insn.
1506 2018-01-17 Andrew Waterman <andrew@sifive.com>
1508 * config/riscv/riscv.c (riscv_conditional_register_usage): If
1509 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
1511 2018-01-17 David Malcolm <dmalcolm@redhat.com>
1514 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
1515 call the lto_location_cache before reading the
1516 DECL_SOURCE_LOCATION of the types.
1518 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
1519 Richard Sandiford <richard.sandiford@linaro.org>
1521 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
1522 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
1523 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
1524 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
1525 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
1527 * config/aarch64/constraints.md (aarch64_movti_operand):
1529 * config/aarch64/predicates.md (Uti): Add new constraint.
1531 2018-01-17 Carl Love <cel@us.ibm.com>
1532 * config/rs6000/vsx.md (define_expand xl_len_r,
1533 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
1534 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
1536 (define_expand, define_insn): Move the shift left from the
1537 define_insn to the define_expand for lxvl and stxvl instructions.
1538 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
1539 and XL_LEN_R definitions to PURE.
1541 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
1543 * config/i386/i386.c (indirect_thunk_name): Declare regno
1544 as unsigned int. Compare regno with INVALID_REGNUM.
1545 (output_indirect_thunk): Ditto.
1546 (output_indirect_thunk_function): Ditto.
1547 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
1548 in the call to output_indirect_thunk_function.
1550 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
1553 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
1554 rather than the size of inner_type to determine the stack slot size
1555 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
1557 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
1560 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
1563 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
1565 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
1566 endian Linux systems to optionally enable multilibs for selecting
1567 the long double type if the user configured an explicit type.
1568 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
1569 have no long double multilibs if not defined.
1570 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
1571 warn if the user used -mabi={ieee,ibm}longdouble and we built
1572 multilibs for long double.
1573 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
1574 appropriate multilib option.
1575 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
1577 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
1578 for building long double multilibs.
1579 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
1581 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
1583 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
1586 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
1588 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
1591 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
1594 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
1597 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1599 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
1600 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
1602 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
1604 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
1605 different rtl trees depending on TARGET_64BIT.
1606 (rs6000_gen_lvx): Likewise.
1608 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
1610 * config/visium/visium.md (nop): Tweak comment.
1611 (hazard_nop): Likewise.
1613 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
1615 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
1616 -mspeculate-indirect-jumps.
1617 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
1618 for -mno-speculate-indirect-jumps.
1619 (*call_indirect_elfv2<mode>_nospec): New define_insn.
1620 (*call_value_indirect_elfv2<mode>): Disable for
1621 -mno-speculate-indirect-jumps.
1622 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
1623 (indirect_jump): Emit different RTL for
1624 -mno-speculate-indirect-jumps.
1625 (*indirect_jump<mode>): Disable for
1626 -mno-speculate-indirect-jumps.
1627 (*indirect_jump<mode>_nospec): New define_insn.
1628 (tablejump): Emit different RTL for
1629 -mno-speculate-indirect-jumps.
1630 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
1631 (tablejumpsi_nospec): New define_expand.
1632 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
1633 (tablejumpdi_nospec): New define_expand.
1634 (*tablejump<mode>_internal1): Disable for
1635 -mno-speculate-indirect-jumps.
1636 (*tablejump<mode>_internal1_nospec): New define_insn.
1637 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
1640 2018-01-16 Artyom Skrobov tyomitch@gmail.com
1642 * caller-save.c (insert_save): Drop unnecessary parameter. All
1645 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1646 Richard Biener <rguenth@suse.de>
1649 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
1650 return early, inline manually is_gimple_sizepos. Make sure if we
1651 call gimplify_expr we don't end up with a gimple constant.
1652 * tree.c (variably_modified_type_p): Don't return true for
1653 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
1654 * gimplify.h (is_gimple_sizepos): Remove.
1656 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1658 PR tree-optimization/83857
1659 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
1660 vectorizable_live_operation for pure SLP statements.
1661 (vectorizable_live_operation): Handle PHIs.
1663 2018-01-16 Richard Biener <rguenther@suse.de>
1665 PR tree-optimization/83867
1666 * tree-vect-stmts.c (vect_transform_stmt): Precompute
1667 nested_in_vect_loop_p since the scalar stmt may get invalidated.
1669 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1672 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
1673 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
1674 If off is not INTEGER_CST, issue a may not be aligned warning
1675 rather than isn't aligned. Use isn%'t rather than isn't.
1676 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
1678 <case MULT_EXPR>: Improve the case when bottom and one of the
1679 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
1680 operand, in that case check if the other operand is multiple of
1681 bottom divided by the INTEGER_CST operand.
1683 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1686 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1687 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1688 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1689 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1690 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1691 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1692 (pa_function_arg_size): New function.
1694 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1696 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1697 in a separate statement.
1699 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1701 PR tree-optimization/83847
1702 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1703 group gathers and scatters.
1705 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1707 PR rtl-optimization/86620
1708 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1710 PR rtl-optimization/83213
1711 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1712 to last if both are JUMP_INSNs.
1714 PR tree-optimization/83843
1715 * gimple-ssa-store-merging.c
1716 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1717 store_immediate_info for bswap/nop orig_stores.
1719 2018-01-15 Andrew Waterman <andrew@sifive.com>
1721 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1723 <UDIV>: Increase cost if !TARGET_DIV.
1725 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1727 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1728 (define_attr "cr_logical_3op"): New.
1729 (cceq_ior_compare): Adjust.
1730 (cceq_ior_compare_complement): Adjust.
1731 (*cceq_rev_compare): Adjust.
1732 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1733 (is_cracked_insn): Adjust.
1734 (insn_must_be_first_in_group): Adjust.
1735 * config/rs6000/40x.md: Adjust.
1736 * config/rs6000/440.md: Adjust.
1737 * config/rs6000/476.md: Adjust.
1738 * config/rs6000/601.md: Adjust.
1739 * config/rs6000/603.md: Adjust.
1740 * config/rs6000/6xx.md: Adjust.
1741 * config/rs6000/7450.md: Adjust.
1742 * config/rs6000/7xx.md: Adjust.
1743 * config/rs6000/8540.md: Adjust.
1744 * config/rs6000/cell.md: Adjust.
1745 * config/rs6000/e300c2c3.md: Adjust.
1746 * config/rs6000/e500mc.md: Adjust.
1747 * config/rs6000/e500mc64.md: Adjust.
1748 * config/rs6000/e5500.md: Adjust.
1749 * config/rs6000/e6500.md: Adjust.
1750 * config/rs6000/mpc.md: Adjust.
1751 * config/rs6000/power4.md: Adjust.
1752 * config/rs6000/power5.md: Adjust.
1753 * config/rs6000/power6.md: Adjust.
1754 * config/rs6000/power7.md: Adjust.
1755 * config/rs6000/power8.md: Adjust.
1756 * config/rs6000/power9.md: Adjust.
1757 * config/rs6000/rs64.md: Adjust.
1758 * config/rs6000/titan.md: Adjust.
1760 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1762 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1763 ix86_indirect_branch_register logic.
1765 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1767 * config/i386/constraints.md (Bs): Update
1768 ix86_indirect_branch_register check. Don't check
1769 ix86_indirect_branch_register with GOT_memory_operand.
1771 * config/i386/predicates.md (GOT_memory_operand): Don't check
1772 ix86_indirect_branch_register here.
1773 (GOT32_symbol_operand): Likewise.
1775 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1777 * config/i386/predicates.md (constant_call_address_operand):
1778 Rewrite ix86_indirect_branch_register logic.
1779 (sibcall_insn_operand): Likewise.
1781 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1783 * config/i386/constraints.md (Bs): Replace
1784 ix86_indirect_branch_thunk_register with
1785 ix86_indirect_branch_register.
1787 * config/i386/i386.md (indirect_jump): Likewise.
1788 (tablejump): Likewise.
1789 (*sibcall_memory): Likewise.
1790 (*sibcall_value_memory): Likewise.
1791 Peepholes of indirect call and jump via memory: Likewise.
1792 * config/i386/i386.opt: Likewise.
1793 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1794 (GOT_memory_operand): Likewise.
1795 (call_insn_operand): Likewise.
1796 (sibcall_insn_operand): Likewise.
1797 (GOT32_symbol_operand): Likewise.
1799 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1802 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1803 type rather than type addr's type points to.
1804 (expand_omp_atomic_mutex): Likewise.
1805 (expand_omp_atomic): Likewise.
1807 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1810 * config/i386/i386.c (output_indirect_thunk_function): Use
1811 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1812 for __x86_return_thunk.
1814 2018-01-15 Richard Biener <rguenther@suse.de>
1817 * expmed.c (extract_bit_field_1): Fix typo.
1819 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1822 * config/arm/iterators.md (VF): New mode iterator.
1823 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1824 Remove integer-related logic from pattern.
1825 (neon_vabd<mode>_3): Likewise.
1827 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1830 * common.opt (fstrict-overflow): No longer an alias.
1831 (fwrapv-pointer): New option.
1832 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1833 also for pointer types based on flag_wrapv_pointer.
1834 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1835 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1836 opts->x_flag_wrapv got set.
1837 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1838 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1839 POINTER_TYPE_OVERFLOW_UNDEFINED.
1840 * match.pd: Likewise in address comparison pattern.
1841 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1843 2018-01-15 Richard Biener <rguenther@suse.de>
1846 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1847 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1848 Reset type names to their identifier if their TYPE_DECL doesn't
1849 have linkage (and thus is used for ODR and devirt).
1850 (save_debug_info_for_decl): Remove.
1851 (save_debug_info_for_type): Likewise.
1852 (add_tree_to_fld_list): Adjust.
1853 * tree-pretty-print.c (dump_generic_node): Make dumping of
1854 type names more robust.
1856 2018-01-15 Richard Biener <rguenther@suse.de>
1858 * BASE-VER: Bump to 8.0.1.
1860 2018-01-14 Martin Sebor <msebor@redhat.com>
1863 * builtins.c (check_access): Avoid warning when the no-warning bit
1866 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1868 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1869 * ira-color (allocno_hard_regs_compare): Likewise.
1871 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1874 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1875 Use .pushsection/.popsection.
1877 2018-01-14 Martin Sebor <msebor@redhat.com>
1880 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1882 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1884 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1885 entry from extra_headers.
1886 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1887 extra_headers, make the list bitwise identical to the i?86-*-* one.
1889 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1891 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1892 -mcmodel=large with -mindirect-branch=thunk,
1893 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1894 -mfunction-return=thunk-extern.
1895 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1896 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1897 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1899 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1901 * config/i386/i386.c (print_reg): Print the name of the full
1902 integer register without '%'.
1903 (ix86_print_operand): Handle 'V'.
1904 * doc/extend.texi: Document 'V' modifier.
1906 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1908 * config/i386/constraints.md (Bs): Disallow memory operand for
1909 -mindirect-branch-register.
1911 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1912 (GOT_memory_operand): Likewise.
1913 (call_insn_operand): Likewise.
1914 (sibcall_insn_operand): Likewise.
1915 (GOT32_symbol_operand): Likewise.
1916 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1917 for -mindirect-branch-register.
1918 (tablejump): Likewise.
1919 (*sibcall_memory): Likewise.
1920 (*sibcall_value_memory): Likewise.
1921 Disallow peepholes of indirect call and jump via memory for
1922 -mindirect-branch-register.
1923 (*call_pop): Replace m with Bw.
1924 (*call_value_pop): Likewise.
1925 (*sibcall_pop_memory): Replace m with Bs.
1926 * config/i386/i386.opt (mindirect-branch-register): New option.
1927 * doc/invoke.texi: Document -mindirect-branch-register option.
1929 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1931 * config/i386/i386-protos.h (ix86_output_function_return): New.
1932 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1933 set function_return_type.
1934 (indirect_thunk_name): Add ret_p to indicate thunk for function
1936 (output_indirect_thunk_function): Pass false to
1937 indirect_thunk_name.
1938 (ix86_output_indirect_branch_via_reg): Likewise.
1939 (ix86_output_indirect_branch_via_push): Likewise.
1940 (output_indirect_thunk_function): Create alias for function
1941 return thunk if regno < 0.
1942 (ix86_output_function_return): New function.
1943 (ix86_handle_fndecl_attribute): Handle function_return.
1944 (ix86_attribute_table): Add function_return.
1945 * config/i386/i386.h (machine_function): Add
1946 function_return_type.
1947 * config/i386/i386.md (simple_return_internal): Use
1948 ix86_output_function_return.
1949 (simple_return_internal_long): Likewise.
1950 * config/i386/i386.opt (mfunction-return=): New option.
1951 (indirect_branch): Mention -mfunction-return=.
1952 * doc/extend.texi: Document function_return function attribute.
1953 * doc/invoke.texi: Document -mfunction-return= option.
1955 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1957 * config/i386/i386-opts.h (indirect_branch): New.
1958 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1959 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1960 with local indirect jump when converting indirect call and jump.
1961 (ix86_set_indirect_branch_type): New.
1962 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1963 (indirectlabelno): New.
1964 (indirect_thunk_needed): Likewise.
1965 (indirect_thunk_bnd_needed): Likewise.
1966 (indirect_thunks_used): Likewise.
1967 (indirect_thunks_bnd_used): Likewise.
1968 (INDIRECT_LABEL): Likewise.
1969 (indirect_thunk_name): Likewise.
1970 (output_indirect_thunk): Likewise.
1971 (output_indirect_thunk_function): Likewise.
1972 (ix86_output_indirect_branch_via_reg): Likewise.
1973 (ix86_output_indirect_branch_via_push): Likewise.
1974 (ix86_output_indirect_branch): Likewise.
1975 (ix86_output_indirect_jmp): Likewise.
1976 (ix86_code_end): Call output_indirect_thunk_function if needed.
1977 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1979 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1980 (ix86_attribute_table): Add indirect_branch.
1981 * config/i386/i386.h (machine_function): Add indirect_branch_type
1982 and has_local_indirect_jump.
1983 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1985 (tablejump): Likewise.
1986 (*indirect_jump): Use ix86_output_indirect_jmp.
1987 (*tablejump_1): Likewise.
1988 (simple_return_indirect_internal): Likewise.
1989 * config/i386/i386.opt (mindirect-branch=): New option.
1990 (indirect_branch): New.
1993 (thunk-inline): Likewise.
1994 (thunk-extern): Likewise.
1995 * doc/extend.texi: Document indirect_branch function attribute.
1996 * doc/invoke.texi: Document -mindirect-branch= option.
1998 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
2001 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
2003 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
2005 * ipa-inline.c (want_inline_small_function_p): Return false if
2006 inlining has already failed with CIF_FINAL_ERROR.
2007 (update_caller_keys): Call want_inline_small_function_p before
2009 (update_callee_keys): Likewise.
2011 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
2013 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
2015 (rs6000_quadword_masked_address_p): Likewise.
2016 (quad_aligned_load_p): Likewise.
2017 (quad_aligned_store_p): Likewise.
2018 (const_load_sequence_p): Add comment to describe the outer-most loop.
2019 (mimic_memory_attributes_and_flags): New function.
2020 (rs6000_gen_stvx): Likewise.
2021 (replace_swapped_aligned_store): Likewise.
2022 (rs6000_gen_lvx): Likewise.
2023 (replace_swapped_aligned_load): Likewise.
2024 (replace_swapped_load_constant): Capitalize argument name in
2025 comment describing this function.
2026 (rs6000_analyze_swaps): Add a third pass to search for vector loads
2027 and stores that access quad-word aligned addresses and replace
2028 with stvx or lvx instructions when appropriate.
2029 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
2030 New function prototype.
2031 (rs6000_quadword_masked_address_p): Likewise.
2032 (rs6000_gen_lvx): Likewise.
2033 (rs6000_gen_stvx): Likewise.
2034 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
2035 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
2036 when memory address is aligned.
2037 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
2038 this split to select lvx instruction when memory address is aligned.
2039 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
2040 instruction when memory address is aligned.
2041 (*vsx_le_perm_load_v16qi): Likewise.
2042 (four unnamed splitters): Modify to select the stvx instruction
2043 when memory is aligned.
2045 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
2047 * predict.c (determine_unlikely_bbs): Handle correctly BBs
2048 which appears in the queue multiple times.
2050 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2051 Alan Hayward <alan.hayward@arm.com>
2052 David Sherwood <david.sherwood@arm.com>
2054 * tree-vectorizer.h (vec_lower_bound): New structure.
2055 (_loop_vec_info): Add check_nonzero and lower_bounds.
2056 (LOOP_VINFO_CHECK_NONZERO): New macro.
2057 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
2058 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
2059 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
2060 fields. Make seg_len the distance travelled, not including the
2062 (dr_direction_indicator): Declare.
2063 (dr_zero_step_indicator): Likewise.
2064 (dr_known_forward_stride_p): Likewise.
2065 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
2067 (runtime_alias_check_p): Allow runtime alias checks with
2069 (operator ==): Compare access_size and align.
2070 (prune_runtime_alias_test_list): Rework for new distinction between
2071 the access_size and seg_len.
2072 (create_intersect_range_checks_index): Likewise. Cope with polynomial
2074 (get_segment_min_max): New function.
2075 (create_intersect_range_checks): Use it.
2076 (dr_step_indicator): New function.
2077 (dr_direction_indicator): Likewise.
2078 (dr_zero_step_indicator): Likewise.
2079 (dr_known_forward_stride_p): Likewise.
2080 * tree-loop-distribution.c (data_ref_segment_size): Return
2081 DR_STEP * (niters - 1).
2082 (compute_alias_check_pairs): Update call to the dr_with_seg_len
2084 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
2085 (vect_preserves_scalar_order_p): New function, split out from...
2086 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
2087 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
2088 (vect_vfa_access_size): New function.
2089 (vect_vfa_align): Likewise.
2090 (vect_compile_time_alias): Take access_size_a and access_b arguments.
2091 (dump_lower_bound): New function.
2092 (vect_check_lower_bound): Likewise.
2093 (vect_small_gap_p): Likewise.
2094 (vectorizable_with_step_bound_p): Likewise.
2095 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
2096 depencies if the vectorization factor is 1. Convert the checks
2097 for nonzero steps into checks on the bounds of DR_STEP. Try using
2098 a bunds check for variable steps if the minimum required step is
2099 relatively small. Update calls to the dr_with_seg_len
2100 constructor and to vect_compile_time_alias.
2101 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
2103 (vect_loop_versioning): Call it.
2104 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
2106 (vect_estimate_min_profitable_iters): Account for any bounds checks.
2108 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2109 Alan Hayward <alan.hayward@arm.com>
2110 David Sherwood <david.sherwood@arm.com>
2112 * doc/sourcebuild.texi (vect_scatter_store): Document.
2113 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
2115 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
2117 * genopinit.c (main): Add supports_vec_scatter_store and
2118 supports_vec_scatter_store_cached to target_optabs.
2119 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
2120 IFN_MASK_SCATTER_STORE.
2121 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
2123 * internal-fn.h (internal_store_fn_p): Declare.
2124 (internal_fn_stored_value_index): Likewise.
2125 * internal-fn.c (scatter_store_direct): New macro.
2126 (expand_scatter_store_optab_fn): New function.
2127 (direct_scatter_store_optab_supported_p): New macro.
2128 (internal_store_fn_p): New function.
2129 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
2130 IFN_MASK_SCATTER_STORE.
2131 (internal_fn_mask_index): Likewise.
2132 (internal_fn_stored_value_index): New function.
2133 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
2135 * optabs-query.h (supports_vec_scatter_store_p): Declare.
2136 * optabs-query.c (supports_vec_scatter_store_p): New function.
2137 * tree-vectorizer.h (vect_get_store_rhs): Declare.
2138 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
2139 true for scatter stores.
2140 (vect_gather_scatter_fn_p): Handle scatter stores too.
2141 (vect_check_gather_scatter): Consider using scatter stores if
2142 supports_vec_scatter_store_p.
2143 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
2145 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2146 internal_fn_stored_value_index.
2147 (check_load_store_masking): Handle scatter stores too.
2148 (vect_get_store_rhs): Make public.
2149 (vectorizable_call): Use internal_store_fn_p.
2150 (vectorizable_store): Handle scatter store internal functions.
2151 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
2152 when deciding whether the end of the group has been reached.
2153 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
2154 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
2155 (mask_scatter_store<mode>): New insns.
2157 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2158 Alan Hayward <alan.hayward@arm.com>
2159 David Sherwood <david.sherwood@arm.com>
2161 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
2162 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
2163 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
2165 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
2166 Use vect_truncate_gather_scatter_offset if we can't treat the
2167 operation as a normal gather load or scatter store.
2168 (get_group_load_store_type): Take the gather_scatter_info
2169 as argument. Try using a gather load or scatter store for
2170 single-element groups.
2171 (get_load_store_type): Update calls to get_group_load_store_type
2172 and vect_use_strided_gather_scatters_p.
2174 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2175 Alan Hayward <alan.hayward@arm.com>
2176 David Sherwood <david.sherwood@arm.com>
2178 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
2179 optional tree argument.
2180 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
2182 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
2183 but continue to use the current value as a fallback.
2184 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
2185 to compare the updates.
2186 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
2187 (get_load_store_type): Use it when handling a strided access.
2188 (vect_get_strided_load_store_ops): New function.
2189 (vect_get_data_ptr_increment): Likewise.
2190 (vectorizable_load): Handle strided gather loads. Always pass
2191 a step to vect_create_data_ref_ptr and bump_vector_ptr.
2193 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2194 Alan Hayward <alan.hayward@arm.com>
2195 David Sherwood <david.sherwood@arm.com>
2197 * doc/md.texi (gather_load@var{m}): Document.
2198 (mask_gather_load@var{m}): Likewise.
2199 * genopinit.c (main): Add supports_vec_gather_load and
2200 supports_vec_gather_load_cached to target_optabs.
2201 * optabs-tree.c (init_tree_optimization_optabs): Use
2202 ggc_cleared_alloc to allocate target_optabs.
2203 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
2204 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
2206 * internal-fn.h (internal_load_fn_p): Declare.
2207 (internal_gather_scatter_fn_p): Likewise.
2208 (internal_fn_mask_index): Likewise.
2209 (internal_gather_scatter_fn_supported_p): Likewise.
2210 * internal-fn.c (gather_load_direct): New macro.
2211 (expand_gather_load_optab_fn): New function.
2212 (direct_gather_load_optab_supported_p): New macro.
2213 (direct_internal_fn_optab): New function.
2214 (internal_load_fn_p): Likewise.
2215 (internal_gather_scatter_fn_p): Likewise.
2216 (internal_fn_mask_index): Likewise.
2217 (internal_gather_scatter_fn_supported_p): Likewise.
2218 * optabs-query.c (supports_at_least_one_mode_p): New function.
2219 (supports_vec_gather_load_p): Likewise.
2220 * optabs-query.h (supports_vec_gather_load_p): Declare.
2221 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
2222 and memory_type field.
2223 (NUM_PATTERNS): Bump to 15.
2224 * tree-vect-data-refs.c: Include internal-fn.h.
2225 (vect_gather_scatter_fn_p): New function.
2226 (vect_describe_gather_scatter_call): Likewise.
2227 (vect_check_gather_scatter): Try using internal functions for
2228 gather loads. Recognize existing calls to a gather load function.
2229 (vect_analyze_data_refs): Consider using gather loads if
2230 supports_vec_gather_load_p.
2231 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
2232 (vect_get_gather_scatter_offset_type): Likewise.
2233 (vect_convert_mask_for_vectype): Likewise.
2234 (vect_add_conversion_to_patterm): Likewise.
2235 (vect_try_gather_scatter_pattern): Likewise.
2236 (vect_recog_gather_scatter_pattern): New pattern recognizer.
2237 (vect_vect_recog_func_ptrs): Add it.
2238 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
2239 internal_fn_mask_index and internal_gather_scatter_fn_p.
2240 (check_load_store_masking): Take the gather_scatter_info as an
2241 argument and handle gather loads.
2242 (vect_get_gather_scatter_ops): New function.
2243 (vectorizable_call): Check internal_load_fn_p.
2244 (vectorizable_load): Likewise. Handle gather load internal
2246 (vectorizable_store): Update call to check_load_store_masking.
2247 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
2248 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
2249 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
2250 (aarch64_gather_scale_operand_d): New predicates.
2251 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
2252 (mask_gather_load<mode>): New insns.
2254 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2255 Alan Hayward <alan.hayward@arm.com>
2256 David Sherwood <david.sherwood@arm.com>
2258 * optabs.def (fold_left_plus_optab): New optab.
2259 * doc/md.texi (fold_left_plus_@var{m}): Document.
2260 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
2261 * internal-fn.c (fold_left_direct): Define.
2262 (expand_fold_left_optab_fn): Likewise.
2263 (direct_fold_left_optab_supported_p): Likewise.
2264 * fold-const-call.c (fold_const_fold_left): New function.
2265 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
2266 * tree-parloops.c (valid_reduction_p): New function.
2267 (gather_scalar_reductions): Use it.
2268 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
2269 (vect_finish_replace_stmt): Declare.
2270 * tree-vect-loop.c (fold_left_reduction_fn): New function.
2271 (needs_fold_left_reduction_p): New function, split out from...
2272 (vect_is_simple_reduction): ...here. Accept reductions that
2273 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
2274 (vect_force_simple_reduction): Also store the reduction type in
2275 the assignment's STMT_VINFO_REDUC_TYPE.
2276 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
2277 (merge_with_identity): New function.
2278 (vect_expand_fold_left): Likewise.
2279 (vectorize_fold_left_reduction): Likewise.
2280 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
2281 scalar phi in place for it. Check for target support and reject
2282 cases that would reassociate the operation. Defer the transform
2283 phase to vectorize_fold_left_reduction.
2284 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
2285 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
2286 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
2288 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2290 * tree-if-conv.c (predicate_mem_writes): Remove redundant
2291 call to ifc_temp_var.
2293 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2294 Alan Hayward <alan.hayward@arm.com>
2295 David Sherwood <david.sherwood@arm.com>
2297 * target.def (legitimize_address_displacement): Take the original
2298 offset as a poly_int.
2299 * targhooks.h (default_legitimize_address_displacement): Update
2301 * targhooks.c (default_legitimize_address_displacement): Likewise.
2302 * doc/tm.texi: Regenerate.
2303 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
2304 as an argument, moving assert of ad->disp == ad->disp_term to...
2305 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
2306 Try calling targetm.legitimize_address_displacement before expanding
2307 the address rather than afterwards, and adjust for the new interface.
2308 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
2309 Match the new hook interface. Handle SVE addresses.
2310 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
2313 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2315 * Makefile.in (OBJS): Add early-remat.o.
2316 * target.def (select_early_remat_modes): New hook.
2317 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
2318 * doc/tm.texi: Regenerate.
2319 * targhooks.h (default_select_early_remat_modes): Declare.
2320 * targhooks.c (default_select_early_remat_modes): New function.
2321 * timevar.def (TV_EARLY_REMAT): New timevar.
2322 * passes.def (pass_early_remat): New pass.
2323 * tree-pass.h (make_pass_early_remat): Declare.
2324 * early-remat.c: New file.
2325 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
2327 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
2329 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2330 Alan Hayward <alan.hayward@arm.com>
2331 David Sherwood <david.sherwood@arm.com>
2333 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
2334 vfm1 with a bound_epilog parameter.
2335 (vect_do_peeling): Update calls accordingly, and move the prologue
2336 call earlier in the function. Treat the base bound_epilog as 0 for
2337 fully-masked loops and retain vf - 1 for other loops. Add 1 to
2338 this base when peeling for gaps.
2339 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
2340 with fully-masked loops.
2341 (vect_estimate_min_profitable_iters): Handle the single peeled
2342 iteration in that case.
2344 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2345 Alan Hayward <alan.hayward@arm.com>
2346 David Sherwood <david.sherwood@arm.com>
2348 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
2349 single-element interleaving even if the size is not a power of 2.
2350 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
2351 accesses for single-element interleaving if the group size is
2354 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2355 Alan Hayward <alan.hayward@arm.com>
2356 David Sherwood <david.sherwood@arm.com>
2358 * doc/md.texi (fold_extract_last_@var{m}): Document.
2359 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
2360 * optabs.def (fold_extract_last_optab): New optab.
2361 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
2362 * internal-fn.c (fold_extract_direct): New macro.
2363 (expand_fold_extract_optab_fn): Likewise.
2364 (direct_fold_extract_optab_supported_p): Likewise.
2365 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
2366 * tree-vect-loop.c (vect_model_reduction_cost): Handle
2367 EXTRACT_LAST_REDUCTION.
2368 (get_initial_def_for_reduction): Do not create an initial vector
2369 for EXTRACT_LAST_REDUCTION reductions.
2370 (vectorizable_reduction): Leave the scalar phi in place for
2371 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
2372 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
2373 epilogue code for EXTRACT_LAST_REDUCTION and defer the
2374 transform phase to vectorizable_condition.
2375 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
2377 (vect_finish_stmt_generation): ...here.
2378 (vect_finish_replace_stmt): New function.
2379 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
2380 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
2382 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
2384 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2385 Alan Hayward <alan.hayward@arm.com>
2386 David Sherwood <david.sherwood@arm.com>
2388 * doc/md.texi (extract_last_@var{m}): Document.
2389 * optabs.def (extract_last_optab): New optab.
2390 * internal-fn.def (EXTRACT_LAST): New internal function.
2391 * internal-fn.c (cond_unary_direct): New macro.
2392 (expand_cond_unary_optab_fn): Likewise.
2393 (direct_cond_unary_optab_supported_p): Likewise.
2394 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
2395 loops using EXTRACT_LAST.
2396 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
2397 (extract_last_<mode>): ...this optab.
2398 (vec_extract<mode><Vel>): Update accordingly.
2400 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2401 Alan Hayward <alan.hayward@arm.com>
2402 David Sherwood <david.sherwood@arm.com>
2404 * target.def (empty_mask_is_expensive): New hook.
2405 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
2406 * doc/tm.texi: Regenerate.
2407 * targhooks.h (default_empty_mask_is_expensive): Declare.
2408 * targhooks.c (default_empty_mask_is_expensive): New function.
2409 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
2410 if the target says that empty masks are expensive.
2411 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
2413 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
2415 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2416 Alan Hayward <alan.hayward@arm.com>
2417 David Sherwood <david.sherwood@arm.com>
2419 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
2420 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
2421 (vect_use_loop_mask_for_alignment_p): New function.
2422 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
2423 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
2424 niters_skip argument. Make sure that the first niters_skip elements
2425 of the first iteration are inactive.
2426 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
2427 Update call to vect_set_loop_masks_directly.
2428 (get_misalign_in_elems): New function, split out from...
2429 (vect_gen_prolog_loop_niters): ...here.
2430 (vect_update_init_of_dr): Take a code argument that specifies whether
2431 the adjustment should be added or subtracted.
2432 (vect_update_init_of_drs): Likewise.
2433 (vect_prepare_for_masked_peels): New function.
2434 (vect_do_peeling): Skip prologue peeling if we're using a mask
2435 instead. Update call to vect_update_inits_of_drs.
2436 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2438 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
2439 alignment. Do not include the number of peeled iterations in
2440 the minimum threshold in that case.
2441 (vectorizable_induction): Adjust the start value down by
2442 LOOP_VINFO_MASK_SKIP_NITERS iterations.
2443 (vect_transform_loop): Call vect_prepare_for_masked_peels.
2444 Take the number of skipped iterations into account when calculating
2446 * tree-vect-stmts.c (vect_gen_while_not): New function.
2448 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2449 Alan Hayward <alan.hayward@arm.com>
2450 David Sherwood <david.sherwood@arm.com>
2452 * doc/sourcebuild.texi (vect_fully_masked): Document.
2453 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
2455 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
2457 (vect_analyze_loop_2): ...here. Don't check the vectorization
2458 factor against the number of loop iterations if the loop is
2461 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2462 Alan Hayward <alan.hayward@arm.com>
2463 David Sherwood <david.sherwood@arm.com>
2465 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
2466 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
2467 (dump_groups): Update accordingly.
2468 (iv_use::mem_type): New member variable.
2469 (address_p): New function.
2470 (record_use): Add a mem_type argument and initialize the new
2472 (record_group_use): Add a mem_type argument. Use address_p.
2473 Remove obsolete null checks of base_object. Update call to record_use.
2474 (find_interesting_uses_op): Update call to record_group_use.
2475 (find_interesting_uses_cond): Likewise.
2476 (find_interesting_uses_address): Likewise.
2477 (get_mem_type_for_internal_fn): New function.
2478 (find_address_like_use): Likewise.
2479 (find_interesting_uses_stmt): Try find_address_like_use before
2480 calling find_interesting_uses_op.
2481 (addr_offset_valid_p): Use the iv mem_type field as the type
2482 of the addressed memory.
2483 (add_autoinc_candidates): Likewise.
2484 (get_address_cost): Likewise.
2485 (split_small_address_groups_p): Use address_p.
2486 (split_address_groups): Likewise.
2487 (add_iv_candidate_for_use): Likewise.
2488 (autoinc_possible_for_pair): Likewise.
2489 (rewrite_groups): Likewise.
2490 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
2491 (determine_group_iv_cost): Update after split of USE_ADDRESS.
2492 (get_alias_ptr_type_for_ptr_address): New function.
2493 (rewrite_use_address): Rewrite address uses in calls that were
2494 identified by find_address_like_use.
2496 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2497 Alan Hayward <alan.hayward@arm.com>
2498 David Sherwood <david.sherwood@arm.com>
2500 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
2502 * gimple-expr.h (is_gimple_addressable: Likewise.
2503 * gimple-expr.c (is_gimple_address): Likewise.
2504 * internal-fn.c (expand_call_mem_ref): New function.
2505 (expand_mask_load_optab_fn): Use it.
2506 (expand_mask_store_optab_fn): Likewise.
2508 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2509 Alan Hayward <alan.hayward@arm.com>
2510 David Sherwood <david.sherwood@arm.com>
2512 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
2513 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
2514 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
2515 (cond_umax@var{mode}): Document.
2516 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
2517 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
2518 (cond_umin_optab, cond_umax_optab): New optabs.
2519 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
2520 (COND_IOR, COND_XOR): New internal functions.
2521 * internal-fn.h (get_conditional_internal_fn): Declare.
2522 * internal-fn.c (cond_binary_direct): New macro.
2523 (expand_cond_binary_optab_fn): Likewise.
2524 (direct_cond_binary_optab_supported_p): Likewise.
2525 (get_conditional_internal_fn): New function.
2526 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
2527 Cope with reduction statements that are vectorized as calls rather
2529 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
2530 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
2531 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
2532 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
2533 (UNSPEC_COND_EOR): New unspecs.
2534 (optab): Add mappings for them.
2535 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
2536 (sve_int_op, sve_fp_op): New int attributes.
2538 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2539 Alan Hayward <alan.hayward@arm.com>
2540 David Sherwood <david.sherwood@arm.com>
2542 * optabs.def (while_ult_optab): New optab.
2543 * doc/md.texi (while_ult@var{m}@var{n}): Document.
2544 * internal-fn.def (WHILE_ULT): New internal function.
2545 * internal-fn.h (direct_internal_fn_supported_p): New override
2546 that takes two types as argument.
2547 * internal-fn.c (while_direct): New macro.
2548 (expand_while_optab_fn): New function.
2549 (convert_optab_supported_p): Likewise.
2550 (direct_while_optab_supported_p): New macro.
2551 * wide-int.h (wi::udiv_ceil): New function.
2552 * tree-vectorizer.h (rgroup_masks): New structure.
2553 (vec_loop_masks): New typedef.
2554 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
2556 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
2557 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
2558 (vect_max_vf): New function.
2559 (slpeel_make_loop_iterate_ntimes): Delete.
2560 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
2561 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
2562 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
2563 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
2564 internal-fn.h, stor-layout.h and optabs-query.h.
2565 (vect_set_loop_mask): New function.
2566 (add_preheader_seq): Likewise.
2567 (add_header_seq): Likewise.
2568 (interleave_supported_p): Likewise.
2569 (vect_maybe_permute_loop_masks): Likewise.
2570 (vect_set_loop_masks_directly): Likewise.
2571 (vect_set_loop_condition_masked): Likewise.
2572 (vect_set_loop_condition_unmasked): New function, split out from
2573 slpeel_make_loop_iterate_ntimes.
2574 (slpeel_make_loop_iterate_ntimes): Rename to..
2575 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
2576 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
2577 (vect_do_peeling): Update call accordingly.
2578 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
2580 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
2581 mask_compare_type, can_fully_mask_p and fully_masked_p.
2582 (release_vec_loop_masks): New function.
2583 (_loop_vec_info): Use it to free the loop masks.
2584 (can_produce_all_loop_masks_p): New function.
2585 (vect_get_max_nscalars_per_iter): Likewise.
2586 (vect_verify_full_masking): Likewise.
2587 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
2588 retries, and free the mask rgroups before retrying. Check loop-wide
2589 reasons for disallowing fully-masked loops. Make the final decision
2590 about whether use a fully-masked loop or not.
2591 (vect_estimate_min_profitable_iters): Do not assume that peeling
2592 for the number of iterations will be needed for fully-masked loops.
2593 (vectorizable_reduction): Disable fully-masked loops.
2594 (vectorizable_live_operation): Likewise.
2595 (vect_halve_mask_nunits): New function.
2596 (vect_double_mask_nunits): Likewise.
2597 (vect_record_loop_mask): Likewise.
2598 (vect_get_loop_mask): Likewise.
2599 (vect_transform_loop): Handle the case in which the final loop
2600 iteration might handle a partial vector. Call vect_set_loop_condition
2601 instead of slpeel_make_loop_iterate_ntimes.
2602 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
2603 (check_load_store_masking): New function.
2604 (prepare_load_store_mask): Likewise.
2605 (vectorizable_store): Handle fully-masked loops.
2606 (vectorizable_load): Likewise.
2607 (supportable_widening_operation): Use vect_halve_mask_nunits for
2609 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
2610 (vect_gen_while): New function.
2611 * config/aarch64/aarch64.md (umax<mode>3): New expander.
2612 (aarch64_uqdec<mode>): New insn.
2614 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2615 Alan Hayward <alan.hayward@arm.com>
2616 David Sherwood <david.sherwood@arm.com>
2618 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
2619 (reduc_xor_scal_optab): New optabs.
2620 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
2621 (reduc_xor_scal_@var{m}): Document.
2622 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
2623 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
2625 * fold-const-call.c (fold_const_call): Handle them.
2626 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
2627 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
2628 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
2629 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
2630 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
2631 (UNSPEC_XORV): New unspecs.
2632 (optab): Add entries for them.
2633 (BITWISEV): New int iterator.
2634 (bit_reduc_op): New int attributes.
2636 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2637 Alan Hayward <alan.hayward@arm.com>
2638 David Sherwood <david.sherwood@arm.com>
2640 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
2641 * internal-fn.def (VEC_SHL_INSERT): New internal function.
2642 * optabs.def (vec_shl_insert_optab): New optab.
2643 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
2644 (duplicate_and_interleave): Likewise.
2645 * tree-vect-loop.c: Include internal-fn.h.
2646 (neutral_op_for_slp_reduction): New function, split out from
2647 get_initial_defs_for_reduction.
2648 (get_initial_def_for_reduction): Handle option 2 for variable-length
2649 vectors by loading the neutral value into a vector and then shifting
2650 the initial value into element 0.
2651 (get_initial_defs_for_reduction): Replace the code argument with
2652 the neutral value calculated by neutral_op_for_slp_reduction.
2653 Use gimple_build_vector for constant-length vectors.
2654 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
2655 but the first group_size elements have a neutral value.
2656 Use duplicate_and_interleave otherwise.
2657 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
2658 Update call to get_initial_defs_for_reduction. Handle SLP
2659 reductions for variable-length vectors by creating one vector
2660 result for each scalar result, with the elements associated
2661 with other scalar results stubbed out with the neutral value.
2662 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
2663 Require IFN_VEC_SHL_INSERT for double reductions on
2664 variable-length vectors, or SLP reductions that have
2665 a neutral value. Require can_duplicate_and_interleave_p
2666 support for variable-length unchained SLP reductions if there
2667 is no neutral value, such as for MIN/MAX reductions. Also require
2668 the number of vector elements to be a multiple of the number of
2669 SLP statements when doing variable-length unchained SLP reductions.
2670 Update call to vect_create_epilog_for_reduction.
2671 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
2672 and remove initial values.
2673 (duplicate_and_interleave): Make public.
2674 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
2675 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
2677 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2678 Alan Hayward <alan.hayward@arm.com>
2679 David Sherwood <david.sherwood@arm.com>
2681 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
2682 (can_duplicate_and_interleave_p): New function.
2683 (vect_get_and_check_slp_defs): Take the vector of statements
2684 rather than just the current one. Remove excess parentheses.
2685 Restriction rejectinon of vect_constant_def and vect_external_def
2686 for variable-length vectors to boolean types, or types for which
2687 can_duplicate_and_interleave_p is false.
2688 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2689 (duplicate_and_interleave): New function.
2690 (vect_get_constant_vectors): Use gimple_build_vector for
2691 constant-length vectors and suitable variable-length constant
2692 vectors. Use duplicate_and_interleave for other variable-length
2693 vectors. Don't defer the update when inserting new statements.
2695 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2696 Alan Hayward <alan.hayward@arm.com>
2697 David Sherwood <david.sherwood@arm.com>
2699 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2700 min_profitable_iters doesn't go negative.
2702 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2703 Alan Hayward <alan.hayward@arm.com>
2704 David Sherwood <david.sherwood@arm.com>
2706 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2707 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2708 * optabs.def (vec_mask_load_lanes_optab): New optab.
2709 (vec_mask_store_lanes_optab): Likewise.
2710 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2711 (MASK_STORE_LANES): Likewise.
2712 * internal-fn.c (mask_load_lanes_direct): New macro.
2713 (mask_store_lanes_direct): Likewise.
2714 (expand_mask_load_optab_fn): Handle masked operations.
2715 (expand_mask_load_lanes_optab_fn): New macro.
2716 (expand_mask_store_optab_fn): Handle masked operations.
2717 (expand_mask_store_lanes_optab_fn): New macro.
2718 (direct_mask_load_lanes_optab_supported_p): Likewise.
2719 (direct_mask_store_lanes_optab_supported_p): Likewise.
2720 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2722 (vect_load_lanes_supported): Likewise.
2723 * tree-vect-data-refs.c (strip_conversion): New function.
2724 (can_group_stmts_p): Likewise.
2725 (vect_analyze_data_ref_accesses): Use it instead of checking
2726 for a pair of assignments.
2727 (vect_store_lanes_supported): Take a masked_p parameter.
2728 (vect_load_lanes_supported): Likewise.
2729 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2730 vect_store_lanes_supported and vect_load_lanes_supported.
2731 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2732 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2733 parameter. Don't allow gaps for masked accesses.
2734 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2735 and vect_load_lanes_supported.
2736 (get_load_store_type): Take a masked_p parameter and update
2737 call to get_group_load_store_type.
2738 (vectorizable_store): Update call to get_load_store_type.
2739 Handle IFN_MASK_STORE_LANES.
2740 (vectorizable_load): Update call to get_load_store_type.
2741 Handle IFN_MASK_LOAD_LANES.
2743 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2744 Alan Hayward <alan.hayward@arm.com>
2745 David Sherwood <david.sherwood@arm.com>
2747 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2749 * config/aarch64/aarch64-protos.h
2750 (aarch64_sve_struct_memory_operand_p): Declare.
2751 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2752 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2753 (VPRED, vpred): Handle SVE structure modes.
2754 * config/aarch64/constraints.md (Utx): New constraint.
2755 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2756 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2757 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2758 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2759 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2760 structure modes. Split into pieces after RA.
2761 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2762 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2764 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2765 SVE structure modes.
2766 (aarch64_classify_address): Likewise.
2767 (sizetochar): Move earlier in file.
2768 (aarch64_print_operand): Handle SVE register lists.
2769 (aarch64_array_mode): New function.
2770 (aarch64_sve_struct_memory_operand_p): Likewise.
2771 (TARGET_ARRAY_MODE): Redefine.
2773 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2774 Alan Hayward <alan.hayward@arm.com>
2775 David Sherwood <david.sherwood@arm.com>
2777 * target.def (array_mode): New target hook.
2778 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2779 * doc/tm.texi: Regenerate.
2780 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2781 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2782 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2784 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2787 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2788 Alan Hayward <alan.hayward@arm.com>
2789 David Sherwood <david.sherwood@arm.com>
2791 * fold-const.c (fold_binary_loc): Check the argument types
2792 rather than the result type when testing for a vector operation.
2794 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2796 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2797 * doc/tm.texi: Regenerate.
2799 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2800 Alan Hayward <alan.hayward@arm.com>
2801 David Sherwood <david.sherwood@arm.com>
2803 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2804 (sve): Document new AArch64 extension.
2805 * doc/md.texi (w): Extend the description of the AArch64
2806 constraint to include SVE vectors.
2807 (Upl, Upa): Document new AArch64 predicate constraints.
2808 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2810 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2811 (msve-vector-bits=): New option.
2812 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2813 SVE when these are disabled.
2814 (sve): New extension.
2815 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2816 modes. Adjust their number of units based on aarch64_sve_vg.
2817 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2818 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2819 aarch64_addr_query_type.
2820 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2821 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2822 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2823 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2824 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2825 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2826 (aarch64_simd_imm_zero_p): Delete.
2827 (aarch64_check_zero_based_sve_index_immediate): Declare.
2828 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2829 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2830 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2831 (aarch64_sve_float_mul_immediate_p): Likewise.
2832 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2834 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2835 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2836 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2837 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2838 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2839 (aarch64_regmode_natural_size): Likewise.
2840 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2841 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2843 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2844 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2845 for VG and the SVE predicate registers.
2846 (V_ALIASES): Add a "z"-prefixed alias.
2847 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2848 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2849 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2850 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2851 (REG_CLASS_NAMES): Add entries for them.
2852 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2853 and the predicate registers.
2854 (aarch64_sve_vg): Declare.
2855 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2856 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2857 (REGMODE_NATURAL_SIZE): Define.
2858 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2860 * config/aarch64/aarch64.c: Include cfgrtl.h.
2861 (simd_immediate_info): Add a constructor for series vectors,
2862 and an associated step field.
2863 (aarch64_sve_vg): New variable.
2864 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2865 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2866 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2867 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2868 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2869 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2870 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2871 (aarch64_get_mask_mode): New functions.
2872 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2873 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2874 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2875 predicate modes and predicate registers. Explicitly restrict
2876 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2877 to store a vector mode if it is recognized by
2878 aarch64_classify_vector_mode.
2879 (aarch64_regmode_natural_size): New function.
2880 (aarch64_hard_regno_caller_save_mode): Return the original mode
2882 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2883 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2884 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2885 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2887 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2888 does not overlap dest if the function is frame-related. Handle
2890 (aarch64_split_add_offset): New function.
2891 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2892 them aarch64_add_offset.
2893 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2894 and update call to aarch64_sub_sp.
2895 (aarch64_add_cfa_expression): New function.
2896 (aarch64_expand_prologue): Pass extra temporary registers to the
2897 functions above. Handle the case in which we need to emit new
2898 DW_CFA_expressions for registers that were originally saved
2899 relative to the stack pointer, but now have to be expressed
2900 relative to the frame pointer.
2901 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2903 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2904 IP0 and IP1 values for SVE frames.
2905 (aarch64_expand_vec_series): New function.
2906 (aarch64_expand_sve_widened_duplicate): Likewise.
2907 (aarch64_expand_sve_const_vector): Likewise.
2908 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2909 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2910 into the register, rather than emitting a SET directly.
2911 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2912 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2913 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2914 (offset_9bit_signed_scaled_p): New functions.
2915 (aarch64_replicate_bitmask_imm): New function.
2916 (aarch64_bitmask_imm): Use it.
2917 (aarch64_cannot_force_const_mem): Reject expressions involving
2918 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2919 (aarch64_classify_index): Handle SVE indices, by requiring
2920 a plain register index with a scale that matches the element size.
2921 (aarch64_classify_address): Handle SVE addresses. Assert that
2922 the mode of the address is VOIDmode or an integer mode.
2923 Update call to aarch64_classify_symbol.
2924 (aarch64_classify_symbolic_expression): Update call to
2925 aarch64_classify_symbol.
2926 (aarch64_const_vec_all_in_range_p): New function.
2927 (aarch64_print_vector_float_operand): Likewise.
2928 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2929 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2930 and the FP immediates 1.0 and 0.5.
2931 (aarch64_print_address_internal): Handle SVE addresses.
2932 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2933 (aarch64_regno_regclass): Handle predicate registers.
2934 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2936 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2937 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2938 (aarch64_convert_sve_vector_bits): New function.
2939 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2940 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2942 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2943 Handle SVE vector and predicate modes. Accept VL-based constants
2944 that need only one temporary register, and VL offsets that require
2945 no temporary registers.
2946 (aarch64_conditional_register_usage): Mark the predicate registers
2947 as fixed if SVE isn't available.
2948 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2949 Return true for SVE vector and predicate modes.
2950 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2951 rather than an unsigned int. Handle SVE modes.
2952 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2954 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2956 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2957 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2958 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2959 (aarch64_sve_float_mul_immediate_p): New functions.
2960 (aarch64_sve_valid_immediate): New function.
2961 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2962 Explicitly reject structure modes. Check for INDEX constants.
2963 Handle PTRUE and PFALSE constants.
2964 (aarch64_check_zero_based_sve_index_immediate): New function.
2965 (aarch64_simd_imm_zero_p): Delete.
2966 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2967 vector modes. Accept constants in the range of CNT[BHWD].
2968 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2969 ask for an Advanced SIMD mode.
2970 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2971 (aarch64_simd_vector_alignment): Handle SVE predicates.
2972 (aarch64_vectorize_preferred_vector_alignment): New function.
2973 (aarch64_simd_vector_alignment_reachable): Use it instead of
2975 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2976 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2978 (MAX_VECT_LEN): Delete.
2979 (expand_vec_perm_d): Add a vec_flags field.
2980 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2981 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2982 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2984 (aarch64_evpc_rev): Rename to...
2985 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2986 (aarch64_evpc_rev_global): New function.
2987 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2988 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2990 (aarch64_evpc_sve_tbl): New function.
2991 (aarch64_expand_vec_perm_const_1): Update after rename of
2992 aarch64_evpc_rev. Handle SVE permutes too, trying
2993 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2994 than aarch64_evpc_tbl.
2995 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2996 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2997 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2998 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2999 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
3000 (aarch64_expand_sve_vcond): New functions.
3001 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
3002 of aarch64_vector_mode_p.
3003 (aarch64_dwarf_poly_indeterminate_value): New function.
3004 (aarch64_compute_pressure_classes): Likewise.
3005 (aarch64_can_change_mode_class): Likewise.
3006 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
3007 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
3008 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
3009 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
3010 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
3011 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
3012 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
3013 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
3015 (Dn, Dl, Dr): Accept const as well as const_vector.
3016 (Dz): Likewise. Compare against CONST0_RTX.
3017 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
3018 of "vector" where appropriate.
3019 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
3020 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
3021 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
3022 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
3023 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
3024 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
3025 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
3026 (v_int_equiv): Extend to SVE modes.
3027 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
3029 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
3030 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
3031 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
3032 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
3033 (SVE_COND_FP_CMP): New int iterators.
3034 (perm_hilo): Handle the new unpack unspecs.
3035 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
3037 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
3038 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
3039 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
3040 (aarch64_equality_operator, aarch64_constant_vector_operand)
3041 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
3042 (aarch64_sve_nonimmediate_operand): Likewise.
3043 (aarch64_sve_general_operand): Likewise.
3044 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
3045 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
3046 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
3047 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
3048 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
3049 (aarch64_sve_float_arith_immediate): Likewise.
3050 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
3051 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
3052 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
3053 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
3054 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
3055 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
3056 (aarch64_sve_float_arith_operand): Likewise.
3057 (aarch64_sve_float_arith_with_sub_operand): Likewise.
3058 (aarch64_sve_float_mul_operand): Likewise.
3059 (aarch64_sve_vec_perm_operand): Likewise.
3060 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
3061 (aarch64_mov_operand): Accept const_poly_int and const_vector.
3062 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
3063 as well as const_vector.
3064 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
3065 in file. Use CONST0_RTX and CONSTM1_RTX.
3066 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
3067 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
3068 Use aarch64_simd_imm_zero.
3069 * config/aarch64/aarch64-sve.md: New file.
3070 * config/aarch64/aarch64.md: Include it.
3071 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
3072 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
3073 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
3074 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
3075 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
3076 (sve): New attribute.
3077 (enabled): Disable instructions with the sve attribute unless
3079 (movqi, movhi): Pass CONST_POLY_INT operaneds through
3080 aarch64_expand_mov_immediate.
3081 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
3082 CNT[BHSD] immediates.
3083 (movti): Split CONST_POLY_INT moves into two halves.
3084 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
3085 Split additions that need a temporary here if the destination
3086 is the stack pointer.
3087 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
3088 (*add<mode>3_poly_1): New instruction.
3089 (set_clobber_cc): New expander.
3091 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3093 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
3094 parameter and use it instead of GET_MODE_SIZE (innermode). Use
3095 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
3096 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
3097 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
3098 Change innermode from fixed_mode_size to machine_mode.
3099 (simplify_subreg): Update call accordingly. Handle a constant-sized
3100 subreg of a variable-length CONST_VECTOR.
3102 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
3103 Alan Hayward <alan.hayward@arm.com>
3104 David Sherwood <david.sherwood@arm.com>
3106 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
3107 (add_offset_to_base): New function, split out from...
3108 (create_mem_ref): ...here. When handling a scale other than 1,
3109 check first whether the address is valid without the offset.
3110 Add it into the base if so, leaving the index and scale as-is.
3112 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3115 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
3116 fold_for_warn before checking if arg2 is INTEGER_CST.
3118 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
3120 * config/rs6000/predicates.md (load_multiple_operation): Delete.
3121 (store_multiple_operation): Delete.
3122 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
3123 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
3124 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
3125 guarded by TARGET_STRING.
3126 (rs6000_output_load_multiple): Delete.
3127 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
3128 OPTION_MASK_STRING / TARGET_STRING handling.
3129 (print_operand) <'N', 'O'>: Add comment that these are unused now.
3130 (const rs6000_opt_masks) <"string">: Change mask to 0.
3131 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
3132 (MASK_STRING): Delete.
3133 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
3135 (load_multiple): Delete.
3142 (store_multiple): Delete.
3149 (movmemsi_8reg): Delete.
3150 (corresponding unnamed define_insn): Delete.
3151 (movmemsi_6reg): Delete.
3152 (corresponding unnamed define_insn): Delete.
3153 (movmemsi_4reg): Delete.
3154 (corresponding unnamed define_insn): Delete.
3155 (movmemsi_2reg): Delete.
3156 (corresponding unnamed define_insn): Delete.
3157 (movmemsi_1reg): Delete.
3158 (corresponding unnamed define_insn): Delete.
3159 * config/rs6000/rs6000.opt (mno-string): New.
3160 (mstring): Replace by deprecation warning stub.
3161 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
3163 2018-01-12 Jakub Jelinek <jakub@redhat.com>
3165 * regrename.c (regrename_do_replace): If replacing the same
3166 reg multiple times, try to reuse last created gen_raw_REG.
3169 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
3170 main to workaround a bug in GDB.
3172 2018-01-12 Tom de Vries <tom@codesourcery.com>
3175 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
3177 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
3179 PR rtl-optimization/80481
3180 * ira-color.c (get_cap_member): New function.
3181 (allocnos_conflict_by_live_ranges_p): Use it.
3182 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
3183 (setup_slot_coalesced_allocno_live_ranges): Ditto.
3185 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
3188 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
3189 (*saddl_se_1): Ditto.
3191 (*ssubl_se_1): Ditto.
3193 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3195 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
3196 rather than wi::to_widest for DR_INITs.
3197 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
3198 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
3199 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
3201 (vect_analyze_group_access_1): Note that here.
3203 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3205 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
3206 polynomial type sizes.
3208 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
3210 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
3211 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
3212 (gimple_add_tmp_var): Likewise.
3214 2018-01-12 Martin Liska <mliska@suse.cz>
3216 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
3217 (gimple_alloc_sizes): Likewise.
3218 (dump_gimple_statistics): Use PRIu64 in printf format.
3219 * gimple.h: Change uint64_t to int.
3221 2018-01-12 Martin Liska <mliska@suse.cz>
3223 * tree-core.h: Use uint64_t instead of int.
3224 * tree.c (tree_node_counts): Likewise.
3225 (tree_node_sizes): Likewise.
3226 (dump_tree_statistics): Use PRIu64 in printf format.
3228 2018-01-12 Martin Liska <mliska@suse.cz>
3230 * Makefile.in: As qsort_chk is implemented in vec.c, add
3231 vec.o to linkage of gencfn-macros.
3232 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
3233 passing the info to record_node_allocation_statistics.
3234 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
3236 * ggc-common.c (struct ggc_usage): Add operator== and use
3237 it in operator< and compare function.
3238 * mem-stats.h (struct mem_usage): Likewise.
3239 * vec.c (struct vec_usage): Remove operator< and compare
3240 function. Can be simply inherited.
3242 2018-01-12 Martin Jambor <mjambor@suse.cz>
3245 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
3246 * tree-ssa-math-opts.c: Include domwalk.h.
3247 (convert_mult_to_fma_1): New function.
3248 (fma_transformation_info): New type.
3249 (fma_deferring_state): Likewise.
3250 (cancel_fma_deferring): New function.
3251 (result_of_phi): Likewise.
3252 (last_fma_candidate_feeds_initial_phi): Likewise.
3253 (convert_mult_to_fma): Added deferring logic, split actual
3254 transformation to convert_mult_to_fma_1.
3255 (math_opts_dom_walker): New type.
3256 (math_opts_dom_walker::after_dom_children): New method, body moved
3257 here from pass_optimize_widening_mul::execute, added deferring logic
3259 (pass_optimize_widening_mul::execute): Moved most of code to
3260 math_opts_dom_walker::after_dom_children.
3261 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
3262 * config/i386/i386.c (ix86_option_override_internal): Added
3263 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
3265 2018-01-12 Richard Biener <rguenther@suse.de>
3268 * dwarf2out.c (gen_variable_die): Do not reset old_die for
3269 inline instance vars.
3271 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
3274 * config/rx/rx.c (rx_is_restricted_memory_address):
3277 2018-01-12 Richard Biener <rguenther@suse.de>
3279 PR tree-optimization/80846
3280 * target.def (split_reduction): New target hook.
3281 * targhooks.c (default_split_reduction): New function.
3282 * targhooks.h (default_split_reduction): Declare.
3283 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
3284 target requests first reduce vectors by combining low and high
3286 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
3287 (get_vectype_for_scalar_type_and_size): Export.
3288 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
3289 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
3290 * doc/tm.texi: Regenerate.
3291 * config/i386/i386.c (ix86_split_reduction): Implement
3292 TARGET_VECTORIZE_SPLIT_REDUCTION.
3294 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3297 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
3298 in PIC mode except for TARGET_VXWORKS_RTP.
3299 * config/sparc/sparc.c: Include cfgrtl.h.
3300 (TARGET_INIT_PIC_REG): Define.
3301 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
3302 (sparc_pic_register_p): New predicate.
3303 (sparc_legitimate_address_p): Use it.
3304 (sparc_legitimize_pic_address): Likewise.
3305 (sparc_delegitimize_address): Likewise.
3306 (sparc_mode_dependent_address_p): Likewise.
3307 (gen_load_pcrel_sym): Remove 4th parameter.
3308 (load_got_register): Adjust call to above. Remove obsolete stuff.
3309 (sparc_expand_prologue): Do not call load_got_register here.
3310 (sparc_flat_expand_prologue): Likewise.
3311 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
3312 (sparc_use_pseudo_pic_reg): New function.
3313 (sparc_init_pic_reg): Likewise.
3314 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
3315 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
3317 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
3319 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
3320 Add item for branch_cost.
3322 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
3324 PR rtl-optimization/83565
3325 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
3326 not extend the result to a larger mode for rotate operations.
3327 (num_sign_bit_copies1): Likewise.
3329 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3332 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
3334 Use values-Xc.o for -pedantic.
3335 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
3337 2018-01-12 Martin Liska <mliska@suse.cz>
3340 * ipa-devirt.c (final_warning_record::grow_type_warnings):
3342 (possible_polymorphic_call_targets): Use it.
3343 (ipa_devirt): Likewise.
3345 2018-01-12 Martin Liska <mliska@suse.cz>
3347 * profile-count.h (enum profile_quality): Use 0 as invalid
3348 enum value of profile_quality.
3350 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
3352 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
3353 -mext-string options.
3355 2018-01-12 Richard Biener <rguenther@suse.de>
3357 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
3358 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
3359 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
3361 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
3363 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
3365 * configure.ac (--with-long-double-format): Add support for the
3366 configuration option to change the default long double format on
3368 * config.gcc (powerpc*-linux*-*): Likewise.
3369 * configure: Regenerate.
3370 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
3371 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
3372 used without modification.
3374 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3376 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
3377 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
3378 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
3379 MISC_BUILTIN_SPEC_BARRIER.
3380 (rs6000_init_builtins): Likewise.
3381 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
3383 (speculation_barrier): New define_insn.
3384 * doc/extend.texi: Document __builtin_speculation_barrier.
3386 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3389 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
3390 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
3391 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
3393 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
3394 integral modes instead of "ss" and "sd".
3395 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
3396 vectors with 32-bit and 64-bit elements.
3397 (vecdupssescalarmodesuffix): New mode attribute.
3398 (vec_dup<mode>): Use it.
3400 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
3403 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
3404 frame if argument is passed on stack.
3406 2018-01-11 Jakub Jelinek <jakub@redhat.com>
3409 * ree.c (combine_reaching_defs): Optimize also
3410 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
3411 reg2=any_extend(exp); reg1=reg2;, formatting fix.
3413 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3416 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
3418 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
3421 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
3422 after they are computed.
3424 2018-01-11 Bin Cheng <bin.cheng@arm.com>
3426 PR tree-optimization/83695
3427 * gimple-loop-linterchange.cc
3428 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
3429 reset cached scev information after interchange.
3430 (pass_linterchange::execute): Remove call to scev_reset_htab.
3432 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3434 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
3435 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
3436 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
3437 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
3438 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
3439 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
3440 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
3441 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
3442 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
3443 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
3444 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
3445 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
3446 (V_lane_reg): Likewise.
3447 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
3449 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
3450 (vfmal_lane_low<mode>_intrinsic,
3451 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
3452 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
3453 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
3454 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
3455 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
3456 vfmsl_lane_high<mode>_intrinsic): New define_insns.
3458 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3460 * config/arm/arm-cpus.in (fp16fml): New feature.
3461 (ALL_SIMD): Add fp16fml.
3462 (armv8.2-a): Add fp16fml as an option.
3463 (armv8.3-a): Likewise.
3464 (armv8.4-a): Add fp16fml as part of fp16.
3465 * config/arm/arm.h (TARGET_FP16FML): Define.
3466 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
3468 * config/arm/arm-modes.def (V2HF): Define.
3469 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
3470 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
3471 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
3472 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
3473 vfmsl_low, vfmsl_high): New set of builtins.
3474 * config/arm/iterators.md (PLUSMINUS): New code iterator.
3475 (vfml_op): New code attribute.
3476 (VFMLHALVES): New int iterator.
3477 (VFML, VFMLSEL): New mode attributes.
3478 (V_reg): Define mapping for V2HF.
3479 (V_hi, V_lo): New mode attributes.
3480 (VF_constraint): Likewise.
3481 (vfml_half, vfml_half_selector): New int attributes.
3482 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
3484 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
3485 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
3487 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
3488 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
3489 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
3490 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
3492 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
3493 Document new effective target and option set.
3495 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3497 * config/arm/arm-cpus.in (armv8_4): New feature.
3498 (ARMv8_4a): New fgroup.
3499 (armv8.4-a): New arch.
3500 * config/arm/arm-tables.opt: Regenerate.
3501 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
3502 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
3503 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
3504 Add matching rules for -march=armv8.4-a and extensions.
3505 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
3507 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
3510 * config/rx/rx.md (BW): New mode attribute.
3511 (sync_lock_test_and_setsi): Add mode suffix to insn output.
3513 2018-01-11 Richard Biener <rguenther@suse.de>
3515 PR tree-optimization/83435
3516 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
3517 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
3518 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
3520 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3521 Alan Hayward <alan.hayward@arm.com>
3522 David Sherwood <david.sherwood@arm.com>
3524 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
3526 (aarch64_classify_address): Initialize it. Track polynomial offsets.
3527 (aarch64_print_address_internal): Use it to check for a zero offset.
3529 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3530 Alan Hayward <alan.hayward@arm.com>
3531 David Sherwood <david.sherwood@arm.com>
3533 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
3534 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
3535 Return a poly_int64 rather than a HOST_WIDE_INT.
3536 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
3537 rather than a HOST_WIDE_INT.
3538 * config/aarch64/aarch64.h (aarch64_frame): Protect with
3539 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
3540 hard_fp_offset, frame_size, initial_adjust, callee_offset and
3541 final_offset from HOST_WIDE_INT to poly_int64.
3542 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
3543 to_constant when getting the number of units in an Advanced SIMD
3545 (aarch64_builtin_vectorized_function): Check for a constant number
3547 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
3549 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
3550 attribute instead of GET_MODE_NUNITS.
3551 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
3552 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
3553 GET_MODE_SIZE for fixed-size registers.
3554 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
3555 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
3556 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
3557 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
3558 (aarch64_print_operand, aarch64_print_address_internal)
3559 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
3560 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
3561 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
3562 Handle polynomial GET_MODE_SIZE.
3563 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
3564 wider than SImode without modification.
3565 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
3566 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
3567 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
3568 passing and returning SVE modes.
3569 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
3570 rather than GEN_INT.
3571 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
3572 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
3573 (aarch64_allocate_and_probe_stack_space): Likewise.
3574 (aarch64_layout_frame): Cope with polynomial offsets.
3575 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
3576 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
3578 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
3579 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
3580 poly_int64 rather than a HOST_WIDE_INT.
3581 (aarch64_get_separate_components, aarch64_process_components)
3582 (aarch64_expand_prologue, aarch64_expand_epilogue)
3583 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
3584 (aarch64_anchor_offset): New function, split out from...
3585 (aarch64_legitimize_address): ...here.
3586 (aarch64_builtin_vectorization_cost): Handle polynomial
3587 TYPE_VECTOR_SUBPARTS.
3588 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
3590 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
3591 number of elements from the PARALLEL rather than the mode.
3592 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
3593 rather than GET_MODE_BITSIZE.
3594 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
3595 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
3596 (aarch64_expand_vec_perm_const_1): Handle polynomial
3597 d->perm.length () and d->perm elements.
3598 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
3599 Apply to_constant to d->perm elements.
3600 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
3601 polynomial CONST_VECTOR_NUNITS.
3602 (aarch64_move_pointer): Take amount as a poly_int64 rather
3604 (aarch64_progress_pointer): Avoid temporary variable.
3605 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
3606 the mode attribute instead of GET_MODE.
3608 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3609 Alan Hayward <alan.hayward@arm.com>
3610 David Sherwood <david.sherwood@arm.com>
3612 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
3613 x exists before using it.
3614 (aarch64_add_constant_internal): Rename to...
3615 (aarch64_add_offset_1): ...this. Replace regnum with separate
3616 src and dest rtxes. Handle the case in which they're different,
3617 including when the offset is zero. Replace scratchreg with an rtx.
3618 Use 2 additions if there is no spare register into which we can
3619 move a 16-bit constant.
3620 (aarch64_add_constant): Delete.
3621 (aarch64_add_offset): Replace reg with separate src and dest
3622 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
3623 Use aarch64_add_offset_1.
3624 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
3625 an rtx rather than an int. Take the delta as a poly_int64
3626 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
3627 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
3628 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
3629 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
3630 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
3632 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
3633 aarch64_add_constant.
3635 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3637 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
3638 Use scalar_float_mode.
3640 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3642 * config/aarch64/aarch64-simd.md
3643 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
3644 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
3645 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
3646 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
3647 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
3648 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
3649 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
3650 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
3651 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
3652 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
3654 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3657 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
3658 targ_options->x_arm_arch_string is non NULL.
3660 2018-01-11 Tamar Christina <tamar.christina@arm.com>
3662 * config/aarch64/aarch64.h
3663 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
3665 2018-01-11 Sudakshina Das <sudi.das@arm.com>
3668 * expmed.c (emit_store_flag_force): Swap if const op0
3669 and change VOIDmode to mode of op0.
3671 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
3673 PR rtl-optimization/83761
3674 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
3675 than bytes to mode_for_size.
3677 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3680 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
3681 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3684 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3687 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3688 when in layout mode.
3689 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3690 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3693 2018-01-10 Michael Collison <michael.collison@arm.com>
3695 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3696 * config/aarch64/aarch64-option-extension.def: Add
3697 AARCH64_OPT_EXTENSION of 'fp16fml'.
3698 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3699 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3700 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3701 * config/aarch64/constraints.md (Ui7): New constraint.
3702 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3703 (VFMLA_SEL_W): Ditto.
3706 (VFMLA16_LOW): New int iterator.
3707 (VFMLA16_HIGH): Ditto.
3708 (UNSPEC_FMLAL): New unspec.
3709 (UNSPEC_FMLSL): Ditto.
3710 (UNSPEC_FMLAL2): Ditto.
3711 (UNSPEC_FMLSL2): Ditto.
3712 (f16mac): New code attribute.
3713 * config/aarch64/aarch64-simd-builtins.def
3714 (aarch64_fmlal_lowv2sf): Ditto.
3715 (aarch64_fmlsl_lowv2sf): Ditto.
3716 (aarch64_fmlalq_lowv4sf): Ditto.
3717 (aarch64_fmlslq_lowv4sf): Ditto.
3718 (aarch64_fmlal_highv2sf): Ditto.
3719 (aarch64_fmlsl_highv2sf): Ditto.
3720 (aarch64_fmlalq_highv4sf): Ditto.
3721 (aarch64_fmlslq_highv4sf): Ditto.
3722 (aarch64_fmlal_lane_lowv2sf): Ditto.
3723 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3724 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3725 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3726 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3727 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3728 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3729 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3730 (aarch64_fmlal_lane_highv2sf): Ditto.
3731 (aarch64_fmlsl_lane_highv2sf): Ditto.
3732 (aarch64_fmlal_laneq_highv2sf): Ditto.
3733 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3734 (aarch64_fmlalq_lane_highv4sf): Ditto.
3735 (aarch64_fmlsl_lane_highv4sf): Ditto.
3736 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3737 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3738 * config/aarch64/aarch64-simd.md:
3739 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3740 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3741 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3742 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3743 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3744 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3745 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3746 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3747 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3748 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3749 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3750 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3751 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3752 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3753 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3754 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3755 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3756 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3757 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3758 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3759 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3760 (vfmlsl_low_u32): Ditto.
3761 (vfmlalq_low_u32): Ditto.
3762 (vfmlslq_low_u32): Ditto.
3763 (vfmlal_high_u32): Ditto.
3764 (vfmlsl_high_u32): Ditto.
3765 (vfmlalq_high_u32): Ditto.
3766 (vfmlslq_high_u32): Ditto.
3767 (vfmlal_lane_low_u32): Ditto.
3768 (vfmlsl_lane_low_u32): Ditto.
3769 (vfmlal_laneq_low_u32): Ditto.
3770 (vfmlsl_laneq_low_u32): Ditto.
3771 (vfmlalq_lane_low_u32): Ditto.
3772 (vfmlslq_lane_low_u32): Ditto.
3773 (vfmlalq_laneq_low_u32): Ditto.
3774 (vfmlslq_laneq_low_u32): Ditto.
3775 (vfmlal_lane_high_u32): Ditto.
3776 (vfmlsl_lane_high_u32): Ditto.
3777 (vfmlal_laneq_high_u32): Ditto.
3778 (vfmlsl_laneq_high_u32): Ditto.
3779 (vfmlalq_lane_high_u32): Ditto.
3780 (vfmlslq_lane_high_u32): Ditto.
3781 (vfmlalq_laneq_high_u32): Ditto.
3782 (vfmlslq_laneq_high_u32): Ditto.
3783 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3784 (AARCH64_FL_FOR_ARCH8_4): New.
3785 (AARCH64_ISA_F16FML): New ISA flag.
3786 (TARGET_F16FML): New feature flag for fp16fml.
3787 (doc/invoke.texi): Document new fp16fml option.
3789 2018-01-10 Michael Collison <michael.collison@arm.com>
3791 * config/aarch64/aarch64-builtins.c:
3792 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3793 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3794 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3795 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3796 (AARCH64_ISA_SHA3): New ISA flag.
3797 (TARGET_SHA3): New feature flag for sha3.
3798 * config/aarch64/iterators.md (sha512_op): New int attribute.
3799 (CRYPTO_SHA512): New int iterator.
3800 (UNSPEC_SHA512H): New unspec.
3801 (UNSPEC_SHA512H2): Ditto.
3802 (UNSPEC_SHA512SU0): Ditto.
3803 (UNSPEC_SHA512SU1): Ditto.
3804 * config/aarch64/aarch64-simd-builtins.def
3805 (aarch64_crypto_sha512hqv2di): New builtin.
3806 (aarch64_crypto_sha512h2qv2di): Ditto.
3807 (aarch64_crypto_sha512su0qv2di): Ditto.
3808 (aarch64_crypto_sha512su1qv2di): Ditto.
3809 (aarch64_eor3qv8hi): Ditto.
3810 (aarch64_rax1qv2di): Ditto.
3811 (aarch64_xarqv2di): Ditto.
3812 (aarch64_bcaxqv8hi): Ditto.
3813 * config/aarch64/aarch64-simd.md:
3814 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3815 (aarch64_crypto_sha512su0qv2di): Ditto.
3816 (aarch64_crypto_sha512su1qv2di): Ditto.
3817 (aarch64_eor3qv8hi): Ditto.
3818 (aarch64_rax1qv2di): Ditto.
3819 (aarch64_xarqv2di): Ditto.
3820 (aarch64_bcaxqv8hi): Ditto.
3821 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3822 (vsha512h2q_u64): Ditto.
3823 (vsha512su0q_u64): Ditto.
3824 (vsha512su1q_u64): Ditto.
3825 (veor3q_u16): Ditto.
3826 (vrax1q_u64): Ditto.
3828 (vbcaxq_u16): Ditto.
3829 * config/arm/types.md (crypto_sha512): New type attribute.
3830 (crypto_sha3): Ditto.
3831 (doc/invoke.texi): Document new sha3 option.
3833 2018-01-10 Michael Collison <michael.collison@arm.com>
3835 * config/aarch64/aarch64-builtins.c:
3836 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3837 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3838 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3839 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3840 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3841 (AARCH64_ISA_SM4): New ISA flag.
3842 (TARGET_SM4): New feature flag for sm4.
3843 * config/aarch64/aarch64-simd-builtins.def
3844 (aarch64_sm3ss1qv4si): Ditto.
3845 (aarch64_sm3tt1aq4si): Ditto.
3846 (aarch64_sm3tt1bq4si): Ditto.
3847 (aarch64_sm3tt2aq4si): Ditto.
3848 (aarch64_sm3tt2bq4si): Ditto.
3849 (aarch64_sm3partw1qv4si): Ditto.
3850 (aarch64_sm3partw2qv4si): Ditto.
3851 (aarch64_sm4eqv4si): Ditto.
3852 (aarch64_sm4ekeyqv4si): Ditto.
3853 * config/aarch64/aarch64-simd.md:
3854 (aarch64_sm3ss1qv4si): Ditto.
3855 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3856 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3857 (aarch64_sm4eqv4si): Ditto.
3858 (aarch64_sm4ekeyqv4si): Ditto.
3859 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3860 (sm3part_op): Ditto.
3861 (CRYPTO_SM3TT): Ditto.
3862 (CRYPTO_SM3PART): Ditto.
3863 (UNSPEC_SM3SS1): New unspec.
3864 (UNSPEC_SM3TT1A): Ditto.
3865 (UNSPEC_SM3TT1B): Ditto.
3866 (UNSPEC_SM3TT2A): Ditto.
3867 (UNSPEC_SM3TT2B): Ditto.
3868 (UNSPEC_SM3PARTW1): Ditto.
3869 (UNSPEC_SM3PARTW2): Ditto.
3870 (UNSPEC_SM4E): Ditto.
3871 (UNSPEC_SM4EKEY): Ditto.
3872 * config/aarch64/constraints.md (Ui2): New constraint.
3873 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3874 * config/arm/types.md (crypto_sm3): New type attribute.
3875 (crypto_sm4): Ditto.
3876 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3877 (vsm3tt1aq_u32): Ditto.
3878 (vsm3tt1bq_u32): Ditto.
3879 (vsm3tt2aq_u32): Ditto.
3880 (vsm3tt2bq_u32): Ditto.
3881 (vsm3partw1q_u32): Ditto.
3882 (vsm3partw2q_u32): Ditto.
3883 (vsm4eq_u32): Ditto.
3884 (vsm4ekeyq_u32): Ditto.
3885 (doc/invoke.texi): Document new sm4 option.
3887 2018-01-10 Michael Collison <michael.collison@arm.com>
3889 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3890 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3891 (AARCH64_FL_FOR_ARCH8_4): New.
3892 (AARCH64_FL_V8_4): New flag.
3893 (doc/invoke.texi): Document new armv8.4-a option.
3895 2018-01-10 Michael Collison <michael.collison@arm.com>
3897 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3898 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3899 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3900 * config/aarch64/aarch64-option-extension.def: Add
3901 AARCH64_OPT_EXTENSION of 'sha2'.
3902 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3903 (crypto): Disable sha2 and aes if crypto disabled.
3904 (crypto): Enable aes and sha2 if enabled.
3905 (simd): Disable sha2 and aes if simd disabled.
3906 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3908 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3909 (TARGET_SHA2): New feature flag for sha2.
3910 (TARGET_AES): New feature flag for aes.
3911 * config/aarch64/aarch64-simd.md:
3912 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3913 conditional on TARGET_AES.
3914 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3915 (aarch64_crypto_sha1hsi): Make pattern conditional
3917 (aarch64_crypto_sha1hv4si): Ditto.
3918 (aarch64_be_crypto_sha1hv4si): Ditto.
3919 (aarch64_crypto_sha1su1v4si): Ditto.
3920 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3921 (aarch64_crypto_sha1su0v4si): Ditto.
3922 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3923 (aarch64_crypto_sha256su0v4si): Ditto.
3924 (aarch64_crypto_sha256su1v4si): Ditto.
3925 (doc/invoke.texi): Document new aes and sha2 options.
3927 2018-01-10 Martin Sebor <msebor@redhat.com>
3929 PR tree-optimization/83781
3930 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3933 2018-01-11 Martin Sebor <msebor@gmail.com>
3934 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3936 PR tree-optimization/83501
3937 PR tree-optimization/81703
3939 * tree-ssa-strlen.c (get_string_cst): Rename...
3940 (get_string_len): ...to this. Handle global constants.
3941 (handle_char_store): Adjust.
3943 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3944 Jim Wilson <jimw@sifive.com>
3946 * config/riscv/riscv-protos.h (riscv_output_return): New.
3947 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3948 (riscv_attribute_table, riscv_output_return),
3949 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3950 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3951 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3953 (riscv_expand_prologue): Add early return for naked function.
3954 (riscv_expand_epilogue): Likewise.
3955 (riscv_function_ok_for_sibcall): Return false for naked function.
3956 (riscv_set_current_function): New.
3957 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3958 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3959 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3960 * doc/extend.texi (RISC-V Function Attributes): New.
3962 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3964 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3965 check for 128-bit long double before checking TCmode.
3966 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3967 128-bit long doubles before checking TFmode or TCmode.
3968 (FLOAT128_IBM_P): Likewise.
3970 2018-01-10 Martin Sebor <msebor@redhat.com>
3972 PR tree-optimization/83671
3973 * builtins.c (c_strlen): Unconditionally return zero for the empty
3975 Use -Warray-bounds for warnings.
3976 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3977 for non-constant array indices with COMPONENT_REF, arrays of
3978 arrays, and pointers to arrays.
3979 (gimple_fold_builtin_strlen): Determine and set length range for
3980 non-constant character arrays.
3982 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3985 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3988 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3990 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3992 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3995 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3996 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3997 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3998 indexed_or_indirect_operand predicate.
3999 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
4000 (*vsx_le_perm_load_v8hi): Likewise.
4001 (*vsx_le_perm_load_v16qi): Likewise.
4002 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
4003 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
4004 (*vsx_le_perm_store_v8hi): Likewise.
4005 (*vsx_le_perm_store_v16qi): Likewise.
4006 (eight unnamed splitters): Likewise.
4008 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
4010 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
4011 * config/rs6000/emmintrin.h: Likewise.
4012 * config/rs6000/mmintrin.h: Likewise.
4013 * config/rs6000/xmmintrin.h: Likewise.
4015 2018-01-10 David Malcolm <dmalcolm@redhat.com>
4018 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
4020 * tree.c (tree_nop_conversion): Return true for location wrapper
4022 (maybe_wrap_with_location): New function.
4023 (selftest::check_strip_nops): New function.
4024 (selftest::test_location_wrappers): New function.
4025 (selftest::tree_c_tests): Call it.
4026 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
4027 (maybe_wrap_with_location): New decl.
4028 (EXPR_LOCATION_WRAPPER_P): New macro.
4029 (location_wrapper_p): New inline function.
4030 (tree_strip_any_location_wrapper): New inline function.
4032 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
4035 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
4036 stack_realign_offset for the largest alignment of stack slot
4038 (ix86_find_max_used_stack_alignment): New function.
4039 (ix86_finalize_stack_frame_flags): Use it. Set
4040 max_used_stack_alignment if we don't realign stack.
4041 * config/i386/i386.h (machine_function): Add
4042 max_used_stack_alignment.
4044 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
4046 * config/arm/arm.opt (-mbranch-cost): New option.
4047 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
4050 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
4053 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
4054 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
4056 2018-01-10 Richard Biener <rguenther@suse.de>
4059 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
4060 early out so it also covers the case where we have a non-NULL
4063 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4065 PR tree-optimization/83753
4066 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
4067 for non-strided grouped accesses if the number of elements is 1.
4069 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
4072 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
4073 * i386.h (TARGET_USE_GATHER): Define.
4074 * x86-tune.def (X86_TUNE_USE_GATHER): New.
4076 2018-01-10 Martin Liska <mliska@suse.cz>
4079 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
4080 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
4082 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
4083 CLEANUP_NO_PARTITIONING is not set.
4085 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
4087 * doc/rtl.texi: Remove documentation of (const ...) wrappers
4088 for vectors, as a partial revert of r254296.
4089 * rtl.h (const_vec_p): Delete.
4090 (const_vec_duplicate_p): Don't test for vector CONSTs.
4091 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
4092 * expmed.c (make_tree): Likewise.
4095 * common.md (E, F): Use CONSTANT_P instead of checking for
4097 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
4098 checking for CONST_VECTOR.
4100 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4103 * predict.c (force_edge_cold): Handle in more sane way edges
4106 2018-01-09 Carl Love <cel@us.ibm.com>
4108 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
4110 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
4111 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
4112 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
4113 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
4114 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
4115 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
4116 * config/rs6000/rs6000-protos.h: Add extern defition for
4117 rs6000_generate_float2_double_code.
4118 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
4120 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
4121 (float2_v2df): Add define_expand.
4123 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
4126 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
4127 op_mode in the force_to_mode call.
4129 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4131 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
4132 instead of checking each element individually.
4133 (aarch64_evpc_uzp): Likewise.
4134 (aarch64_evpc_zip): Likewise.
4135 (aarch64_evpc_ext): Likewise.
4136 (aarch64_evpc_rev): Likewise.
4137 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
4138 instead of checking each element individually. Return true without
4140 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
4141 whether all selected elements come from the same input, instead of
4142 checking each element individually. Remove calls to gen_rtx_REG,
4143 start_sequence and end_sequence and instead assert that no rtl is
4146 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4148 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
4149 order of HIGH and CONST checks.
4151 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
4153 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
4154 if the destination isn't an SSA_NAME.
4156 2018-01-09 Richard Biener <rguenther@suse.de>
4158 PR tree-optimization/83668
4159 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
4161 (canonicalize_loop_form): ... here, renamed from ...
4162 (canonicalize_loop_closed_ssa_form): ... this and amended to
4163 swap successor edges for loop exit blocks to make us use
4164 the RPO order we need for initial schedule generation.
4166 2018-01-09 Joseph Myers <joseph@codesourcery.com>
4168 PR tree-optimization/64811
4169 * match.pd: When optimizing comparisons with Inf, avoid
4170 introducing or losing exceptions from comparisons with NaN.
4172 2018-01-09 Martin Liska <mliska@suse.cz>
4175 * asan.c (shadow_mem_size): Add gcc_assert.
4177 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
4179 Don't save registers in main().
4182 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
4183 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
4184 * config/avr/avr.c (avr_set_current_function): Don't error if
4185 naked, OS_task or OS_main are specified at the same time.
4186 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
4188 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
4190 * common/config/avr/avr-common.c (avr_option_optimization_table):
4191 Switch on -mmain-is-OS_task for optimizing compilations.
4193 2018-01-09 Richard Biener <rguenther@suse.de>
4195 PR tree-optimization/83572
4196 * graphite.c: Include cfganal.h.
4197 (graphite_transform_loops): Connect infinite loops to exit
4198 and remove fake edges at the end.
4200 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4202 * ipa-inline.c (edge_badness): Revert accidental checkin.
4204 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
4207 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
4208 symbols; not inline clones.
4210 2018-01-09 Jakub Jelinek <jakub@redhat.com>
4213 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
4214 hard registers. Formatting fixes.
4216 PR preprocessor/83722
4217 * gcc.c (try_generate_repro): Pass
4218 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
4219 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
4222 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
4223 Kito Cheng <kito.cheng@gmail.com>
4225 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
4226 (riscv_leaf_function_p): Delete.
4227 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
4229 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4231 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
4233 (do_ifelse): New function.
4234 (do_isel): New function.
4235 (do_sub3): New function.
4236 (do_add3): New function.
4237 (do_load_mask_compare): New function.
4238 (do_overlap_load_compare): New function.
4239 (expand_compare_loop): New function.
4240 (expand_block_compare): Call expand_compare_loop() when appropriate.
4241 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
4243 (-mblock-compare-inline-loop-limit): New option.
4245 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
4248 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
4249 Reverse order of second and third operands in first alternative.
4250 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
4251 of first and second elements in UNSPEC_VPERMR vector.
4252 (altivec_expand_vec_perm_le): Likewise.
4254 2018-01-08 Jeff Law <law@redhat.com>
4256 PR rtl-optimizatin/81308
4257 * tree-switch-conversion.c (cfg_altered): New file scoped static.
4258 (process_switch): If group_case_labels makes a change, then set
4260 (pass_convert_switch::execute): If a switch is converted, then
4261 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
4263 PR rtl-optimization/81308
4264 * recog.c (split_all_insns): Conditionally cleanup the CFG after
4267 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
4269 PR target/83663 - Revert r255946
4270 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
4271 generation for cases where splatting a value is not useful.
4272 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
4273 across a vec_duplicate and a paradoxical subreg forming a vector
4274 mode to a vec_concat.
4276 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
4278 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
4279 -march=armv8.3-a variants.
4280 * config/arm/t-multilib: Likewise.
4281 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
4283 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
4285 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
4287 (cceq_ior_compare_complement): Give it a name so I can use it, and
4288 change boolean_or_operator predicate to boolean_operator so it can
4289 be used to generate a crand.
4290 (eqne): New code iterator.
4291 (bd/bd_neg): New code_attrs.
4292 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
4293 a single define_insn.
4294 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
4295 decrement (bdnzt/bdnzf/bdzt/bdzf).
4296 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
4297 with the new names of the branch decrement patterns, and added the
4298 names of the branch decrement conditional patterns.
4300 2018-01-08 Richard Biener <rguenther@suse.de>
4302 PR tree-optimization/83563
4303 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
4306 2018-01-08 Richard Biener <rguenther@suse.de>
4309 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
4311 2018-01-08 Richard Biener <rguenther@suse.de>
4313 PR tree-optimization/83685
4314 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
4315 references to abnormals.
4317 2018-01-08 Richard Biener <rguenther@suse.de>
4320 * dwarf2out.c (output_indirect_strings): Handle empty
4321 skeleton_debug_str_hash.
4322 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
4324 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4326 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
4327 (emit_store_direct): Likewise.
4328 (arc_trampoline_adjust_address): Likewise.
4329 (arc_asm_trampoline_template): New function.
4330 (arc_initialize_trampoline): Use asm_trampoline_template.
4331 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
4332 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
4333 * config/arc/arc.md (flush_icache): Delete pattern.
4335 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
4337 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
4338 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
4341 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4344 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
4345 by not USED_FOR_TARGET.
4346 (make_pass_resolve_sw_modes): Likewise.
4348 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
4350 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
4353 2018-01-08 Richard Biener <rguenther@suse.de>
4356 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
4358 2018-01-08 Richard Biener <rguenther@suse.de>
4361 * match.pd ((t * 2) / 2) -> t): Add missing :c.
4363 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
4366 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
4367 basic blocks with a small number of successors.
4368 (convert_control_dep_chain_into_preds): Improve handling of
4370 (dump_predicates): Split apart into...
4371 (dump_pred_chain): ...here...
4372 (dump_pred_info): ...and here.
4373 (can_one_predicate_be_invalidated_p): Add debugging printfs.
4374 (can_chain_union_be_invalidated_p): Improve check for invalidation
4376 (uninit_uses_cannot_happen): Avoid unnecessary if
4377 convert_control_dep_chain_into_preds yielded nothing.
4379 2018-01-06 Martin Sebor <msebor@redhat.com>
4381 PR tree-optimization/83640
4382 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
4383 subtracting negative offset from size.
4384 (builtin_access::overlap): Adjust offset bounds of the access to fall
4385 within the size of the object if possible.
4387 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
4389 PR rtl-optimization/83699
4390 * expmed.c (extract_bit_field_1): Restrict the vector usage of
4391 extract_bit_field_as_subreg to cases in which the extracted
4392 value is also a vector.
4394 * lra-constraints.c (process_alt_operands): Test for the equivalence
4395 substitutions when detecting a possible reload cycle.
4397 2018-01-06 Jakub Jelinek <jakub@redhat.com>
4400 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
4401 by default if flag_selective_schedling{,2}. Formatting fixes.
4403 PR rtl-optimization/83682
4404 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
4405 if it has non-VECTOR_MODE element mode.
4406 (vec_duplicate_p): Likewise.
4409 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
4410 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
4412 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4415 * config/i386/i386-builtin.def
4416 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
4417 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
4418 Require also OPTION_MASK_ISA_AVX512F in addition to
4419 OPTION_MASK_ISA_GFNI.
4420 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
4421 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
4422 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
4423 to OPTION_MASK_ISA_GFNI.
4424 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
4425 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
4426 OPTION_MASK_ISA_AVX512BW.
4427 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
4428 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
4429 addition to OPTION_MASK_ISA_GFNI.
4430 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
4431 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
4432 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
4433 to OPTION_MASK_ISA_GFNI.
4434 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
4435 a requirement for all ISAs rather than any of them with a few
4437 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
4439 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
4440 bitmasks to be enabled with 3 exceptions, instead of requiring any
4441 enabled ISA with lots of exceptions.
4442 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
4443 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
4444 Change avx512bw in isa attribute to avx512f.
4445 * config/i386/sgxintrin.h: Add license boilerplate.
4446 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
4447 to __AVX512F__ and __AVX512VL to __AVX512VL__.
4448 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
4449 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
4451 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
4452 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
4453 temporarily sse2 rather than sse if not enabled already.
4456 * config/i386/sse.md (VI248_VLBW): Rename to ...
4457 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
4458 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
4459 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
4460 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
4461 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
4462 mode iterator instead of VI248_VLBW.
4464 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
4466 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
4467 (record_modified): Skip clobbers; add debug output.
4468 (param_change_prob): Use sreal frequencies.
4470 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4472 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
4473 punt for user-aligned variables.
4475 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
4477 * tree-chrec.c (chrec_contains_symbols): Return true for
4480 2018-01-05 Sudakshina Das <sudi.das@arm.com>
4483 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
4484 of (x|y) == x for BICS pattern.
4486 2018-01-05 Jakub Jelinek <jakub@redhat.com>
4488 PR tree-optimization/83605
4489 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
4490 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
4493 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
4495 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
4496 * config/epiphany/rtems.h: New file.
4498 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4499 Uros Bizjak <ubizjak@gmail.com>
4502 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
4503 QIreg_operand instead of register_operand predicate.
4504 * config/i386/i386.c (ix86_rop_should_change_byte_p,
4505 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
4506 comments instead of -fmitigate[-_]rop.
4508 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
4511 * cgraphunit.c (symbol_table::compile): Switch to text_section
4512 before calling assembly_start debug hook.
4513 * run-rtl-passes.c (run_rtl_passes): Likewise.
4516 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4518 * tree-vrp.c (extract_range_from_binary_expr_1): Check
4519 range_int_cst_p rather than !symbolic_range_p before calling
4520 extract_range_from_multiplicative_op_1.
4522 2018-01-04 Jeff Law <law@redhat.com>
4524 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
4525 redundant test in assertion.
4527 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4529 * doc/rtl.texi: Document machine_mode wrapper classes.
4531 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4533 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
4536 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4538 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
4539 the VEC_PERM_EXPR fold to fail.
4541 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4544 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
4545 to switched_sections.
4547 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4550 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
4553 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
4556 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
4557 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
4559 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4562 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
4563 is BLKmode and bitpos not zero or mode change is needed.
4565 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
4568 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
4571 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
4574 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
4575 instead of MULT rtx. Update all corresponding splitters.
4577 (*ssub<modesuffix>): Ditto.
4579 (*cmp_sadd_di): Update split patterns.
4580 (*cmp_sadd_si): Ditto.
4581 (*cmp_sadd_sidi): Ditto.
4582 (*cmp_ssub_di): Ditto.
4583 (*cmp_ssub_si): Ditto.
4584 (*cmp_ssub_sidi): Ditto.
4585 * config/alpha/predicates.md (const23_operand): New predicate.
4586 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
4587 Look for ASHIFT, not MULT inner operand.
4588 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
4590 2018-01-04 Martin Liska <mliska@suse.cz>
4592 PR gcov-profile/83669
4593 * gcov.c (output_intermediate_file): Add version to intermediate
4595 * doc/gcov.texi: Document new field 'version' in intermediate
4596 file format. Fix location of '-k' option of gcov command.
4598 2018-01-04 Martin Liska <mliska@suse.cz>
4601 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
4603 2018-01-04 Jakub Jelinek <jakub@redhat.com>
4605 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
4607 2018-01-03 Martin Sebor <msebor@redhat.com>
4609 PR tree-optimization/83655
4610 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
4611 checking calls with invalid arguments.
4613 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4615 * tree-vect-stmts.c (vect_get_store_rhs): New function.
4616 (vectorizable_mask_load_store): Delete.
4617 (vectorizable_call): Return false for masked loads and stores.
4618 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
4619 instead of gimple_assign_rhs1.
4620 (vectorizable_load): Handle IFN_MASK_LOAD.
4621 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
4623 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4625 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
4627 (vectorizable_mask_load_store): ...here.
4628 (vectorizable_load): ...and here.
4630 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4632 * tree-vect-stmts.c (vect_build_all_ones_mask)
4633 (vect_build_zero_merge_argument): New functions, split out from...
4634 (vectorizable_load): ...here.
4636 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4638 * tree-vect-stmts.c (vect_check_store_rhs): New function,
4640 (vectorizable_mask_load_store): ...here.
4641 (vectorizable_store): ...and here.
4643 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4645 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
4647 (vectorizable_mask_load_store): ...here.
4649 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4651 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
4652 (vect_model_store_cost): Take a vec_load_store_type instead of a
4654 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
4655 (vect_model_store_cost): Take a vec_load_store_type instead of a
4657 (vectorizable_mask_load_store): Update accordingly.
4658 (vectorizable_store): Likewise.
4659 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
4661 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4663 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
4664 IFN_MASK_LOAD calls here rather than...
4665 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
4667 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4668 Alan Hayward <alan.hayward@arm.com>
4669 David Sherwood <david.sherwood@arm.com>
4671 * expmed.c (extract_bit_field_1): For vector extracts,
4672 fall back to extract_bit_field_as_subreg if vec_extract
4675 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4676 Alan Hayward <alan.hayward@arm.com>
4677 David Sherwood <david.sherwood@arm.com>
4679 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
4680 they are variable or constant sized.
4681 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
4682 slots for constant-sized data.
4684 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4685 Alan Hayward <alan.hayward@arm.com>
4686 David Sherwood <david.sherwood@arm.com>
4688 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4689 handling COND_EXPRs with boolean comparisons, try to find a better
4690 basis for the mask type than the boolean itself.
4692 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4694 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4695 is calculated and how it can be overridden.
4696 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4697 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4699 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4702 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4703 Alan Hayward <alan.hayward@arm.com>
4704 David Sherwood <david.sherwood@arm.com>
4706 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4707 Remove the mode argument.
4708 (aarch64_simd_valid_immediate): Remove the mode and inverse
4710 * config/aarch64/iterators.md (bitsize): New iterator.
4711 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4712 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4713 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4714 aarch64_simd_valid_immediate.
4715 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4716 (aarch64_reg_or_bic_imm): Likewise.
4717 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4718 with an insn_type enum and msl with a modifier_type enum.
4719 Replace element_width with a scalar_mode. Change the shift
4720 to unsigned int. Add constructors for scalar_float_mode and
4721 scalar_int_mode elements.
4722 (aarch64_vect_float_const_representable_p): Delete.
4723 (aarch64_can_const_movi_rtx_p)
4724 (aarch64_simd_scalar_immediate_valid_for_move)
4725 (aarch64_simd_make_constant): Update call to
4726 aarch64_simd_valid_immediate.
4727 (aarch64_advsimd_valid_immediate_hs): New function.
4728 (aarch64_advsimd_valid_immediate): Likewise.
4729 (aarch64_simd_valid_immediate): Remove mode and inverse
4730 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4731 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4732 and aarch64_float_const_representable_p on the result.
4733 (aarch64_output_simd_mov_immediate): Remove mode argument.
4734 Update call to aarch64_simd_valid_immediate and use of
4735 simd_immediate_info.
4736 (aarch64_output_scalar_simd_mov_immediate): Update call
4739 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4740 Alan Hayward <alan.hayward@arm.com>
4741 David Sherwood <david.sherwood@arm.com>
4743 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4744 (mode_nunits): Likewise CONST_MODE_NUNITS.
4745 * machmode.def (ADJUST_NUNITS): Document.
4746 * genmodes.c (mode_data::need_nunits_adj): New field.
4747 (blank_mode): Update accordingly.
4748 (adj_nunits): New variable.
4749 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4751 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4752 listed in adj_nunits.
4753 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4754 listed in adj_nunits. Don't emit case statements for such modes.
4755 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4756 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4757 nothing if adj_nunits is nonnull.
4758 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4759 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4760 (emit_mode_fbit): Update use of print_maybe_const_decl.
4761 (emit_move_size): Likewise. Treat the array as non-const
4763 (emit_mode_adjustments): Handle adj_nunits.
4765 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4767 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4768 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4769 (VECTOR_MODES): Use it.
4770 (make_vector_modes): Take the prefix as an argument.
4772 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4773 Alan Hayward <alan.hayward@arm.com>
4774 David Sherwood <david.sherwood@arm.com>
4776 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4777 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4778 for MODE_VECTOR_BOOL.
4779 * machmode.def (VECTOR_BOOL_MODE): Document.
4780 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4781 (make_vector_bool_mode): New function.
4782 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4784 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4785 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4787 * stor-layout.c (int_mode_for_mode): Likewise.
4788 * tree.c (build_vector_type_for_mode): Likewise.
4789 * varasm.c (output_constant_pool_2): Likewise.
4790 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4791 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4792 for MODE_VECTOR_BOOL.
4793 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4794 of mode class checks.
4795 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4796 instead of a list of mode class checks.
4797 (expand_vector_scalar_condition): Likewise.
4798 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4800 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4801 Alan Hayward <alan.hayward@arm.com>
4802 David Sherwood <david.sherwood@arm.com>
4804 * machmode.h (mode_size): Change from unsigned short to
4806 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4807 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4808 or if measurement_type is not polynomial.
4809 (fixed_size_mode::includes_p): Check for constant-sized modes.
4810 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4811 return a poly_uint16 rather than an unsigned short.
4812 (emit_mode_size): Change the type of mode_size from unsigned short
4813 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4814 (emit_mode_adjustments): Cope with polynomial vector sizes.
4815 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4817 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4819 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4820 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4821 * caller-save.c (setup_save_areas): Likewise.
4822 (replace_reg_with_saved_mem): Likewise.
4823 * calls.c (emit_library_call_value_1): Likewise.
4824 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4825 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4826 (gen_lowpart_for_combine): Likewise.
4827 * convert.c (convert_to_integer_1): Likewise.
4828 * cse.c (equiv_constant, cse_insn): Likewise.
4829 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4830 (cselib_subst_to_values): Likewise.
4831 * dce.c (word_dce_process_block): Likewise.
4832 * df-problems.c (df_word_lr_mark_ref): Likewise.
4833 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4834 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4835 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4836 (rtl_for_decl_location): Likewise.
4837 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4838 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4839 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4840 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4841 (expand_expr_real_1): Likewise.
4842 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4843 (pad_below): Likewise.
4844 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4845 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4846 * ira.c (get_subreg_tracking_sizes): Likewise.
4847 * ira-build.c (ira_create_allocno_objects): Likewise.
4848 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4849 (ira_sort_regnos_for_alter_reg): Likewise.
4850 * ira-costs.c (record_operand_costs): Likewise.
4851 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4852 (resolve_simple_move): Likewise.
4853 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4854 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4855 (lra_constraints): Likewise.
4856 (CONST_POOL_OK_P): Reject variable-sized modes.
4857 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4858 (add_pseudo_to_slot, lra_spill): Likewise.
4859 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4860 * optabs-query.c (get_best_extraction_insn): Likewise.
4861 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4862 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4863 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4864 * recog.c (offsettable_address_addr_space_p): Likewise.
4865 * regcprop.c (maybe_mode_change): Likewise.
4866 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4867 * regrename.c (build_def_use): Likewise.
4868 * regstat.c (dump_reg_info): Likewise.
4869 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4870 (find_reloads, find_reloads_subreg_address): Likewise.
4871 * reload1.c (eliminate_regs_1): Likewise.
4872 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4873 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4874 (simplify_binary_operation_1, simplify_subreg): Likewise.
4875 * targhooks.c (default_function_arg_padding): Likewise.
4876 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4877 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4878 (verify_gimple_assign_ternary): Likewise.
4879 * tree-inline.c (estimate_move_cost): Likewise.
4880 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4881 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4882 (get_address_cost_ainc): Likewise.
4883 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4884 (vect_supportable_dr_alignment): Likewise.
4885 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4886 (vectorizable_reduction): Likewise.
4887 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4888 (vectorizable_operation, vectorizable_load): Likewise.
4889 * tree.c (build_same_sized_truth_vector_type): Likewise.
4890 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4891 * var-tracking.c (emit_note_insn_var_location): Likewise.
4892 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4893 (ADDR_VEC_ALIGN): Likewise.
4895 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4896 Alan Hayward <alan.hayward@arm.com>
4897 David Sherwood <david.sherwood@arm.com>
4899 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4901 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4902 or if measurement_type is polynomial.
4903 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4904 * combine.c (make_extraction): Likewise.
4905 * dse.c (find_shift_sequence): Likewise.
4906 * dwarf2out.c (mem_loc_descriptor): Likewise.
4907 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4908 (extract_bit_field, extract_low_bits): Likewise.
4909 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4910 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4911 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4912 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4913 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4914 * reload.c (find_reloads): Likewise.
4915 * reload1.c (alter_reg): Likewise.
4916 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4917 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4918 * tree-if-conv.c (predicate_mem_writes): Likewise.
4919 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4920 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4921 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4922 * valtrack.c (dead_debug_insert_temp): Likewise.
4923 * varasm.c (mergeable_constant_section): Likewise.
4924 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4926 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4927 Alan Hayward <alan.hayward@arm.com>
4928 David Sherwood <david.sherwood@arm.com>
4930 * expr.c (expand_assignment): Cope with polynomial mode sizes
4931 when assigning to a CONCAT.
4933 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4934 Alan Hayward <alan.hayward@arm.com>
4935 David Sherwood <david.sherwood@arm.com>
4937 * machmode.h (mode_precision): Change from unsigned short to
4939 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4941 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4942 or if measurement_type is not polynomial.
4943 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4944 in which the mode is already known to be a scalar_int_mode.
4945 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4946 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4948 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4949 for GET_MODE_PRECISION.
4950 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4951 for GET_MODE_PRECISION.
4952 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4954 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4955 (expand_field_assignment, make_extraction): Likewise.
4956 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4957 (get_last_value): Likewise.
4958 * convert.c (convert_to_integer_1): Likewise.
4959 * cse.c (cse_insn): Likewise.
4960 * expr.c (expand_expr_real_1): Likewise.
4961 * lra-constraints.c (simplify_operand_subreg): Likewise.
4962 * optabs-query.c (can_atomic_load_p): Likewise.
4963 * optabs.c (expand_atomic_load): Likewise.
4964 (expand_atomic_store): Likewise.
4965 * ree.c (combine_reaching_defs): Likewise.
4966 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4967 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4968 * tree.h (type_has_mode_precision_p): Likewise.
4969 * ubsan.c (instrument_si_overflow): Likewise.
4971 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4972 Alan Hayward <alan.hayward@arm.com>
4973 David Sherwood <david.sherwood@arm.com>
4975 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4976 polynomial numbers of units.
4977 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4978 (valid_vector_subparts_p): New function.
4979 (build_vector_type): Remove temporary shim and take the number
4980 of units as a poly_uint64 rather than an int.
4981 (build_opaque_vector_type): Take the number of units as a
4982 poly_uint64 rather than an int.
4983 * tree.c (build_vector_from_ctor): Handle polynomial
4984 TYPE_VECTOR_SUBPARTS.
4985 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4986 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4987 (build_vector_from_val): If the number of units is variable,
4988 use build_vec_duplicate_cst for constant operands and
4989 VEC_DUPLICATE_EXPR otherwise.
4990 (make_vector_type): Remove temporary is_constant ().
4991 (build_vector_type, build_opaque_vector_type): Take the number of
4992 units as a poly_uint64 rather than an int.
4993 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4995 * cfgexpand.c (expand_debug_expr): Likewise.
4996 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4997 (store_constructor, expand_expr_real_1): Likewise.
4998 (const_scalar_mask_from_tree): Likewise.
4999 * fold-const-call.c (fold_const_reduction): Likewise.
5000 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
5001 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
5002 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
5003 (fold_relational_const): Likewise.
5004 (native_interpret_vector): Likewise. Change the size from an
5005 int to an unsigned int.
5006 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
5007 TYPE_VECTOR_SUBPARTS.
5008 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
5009 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
5010 duplicating a non-constant operand into a variable-length vector.
5011 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
5012 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
5013 * ipa-icf.c (sem_variable::equals): Likewise.
5014 * match.pd: Likewise.
5015 * omp-simd-clone.c (simd_clone_subparts): Likewise.
5016 * print-tree.c (print_node): Likewise.
5017 * stor-layout.c (layout_type): Likewise.
5018 * targhooks.c (default_builtin_vectorization_cost): Likewise.
5019 * tree-cfg.c (verify_gimple_comparison): Likewise.
5020 (verify_gimple_assign_binary): Likewise.
5021 (verify_gimple_assign_ternary): Likewise.
5022 (verify_gimple_assign_single): Likewise.
5023 * tree-pretty-print.c (dump_generic_node): Likewise.
5024 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5025 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
5026 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
5027 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
5028 (vect_shift_permute_load_chain): Likewise.
5029 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
5030 (expand_vector_condition, optimize_vector_constructor): Likewise.
5031 (lower_vec_perm, get_compute_type): Likewise.
5032 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
5033 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
5034 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
5035 (vect_recog_mask_conversion_pattern): Likewise.
5036 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
5037 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
5038 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5039 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
5040 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
5041 (vectorizable_shift, vectorizable_operation, vectorizable_store)
5042 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
5043 (supportable_widening_operation): Likewise.
5044 (supportable_narrowing_operation): Likewise.
5045 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
5047 * varasm.c (output_constant): Likewise.
5049 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5050 Alan Hayward <alan.hayward@arm.com>
5051 David Sherwood <david.sherwood@arm.com>
5053 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
5054 so that both the length == 3 and length != 3 cases set up their
5055 own permute vectors. Add comments explaining why we know the
5056 number of elements is constant.
5057 (vect_permute_load_chain): Likewise.
5059 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5060 Alan Hayward <alan.hayward@arm.com>
5061 David Sherwood <david.sherwood@arm.com>
5063 * machmode.h (mode_nunits): Change from unsigned char to
5065 (ONLY_FIXED_SIZE_MODES): New macro.
5066 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
5067 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
5068 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
5070 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
5071 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
5072 or if measurement_type is not polynomial.
5073 * genmodes.c (ZERO_COEFFS): New macro.
5074 (emit_mode_nunits_inline): Make mode_nunits_inline return a
5076 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
5077 Use ZERO_COEFFS when emitting initializers.
5078 * data-streamer.h (bp_pack_poly_value): New function.
5079 (bp_unpack_poly_value): Likewise.
5080 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
5081 for GET_MODE_NUNITS.
5082 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
5083 for GET_MODE_NUNITS.
5084 * tree.c (make_vector_type): Remove temporary shim and make
5085 the real function take the number of units as a poly_uint64
5087 (build_vector_type_for_mode): Handle polynomial nunits.
5088 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
5089 * emit-rtl.c (const_vec_series_p_1): Likewise.
5090 (gen_rtx_CONST_VECTOR): Likewise.
5091 * fold-const.c (test_vec_duplicate_folding): Likewise.
5092 * genrecog.c (validate_pattern): Likewise.
5093 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
5094 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
5095 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
5096 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
5097 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
5098 * rtlanal.c (subreg_get_info): Likewise.
5099 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5100 (vect_grouped_load_supported): Likewise.
5101 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
5102 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
5103 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
5104 (simplify_const_unary_operation, simplify_binary_operation_1)
5105 (simplify_const_binary_operation, simplify_ternary_operation)
5106 (test_vector_ops_duplicate, test_vector_ops): Likewise.
5107 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
5108 instead of CONST_VECTOR_NUNITS.
5109 * varasm.c (output_constant_pool_2): Likewise.
5110 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
5111 explicit-encoded elements in the XVEC for variable-length vectors.
5113 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5115 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
5117 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5118 Alan Hayward <alan.hayward@arm.com>
5119 David Sherwood <david.sherwood@arm.com>
5121 * coretypes.h (fixed_size_mode): Declare.
5122 (fixed_size_mode_pod): New typedef.
5123 * builtins.h (target_builtins::x_apply_args_mode)
5124 (target_builtins::x_apply_result_mode): Change type to
5125 fixed_size_mode_pod.
5126 * builtins.c (apply_args_size, apply_result_size, result_vector)
5127 (expand_builtin_apply_args_1, expand_builtin_apply)
5128 (expand_builtin_return): Update accordingly.
5130 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5132 * cse.c (hash_rtx_cb): Hash only the encoded elements.
5133 * cselib.c (cselib_hash_rtx): Likewise.
5134 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
5135 CONST_VECTOR encoding.
5137 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5138 Jeff Law <law@redhat.com>
5141 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
5142 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
5143 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
5144 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
5147 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
5148 explicitly probe *sp in a noreturn function if there were any callee
5149 register saves or frame pointer is needed.
5151 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5154 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
5155 BLKmode for ternary, binary or unary expressions.
5158 * var-tracking.c (delete_vta_debug_insn): New inline function.
5159 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
5160 insns from get_insns () to NULL instead of each bb separately.
5161 Use delete_vta_debug_insn. No longer static.
5162 (vt_debug_insns_local, variable_tracking_main_1): Adjust
5163 delete_vta_debug_insns callers.
5164 * rtl.h (delete_vta_debug_insns): Declare.
5165 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
5166 instead of variable_tracking_main.
5168 2018-01-03 Martin Sebor <msebor@redhat.com>
5170 PR tree-optimization/83603
5171 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
5172 arguments past the endof the argument list in functions declared
5173 without a prototype.
5174 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
5175 Avoid checking when arguments are null.
5177 2018-01-03 Martin Sebor <msebor@redhat.com>
5180 * doc/extend.texi (attribute const): Fix a typo.
5181 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
5182 issuing -Wsuggest-attribute for void functions.
5184 2018-01-03 Martin Sebor <msebor@redhat.com>
5186 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
5187 offset_int::from instead of wide_int::to_shwi.
5188 (maybe_diag_overlap): Remove assertion.
5189 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
5190 * gimple-ssa-sprintf.c (format_directive): Same.
5191 (parse_directive): Same.
5192 (sprintf_dom_walker::compute_format_length): Same.
5193 (try_substitute_return_value): Same.
5195 2018-01-03 Jeff Law <law@redhat.com>
5198 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
5199 non-constant residual for zero at runtime and avoid probing in
5200 that case. Reorganize code for trailing problem to mirror handling
5203 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
5205 PR tree-optimization/83501
5206 * tree-ssa-strlen.c (get_string_cst): New.
5207 (handle_char_store): Call get_string_cst.
5209 2018-01-03 Martin Liska <mliska@suse.cz>
5211 PR tree-optimization/83593
5212 * tree-ssa-strlen.c: Include tree-cfg.h.
5213 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
5214 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
5215 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
5217 (strlen_dom_walker::before_dom_children): Call
5218 gimple_purge_dead_eh_edges. Dump tranformation with details
5220 (strlen_dom_walker::before_dom_children): Update call by adding
5221 new argument cleanup_eh.
5222 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
5224 2018-01-03 Martin Liska <mliska@suse.cz>
5227 * cif-code.def (VARIADIC_THUNK): New enum value.
5228 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
5231 2018-01-03 Jan Beulich <jbeulich@suse.com>
5233 * sse.md (mov<mode>_internal): Tighten condition for when to use
5234 vmovdqu<ssescalarsize> for TI and OI modes.
5236 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5238 Update copyright years.
5240 2018-01-03 Martin Liska <mliska@suse.cz>
5243 * ipa-visibility.c (function_and_variable_visibility): Skip
5244 functions with noipa attribure.
5246 2018-01-03 Jakub Jelinek <jakub@redhat.com>
5248 * gcc.c (process_command): Update copyright notice dates.
5249 * gcov-dump.c (print_version): Ditto.
5250 * gcov.c (print_version): Ditto.
5251 * gcov-tool.c (print_version): Ditto.
5252 * gengtype.c (create_file): Ditto.
5253 * doc/cpp.texi: Bump @copying's copyright year.
5254 * doc/cppinternals.texi: Ditto.
5255 * doc/gcc.texi: Ditto.
5256 * doc/gccint.texi: Ditto.
5257 * doc/gcov.texi: Ditto.
5258 * doc/install.texi: Ditto.
5259 * doc/invoke.texi: Ditto.
5261 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5263 * vector-builder.h (vector_builder::m_full_nelts): Change from
5264 unsigned int to poly_uint64.
5265 (vector_builder::full_nelts): Update prototype accordingly.
5266 (vector_builder::new_vector): Likewise.
5267 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
5268 (vector_builder::operator ==): Likewise.
5269 (vector_builder::finalize): Likewise.
5270 * int-vector-builder.h (int_vector_builder::int_vector_builder):
5271 Take the number of elements as a poly_uint64 rather than an
5273 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
5274 from unsigned int to poly_uint64.
5275 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
5276 (vec_perm_indices::new_vector): Likewise.
5277 (vec_perm_indices::length): Likewise.
5278 (vec_perm_indices::nelts_per_input): Likewise.
5279 (vec_perm_indices::input_nelts): Likewise.
5280 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
5281 number of elements per input as a poly_uint64 rather than an
5282 unsigned int. Use the original encoding for variable-length
5283 vectors, rather than clamping each individual element.
5284 For the second and subsequent elements in each pattern,
5285 clamp the step and base before clamping their sum.
5286 (vec_perm_indices::series_p): Handle polynomial element counts.
5287 (vec_perm_indices::all_in_range_p): Likewise.
5288 (vec_perm_indices_to_tree): Likewise.
5289 (vec_perm_indices_to_rtx): Likewise.
5290 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
5291 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
5292 (tree_vector_builder::new_binary_operation): Handle polynomial
5293 element counts. Return false if we need to know the number
5294 of elements at compile time.
5295 * fold-const.c (fold_vec_perm): Punt if the number of elements
5296 isn't known at compile time.
5298 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5300 * vec-perm-indices.h (vec_perm_builder): Change element type
5301 from HOST_WIDE_INT to poly_int64.
5302 (vec_perm_indices::element_type): Update accordingly.
5303 (vec_perm_indices::clamp): Handle polynomial element_types.
5304 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5305 (vec_perm_indices::all_in_range_p): Likewise.
5306 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
5308 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
5309 polynomial vec_perm_indices element types.
5310 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
5311 * fold-const.c (fold_vec_perm): Likewise.
5312 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
5313 * tree-vect-generic.c (lower_vec_perm): Likewise.
5314 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5315 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
5316 element type to HOST_WIDE_INT.
5318 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5319 Alan Hayward <alan.hayward@arm.com>
5320 David Sherwood <david.sherwood@arm.com>
5322 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
5323 rather than an int. Use plus_constant.
5324 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
5325 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
5327 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5328 Alan Hayward <alan.hayward@arm.com>
5329 David Sherwood <david.sherwood@arm.com>
5331 * calls.c (emit_call_1, expand_call): Change struct_value_size from
5332 a HOST_WIDE_INT to a poly_int64.
5334 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5335 Alan Hayward <alan.hayward@arm.com>
5336 David Sherwood <david.sherwood@arm.com>
5338 * calls.c (load_register_parameters): Cope with polynomial
5339 mode sizes. Require a constant size for BLKmode parameters
5340 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
5341 forces a parameter to be padded at the lsb end in order to
5342 fill a complete number of words, require the parameter size
5343 to be ordered wrt UNITS_PER_WORD.
5345 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5346 Alan Hayward <alan.hayward@arm.com>
5347 David Sherwood <david.sherwood@arm.com>
5349 * reload1.c (spill_stack_slot_width): Change element type
5350 from unsigned int to poly_uint64_pod.
5351 (alter_reg): Treat mode sizes as polynomial.
5353 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5354 Alan Hayward <alan.hayward@arm.com>
5355 David Sherwood <david.sherwood@arm.com>
5357 * reload.c (complex_word_subreg_p): New function.
5358 (reload_inner_reg_of_subreg, push_reload): Use it.
5360 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5361 Alan Hayward <alan.hayward@arm.com>
5362 David Sherwood <david.sherwood@arm.com>
5364 * lra-constraints.c (process_alt_operands): Reject matched
5365 operands whose sizes aren't ordered.
5366 (match_reload): Refer to this check here.
5368 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5369 Alan Hayward <alan.hayward@arm.com>
5370 David Sherwood <david.sherwood@arm.com>
5372 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
5373 that the mode size is in the set {1, 2, 4, 8, 16}.
5375 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5376 Alan Hayward <alan.hayward@arm.com>
5377 David Sherwood <david.sherwood@arm.com>
5379 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
5380 Use plus_constant instead of gen_rtx_PLUS.
5382 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5383 Alan Hayward <alan.hayward@arm.com>
5384 David Sherwood <david.sherwood@arm.com>
5386 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
5387 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
5388 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
5389 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
5390 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
5391 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
5392 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
5393 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
5394 * config/i386/i386.c (ix86_push_rounding): ...this new function.
5395 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
5397 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
5398 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
5399 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
5400 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
5401 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
5402 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
5403 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
5404 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
5405 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
5406 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
5408 * expr.c (emit_move_resolve_push): Treat the input and result
5409 of PUSH_ROUNDING as a poly_int64.
5410 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
5411 (emit_push_insn): Likewise.
5412 * lra-eliminations.c (mark_not_eliminable): Likewise.
5413 * recog.c (push_operand): Likewise.
5414 * reload1.c (elimination_effects): Likewise.
5415 * rtlanal.c (nonzero_bits1): Likewise.
5416 * calls.c (store_one_arg): Likewise. Require the padding to be
5417 known at compile time.
5419 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5420 Alan Hayward <alan.hayward@arm.com>
5421 David Sherwood <david.sherwood@arm.com>
5423 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
5424 Use plus_constant instead of gen_rtx_PLUS.
5426 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5427 Alan Hayward <alan.hayward@arm.com>
5428 David Sherwood <david.sherwood@arm.com>
5430 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
5433 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5434 Alan Hayward <alan.hayward@arm.com>
5435 David Sherwood <david.sherwood@arm.com>
5437 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
5438 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
5439 via stack temporaries. Treat the mode size as polynomial too.
5441 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5442 Alan Hayward <alan.hayward@arm.com>
5443 David Sherwood <david.sherwood@arm.com>
5445 * expr.c (expand_expr_real_2): When handling conversions involving
5446 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
5447 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
5448 as a poly_uint64 too.
5450 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5451 Alan Hayward <alan.hayward@arm.com>
5452 David Sherwood <david.sherwood@arm.com>
5454 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
5456 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5457 Alan Hayward <alan.hayward@arm.com>
5458 David Sherwood <david.sherwood@arm.com>
5460 * combine.c (can_change_dest_mode): Handle polynomial
5461 REGMODE_NATURAL_SIZE.
5462 * expmed.c (store_bit_field_1): Likewise.
5463 * expr.c (store_constructor): Likewise.
5464 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
5465 and polynomial REGMODE_NATURAL_SIZE.
5466 (gen_lowpart_common): Likewise.
5467 * reginfo.c (record_subregs_of_mode): Likewise.
5468 * rtlanal.c (read_modify_subreg_p): Likewise.
5470 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5471 Alan Hayward <alan.hayward@arm.com>
5472 David Sherwood <david.sherwood@arm.com>
5474 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
5475 numbers of elements.
5477 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5478 Alan Hayward <alan.hayward@arm.com>
5479 David Sherwood <david.sherwood@arm.com>
5481 * match.pd: Cope with polynomial numbers of vector elements.
5483 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5484 Alan Hayward <alan.hayward@arm.com>
5485 David Sherwood <david.sherwood@arm.com>
5487 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
5488 in a POINTER_PLUS_EXPR.
5490 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5491 Alan Hayward <alan.hayward@arm.com>
5492 David Sherwood <david.sherwood@arm.com>
5494 * omp-simd-clone.c (simd_clone_subparts): New function.
5495 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
5496 (ipa_simd_modify_function_body): Likewise.
5498 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5499 Alan Hayward <alan.hayward@arm.com>
5500 David Sherwood <david.sherwood@arm.com>
5502 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
5503 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
5504 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
5505 (expand_vector_condition, vector_element): Likewise.
5506 (subparts_gt): New function.
5507 (get_compute_type): Use subparts_gt.
5508 (count_type_subparts): Delete.
5509 (expand_vector_operations_1): Use subparts_gt instead of
5510 count_type_subparts.
5512 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5513 Alan Hayward <alan.hayward@arm.com>
5514 David Sherwood <david.sherwood@arm.com>
5516 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
5517 (vect_compile_time_alias): ...this new function. Do the calculation
5518 on poly_ints rather than trees.
5519 (vect_prune_runtime_alias_test_list): Update call accordingly.
5521 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5522 Alan Hayward <alan.hayward@arm.com>
5523 David Sherwood <david.sherwood@arm.com>
5525 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
5527 (vect_schedule_slp_instance): Likewise.
5529 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5530 Alan Hayward <alan.hayward@arm.com>
5531 David Sherwood <david.sherwood@arm.com>
5533 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
5534 constant and extern definitions for variable-length vectors.
5535 (vect_get_constant_vectors): Note that the number of units
5536 is known to be constant.
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-vect-stmts.c (vectorizable_conversion): Treat the number
5543 of units as polynomial. Choose between WIDE and NARROW based
5546 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5547 Alan Hayward <alan.hayward@arm.com>
5548 David Sherwood <david.sherwood@arm.com>
5550 * tree-vect-stmts.c (simd_clone_subparts): New function.
5551 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
5553 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5554 Alan Hayward <alan.hayward@arm.com>
5555 David Sherwood <david.sherwood@arm.com>
5557 * tree-vect-stmts.c (vectorizable_call): Treat the number of
5558 vectors as polynomial. Use build_index_vector for
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 * tree-vect-stmts.c (get_load_store_type): Treat the number of
5566 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
5567 for variable-length vectors.
5568 (vectorizable_mask_load_store): Treat the number of units as
5569 polynomial, asserting that it is constant if the condition has
5570 already been enforced.
5571 (vectorizable_store, vectorizable_load): Likewise.
5573 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5574 Alan Hayward <alan.hayward@arm.com>
5575 David Sherwood <david.sherwood@arm.com>
5577 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
5578 of units as polynomial. Punt if we can't tell at compile time
5579 which vector contains the final result.
5581 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5582 Alan Hayward <alan.hayward@arm.com>
5583 David Sherwood <david.sherwood@arm.com>
5585 * tree-vect-loop.c (vectorizable_induction): Treat the number
5586 of units as polynomial. Punt on SLP inductions. Use an integer
5587 VEC_SERIES_EXPR for variable-length integer reductions. Use a
5588 cast of such a series for variable-length floating-point
5591 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5592 Alan Hayward <alan.hayward@arm.com>
5593 David Sherwood <david.sherwood@arm.com>
5595 * tree.h (build_index_vector): Declare.
5596 * tree.c (build_index_vector): New function.
5597 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
5598 of units as polynomial, forcibly converting it to a constant if
5599 vectorizable_reduction has already enforced the condition.
5600 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
5601 to create a {1,2,3,...} vector.
5602 (vectorizable_reduction): Treat the number of units as polynomial.
5603 Choose vectype_in based on the largest scalar element size rather
5604 than the smallest number of units. Enforce the restrictions
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 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
5612 number of units as polynomial.
5614 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5615 Alan Hayward <alan.hayward@arm.com>
5616 David Sherwood <david.sherwood@arm.com>
5618 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
5619 * target.def (autovectorize_vector_sizes): Return the vector sizes
5620 by pointer, using vector_sizes rather than a bitmask.
5621 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
5622 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
5623 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
5625 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
5626 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
5627 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
5628 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
5629 * omp-general.c (omp_max_vf): Likewise.
5630 * omp-low.c (omp_clause_aligned_alignment): Likewise.
5631 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
5632 * tree-vect-loop.c (vect_analyze_loop): Likewise.
5633 * tree-vect-slp.c (vect_slp_bb): Likewise.
5634 * doc/tm.texi: Regenerate.
5635 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
5637 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
5638 the vector size as a poly_uint64 rather than an unsigned int.
5639 (current_vector_size): Change from an unsigned int to a poly_uint64.
5640 (get_vectype_for_scalar_type): Update accordingly.
5641 * tree.h (build_truth_vector_type): Take the size and number of
5642 units as a poly_uint64 rather than an unsigned int.
5643 (build_vector_type): Add a temporary overload that takes
5644 the number of units as a poly_uint64 rather than an unsigned int.
5645 * tree.c (make_vector_type): Likewise.
5646 (build_truth_vector_type): Take the number of units as a poly_uint64
5647 rather than an unsigned int.
5649 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5650 Alan Hayward <alan.hayward@arm.com>
5651 David Sherwood <david.sherwood@arm.com>
5653 * target.def (get_mask_mode): Take the number of units and length
5654 as poly_uint64s rather than unsigned ints.
5655 * targhooks.h (default_get_mask_mode): Update accordingly.
5656 * targhooks.c (default_get_mask_mode): Likewise.
5657 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
5658 * doc/tm.texi: Regenerate.
5660 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5661 Alan Hayward <alan.hayward@arm.com>
5662 David Sherwood <david.sherwood@arm.com>
5664 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
5665 * omp-general.c (omp_max_vf): Likewise.
5666 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
5667 (expand_omp_simd): Handle polynomial safelen.
5668 * omp-low.c (omplow_simd_context): Add a default constructor.
5669 (omplow_simd_context::max_vf): Change from int to poly_uint64.
5670 (lower_rec_simd_input_clauses): Update accordingly.
5671 (lower_rec_input_clauses): Likewise.
5673 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5674 Alan Hayward <alan.hayward@arm.com>
5675 David Sherwood <david.sherwood@arm.com>
5677 * tree-vectorizer.h (vect_nunits_for_cost): New function.
5678 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
5679 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
5680 (vect_analyze_slp_cost): Likewise.
5681 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
5682 (vect_model_load_cost): Likewise.
5684 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5685 Alan Hayward <alan.hayward@arm.com>
5686 David Sherwood <david.sherwood@arm.com>
5688 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5689 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5690 from an unsigned int * to a poly_uint64_pod *.
5691 (calculate_unrolling_factor): New function.
5692 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5694 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5695 Alan Hayward <alan.hayward@arm.com>
5696 David Sherwood <david.sherwood@arm.com>
5698 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5699 from an unsigned int to a poly_uint64.
5700 (_loop_vec_info::slp_unrolling_factor): Likewise.
5701 (_loop_vec_info::vectorization_factor): Change from an int
5703 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5704 (vect_get_num_vectors): New function.
5705 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5706 (vect_get_num_copies): Use vect_get_num_vectors.
5707 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5708 to an unsigned int *.
5709 (vect_analyze_data_refs): Change min_vf from an int * to a
5711 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5712 than an unsigned HOST_WIDE_INT.
5713 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5714 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5715 to an unsigned int *.
5716 (vect_analyze_data_ref_dependences): Likewise.
5717 (vect_compute_data_ref_alignment): Handle polynomial vf.
5718 (vect_enhance_data_refs_alignment): Likewise.
5719 (vect_prune_runtime_alias_test_list): Likewise.
5720 (vect_shift_permute_load_chain): Likewise.
5721 (vect_supportable_dr_alignment): Likewise.
5722 (dependence_distance_ge_vf): Take the vectorization factor as a
5723 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5724 (vect_analyze_data_refs): Change min_vf from an int * to a
5726 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5727 vfm1 as a poly_uint64 rather than an int. Make the same change
5728 for the returned bound_scalar.
5729 (vect_gen_vector_loop_niters): Handle polynomial vf.
5730 (vect_do_peeling): Likewise. Update call to
5731 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5732 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5734 * tree-vect-loop.c (vect_determine_vectorization_factor)
5735 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5736 (vect_get_known_peeling_cost): Likewise.
5737 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5738 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5739 (vect_transform_loop): Likewise. Use the lowest possible VF when
5740 updating the upper bounds of the loop.
5741 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5743 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5744 polynomial unroll factors.
5745 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5746 (vect_make_slp_decision): Likewise.
5747 (vect_supported_load_permutation_p): Likewise, and polynomial
5749 (vect_analyze_slp_cost): Handle polynomial vf.
5750 (vect_slp_analyze_node_operations): Likewise.
5751 (vect_slp_analyze_bb_1): Likewise.
5752 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5753 than an unsigned HOST_WIDE_INT.
5754 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5755 (vectorizable_load): Handle polynomial vf.
5756 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5758 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5760 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5761 Alan Hayward <alan.hayward@arm.com>
5762 David Sherwood <david.sherwood@arm.com>
5764 * match.pd: Handle bit operations involving three constants
5765 and try to fold one pair.
5767 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5769 * tree-vect-loop-manip.c: Include gimple-fold.h.
5770 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5771 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5772 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5773 Add a path that uses a step of VF instead of 1, but disable it
5775 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5776 and niters_no_overflow parameters. Update calls to
5777 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5778 Create a new SSA name if the latter choses to use a ste other
5779 than zero, and return it via niters_vector_mult_vf_var.
5780 * tree-vect-loop.c (vect_transform_loop): Update calls to
5781 vect_do_peeling, vect_gen_vector_loop_niters and
5782 slpeel_make_loop_iterate_ntimes.
5783 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5784 (vect_gen_vector_loop_niters): Update declarations after above changes.
5786 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5788 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5789 128-bit round to integer instructions.
5790 (ceil<mode>2): Likewise.
5791 (btrunc<mode>2): Likewise.
5792 (round<mode>2): Likewise.
5794 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5796 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5797 unaligned VSX load/store on P8/P9.
5798 (expand_block_clear): Allow the use of unaligned VSX
5799 load/store on P8/P9.
5801 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5803 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5805 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5806 swap associated with both a load and a store.
5808 2018-01-02 Andrew Waterman <andrew@sifive.com>
5810 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5811 * config/riscv/riscv.md (clear_cache): Use it.
5813 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5815 * web.c: Remove out-of-date comment.
5817 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5819 * expr.c (fixup_args_size_notes): Check that any existing
5820 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5821 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5822 (emit_single_push_insn): ...here.
5824 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5826 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5827 (const_vector_encoded_nelts): New function.
5828 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5829 (const_vector_int_elt, const_vector_elt): Declare.
5830 * emit-rtl.c (const_vector_int_elt_1): New function.
5831 (const_vector_elt): Likewise.
5832 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5833 of CONST_VECTOR_ELT.
5835 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5837 * expr.c: Include rtx-vector-builder.h.
5838 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5839 directly on the tree encoding.
5840 (const_vector_from_tree): Likewise.
5841 * optabs.c: Include rtx-vector-builder.h.
5842 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5843 sequence of "u" values.
5844 * vec-perm-indices.c: Include rtx-vector-builder.h.
5845 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5846 directly on the vec_perm_indices encoding.
5848 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5850 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5851 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5852 * rtx-vector-builder.h: New file.
5853 * rtx-vector-builder.c: Likewise.
5854 * rtl.h (rtx_def::u2): Add a const_vector field.
5855 (CONST_VECTOR_NPATTERNS): New macro.
5856 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5857 (CONST_VECTOR_DUPLICATE_P): Likewise.
5858 (CONST_VECTOR_STEPPED_P): Likewise.
5859 (CONST_VECTOR_ENCODED_ELT): Likewise.
5860 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5861 (unwrap_const_vec_duplicate): Likewise.
5862 (const_vec_series_p): Check for a non-duplicated vector encoding.
5863 Say that the function only returns true for integer vectors.
5864 * emit-rtl.c: Include rtx-vector-builder.h.
5865 (gen_const_vec_duplicate_1): Delete.
5866 (gen_const_vector): Call gen_const_vec_duplicate instead of
5867 gen_const_vec_duplicate_1.
5868 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5869 (gen_const_vec_duplicate): Use rtx_vector_builder.
5870 (gen_const_vec_series): Likewise.
5871 (gen_rtx_CONST_VECTOR): Likewise.
5872 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5873 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5874 Build a new vector rather than modifying a CONST_VECTOR in-place.
5875 (handle_special_swappables): Update call accordingly.
5876 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5877 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5878 Build a new vector rather than modifying a CONST_VECTOR in-place.
5879 (handle_special_swappables): Update call accordingly.
5881 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5883 * simplify-rtx.c (simplify_const_binary_operation): Use
5884 CONST_VECTOR_ELT instead of XVECEXP.
5886 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5888 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5889 the selector elements to be different from the data elements
5890 if the selector is a VECTOR_CST.
5891 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5892 ssizetype for the selector.
5894 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5896 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5897 before testing each element individually.
5898 * tree-vect-generic.c (lower_vec_perm): Likewise.
5900 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5902 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5903 * selftest-run-tests.c (selftest::run_tests): Call it.
5904 * vector-builder.h (vector_builder::operator ==): New function.
5905 (vector_builder::operator !=): Likewise.
5906 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5907 (vec_perm_indices::all_from_input_p): New function.
5908 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5909 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5910 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5911 instead of reading the VECTOR_CST directly. Detect whether both
5912 vector inputs are the same before constructing the vec_perm_indices,
5913 and update the number of inputs argument accordingly. Use the
5914 utility functions added above. Only construct sel2 if we need to.
5916 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5918 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5919 the broadcast of the low byte.
5920 (expand_mult_highpart): Use an explicit encoding for the permutes.
5921 * optabs-query.c (can_mult_highpart_p): Likewise.
5922 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5923 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5924 (vectorizable_bswap): Likewise.
5925 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5926 explicit encoding for the power-of-2 permutes.
5927 (vect_permute_store_chain): Likewise.
5928 (vect_grouped_load_supported): Likewise.
5929 (vect_permute_load_chain): Likewise.
5931 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5933 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5934 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5935 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5936 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5937 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5938 (vect_gen_perm_mask_any): Likewise.
5940 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5942 * int-vector-builder.h: New file.
5943 * vec-perm-indices.h: Include int-vector-builder.h.
5944 (vec_perm_indices): Redefine as an int_vector_builder.
5945 (auto_vec_perm_indices): Delete.
5946 (vec_perm_builder): Redefine as a stand-alone class.
5947 (vec_perm_indices::vec_perm_indices): New function.
5948 (vec_perm_indices::clamp): Likewise.
5949 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5950 (vec_perm_indices::new_vector): New function.
5951 (vec_perm_indices::new_expanded_vector): Update for new
5952 vec_perm_indices class.
5953 (vec_perm_indices::rotate_inputs): New function.
5954 (vec_perm_indices::all_in_range_p): Operate directly on the
5955 encoded form, without computing elided elements.
5956 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5957 encoding. Update for new vec_perm_indices class.
5958 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5959 the given vec_perm_builder.
5960 (expand_vec_perm_var): Update vec_perm_builder constructor.
5961 (expand_mult_highpart): Use vec_perm_builder instead of
5962 auto_vec_perm_indices.
5963 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5964 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5965 or double series encoding as appropriate.
5966 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5967 vec_perm_indices instead of auto_vec_perm_indices.
5968 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5969 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5970 (vect_permute_store_chain): Likewise.
5971 (vect_grouped_load_supported): Likewise.
5972 (vect_permute_load_chain): Likewise.
5973 (vect_shift_permute_load_chain): Likewise.
5974 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5975 (vect_transform_slp_perm_load): Likewise.
5976 (vect_schedule_slp_instance): Likewise.
5977 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5978 (vectorizable_mask_load_store): Likewise.
5979 (vectorizable_bswap): Likewise.
5980 (vectorizable_store): Likewise.
5981 (vectorizable_load): Likewise.
5982 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5983 vec_perm_indices instead of auto_vec_perm_indices. Use
5984 tree_to_vec_perm_builder to read the vector from a tree.
5985 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5986 vec_perm_builder instead of a vec_perm_indices.
5987 (have_whole_vector_shift): Use vec_perm_builder and
5988 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5989 truncation to calc_vec_perm_mask_for_shift.
5990 (vect_create_epilog_for_reduction): Likewise.
5991 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5992 from auto_vec_perm_indices to vec_perm_indices.
5993 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5994 instead of changing individual elements.
5995 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5996 the vector in d.perm.
5997 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5998 from auto_vec_perm_indices to vec_perm_indices.
5999 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
6000 instead of changing individual elements.
6001 (arm_vectorize_vec_perm_const): Use new_vector to install
6002 the vector in d.perm.
6003 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
6004 Update vec_perm_builder constructor.
6005 (rs6000_expand_interleave): Likewise.
6006 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
6007 (rs6000_expand_interleave): Likewise.
6009 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6011 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
6012 to qimode could truncate the indices.
6013 * optabs.c (expand_vec_perm_var): Likewise.
6015 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6017 * Makefile.in (OBJS): Add vec-perm-indices.o.
6018 * vec-perm-indices.h: New file.
6019 * vec-perm-indices.c: Likewise.
6020 * target.h (vec_perm_indices): Replace with a forward class
6022 (auto_vec_perm_indices): Move to vec-perm-indices.h.
6023 * optabs.h: Include vec-perm-indices.h.
6024 (expand_vec_perm): Delete.
6025 (selector_fits_mode_p, expand_vec_perm_var): Declare.
6026 (expand_vec_perm_const): Declare.
6027 * target.def (vec_perm_const_ok): Replace with...
6028 (vec_perm_const): ...this new hook.
6029 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
6030 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
6031 * doc/tm.texi: Regenerate.
6032 * optabs.def (vec_perm_const): Delete.
6033 * doc/md.texi (vec_perm_const): Likewise.
6034 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
6035 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
6036 expand_vec_perm for constant permutation vectors. Assert that
6037 the mode of variable permutation vectors is the integer equivalent
6038 of the mode that is being permuted.
6039 * optabs-query.h (selector_fits_mode_p): Declare.
6040 * optabs-query.c: Include vec-perm-indices.h.
6041 (selector_fits_mode_p): New function.
6042 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
6043 is defined, instead of checking whether the vec_perm_const_optab
6044 exists. Use targetm.vectorize.vec_perm_const instead of
6045 targetm.vectorize.vec_perm_const_ok. Check whether the indices
6046 fit in the vector mode before using a variable permute.
6047 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
6048 vec_perm_indices instead of an rtx.
6049 (expand_vec_perm): Replace with...
6050 (expand_vec_perm_const): ...this new function. Take the selector
6051 as a vec_perm_indices rather than an rtx. Also take the mode of
6052 the selector. Update call to shift_amt_for_vec_perm_mask.
6053 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
6054 Use vec_perm_indices::new_expanded_vector to expand the original
6055 selector into bytes. Check whether the indices fit in the vector
6056 mode before using a variable permute.
6057 (expand_vec_perm_var): Make global.
6058 (expand_mult_highpart): Use expand_vec_perm_const.
6059 * fold-const.c: Includes vec-perm-indices.h.
6060 * tree-ssa-forwprop.c: Likewise.
6061 * tree-vect-data-refs.c: Likewise.
6062 * tree-vect-generic.c: Likewise.
6063 * tree-vect-loop.c: Likewise.
6064 * tree-vect-slp.c: Likewise.
6065 * tree-vect-stmts.c: Likewise.
6066 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
6068 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
6069 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
6070 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
6071 (aarch64_vectorize_vec_perm_const): ...this new function.
6072 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6073 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6074 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
6075 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
6076 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6077 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6078 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
6080 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
6081 check for NEON modes.
6082 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
6083 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
6084 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
6085 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
6087 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
6088 the old VEC_PERM_CONST conditions.
6089 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
6090 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
6091 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
6092 (ia64_vectorize_vec_perm_const_ok): Merge into...
6093 (ia64_vectorize_vec_perm_const): ...this new function.
6094 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
6095 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
6096 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
6097 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
6098 * config/mips/mips.c (mips_expand_vec_perm_const)
6099 (mips_vectorize_vec_perm_const_ok): Merge into...
6100 (mips_vectorize_vec_perm_const): ...this new function.
6101 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
6102 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
6103 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
6104 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
6105 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
6106 (rs6000_expand_vec_perm_const): Delete.
6107 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
6109 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6110 (altivec_expand_vec_perm_const_le): Take each operand individually.
6111 Operate on constant selectors rather than rtxes.
6112 (altivec_expand_vec_perm_const): Likewise. Update call to
6113 altivec_expand_vec_perm_const_le.
6114 (rs6000_expand_vec_perm_const): Delete.
6115 (rs6000_vectorize_vec_perm_const_ok): Delete.
6116 (rs6000_vectorize_vec_perm_const): New function.
6117 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6118 an element count and rtx array.
6119 (rs6000_expand_extract_even): Update call accordingly.
6120 (rs6000_expand_interleave): Likewise.
6121 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
6122 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
6123 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
6124 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
6125 (rs6000_expand_vec_perm_const): Delete.
6126 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
6127 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6128 (altivec_expand_vec_perm_const_le): Take each operand individually.
6129 Operate on constant selectors rather than rtxes.
6130 (altivec_expand_vec_perm_const): Likewise. Update call to
6131 altivec_expand_vec_perm_const_le.
6132 (rs6000_expand_vec_perm_const): Delete.
6133 (rs6000_vectorize_vec_perm_const_ok): Delete.
6134 (rs6000_vectorize_vec_perm_const): New function. Remove stray
6135 reference to the SPE evmerge intructions.
6136 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
6137 an element count and rtx array.
6138 (rs6000_expand_extract_even): Update call accordingly.
6139 (rs6000_expand_interleave): Likewise.
6140 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
6141 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
6143 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
6145 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6147 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
6148 vector mode and that that mode matches the mode of the data
6150 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
6151 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
6152 directly using expand_vec_perm_1 when forcing selectors into
6154 (expand_vec_perm_var): New function, split out from expand_vec_perm.
6156 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6158 * optabs-query.h (can_vec_perm_p): Delete.
6159 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
6160 * optabs-query.c (can_vec_perm_p): Split into...
6161 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
6162 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
6163 particular selector is valid.
6164 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
6165 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
6166 (vect_grouped_load_supported): Likewise.
6167 (vect_shift_permute_load_chain): Likewise.
6168 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
6169 (vect_transform_slp_perm_load): Likewise.
6170 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
6171 (vectorizable_bswap): Likewise.
6172 (vect_gen_perm_mask_checked): Likewise.
6173 * fold-const.c (fold_ternary_loc): Likewise. Don't take
6174 implementations of variable permutation vectors into account
6175 when deciding which selector to use.
6176 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
6177 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
6178 with a false third argument.
6179 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
6180 to test whether the constant selector is valid and can_vec_perm_var_p
6181 to test whether a variable selector is valid.
6183 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6185 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
6186 * optabs-query.c (can_vec_perm_p): Likewise.
6187 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
6188 instead of vec_perm_indices.
6189 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
6190 (vect_gen_perm_mask_checked): Likewise,
6191 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
6192 (vect_gen_perm_mask_checked): Likewise,
6194 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
6196 * optabs-query.h (qimode_for_vec_perm): Declare.
6197 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
6198 (qimode_for_vec_perm): ...this new function.
6199 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
6201 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
6203 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
6204 does not have a conditional at the top.
6206 2018-01-02 Richard Biener <rguenther@suse.de>
6208 * ipa-inline.c (big_speedup_p): Fix expression.
6210 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6213 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
6216 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
6220 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
6221 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
6222 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
6223 cond_taken_branch_cost 3->4.
6225 2018-01-01 Jakub Jelinek <jakub@redhat.com>
6227 PR tree-optimization/83581
6228 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
6229 TODO_cleanup_cfg if any changes have been made.
6232 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
6233 convert_modes if target mode has the right side, but different mode
6237 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
6238 last argument when extracting from CONCAT. If either from_real or
6239 from_imag is NULL, use expansion through memory. If result is not
6240 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
6241 the parts directly to inner mode, if even that fails, use expansion
6245 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
6246 check for bswap in mode rather than HImode and use that in expand_unop
6249 Copyright (C) 2018 Free Software Foundation, Inc.
6251 Copying and distribution of this file, with or without modification,
6252 are permitted in any medium without royalty provided the copyright
6253 notice and this notice are preserved.